diff options
Diffstat (limited to 'pkg/sentry/fs')
115 files changed, 8193 insertions, 12929 deletions
diff --git a/pkg/sentry/fs/BUILD b/pkg/sentry/fs/BUILD deleted file mode 100644 index d7259b47b..000000000 --- a/pkg/sentry/fs/BUILD +++ /dev/null @@ -1,135 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library", "go_test") - -go_library( - name = "fs", - srcs = [ - "attr.go", - "context.go", - "copy_up.go", - "dentry.go", - "dirent.go", - "dirent_cache.go", - "dirent_cache_limiter.go", - "dirent_list.go", - "dirent_state.go", - "event_list.go", - "file.go", - "file_operations.go", - "file_overlay.go", - "file_state.go", - "filesystems.go", - "flags.go", - "fs.go", - "inode.go", - "inode_inotify.go", - "inode_operations.go", - "inode_overlay.go", - "inotify.go", - "inotify_event.go", - "inotify_watch.go", - "mock.go", - "mount.go", - "mount_overlay.go", - "mounts.go", - "offset.go", - "overlay.go", - "path.go", - "restore.go", - "save.go", - "seek.go", - "splice.go", - "sync.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/amutex", - "//pkg/log", - "//pkg/metric", - "//pkg/p9", - "//pkg/refs", - "//pkg/secio", - "//pkg/sentry/arch", - "//pkg/sentry/context", - "//pkg/sentry/device", - "//pkg/sentry/fs/lock", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sentry/limits", - "//pkg/sentry/memmap", - "//pkg/sentry/platform", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/uniqueid", - "//pkg/sentry/usage", - "//pkg/sentry/usermem", - "//pkg/state", - "//pkg/syserror", - "//pkg/waiter", - "//third_party/gvsync", - ], -) - -go_template_instance( - name = "dirent_list", - out = "dirent_list.go", - package = "fs", - prefix = "dirent", - template = "//pkg/ilist:generic_list", - types = { - "Linker": "*Dirent", - "Element": "*Dirent", - }, -) - -go_template_instance( - name = "event_list", - out = "event_list.go", - package = "fs", - prefix = "event", - template = "//pkg/ilist:generic_list", - types = { - "Linker": "*Event", - "Element": "*Event", - }, -) - -go_test( - name = "fs_x_test", - size = "small", - srcs = [ - "copy_up_test.go", - "file_overlay_test.go", - "inode_overlay_test.go", - "mounts_test.go", - ], - deps = [ - ":fs", - "//pkg/sentry/context", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fs/ramfs", - "//pkg/sentry/fs/tmpfs", - "//pkg/sentry/kernel/contexttest", - "//pkg/sentry/usermem", - "//pkg/syserror", - ], -) - -go_test( - name = "fs_test", - size = "small", - srcs = [ - "dirent_cache_test.go", - "dirent_refs_test.go", - "mount_test.go", - "path_test.go", - ], - embed = [":fs"], - deps = [ - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", - ], -) diff --git a/pkg/sentry/fs/README.md b/pkg/sentry/fs/README.md deleted file mode 100644 index db4a1b730..000000000 --- a/pkg/sentry/fs/README.md +++ /dev/null @@ -1,229 +0,0 @@ -This package provides an implementation of the Linux virtual filesystem. - -[TOC] - -## Overview - -- An `fs.Dirent` caches an `fs.Inode` in memory at a path in the VFS, giving - the `fs.Inode` a relative position with respect to other `fs.Inode`s. - -- If an `fs.Dirent` is referenced by two file descriptors, then those file - descriptors are coherent with each other: they depend on the same - `fs.Inode`. - -- A mount point is an `fs.Dirent` for which `fs.Dirent.mounted` is true. It - exposes the root of a mounted filesystem. - -- The `fs.Inode` produced by a registered filesystem on mount(2) owns an - `fs.MountedFilesystem` from which other `fs.Inode`s will be looked up. For a - remote filesystem, the `fs.MountedFilesystem` owns the connection to that - remote filesystem. - -- In general: - -``` -fs.Inode <------------------------------ -| | -| | -produced by | -exactly one | -| responsible for the -| virtual identity of -v | -fs.MountedFilesystem ------------------- -``` - -Glossary: - -- VFS: virtual filesystem. - -- inode: a virtual file object holding a cached view of a file on a backing - filesystem (includes metadata and page caches). - -- superblock: the virtual state of a mounted filesystem (e.g. the virtual - inode number set). - -- mount namespace: a view of the mounts under a root (during path traversal, - the VFS makes visible/follows the mount point that is in the current task's - mount namespace). - -## Save and restore - -An application's hard dependencies on filesystem state can be broken down into -two categories: - -- The state necessary to execute a traversal on or view the *virtual* - filesystem hierarchy, regardless of what files an application has open. - -- The state necessary to represent open files. - -The first is always necessary to save and restore. An application may never have -any open file descriptors, but across save and restore it should see a coherent -view of any mount namespace. NOTE(b/63601033): Currently only one "initial" -mount namespace is supported. - -The second is so that system calls across save and restore are coherent with -each other (e.g. so that unintended re-reads or overwrites do not occur). - -Specifically this state is: - -- An `fs.MountManager` containing mount points. - -- A `kernel.FDTable` containing pointers to open files. - -Anything else managed by the VFS that can be easily loaded into memory from a -filesystem is synced back to those filesystems and is not saved. Examples are -pages in page caches used for optimizations (i.e. readahead and writeback), and -directory entries used to accelerate path lookups. - -### Mount points - -Saving and restoring a mount point means saving and restoring: - -- The root of the mounted filesystem. - -- Mount flags, which control how the VFS interacts with the mounted - filesystem. - -- Any relevant metadata about the mounted filesystem. - -- All `fs.Inode`s referenced by the application that reside under the mount - point. - -`fs.MountedFilesystem` is metadata about a filesystem that is mounted. It is -referenced by every `fs.Inode` loaded into memory under the mount point -including the `fs.Inode` of the mount point itself. The `fs.MountedFilesystem` -maps file objects on the filesystem to a virtualized `fs.Inode` number and vice -versa. - -To restore all `fs.Inode`s under a given mount point, each `fs.Inode` leverages -its dependency on an `fs.MountedFilesystem`. Since the `fs.MountedFilesystem` -knows how an `fs.Inode` maps to a file object on a backing filesystem, this -mapping can be trivially consulted by each `fs.Inode` when the `fs.Inode` is -restored. - -In detail, a mount point is saved in two steps: - -- First, after the kernel is paused but before state.Save, we walk all mount - namespaces and install a mapping from `fs.Inode` numbers to file paths - relative to the root of the mounted filesystem in each - `fs.MountedFilesystem`. This is subsequently called the set of `fs.Inode` - mappings. - -- Second, during state.Save, each `fs.MountedFilesystem` decides whether to - save the set of `fs.Inode` mappings. In-memory filesystems, like tmpfs, have - no need to save a set of `fs.Inode` mappings, since the `fs.Inode`s can be - entirely encoded in state file. Each `fs.MountedFilesystem` also optionally - saves the device name from when the filesystem was originally mounted. Each - `fs.Inode` saves its virtual identifier and a reference to a - `fs.MountedFilesystem`. - -A mount point is restored in two steps: - -- First, before state.Load, all mount configurations are stored in a global - `fs.RestoreEnvironment`. This tells us what mount points the user wants to - restore and how to re-establish pointers to backing filesystems. - -- Second, during state.Load, each `fs.MountedFilesystem` optionally searches - for a mount in the `fs.RestoreEnvironment` that matches its saved device - name. The `fs.MountedFilesystem` then reestablishes a pointer to the root of - the mounted filesystem. For example, the mount specification provides the - network connection for a mounted remote filesystem client to communicate - with its remote file server. The `fs.MountedFilesystem` also trivially loads - its set of `fs.Inode` mappings. When an `fs.Inode` is encountered, the - `fs.Inode` loads its virtual identifier and its reference a - `fs.MountedFilesystem`. It uses the `fs.MountedFilesystem` to obtain the - root of the mounted filesystem and the `fs.Inode` mappings to obtain the - relative file path to its data. With these, the `fs.Inode` re-establishes a - pointer to its file object. - -A mount point can trivially restore its `fs.Inode`s in parallel since -`fs.Inode`s have a restore dependency on their `fs.MountedFilesystem` and not on -each other. - -### Open files - -An `fs.File` references the following filesystem objects: - -```go -fs.File -> fs.Dirent -> fs.Inode -> fs.MountedFilesystem -``` - -The `fs.Inode` is restored using its `fs.MountedFilesystem`. The -[Mount points](#mount-points) section above describes how this happens in -detail. The `fs.Dirent` restores its pointer to an `fs.Inode`, pointers to -parent and children `fs.Dirents`, and the basename of the file. - -Otherwise an `fs.File` restores flags, an offset, and a unique identifier (only -used internally). - -It may use the `fs.Inode`, which it indirectly holds a reference on through the -`fs.Dirent`, to reestablish an open file handle on the backing filesystem (e.g. -to continue reading and writing). - -## Overlay - -The overlay implementation in the fs package takes Linux overlayfs as a frame of -reference but corrects for several POSIX consistency errors. - -In Linux overlayfs, the `struct inode` used for reading and writing to the same -file may be different. This is because the `struct inode` is dissociated with -the process of copying up the file from the upper to the lower directory. Since -flock(2) and fcntl(2) locks, inotify(7) watches, page caches, and a file's -identity are all stored directly or indirectly off the `struct inode`, these -properties of the `struct inode` may be stale after the first modification. This -can lead to file locking bugs, missed inotify events, and inconsistent data in -shared memory mappings of files, to name a few problems. - -The fs package maintains a single `fs.Inode` to represent a directory entry in -an overlay and defines operations on this `fs.Inode` which synchronize with the -copy up process. This achieves several things: - -+ File locks, inotify watches, and the identity of the file need not be copied - at all. - -+ Memory mappings of files coordinate with the copy up process so that if a - file in the lower directory is memory mapped, all references to it are - invalidated, forcing the application to re-fault on memory mappings of the - file under the upper directory. - -The `fs.Inode` holds metadata about files in the upper and/or lower directories -via an `fs.overlayEntry`. The `fs.overlayEntry` implements the `fs.Mappable` -interface. It multiplexes between upper and lower directory memory mappings and -stores a copy of memory references so they can be transferred to the upper -directory `fs.Mappable` when the file is copied up. - -The lower filesystem in an overlay may contain another (nested) overlay, but the -upper filesystem may not contain another overlay. In other words, nested -overlays form a tree structure that only allows branching in the lower -filesystem. - -Caching decisions in the overlay are delegated to the upper filesystem, meaning -that the Keep and Revalidate methods on the overlay return the same values as -the upper filesystem. A small wrinkle is that the lower filesystem is not -allowed to return `true` from Revalidate, as the overlay can not reload inodes -from the lower filesystem. A lower filesystem that does return `true` from -Revalidate will trigger a panic. - -The `fs.Inode` also holds a reference to a `fs.MountedFilesystem` that -normalizes across the mounted filesystem state of the upper and lower -directories. - -When a file is copied from the lower to the upper directory, attempts to -interact with the file block until the copy completes. All copying synchronizes -with rename(2). - -## Future Work - -### Overlay - -When a file is copied from a lower directory to an upper directory, several -locks are taken: the global renamuMu and the copyMu of the `fs.Inode` being -copied. This blocks operations on the file, including fault handling of memory -mappings. Performance could be improved by copying files into a temporary -directory that resides on the same filesystem as the upper directory and doing -an atomic rename, holding locks only during the rename operation. - -Additionally files are copied up synchronously. For large files, this causes a -noticeable latency. Performance could be improved by pipelining copies at -non-overlapping file offsets. diff --git a/pkg/sentry/fs/anon/BUILD b/pkg/sentry/fs/anon/BUILD deleted file mode 100644 index ae1c9cf76..000000000 --- a/pkg/sentry/fs/anon/BUILD +++ /dev/null @@ -1,21 +0,0 @@ -load("//tools/go_stateify:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "anon", - srcs = [ - "anon.go", - "device.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/anon", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/sentry/context", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/usermem", - ], -) diff --git a/pkg/sentry/fs/anon/anon_state_autogen.go b/pkg/sentry/fs/anon/anon_state_autogen.go new file mode 100755 index 000000000..fcb914212 --- /dev/null +++ b/pkg/sentry/fs/anon/anon_state_autogen.go @@ -0,0 +1,4 @@ +// automatically generated by stateify. + +package anon + diff --git a/pkg/sentry/fs/copy_up_test.go b/pkg/sentry/fs/copy_up_test.go deleted file mode 100644 index 1d80bf15a..000000000 --- a/pkg/sentry/fs/copy_up_test.go +++ /dev/null @@ -1,183 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fs_test - -import ( - "bytes" - "crypto/rand" - "fmt" - "io" - "sync" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/fs" - _ "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs" - "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" - "gvisor.dev/gvisor/pkg/sentry/usermem" -) - -const ( - // origFileSize is the original file size. This many bytes should be - // copied up before the test file is modified. - origFileSize = 4096 - - // truncatedFileSize is the size to truncate all test files. - truncateFileSize = 10 -) - -// TestConcurrentCopyUp is a copy up stress test for an overlay. -// -// It creates a 64-level deep directory tree in the lower filesystem and -// populates the last subdirectory with 64 files containing random content: -// -// /lower -// /sudir0/.../subdir63/ -// /file0 -// ... -// /file63 -// -// The files are truncated concurrently by 4 goroutines per file. -// These goroutines contend with copying up all parent 64 subdirectories -// as well as the final file content. -// -// At the end of the test, we assert that the files respect the new truncated -// size and contain the content we expect. -func TestConcurrentCopyUp(t *testing.T) { - ctx := contexttest.Context(t) - files := makeOverlayTestFiles(t) - - var wg sync.WaitGroup - for _, file := range files { - for i := 0; i < 4; i++ { - wg.Add(1) - go func(o *overlayTestFile) { - if err := o.File.Dirent.Inode.Truncate(ctx, o.File.Dirent, truncateFileSize); err != nil { - t.Fatalf("failed to copy up: %v", err) - } - wg.Done() - }(file) - } - } - wg.Wait() - - for _, file := range files { - got := make([]byte, origFileSize) - n, err := file.File.Readv(ctx, usermem.BytesIOSequence(got)) - if int(n) != truncateFileSize { - t.Fatalf("read %d bytes from file, want %d", n, truncateFileSize) - } - if err != nil && err != io.EOF { - t.Fatalf("read got error %v, want nil", err) - } - if !bytes.Equal(got[:n], file.content[:truncateFileSize]) { - t.Fatalf("file content is %v, want %v", got[:n], file.content[:truncateFileSize]) - } - } -} - -type overlayTestFile struct { - File *fs.File - name string - content []byte -} - -func makeOverlayTestFiles(t *testing.T) []*overlayTestFile { - ctx := contexttest.Context(t) - - // Create a lower tmpfs mount. - fsys, _ := fs.FindFilesystem("tmpfs") - lower, err := fsys.Mount(contexttest.Context(t), "", fs.MountSourceFlags{}, "", nil) - if err != nil { - t.Fatalf("failed to mount tmpfs: %v", err) - } - lowerRoot := fs.NewDirent(ctx, lower, "") - - // Make a deep set of subdirectories that everyone shares. - next := lowerRoot - for i := 0; i < 64; i++ { - name := fmt.Sprintf("subdir%d", i) - err := next.CreateDirectory(ctx, lowerRoot, name, fs.FilePermsFromMode(0777)) - if err != nil { - t.Fatalf("failed to create dir %q: %v", name, err) - } - next, err = next.Walk(ctx, lowerRoot, name) - if err != nil { - t.Fatalf("failed to walk to %q: %v", name, err) - } - } - - // Make a bunch of files in the last directory. - var files []*overlayTestFile - for i := 0; i < 64; i++ { - name := fmt.Sprintf("file%d", i) - f, err := next.Create(ctx, next, name, fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0666)) - if err != nil { - t.Fatalf("failed to create file %q: %v", name, err) - } - defer f.DecRef() - - relname, _ := f.Dirent.FullName(lowerRoot) - - o := &overlayTestFile{ - name: relname, - content: make([]byte, origFileSize), - } - - if _, err := rand.Read(o.content); err != nil { - t.Fatalf("failed to read from /dev/urandom: %v", err) - } - - if _, err := f.Writev(ctx, usermem.BytesIOSequence(o.content)); err != nil { - t.Fatalf("failed to write content to file %q: %v", name, err) - } - - files = append(files, o) - } - - // Create an empty upper tmpfs mount which we will copy up into. - upper, err := fsys.Mount(ctx, "", fs.MountSourceFlags{}, "", nil) - if err != nil { - t.Fatalf("failed to mount tmpfs: %v", err) - } - - // Construct an overlay root. - overlay, err := fs.NewOverlayRoot(ctx, upper, lower, fs.MountSourceFlags{}) - if err != nil { - t.Fatalf("failed to construct overlay root: %v", err) - } - - // Create a MountNamespace to traverse the file system. - mns, err := fs.NewMountNamespace(ctx, overlay) - if err != nil { - t.Fatalf("failed to construct mount manager: %v", err) - } - - // Walk to all of the files in the overlay, open them readable. - for _, f := range files { - maxTraversals := uint(0) - d, err := mns.FindInode(ctx, mns.Root(), mns.Root(), f.name, &maxTraversals) - if err != nil { - t.Fatalf("failed to find %q: %v", f.name, err) - } - defer d.DecRef() - - f.File, err = d.Inode.GetFile(ctx, d, fs.FileFlags{Read: true}) - if err != nil { - t.Fatalf("failed to open file %q readable: %v", f.name, err) - } - } - - return files -} diff --git a/pkg/sentry/fs/dev/BUILD b/pkg/sentry/fs/dev/BUILD deleted file mode 100644 index 80e106e6f..000000000 --- a/pkg/sentry/fs/dev/BUILD +++ /dev/null @@ -1,35 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_stateify:defs.bzl", "go_library") - -go_library( - name = "dev", - srcs = [ - "dev.go", - "device.go", - "fs.go", - "full.go", - "null.go", - "random.go", - "tty.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/dev", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/rand", - "//pkg/sentry/context", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fs/ramfs", - "//pkg/sentry/fs/tmpfs", - "//pkg/sentry/memmap", - "//pkg/sentry/mm", - "//pkg/sentry/pgalloc", - "//pkg/sentry/safemem", - "//pkg/sentry/usermem", - "//pkg/syserror", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/fs/dev/dev_state_autogen.go b/pkg/sentry/fs/dev/dev_state_autogen.go new file mode 100755 index 000000000..a997f3ecf --- /dev/null +++ b/pkg/sentry/fs/dev/dev_state_autogen.go @@ -0,0 +1,130 @@ +// automatically generated by stateify. + +package dev + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *filesystem) beforeSave() {} +func (x *filesystem) save(m state.Map) { + x.beforeSave() +} + +func (x *filesystem) afterLoad() {} +func (x *filesystem) load(m state.Map) { +} + +func (x *fullDevice) beforeSave() {} +func (x *fullDevice) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *fullDevice) afterLoad() {} +func (x *fullDevice) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *fullFileOperations) beforeSave() {} +func (x *fullFileOperations) save(m state.Map) { + x.beforeSave() +} + +func (x *fullFileOperations) afterLoad() {} +func (x *fullFileOperations) load(m state.Map) { +} + +func (x *nullDevice) beforeSave() {} +func (x *nullDevice) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *nullDevice) afterLoad() {} +func (x *nullDevice) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *nullFileOperations) beforeSave() {} +func (x *nullFileOperations) save(m state.Map) { + x.beforeSave() +} + +func (x *nullFileOperations) afterLoad() {} +func (x *nullFileOperations) load(m state.Map) { +} + +func (x *zeroDevice) beforeSave() {} +func (x *zeroDevice) save(m state.Map) { + x.beforeSave() + m.Save("nullDevice", &x.nullDevice) +} + +func (x *zeroDevice) afterLoad() {} +func (x *zeroDevice) load(m state.Map) { + m.Load("nullDevice", &x.nullDevice) +} + +func (x *zeroFileOperations) beforeSave() {} +func (x *zeroFileOperations) save(m state.Map) { + x.beforeSave() +} + +func (x *zeroFileOperations) afterLoad() {} +func (x *zeroFileOperations) load(m state.Map) { +} + +func (x *randomDevice) beforeSave() {} +func (x *randomDevice) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *randomDevice) afterLoad() {} +func (x *randomDevice) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *randomFileOperations) beforeSave() {} +func (x *randomFileOperations) save(m state.Map) { + x.beforeSave() +} + +func (x *randomFileOperations) afterLoad() {} +func (x *randomFileOperations) load(m state.Map) { +} + +func (x *ttyInodeOperations) beforeSave() {} +func (x *ttyInodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *ttyInodeOperations) afterLoad() {} +func (x *ttyInodeOperations) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *ttyFileOperations) beforeSave() {} +func (x *ttyFileOperations) save(m state.Map) { + x.beforeSave() +} + +func (x *ttyFileOperations) afterLoad() {} +func (x *ttyFileOperations) load(m state.Map) { +} + +func init() { + state.Register("dev.filesystem", (*filesystem)(nil), state.Fns{Save: (*filesystem).save, Load: (*filesystem).load}) + state.Register("dev.fullDevice", (*fullDevice)(nil), state.Fns{Save: (*fullDevice).save, Load: (*fullDevice).load}) + state.Register("dev.fullFileOperations", (*fullFileOperations)(nil), state.Fns{Save: (*fullFileOperations).save, Load: (*fullFileOperations).load}) + state.Register("dev.nullDevice", (*nullDevice)(nil), state.Fns{Save: (*nullDevice).save, Load: (*nullDevice).load}) + state.Register("dev.nullFileOperations", (*nullFileOperations)(nil), state.Fns{Save: (*nullFileOperations).save, Load: (*nullFileOperations).load}) + state.Register("dev.zeroDevice", (*zeroDevice)(nil), state.Fns{Save: (*zeroDevice).save, Load: (*zeroDevice).load}) + state.Register("dev.zeroFileOperations", (*zeroFileOperations)(nil), state.Fns{Save: (*zeroFileOperations).save, Load: (*zeroFileOperations).load}) + state.Register("dev.randomDevice", (*randomDevice)(nil), state.Fns{Save: (*randomDevice).save, Load: (*randomDevice).load}) + state.Register("dev.randomFileOperations", (*randomFileOperations)(nil), state.Fns{Save: (*randomFileOperations).save, Load: (*randomFileOperations).load}) + state.Register("dev.ttyInodeOperations", (*ttyInodeOperations)(nil), state.Fns{Save: (*ttyInodeOperations).save, Load: (*ttyInodeOperations).load}) + state.Register("dev.ttyFileOperations", (*ttyFileOperations)(nil), state.Fns{Save: (*ttyFileOperations).save, Load: (*ttyFileOperations).load}) +} diff --git a/pkg/sentry/fs/dirent_cache_test.go b/pkg/sentry/fs/dirent_cache_test.go deleted file mode 100644 index 395c879f5..000000000 --- a/pkg/sentry/fs/dirent_cache_test.go +++ /dev/null @@ -1,247 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fs - -import ( - "testing" -) - -func TestDirentCache(t *testing.T) { - const maxSize = 5 - - c := NewDirentCache(maxSize) - - // Size starts at 0. - if got, want := c.Size(), uint64(0); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // Create a Dirent d. - d := NewNegativeDirent("") - - // c does not contain d. - if got, want := c.contains(d), false; got != want { - t.Errorf("c.contains(d) got %v want %v", got, want) - } - - // Add d to the cache. - c.Add(d) - - // Size is now 1. - if got, want := c.Size(), uint64(1); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // c contains d. - if got, want := c.contains(d), true; got != want { - t.Errorf("c.contains(d) got %v want %v", got, want) - } - - // Add maxSize-1 more elements. d should be oldest element. - for i := 0; i < maxSize-1; i++ { - c.Add(NewNegativeDirent("")) - } - - // Size is maxSize. - if got, want := c.Size(), uint64(maxSize); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // c contains d. - if got, want := c.contains(d), true; got != want { - t.Errorf("c.contains(d) got %v want %v", got, want) - } - - // "Bump" d to the front by re-adding it. - c.Add(d) - - // Size is maxSize. - if got, want := c.Size(), uint64(maxSize); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // c contains d. - if got, want := c.contains(d), true; got != want { - t.Errorf("c.contains(d) got %v want %v", got, want) - } - - // Add maxSize-1 more elements. d should again be oldest element. - for i := 0; i < maxSize-1; i++ { - c.Add(NewNegativeDirent("")) - } - - // Size is maxSize. - if got, want := c.Size(), uint64(maxSize); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // c contains d. - if got, want := c.contains(d), true; got != want { - t.Errorf("c.contains(d) got %v want %v", got, want) - } - - // Add one more element, which will bump d from the cache. - c.Add(NewNegativeDirent("")) - - // Size is maxSize. - if got, want := c.Size(), uint64(maxSize); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // c does not contain d. - if got, want := c.contains(d), false; got != want { - t.Errorf("c.contains(d) got %v want %v", got, want) - } - - // Invalidating causes size to be 0 and list to be empty. - c.Invalidate() - if got, want := c.Size(), uint64(0); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - if got, want := c.list.Empty(), true; got != want { - t.Errorf("c.list.Empty() got %v, want %v", got, want) - } - - // Fill cache with maxSize dirents. - for i := 0; i < maxSize; i++ { - c.Add(NewNegativeDirent("")) - } -} - -func TestDirentCacheLimiter(t *testing.T) { - const ( - globalMaxSize = 5 - maxSize = 3 - ) - - limit := NewDirentCacheLimiter(globalMaxSize) - c1 := NewDirentCache(maxSize) - c1.limit = limit - c2 := NewDirentCache(maxSize) - c2.limit = limit - - // Create a Dirent d. - d := NewNegativeDirent("") - - // Add d to the cache. - c1.Add(d) - if got, want := c1.Size(), uint64(1); got != want { - t.Errorf("c1.Size() got %v, want %v", got, want) - } - - // Add maxSize-1 more elements. d should be oldest element. - for i := 0; i < maxSize-1; i++ { - c1.Add(NewNegativeDirent("")) - } - if got, want := c1.Size(), uint64(maxSize); got != want { - t.Errorf("c1.Size() got %v, want %v", got, want) - } - - // Check that d is still there. - if got, want := c1.contains(d), true; got != want { - t.Errorf("c1.contains(d) got %v want %v", got, want) - } - - // Fill up the other cache, it will start dropping old entries from the cache - // when the global limit is reached. - for i := 0; i < maxSize; i++ { - c2.Add(NewNegativeDirent("")) - } - - // Check is what's remaining from global max. - if got, want := c2.Size(), globalMaxSize-maxSize; int(got) != want { - t.Errorf("c2.Size() got %v, want %v", got, want) - } - - // Check that d was not dropped. - if got, want := c1.contains(d), true; got != want { - t.Errorf("c1.contains(d) got %v want %v", got, want) - } - - // Add an entry that will eventually be dropped. Check is done later... - drop := NewNegativeDirent("") - c1.Add(drop) - - // Check that d is bumped to front even when global limit is reached. - c1.Add(d) - if got, want := c1.contains(d), true; got != want { - t.Errorf("c1.contains(d) got %v want %v", got, want) - } - - // Add 2 more element and check that: - // - d is still in the list: to verify that d was bumped - // - d2/d3 are in the list: older entries are dropped when global limit is - // reached. - // - drop is not in the list: indeed older elements are dropped. - d2 := NewNegativeDirent("") - c1.Add(d2) - d3 := NewNegativeDirent("") - c1.Add(d3) - if got, want := c1.contains(d), true; got != want { - t.Errorf("c1.contains(d) got %v want %v", got, want) - } - if got, want := c1.contains(d2), true; got != want { - t.Errorf("c1.contains(d2) got %v want %v", got, want) - } - if got, want := c1.contains(d3), true; got != want { - t.Errorf("c1.contains(d3) got %v want %v", got, want) - } - if got, want := c1.contains(drop), false; got != want { - t.Errorf("c1.contains(drop) got %v want %v", got, want) - } - - // Drop all entries from one cache. The other will be allowed to grow. - c1.Invalidate() - c2.Add(NewNegativeDirent("")) - if got, want := c2.Size(), uint64(maxSize); got != want { - t.Errorf("c2.Size() got %v, want %v", got, want) - } -} - -// TestNilDirentCache tests that a nil cache supports all cache operations, but -// treats them as noop. -func TestNilDirentCache(t *testing.T) { - // Create a nil cache. - var c *DirentCache - - // Size is zero. - if got, want := c.Size(), uint64(0); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // Call Add. - c.Add(NewNegativeDirent("")) - - // Size is zero. - if got, want := c.Size(), uint64(0); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // Call Remove. - c.Remove(NewNegativeDirent("")) - - // Size is zero. - if got, want := c.Size(), uint64(0); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } - - // Call Invalidate. - c.Invalidate() - - // Size is zero. - if got, want := c.Size(), uint64(0); got != want { - t.Errorf("c.Size() got %v, want %v", got, want) - } -} diff --git a/pkg/sentry/fs/dirent_list.go b/pkg/sentry/fs/dirent_list.go new file mode 100755 index 000000000..750961a48 --- /dev/null +++ b/pkg/sentry/fs/dirent_list.go @@ -0,0 +1,173 @@ +package fs + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type direntElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (direntElementMapper) linkerFor(elem *Dirent) *Dirent { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type direntList struct { + head *Dirent + tail *Dirent +} + +// Reset resets list l to the empty state. +func (l *direntList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *direntList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *direntList) Front() *Dirent { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *direntList) Back() *Dirent { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *direntList) PushFront(e *Dirent) { + direntElementMapper{}.linkerFor(e).SetNext(l.head) + direntElementMapper{}.linkerFor(e).SetPrev(nil) + + if l.head != nil { + direntElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *direntList) PushBack(e *Dirent) { + direntElementMapper{}.linkerFor(e).SetNext(nil) + direntElementMapper{}.linkerFor(e).SetPrev(l.tail) + + if l.tail != nil { + direntElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *direntList) PushBackList(m *direntList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + direntElementMapper{}.linkerFor(l.tail).SetNext(m.head) + direntElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *direntList) InsertAfter(b, e *Dirent) { + a := direntElementMapper{}.linkerFor(b).Next() + direntElementMapper{}.linkerFor(e).SetNext(a) + direntElementMapper{}.linkerFor(e).SetPrev(b) + direntElementMapper{}.linkerFor(b).SetNext(e) + + if a != nil { + direntElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *direntList) InsertBefore(a, e *Dirent) { + b := direntElementMapper{}.linkerFor(a).Prev() + direntElementMapper{}.linkerFor(e).SetNext(a) + direntElementMapper{}.linkerFor(e).SetPrev(b) + direntElementMapper{}.linkerFor(a).SetPrev(e) + + if b != nil { + direntElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *direntList) Remove(e *Dirent) { + prev := direntElementMapper{}.linkerFor(e).Prev() + next := direntElementMapper{}.linkerFor(e).Next() + + if prev != nil { + direntElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + direntElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type direntEntry struct { + next *Dirent + prev *Dirent +} + +// Next returns the entry that follows e in the list. +func (e *direntEntry) Next() *Dirent { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *direntEntry) Prev() *Dirent { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *direntEntry) SetNext(elem *Dirent) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *direntEntry) SetPrev(elem *Dirent) { + e.prev = elem +} diff --git a/pkg/sentry/fs/dirent_refs_test.go b/pkg/sentry/fs/dirent_refs_test.go deleted file mode 100644 index 884e3ff06..000000000 --- a/pkg/sentry/fs/dirent_refs_test.go +++ /dev/null @@ -1,418 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fs - -import ( - "syscall" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" -) - -func newMockDirInode(ctx context.Context, cache *DirentCache) *Inode { - return NewMockInode(ctx, NewMockMountSource(cache), StableAttr{Type: Directory}) -} - -func TestWalkPositive(t *testing.T) { - // refs == 0 -> one reference. - // refs == -1 -> has been destroyed. - - ctx := contexttest.Context(t) - root := NewDirent(ctx, newMockDirInode(ctx, nil), "root") - - if got := root.ReadRefs(); got != 1 { - t.Fatalf("root has a ref count of %d, want %d", got, 1) - } - - name := "d" - d, err := root.walk(ctx, root, name, false) - if err != nil { - t.Fatalf("root.walk(root, %q) got %v, want nil", name, err) - } - - if got := root.ReadRefs(); got != 2 { - t.Fatalf("root has a ref count of %d, want %d", got, 2) - } - - if got := d.ReadRefs(); got != 1 { - t.Fatalf("child name = %q has a ref count of %d, want %d", d.name, got, 1) - } - - d.DecRef() - - if got := root.ReadRefs(); got != 1 { - t.Fatalf("root has a ref count of %d, want %d", got, 1) - } - - if got := d.ReadRefs(); got != 0 { - t.Fatalf("child name = %q has a ref count of %d, want %d", d.name, got, 0) - } - - root.flush() - - if got := len(root.children); got != 0 { - t.Fatalf("root has %d children, want %d", got, 0) - } -} - -func TestWalkNegative(t *testing.T) { - // refs == 0 -> one reference. - // refs == -1 -> has been destroyed. - - ctx := contexttest.Context(t) - root := NewDirent(ctx, NewEmptyDir(ctx, nil), "root") - mn := root.Inode.InodeOperations.(*mockInodeOperationsLookupNegative) - - if got := root.ReadRefs(); got != 1 { - t.Fatalf("root has a ref count of %d, want %d", got, 1) - } - - name := "d" - for i := 0; i < 100; i++ { - _, err := root.walk(ctx, root, name, false) - if err != syscall.ENOENT { - t.Fatalf("root.walk(root, %q) got %v, want %v", name, err, syscall.ENOENT) - } - } - - if got := root.ReadRefs(); got != 1 { - t.Fatalf("root has a ref count of %d, want %d", got, 1) - } - - if got := len(root.children); got != 1 { - t.Fatalf("root has %d children, want %d", got, 1) - } - - w, ok := root.children[name] - if !ok { - t.Fatalf("root wants child at %q", name) - } - - child := w.Get() - if child == nil { - t.Fatalf("root wants to resolve weak reference") - } - - if !child.(*Dirent).IsNegative() { - t.Fatalf("root found positive child at %q, want negative", name) - } - - if got := child.(*Dirent).ReadRefs(); got != 2 { - t.Fatalf("child has a ref count of %d, want %d", got, 2) - } - - child.DecRef() - - if got := child.(*Dirent).ReadRefs(); got != 1 { - t.Fatalf("child has a ref count of %d, want %d", got, 1) - } - - if got := len(root.children); got != 1 { - t.Fatalf("root has %d children, want %d", got, 1) - } - - root.DecRef() - - if got := root.ReadRefs(); got != 0 { - t.Fatalf("root has a ref count of %d, want %d", got, 0) - } - - AsyncBarrier() - - if got := mn.releaseCalled; got != true { - t.Fatalf("root.Close was called %v, want true", got) - } -} - -type mockInodeOperationsLookupNegative struct { - *MockInodeOperations - releaseCalled bool -} - -func NewEmptyDir(ctx context.Context, cache *DirentCache) *Inode { - m := NewMockMountSource(cache) - return NewInode(ctx, &mockInodeOperationsLookupNegative{ - MockInodeOperations: NewMockInodeOperations(ctx), - }, m, StableAttr{Type: Directory}) -} - -func (m *mockInodeOperationsLookupNegative) Lookup(ctx context.Context, dir *Inode, p string) (*Dirent, error) { - return NewNegativeDirent(p), nil -} - -func (m *mockInodeOperationsLookupNegative) Release(context.Context) { - m.releaseCalled = true -} - -func TestHashNegativeToPositive(t *testing.T) { - // refs == 0 -> one reference. - // refs == -1 -> has been destroyed. - - ctx := contexttest.Context(t) - root := NewDirent(ctx, NewEmptyDir(ctx, nil), "root") - - name := "d" - _, err := root.walk(ctx, root, name, false) - if err != syscall.ENOENT { - t.Fatalf("root.walk(root, %q) got %v, want %v", name, err, syscall.ENOENT) - } - - if got := root.exists(ctx, root, name); got != false { - t.Fatalf("got %q exists, want does not exist", name) - } - - f, err := root.Create(ctx, root, name, FileFlags{}, FilePermissions{}) - if err != nil { - t.Fatalf("root.Create(%q, _), got error %v, want nil", name, err) - } - d := f.Dirent - - if d.IsNegative() { - t.Fatalf("got negative Dirent, want positive") - } - - if got := d.ReadRefs(); got != 1 { - t.Fatalf("child %q has a ref count of %d, want %d", name, got, 1) - } - - if got := root.ReadRefs(); got != 2 { - t.Fatalf("root has a ref count of %d, want %d", got, 2) - } - - if got := len(root.children); got != 1 { - t.Fatalf("got %d children, want %d", got, 1) - } - - w, ok := root.children[name] - if !ok { - t.Fatalf("failed to find weak reference to %q", name) - } - - child := w.Get() - if child == nil { - t.Fatalf("want to resolve weak reference") - } - - if child.(*Dirent) != d { - t.Fatalf("got foreign child") - } -} - -func TestRevalidate(t *testing.T) { - // refs == 0 -> one reference. - // refs == -1 -> has been destroyed. - - for _, test := range []struct { - // desc is the test's description. - desc string - - // Whether to make negative Dirents. - makeNegative bool - }{ - { - desc: "Revalidate negative Dirent", - makeNegative: true, - }, - { - desc: "Revalidate positive Dirent", - makeNegative: false, - }, - } { - t.Run(test.desc, func(t *testing.T) { - ctx := contexttest.Context(t) - root := NewDirent(ctx, NewMockInodeRevalidate(ctx, test.makeNegative), "root") - - name := "d" - d1, err := root.walk(ctx, root, name, false) - if !test.makeNegative && err != nil { - t.Fatalf("root.walk(root, %q) got %v, want nil", name, err) - } - d2, err := root.walk(ctx, root, name, false) - if !test.makeNegative && err != nil { - t.Fatalf("root.walk(root, %q) got %v, want nil", name, err) - } - if !test.makeNegative && d1 == d2 { - t.Fatalf("revalidating walk got same *Dirent, want different") - } - if got := len(root.children); got != 1 { - t.Errorf("revalidating walk got %d children, want %d", got, 1) - } - }) - } -} - -type MockInodeOperationsRevalidate struct { - *MockInodeOperations - makeNegative bool -} - -func NewMockInodeRevalidate(ctx context.Context, makeNegative bool) *Inode { - mn := NewMockInodeOperations(ctx) - m := NewMockMountSource(nil) - m.MountSourceOperations.(*MockMountSourceOps).revalidate = true - return NewInode(ctx, &MockInodeOperationsRevalidate{MockInodeOperations: mn, makeNegative: makeNegative}, m, StableAttr{Type: Directory}) -} - -func (m *MockInodeOperationsRevalidate) Lookup(ctx context.Context, dir *Inode, p string) (*Dirent, error) { - if !m.makeNegative { - return m.MockInodeOperations.Lookup(ctx, dir, p) - } - return NewNegativeDirent(p), nil -} - -func TestCreateExtraRefs(t *testing.T) { - // refs == 0 -> one reference. - // refs == -1 -> has been destroyed. - - ctx := contexttest.Context(t) - for _, test := range []struct { - // desc is the test's description. - desc string - - // root is the Dirent to create from. - root *Dirent - - // expected references on walked Dirent. - refs int64 - }{ - { - desc: "Create caching", - root: NewDirent(ctx, NewEmptyDir(ctx, NewDirentCache(1)), "root"), - refs: 2, - }, - { - desc: "Create not caching", - root: NewDirent(ctx, NewEmptyDir(ctx, nil), "root"), - refs: 1, - }, - } { - t.Run(test.desc, func(t *testing.T) { - name := "d" - f, err := test.root.Create(ctx, test.root, name, FileFlags{}, FilePermissions{}) - if err != nil { - t.Fatalf("root.Create(root, %q) failed: %v", name, err) - } - d := f.Dirent - - if got := d.ReadRefs(); got != test.refs { - t.Errorf("dirent has a ref count of %d, want %d", got, test.refs) - } - }) - } -} - -func TestRemoveExtraRefs(t *testing.T) { - // refs == 0 -> one reference. - // refs == -1 -> has been destroyed. - - ctx := contexttest.Context(t) - for _, test := range []struct { - // desc is the test's description. - desc string - - // root is the Dirent to make and remove from. - root *Dirent - }{ - { - desc: "Remove caching", - root: NewDirent(ctx, NewEmptyDir(ctx, NewDirentCache(1)), "root"), - }, - { - desc: "Remove not caching", - root: NewDirent(ctx, NewEmptyDir(ctx, nil), "root"), - }, - } { - t.Run(test.desc, func(t *testing.T) { - name := "d" - f, err := test.root.Create(ctx, test.root, name, FileFlags{}, FilePermissions{}) - if err != nil { - t.Fatalf("root.Create(%q, _) failed: %v", name, err) - } - d := f.Dirent - - if err := test.root.Remove(contexttest.Context(t), test.root, name); err != nil { - t.Fatalf("root.Remove(root, %q) failed: %v", name, err) - } - - if got := d.ReadRefs(); got != 1 { - t.Fatalf("dirent has a ref count of %d, want %d", got, 1) - } - - d.DecRef() - - test.root.flush() - - if got := len(test.root.children); got != 0 { - t.Errorf("root has %d children, want %d", got, 0) - } - }) - } -} - -func TestRenameExtraRefs(t *testing.T) { - // refs == 0 -> one reference. - // refs == -1 -> has been destroyed. - - for _, test := range []struct { - // desc is the test's description. - desc string - - // cache of extra Dirent references, may be nil. - cache *DirentCache - }{ - { - desc: "Rename no caching", - cache: nil, - }, - { - desc: "Rename caching", - cache: NewDirentCache(5), - }, - } { - t.Run(test.desc, func(t *testing.T) { - ctx := contexttest.Context(t) - - dirAttr := StableAttr{Type: Directory} - - oldParent := NewDirent(ctx, NewMockInode(ctx, NewMockMountSource(test.cache), dirAttr), "old_parent") - newParent := NewDirent(ctx, NewMockInode(ctx, NewMockMountSource(test.cache), dirAttr), "new_parent") - - renamed, err := oldParent.Walk(ctx, oldParent, "old_child") - if err != nil { - t.Fatalf("Walk(oldParent, %q) got error %v, want nil", "old_child", err) - } - replaced, err := newParent.Walk(ctx, oldParent, "new_child") - if err != nil { - t.Fatalf("Walk(newParent, %q) got error %v, want nil", "new_child", err) - } - - if err := Rename(contexttest.RootContext(t), oldParent /*root */, oldParent, "old_child", newParent, "new_child"); err != nil { - t.Fatalf("Rename got error %v, want nil", err) - } - - oldParent.flush() - newParent.flush() - - // Expect to have only active references. - if got := renamed.ReadRefs(); got != 1 { - t.Errorf("renamed has ref count %d, want only active references %d", got, 1) - } - if got := replaced.ReadRefs(); got != 1 { - t.Errorf("replaced has ref count %d, want only active references %d", got, 1) - } - }) - } -} diff --git a/pkg/sentry/fs/event_list.go b/pkg/sentry/fs/event_list.go new file mode 100755 index 000000000..c94cb03e1 --- /dev/null +++ b/pkg/sentry/fs/event_list.go @@ -0,0 +1,173 @@ +package fs + +// ElementMapper provides an identity mapping by default. +// +// This can be replaced to provide a struct that maps elements to linker +// objects, if they are not the same. An ElementMapper is not typically +// required if: Linker is left as is, Element is left as is, or Linker and +// Element are the same type. +type eventElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (eventElementMapper) linkerFor(elem *Event) *Event { return elem } + +// List is an intrusive list. Entries can be added to or removed from the list +// in O(1) time and with no additional memory allocations. +// +// The zero value for List is an empty list ready to use. +// +// To iterate over a list (where l is a List): +// for e := l.Front(); e != nil; e = e.Next() { +// // do something with e. +// } +// +// +stateify savable +type eventList struct { + head *Event + tail *Event +} + +// Reset resets list l to the empty state. +func (l *eventList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *eventList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *eventList) Front() *Event { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *eventList) Back() *Event { + return l.tail +} + +// PushFront inserts the element e at the front of list l. +func (l *eventList) PushFront(e *Event) { + eventElementMapper{}.linkerFor(e).SetNext(l.head) + eventElementMapper{}.linkerFor(e).SetPrev(nil) + + if l.head != nil { + eventElementMapper{}.linkerFor(l.head).SetPrev(e) + } else { + l.tail = e + } + + l.head = e +} + +// PushBack inserts the element e at the back of list l. +func (l *eventList) PushBack(e *Event) { + eventElementMapper{}.linkerFor(e).SetNext(nil) + eventElementMapper{}.linkerFor(e).SetPrev(l.tail) + + if l.tail != nil { + eventElementMapper{}.linkerFor(l.tail).SetNext(e) + } else { + l.head = e + } + + l.tail = e +} + +// PushBackList inserts list m at the end of list l, emptying m. +func (l *eventList) PushBackList(m *eventList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + eventElementMapper{}.linkerFor(l.tail).SetNext(m.head) + eventElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *eventList) InsertAfter(b, e *Event) { + a := eventElementMapper{}.linkerFor(b).Next() + eventElementMapper{}.linkerFor(e).SetNext(a) + eventElementMapper{}.linkerFor(e).SetPrev(b) + eventElementMapper{}.linkerFor(b).SetNext(e) + + if a != nil { + eventElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *eventList) InsertBefore(a, e *Event) { + b := eventElementMapper{}.linkerFor(a).Prev() + eventElementMapper{}.linkerFor(e).SetNext(a) + eventElementMapper{}.linkerFor(e).SetPrev(b) + eventElementMapper{}.linkerFor(a).SetPrev(e) + + if b != nil { + eventElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *eventList) Remove(e *Event) { + prev := eventElementMapper{}.linkerFor(e).Prev() + next := eventElementMapper{}.linkerFor(e).Next() + + if prev != nil { + eventElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + eventElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } +} + +// Entry is a default implementation of Linker. Users can add anonymous fields +// of this type to their structs to make them automatically implement the +// methods needed by List. +// +// +stateify savable +type eventEntry struct { + next *Event + prev *Event +} + +// Next returns the entry that follows e in the list. +func (e *eventEntry) Next() *Event { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *eventEntry) Prev() *Event { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *eventEntry) SetNext(elem *Event) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *eventEntry) SetPrev(elem *Event) { + e.prev = elem +} diff --git a/pkg/sentry/fs/ext/BUILD b/pkg/sentry/fs/ext/BUILD deleted file mode 100644 index 8158aa522..000000000 --- a/pkg/sentry/fs/ext/BUILD +++ /dev/null @@ -1,80 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_stateify:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -go_template_instance( - name = "dentry_list", - out = "dentry_list.go", - package = "ext", - prefix = "dentry", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*dentry", - "Linker": "*dentry", - }, -) - -go_library( - name = "ext", - srcs = [ - "block_map_file.go", - "dentry.go", - "dentry_list.go", - "directory.go", - "ext.go", - "extent_file.go", - "filesystem.go", - "inline_file.go", - "inode.go", - "named_pipe.go", - "regular_file.go", - "symlink.go", - "utils.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/ext", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/fd", - "//pkg/sentry/context", - "//pkg/sentry/fs", - "//pkg/sentry/fs/ext/disklayout", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/pipe", - "//pkg/sentry/safemem", - "//pkg/sentry/usermem", - "//pkg/sentry/vfs", - "//pkg/syserror", - ], -) - -go_test( - name = "ext_test", - size = "small", - srcs = [ - "ext_test.go", - "extent_test.go", - ], - data = [ - "//pkg/sentry/fs/ext:assets/bigfile.txt", - "//pkg/sentry/fs/ext:assets/file.txt", - "//pkg/sentry/fs/ext:assets/tiny.ext2", - "//pkg/sentry/fs/ext:assets/tiny.ext3", - "//pkg/sentry/fs/ext:assets/tiny.ext4", - ], - embed = [":ext"], - deps = [ - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", - "//pkg/sentry/fs/ext/disklayout", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - "//runsc/test/testutil", - "@com_github_google_go-cmp//cmp:go_default_library", - "@com_github_google_go-cmp//cmp/cmpopts:go_default_library", - ], -) diff --git a/pkg/sentry/fs/ext/assets/README.md b/pkg/sentry/fs/ext/assets/README.md deleted file mode 100644 index 6f1e81b3a..000000000 --- a/pkg/sentry/fs/ext/assets/README.md +++ /dev/null @@ -1,36 +0,0 @@ -### Tiny Ext(2/3/4) Images - -The images are of size 64Kb which supports 64 1k blocks and 16 inodes. This is -the smallest size mkfs.ext(2/3/4) works with. - -These images were generated using the following commands. - -```bash -fallocate -l 64K tiny.ext$VERSION -mkfs.ext$VERSION -j tiny.ext$VERSION -``` - -where `VERSION` is `2`, `3` or `4`. - -You can mount it using: - -```bash -sudo mount -o loop tiny.ext$VERSION $MOUNTPOINT -``` - -`file.txt`, `bigfile.txt` and `symlink.txt` were added to this image by just -mounting it and copying (while preserving links) those files to the mountpoint -directory using: - -```bash -sudo cp -P {file.txt,symlink.txt,bigfile.txt} $MOUNTPOINT -``` - -The files in this directory mirror the contents and organisation of the files -stored in the image. - -You can umount the filesystem using: - -```bash -sudo umount $MOUNTPOINT -``` diff --git a/pkg/sentry/fs/ext/assets/bigfile.txt b/pkg/sentry/fs/ext/assets/bigfile.txt deleted file mode 100644 index 3857cf516..000000000 --- a/pkg/sentry/fs/ext/assets/bigfile.txt +++ /dev/null @@ -1,41 +0,0 @@ -Lorem ipsum dolor sit amet, consectetur adipiscing elit. Phasellus faucibus eleifend orci, ut ornare nibh faucibus eu. Cras at condimentum massa. Nullam luctus, elit non porttitor congue, sapien diam feugiat sapien, sed eleifend nulla mauris non arcu. Sed lacinia mauris magna, eu mollis libero varius sit amet. Donec mollis, quam convallis commodo posuere, dolor nisi placerat nisi, in faucibus augue mi eu lorem. In pharetra consectetur faucibus. Ut euismod ex efficitur egestas tincidunt. Maecenas condimentum ut ante in rutrum. Vivamus sed arcu tempor, faucibus turpis et, lacinia diam. - -Sed in lacus vel nisl interdum bibendum in sed justo. Nunc tellus risus, molestie vitae arcu sed, molestie tempus ligula. Pellentesque habitant morbi tristique senectus et netus et malesuada fames ac turpis egestas. Nunc risus neque, volutpat et ante non, ullamcorper condimentum ante. Aliquam sed metus in urna condimentum convallis. Vivamus ut libero mauris. Proin mollis posuere consequat. Vestibulum placerat mollis est et pulvinar. - -Donec rutrum odio ac diam pharetra, id fermentum magna cursus. Pellentesque in dapibus elit, et condimentum orci. Class aptent taciti sociosqu ad litora torquent per conubia nostra, per inceptos himenaeos. Suspendisse euismod dapibus est, id vestibulum mauris. Nulla facilisi. Nulla cursus gravida nisi. Phasellus vestibulum rutrum lectus, a dignissim mauris hendrerit vitae. In at elementum mauris. Integer vel efficitur velit. Nullam fringilla sapien mi, quis luctus neque efficitur ac. Aenean nec quam dapibus nunc commodo pharetra. Proin sapien mi, fermentum aliquet vulputate non, aliquet porttitor diam. Quisque lacinia, urna et finibus fermentum, nunc lacus vehicula ex, sed congue metus lectus ac quam. Aliquam erat volutpat. Suspendisse sodales, dolor ut tincidunt finibus, augue erat varius tellus, a interdum erat sem at nunc. Vestibulum cursus iaculis sapien, vitae feugiat dui auctor quis. - -Pellentesque nec maximus nulla, eu blandit diam. Maecenas quis arcu ornare, congue ante at, vehicula ipsum. Praesent feugiat mauris rutrum sem fermentum, nec luctus ipsum placerat. Pellentesque placerat ipsum at dignissim fringilla. Vivamus et posuere sem, eget hendrerit felis. Aenean vulputate, augue vel mollis feugiat, justo ipsum mollis dolor, eu mollis elit neque ut ipsum. Orci varius natoque penatibus et magnis dis parturient montes, nascetur ridiculus mus. Fusce bibendum sem quam, vulputate laoreet mi dapibus imperdiet. Sed a purus non nibh pretium aliquet. Integer eget luctus augue, vitae tincidunt magna. Ut eros enim, egestas eu nulla et, lobortis egestas arcu. Cras id ipsum ac justo lacinia rutrum. Vivamus lectus leo, ultricies sed justo at, pellentesque feugiat magna. Ut sollicitudin neque elit, vel ornare mauris commodo id. - -Duis dapibus orci et sapien finibus finibus. Mauris eleifend, lacus at vestibulum maximus, quam ligula pharetra erat, sit amet dapibus neque elit vitae neque. In bibendum sollicitudin erat, eget ultricies tortor malesuada at. Sed sit amet orci turpis. Donec feugiat ligula nibh, molestie tincidunt lectus elementum id. Donec volutpat maximus nibh, in vulputate felis posuere eu. Cras tincidunt ullamcorper lacus. Phasellus porta lorem auctor, congue magna a, commodo elit. - -Etiam auctor mi quis elit sodales, eu pulvinar arcu condimentum. Aenean imperdiet risus et dapibus tincidunt. Nullam tincidunt dictum dui, sed commodo urna rutrum id. Ut mollis libero vel elit laoreet bibendum. Quisque arcu arcu, tincidunt at ultricies id, vulputate nec metus. In tristique posuere quam sit amet volutpat. Vivamus scelerisque et nunc at dapibus. Fusce finibus libero ut ligula pretium rhoncus. Mauris non elit in arcu finibus imperdiet. Pellentesque nec massa odio. Proin rutrum mauris non sagittis efficitur. Aliquam auctor quam at dignissim faucibus. Ut eget ligula in magna posuere ultricies vitae sit amet turpis. Duis maximus odio nulla. Donec gravida sem tristique tempus scelerisque. - -Interdum et malesuada fames ac ante ipsum primis in faucibus. Fusce pharetra magna vulputate aliquet tempus. Duis id hendrerit arcu. Quisque ut ex elit. Integer velit orci, venenatis ut sapien ac, placerat porttitor dui. Interdum et malesuada fames ac ante ipsum primis in faucibus. Nunc hendrerit cursus diam, hendrerit finibus ipsum scelerisque ut. Class aptent taciti sociosqu ad litora torquent per conubia nostra, per inceptos himenaeos. - -Nulla non euismod neque. Phasellus vel sapien eu metus pulvinar rhoncus. Suspendisse eu mollis tellus, quis vestibulum tortor. Maecenas interdum dolor sed nulla fermentum maximus. Donec imperdiet ullamcorper condimentum. Nam quis nibh ante. Praesent quis tellus ut tortor pulvinar blandit sit amet ut sapien. Vestibulum est orci, pellentesque vitae tristique sit amet, tristique non felis. - -Vivamus sodales pellentesque varius. Sed vel tempus ligula. Nulla tristique nisl vel dui facilisis, ac sodales augue hendrerit. Proin augue nisi, vestibulum quis augue nec, sagittis tincidunt velit. Vestibulum euismod, nulla nec sodales faucibus, urna sapien vulputate magna, id varius metus sapien ut neque. Duis in mollis urna, in scelerisque enim. Pellentesque habitant morbi tristique senectus et netus et malesuada fames ac turpis egestas. Nunc condimentum dictum turpis, et egestas neque dapibus eget. Quisque fringilla, dui eu venenatis eleifend, erat nibh lacinia urna, at lacinia lacus sapien eu dui. Duis eu erat ut mi lacinia convallis a sed ex. - -Fusce elit metus, tincidunt nec eleifend a, hendrerit nec ligula. Duis placerat finibus sollicitudin. In euismod porta tellus, in luctus justo bibendum bibendum. Maecenas at magna eleifend lectus tincidunt suscipit ut a ligula. Nulla tempor accumsan felis, fermentum dapibus est eleifend vitae. Mauris urna sem, fringilla at ultricies non, ultrices in arcu. Class aptent taciti sociosqu ad litora torquent per conubia nostra, per inceptos himenaeos. Nam vehicula nunc at laoreet imperdiet. Nunc tristique ut risus id aliquet. Integer eleifend massa orci. - -Vestibulum sed ante sollicitudin nisi fringilla bibendum nec vel quam. Sed pretium augue eu ligula congue pulvinar. Donec vitae magna tincidunt, pharetra lacus id, convallis nulla. Cras viverra nisl nisl, varius convallis leo vulputate nec. Morbi at consequat dui, sed aliquet metus. Sed suscipit fermentum mollis. Maecenas nec mi sodales, tincidunt purus in, tristique mauris. Orci varius natoque penatibus et magnis dis parturient montes, nascetur ridiculus mus. Donec interdum mi in velit efficitur, quis ultrices ex imperdiet. Sed vestibulum, magna ut tristique pretium, mi ipsum placerat tellus, non tempor enim augue et ex. Pellentesque eget felis quis ante sodales viverra ac sed lacus. Donec suscipit tempus massa, eget laoreet massa molestie at. - -Aenean fringilla dui non aliquet consectetur. Fusce cursus quam nec orci hendrerit faucibus. Donec consequat suscipit enim, non volutpat lectus auctor interdum. Proin lorem purus, maximus vel orci vitae, suscipit egestas turpis. Donec risus urna, congue a sem eu, aliquet placerat odio. Morbi gravida tristique turpis, quis efficitur enim. Nunc interdum gravida ipsum vel facilisis. Nunc congue finibus sollicitudin. Quisque euismod aliquet lectus et tincidunt. Curabitur ultrices sem ut mi fringilla fermentum. Morbi pretium, nisi sit amet dapibus congue, dolor enim consectetur risus, a interdum ligula odio sed odio. Quisque facilisis, mi at suscipit gravida, nunc sapien cursus justo, ut luctus odio nulla quis leo. Integer condimentum lobortis mauris, non egestas tellus lobortis sit amet. - -In sollicitudin velit ac ante vehicula, vitae varius tortor mollis. In hac habitasse platea dictumst. Quisque et orci lorem. Integer malesuada fringilla luctus. Pellentesque malesuada, mi non lobortis porttitor, ante ligula vulputate ante, nec dictum risus eros sit amet sapien. Nulla aliquam lorem libero, ac varius nulla tristique eget. Class aptent taciti sociosqu ad litora torquent per conubia nostra, per inceptos himenaeos. Ut pellentesque mauris orci, vel consequat mi varius a. Ut sit amet elit vulputate, lacinia metus non, fermentum nisl. Pellentesque eu nisi sed quam egestas blandit. Duis sit amet lobortis dolor. Donec consectetur sem interdum, tristique elit sit amet, sodales lacus. Class aptent taciti sociosqu ad litora torquent per conubia nostra, per inceptos himenaeos. Fusce id aliquam augue. Sed pretium congue risus vitae lacinia. Vestibulum non vulputate risus, ut malesuada justo. - -Sed odio elit, consectetur ac mauris quis, consequat commodo libero. Fusce sodales velit vulputate pulvinar fermentum. Donec iaculis nec nisl eget faucibus. Mauris at dictum velit. Donec fermentum lectus eu viverra volutpat. Aliquam consequat facilisis lorem, cursus consequat dui bibendum ullamcorper. Pellentesque nulla magna, imperdiet at magna et, cursus egestas enim. Nullam semper molestie lectus sit amet semper. Duis eget tincidunt est. Integer id neque risus. Integer ultricies hendrerit vestibulum. Donec blandit blandit sagittis. Nunc consectetur vitae nisi consectetur volutpat. - -Nulla id lorem fermentum, efficitur magna a, hendrerit dui. Vivamus sagittis orci gravida, bibendum quam eget, molestie est. Phasellus nec enim tincidunt, volutpat sapien non, laoreet diam. Nulla posuere enim nec porttitor lobortis. Donec auctor odio ut orci eleifend, ut eleifend purus convallis. Interdum et malesuada fames ac ante ipsum primis in faucibus. Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut hendrerit, purus eget viverra tincidunt, sem magna imperdiet libero, et aliquam turpis neque vitae elit. Maecenas semper varius iaculis. Cras non lorem quis quam bibendum eleifend in et libero. Curabitur at purus mauris. Lorem ipsum dolor sit amet, consectetur adipiscing elit. Vivamus porta diam sed elit eleifend gravida. - -Nulla facilisi. Ut ultricies diam vel diam consectetur, vel porta augue molestie. Fusce interdum sapien et metus facilisis pellentesque. Nulla convallis sem at nunc vehicula facilisis. Nam ac rutrum purus. Nunc bibendum, dolor sit amet tempus ullamcorper, lorem leo tempor sem, id fringilla nunc augue scelerisque augue. Nullam sit amet rutrum nisl. Vestibulum ante ipsum primis in faucibus orci luctus et ultrices posuere cubilia Curae; Donec sed mauris gravida eros vehicula sagittis at eget orci. Cras elementum, eros at accumsan bibendum, libero neque blandit purus, vitae vestibulum libero massa ac nibh. Integer at placerat nulla. Mauris eu eleifend orci. Aliquam consequat ligula vitae erat porta lobortis. Duis fermentum elit ac aliquet ornare. - -Mauris eget cursus tellus, eget sodales purus. Aliquam malesuada, augue id vulputate finibus, nisi ex bibendum nisl, sit amet laoreet quam urna a dolor. Nullam ultricies, sapien eu laoreet consequat, erat eros dignissim diam, ultrices sodales lectus mauris et leo. Morbi lacinia eu ante at tempus. Sed iaculis finibus magna malesuada efficitur. Donec faucibus erat sit amet elementum feugiat. Praesent a placerat nisi. Etiam lacinia gravida diam, et sollicitudin sapien tincidunt ut. - -Maecenas felis quam, tincidunt vitae venenatis scelerisque, viverra vitae odio. Phasellus enim neque, ultricies suscipit malesuada sit amet, vehicula sit amet purus. Nulla placerat sit amet dui vel tincidunt. Nam quis neque vel magna commodo egestas. Vestibulum sagittis rutrum lorem ut congue. Maecenas vel ultrices tellus. Donec efficitur, urna ac consequat iaculis, lorem felis pharetra eros, eget faucibus orci lectus sit amet arcu. - -Ut a tempus nisi. Nulla facilisi. Praesent vulputate maximus mi et dapibus. Sed sit amet libero ac augue hendrerit efficitur in a sapien. Mauris placerat velit sit amet tellus sollicitudin faucibus. Donec egestas a magna ac suscipit. Duis enim sapien, mollis sed egestas et, vestibulum vel leo. - -Proin quis dapibus dui. Donec eu tincidunt nunc. Vivamus eget purus consectetur, maximus ante vitae, tincidunt elit. Aenean mattis dolor a gravida aliquam. Praesent quis tellus id sem maximus vulputate nec sed nulla. Lorem ipsum dolor sit amet, consectetur adipiscing elit. Curabitur metus nulla, volutpat volutpat est eu, hendrerit congue erat. Aliquam sollicitudin augue ante. Sed sollicitudin, magna eu consequat elementum, mi augue ullamcorper felis, molestie imperdiet erat metus iaculis est. Proin ac tortor nisi. Pellentesque quis nisi risus. Integer enim sapien, tincidunt quis tortor id, accumsan venenatis mi. Nulla facilisi. - -Cras pretium sit amet quam congue maximus. Morbi lacus libero, imperdiet commodo massa sed, scelerisque placerat libero. Cras nisl nisi, consectetur sed bibendum eu, venenatis at enim. Proin sodales justo at quam aliquam, a consectetur mi ornare. Donec porta ac est sit amet efficitur. Suspendisse vestibulum tortor id neque imperdiet, id lacinia risus vehicula. Phasellus ac eleifend purus. Mauris vel gravida ante. Aliquam vitae lobortis risus. Sed vehicula consectetur tincidunt. Nam et justo vitae purus molestie consequat. Pellentesque ipsum ex, convallis quis blandit non, gravida et urna. Donec diam ligula amet. diff --git a/pkg/sentry/fs/ext/assets/file.txt b/pkg/sentry/fs/ext/assets/file.txt deleted file mode 100644 index 980a0d5f1..000000000 --- a/pkg/sentry/fs/ext/assets/file.txt +++ /dev/null @@ -1 +0,0 @@ -Hello World! diff --git a/pkg/sentry/fs/ext/assets/symlink.txt b/pkg/sentry/fs/ext/assets/symlink.txt deleted file mode 120000 index 4c330738c..000000000 --- a/pkg/sentry/fs/ext/assets/symlink.txt +++ /dev/null @@ -1 +0,0 @@ -file.txt
\ No newline at end of file diff --git a/pkg/sentry/fs/ext/assets/tiny.ext2 b/pkg/sentry/fs/ext/assets/tiny.ext2 Binary files differdeleted file mode 100644 index 381ade9bf..000000000 --- a/pkg/sentry/fs/ext/assets/tiny.ext2 +++ /dev/null diff --git a/pkg/sentry/fs/ext/assets/tiny.ext3 b/pkg/sentry/fs/ext/assets/tiny.ext3 Binary files differdeleted file mode 100644 index 0e97a324c..000000000 --- a/pkg/sentry/fs/ext/assets/tiny.ext3 +++ /dev/null diff --git a/pkg/sentry/fs/ext/assets/tiny.ext4 b/pkg/sentry/fs/ext/assets/tiny.ext4 Binary files differdeleted file mode 100644 index a6859736d..000000000 --- a/pkg/sentry/fs/ext/assets/tiny.ext4 +++ /dev/null diff --git a/pkg/sentry/fs/ext/block_map_file.go b/pkg/sentry/fs/ext/block_map_file.go deleted file mode 100644 index 9aabbd145..000000000 --- a/pkg/sentry/fs/ext/block_map_file.go +++ /dev/null @@ -1,65 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "io" - "sync" - - "gvisor.dev/gvisor/pkg/binary" -) - -// blockMapFile is a type of regular file which uses direct/indirect block -// addressing to store file data. This was deprecated in ext4. -type blockMapFile struct { - regFile regularFile - - // mu serializes changes to fileToPhysBlks. - mu sync.RWMutex - - // fileToPhysBlks maps the file block numbers to the physical block numbers. - // the physical block number for the (i)th file block is stored in the (i)th - // index. This is initialized (at max) with the first 12 entries. The rest - // have to be read in from disk when required. Protected by mu. - fileToPhysBlks []uint32 -} - -// Compiles only if blockMapFile implements fileReader. -var _ fileReader = (*blockMapFile)(nil) - -// Read implements fileReader.getFileReader. -func (f *blockMapFile) getFileReader(dev io.ReaderAt, blkSize uint64, offset uint64) io.Reader { - panic("unimplemented") -} - -// newBlockMapFile is the blockMapFile constructor. It initializes the file to -// physical blocks map with (at most) the first 12 (direct) blocks. -func newBlockMapFile(blkSize uint64, regFile regularFile) (*blockMapFile, error) { - file := &blockMapFile{regFile: regFile} - file.regFile.impl = file - - toFill := uint64(12) - blksUsed := regFile.blksUsed(blkSize) - if blksUsed < toFill { - toFill = blksUsed - } - - blkMap := regFile.inode.diskInode.Data() - file.fileToPhysBlks = make([]uint32, toFill) - for i := uint64(0); i < toFill; i++ { - binary.Unmarshal(blkMap[i*4:(i+1)*4], binary.LittleEndian, &file.fileToPhysBlks[i]) - } - return file, nil -} diff --git a/pkg/sentry/fs/ext/dentry.go b/pkg/sentry/fs/ext/dentry.go deleted file mode 100644 index 19c9b3b2d..000000000 --- a/pkg/sentry/fs/ext/dentry.go +++ /dev/null @@ -1,58 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -// dentry implements vfs.DentryImpl. -type dentry struct { - vfsd vfs.Dentry - - // inode is the inode represented by this dentry. Multiple Dentries may - // share a single non-directory Inode (with hard links). inode is - // immutable. - inode *inode - // dentryEntry links Dentries into their parent directory.childList. - dentryEntry -} - -// Compiles only if dentry implements vfs.DentryImpl. -var _ vfs.DentryImpl = (*dentry)(nil) - -// newDentry is the dentry constructor. -func newDentry(in *inode) *dentry { - d := &dentry{ - inode: in, - } - d.vfsd.Init(d) - return d -} - -// IncRef implements vfs.DentryImpl.IncRef. -func (d *dentry) IncRef(vfsfs *vfs.Filesystem) { - d.inode.incRef() -} - -// TryIncRef implements vfs.DentryImpl.TryIncRef. -func (d *dentry) TryIncRef(vfsfs *vfs.Filesystem) bool { - return d.inode.tryIncRef() -} - -// DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef(vfsfs *vfs.Filesystem) { - d.inode.decRef(vfsfs.Impl().(*filesystem)) -} diff --git a/pkg/sentry/fs/ext/directory.go b/pkg/sentry/fs/ext/directory.go deleted file mode 100644 index ab2b59e44..000000000 --- a/pkg/sentry/fs/ext/directory.go +++ /dev/null @@ -1,36 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -// directory represents a directory inode. It holds the childList in memory. -type directory struct { - inode inode - - // childList is a list containing (1) child Dentries and (2) fake Dentries - // (with inode == nil) that represent the iteration position of - // directoryFDs. childList is used to support directoryFD.IterDirents() - // efficiently. childList is immutable. - childList dentryList - - // TODO(b/134676337): Add directory navigators. -} - -// newDirectroy is the directory constructor. -func newDirectroy(inode inode) *directory { - // TODO(b/134676337): initialize childList. - file := &directory{inode: inode} - file.inode.impl = file - return file -} diff --git a/pkg/sentry/fs/ext/disklayout/BUILD b/pkg/sentry/fs/ext/disklayout/BUILD deleted file mode 100644 index dde15110d..000000000 --- a/pkg/sentry/fs/ext/disklayout/BUILD +++ /dev/null @@ -1,48 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_stateify:defs.bzl", "go_library", "go_test") - -go_library( - name = "disklayout", - srcs = [ - "block_group.go", - "block_group_32.go", - "block_group_64.go", - "dirent.go", - "dirent_new.go", - "dirent_old.go", - "disklayout.go", - "extent.go", - "inode.go", - "inode_new.go", - "inode_old.go", - "superblock.go", - "superblock_32.go", - "superblock_64.go", - "superblock_old.go", - "test_utils.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/ext/disklayout", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/sentry/fs", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - ], -) - -go_test( - name = "disklayout_test", - size = "small", - srcs = [ - "block_group_test.go", - "dirent_test.go", - "extent_test.go", - "inode_test.go", - "superblock_test.go", - ], - embed = [":disklayout"], - deps = ["//pkg/sentry/kernel/time"], -) diff --git a/pkg/sentry/fs/ext/disklayout/block_group.go b/pkg/sentry/fs/ext/disklayout/block_group.go deleted file mode 100644 index ad6f4fef8..000000000 --- a/pkg/sentry/fs/ext/disklayout/block_group.go +++ /dev/null @@ -1,137 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -// BlockGroup represents a Linux ext block group descriptor. An ext file system -// is split into a series of block groups. This provides an access layer to -// information needed to access and use a block group. -// -// Location: -// - The block group descriptor table is always placed in the blocks -// immediately after the block containing the superblock. -// - The 1st block group descriptor in the original table is in the -// (sb.FirstDataBlock() + 1)th block. -// - See SuperBlock docs to see where the block group descriptor table is -// replicated. -// - sb.BgDescSize() must be used as the block group descriptor entry size -// while reading the table from disk. -// -// See https://www.kernel.org/doc/html/latest/filesystems/ext4/globals.html#block-group-descriptors. -type BlockGroup interface { - // InodeTable returns the absolute block number of the block containing the - // inode table. This points to an array of Inode structs. Inode tables are - // statically allocated at mkfs time. The superblock records the number of - // inodes per group (length of this table) and the size of each inode struct. - InodeTable() uint64 - - // BlockBitmap returns the absolute block number of the block containing the - // block bitmap. This bitmap tracks the usage of data blocks within this block - // group and has its own checksum. - BlockBitmap() uint64 - - // InodeBitmap returns the absolute block number of the block containing the - // inode bitmap. This bitmap tracks the usage of this group's inode table - // entries and has its own checksum. - InodeBitmap() uint64 - - // ExclusionBitmap returns the absolute block number of the snapshot exclusion - // bitmap. - ExclusionBitmap() uint64 - - // FreeBlocksCount returns the number of free blocks in the group. - FreeBlocksCount() uint32 - - // FreeInodesCount returns the number of free inodes in the group. - FreeInodesCount() uint32 - - // DirectoryCount returns the number of inodes that represent directories - // under this block group. - DirectoryCount() uint32 - - // UnusedInodeCount returns the number of unused inodes beyond the last used - // inode in this group's inode table. As a result, we needn’t scan past the - // (InodesPerGroup - UnusedInodeCount())th entry in the inode table. - UnusedInodeCount() uint32 - - // BlockBitmapChecksum returns the block bitmap checksum. This is calculated - // using crc32c(FS UUID + group number + entire bitmap). - BlockBitmapChecksum() uint32 - - // InodeBitmapChecksum returns the inode bitmap checksum. This is calculated - // using crc32c(FS UUID + group number + entire bitmap). - InodeBitmapChecksum() uint32 - - // Checksum returns this block group's checksum. - // - // If SbMetadataCsum feature is set: - // - checksum is crc32c(FS UUID + group number + group descriptor - // structure) & 0xFFFF. - // - // If SbGdtCsum feature is set: - // - checksum is crc16(FS UUID + group number + group descriptor - // structure). - // - // SbMetadataCsum and SbGdtCsum should not be both set. - // If they are, Linux warns and asks to run fsck. - Checksum() uint16 - - // Flags returns BGFlags which represents the block group flags. - Flags() BGFlags -} - -// These are the different block group flags. -const ( - // BgInodeUninit indicates that inode table and bitmap are not initialized. - BgInodeUninit uint16 = 0x1 - - // BgBlockUninit indicates that block bitmap is not initialized. - BgBlockUninit uint16 = 0x2 - - // BgInodeZeroed indicates that inode table is zeroed. - BgInodeZeroed uint16 = 0x4 -) - -// BGFlags represents all the different combinations of block group flags. -type BGFlags struct { - InodeUninit bool - BlockUninit bool - InodeZeroed bool -} - -// ToInt converts a BGFlags struct back to its 16-bit representation. -func (f BGFlags) ToInt() uint16 { - var res uint16 - - if f.InodeUninit { - res |= BgInodeUninit - } - if f.BlockUninit { - res |= BgBlockUninit - } - if f.InodeZeroed { - res |= BgInodeZeroed - } - - return res -} - -// BGFlagsFromInt converts the 16-bit flag representation to a BGFlags struct. -func BGFlagsFromInt(flags uint16) BGFlags { - return BGFlags{ - InodeUninit: flags&BgInodeUninit > 0, - BlockUninit: flags&BgBlockUninit > 0, - InodeZeroed: flags&BgInodeZeroed > 0, - } -} diff --git a/pkg/sentry/fs/ext/disklayout/block_group_32.go b/pkg/sentry/fs/ext/disklayout/block_group_32.go deleted file mode 100644 index 3e16c76db..000000000 --- a/pkg/sentry/fs/ext/disklayout/block_group_32.go +++ /dev/null @@ -1,72 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -// BlockGroup32Bit emulates the first half of struct ext4_group_desc in -// fs/ext4/ext4.h. It is the block group descriptor struct for ext2, ext3 and -// 32-bit ext4 filesystems. It implements BlockGroup interface. -type BlockGroup32Bit struct { - BlockBitmapLo uint32 - InodeBitmapLo uint32 - InodeTableLo uint32 - FreeBlocksCountLo uint16 - FreeInodesCountLo uint16 - UsedDirsCountLo uint16 - FlagsRaw uint16 - ExcludeBitmapLo uint32 - BlockBitmapChecksumLo uint16 - InodeBitmapChecksumLo uint16 - ItableUnusedLo uint16 - ChecksumRaw uint16 -} - -// Compiles only if BlockGroup32Bit implements BlockGroup. -var _ BlockGroup = (*BlockGroup32Bit)(nil) - -// InodeTable implements BlockGroup.InodeTable. -func (bg *BlockGroup32Bit) InodeTable() uint64 { return uint64(bg.InodeTableLo) } - -// BlockBitmap implements BlockGroup.BlockBitmap. -func (bg *BlockGroup32Bit) BlockBitmap() uint64 { return uint64(bg.BlockBitmapLo) } - -// InodeBitmap implements BlockGroup.InodeBitmap. -func (bg *BlockGroup32Bit) InodeBitmap() uint64 { return uint64(bg.InodeBitmapLo) } - -// ExclusionBitmap implements BlockGroup.ExclusionBitmap. -func (bg *BlockGroup32Bit) ExclusionBitmap() uint64 { return uint64(bg.ExcludeBitmapLo) } - -// FreeBlocksCount implements BlockGroup.FreeBlocksCount. -func (bg *BlockGroup32Bit) FreeBlocksCount() uint32 { return uint32(bg.FreeBlocksCountLo) } - -// FreeInodesCount implements BlockGroup.FreeInodesCount. -func (bg *BlockGroup32Bit) FreeInodesCount() uint32 { return uint32(bg.FreeInodesCountLo) } - -// DirectoryCount implements BlockGroup.DirectoryCount. -func (bg *BlockGroup32Bit) DirectoryCount() uint32 { return uint32(bg.UsedDirsCountLo) } - -// UnusedInodeCount implements BlockGroup.UnusedInodeCount. -func (bg *BlockGroup32Bit) UnusedInodeCount() uint32 { return uint32(bg.ItableUnusedLo) } - -// BlockBitmapChecksum implements BlockGroup.BlockBitmapChecksum. -func (bg *BlockGroup32Bit) BlockBitmapChecksum() uint32 { return uint32(bg.BlockBitmapChecksumLo) } - -// InodeBitmapChecksum implements BlockGroup.InodeBitmapChecksum. -func (bg *BlockGroup32Bit) InodeBitmapChecksum() uint32 { return uint32(bg.InodeBitmapChecksumLo) } - -// Checksum implements BlockGroup.Checksum. -func (bg *BlockGroup32Bit) Checksum() uint16 { return bg.ChecksumRaw } - -// Flags implements BlockGroup.Flags. -func (bg *BlockGroup32Bit) Flags() BGFlags { return BGFlagsFromInt(bg.FlagsRaw) } diff --git a/pkg/sentry/fs/ext/disklayout/block_group_64.go b/pkg/sentry/fs/ext/disklayout/block_group_64.go deleted file mode 100644 index 9a809197a..000000000 --- a/pkg/sentry/fs/ext/disklayout/block_group_64.go +++ /dev/null @@ -1,93 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -// BlockGroup64Bit emulates struct ext4_group_desc in fs/ext4/ext4.h. -// It is the block group descriptor struct for 64-bit ext4 filesystems. -// It implements BlockGroup interface. It is an extension of the 32-bit -// version of BlockGroup. -type BlockGroup64Bit struct { - // We embed the 32-bit struct here because 64-bit version is just an extension - // of the 32-bit version. - BlockGroup32Bit - - // 64-bit specific fields. - BlockBitmapHi uint32 - InodeBitmapHi uint32 - InodeTableHi uint32 - FreeBlocksCountHi uint16 - FreeInodesCountHi uint16 - UsedDirsCountHi uint16 - ItableUnusedHi uint16 - ExcludeBitmapHi uint32 - BlockBitmapChecksumHi uint16 - InodeBitmapChecksumHi uint16 - _ uint32 // Padding to 64 bytes. -} - -// Compiles only if BlockGroup64Bit implements BlockGroup. -var _ BlockGroup = (*BlockGroup64Bit)(nil) - -// Methods to override. Checksum() and Flags() are not overridden. - -// InodeTable implements BlockGroup.InodeTable. -func (bg *BlockGroup64Bit) InodeTable() uint64 { - return (uint64(bg.InodeTableHi) << 32) | uint64(bg.InodeTableLo) -} - -// BlockBitmap implements BlockGroup.BlockBitmap. -func (bg *BlockGroup64Bit) BlockBitmap() uint64 { - return (uint64(bg.BlockBitmapHi) << 32) | uint64(bg.BlockBitmapLo) -} - -// InodeBitmap implements BlockGroup.InodeBitmap. -func (bg *BlockGroup64Bit) InodeBitmap() uint64 { - return (uint64(bg.InodeBitmapHi) << 32) | uint64(bg.InodeBitmapLo) -} - -// ExclusionBitmap implements BlockGroup.ExclusionBitmap. -func (bg *BlockGroup64Bit) ExclusionBitmap() uint64 { - return (uint64(bg.ExcludeBitmapHi) << 32) | uint64(bg.ExcludeBitmapLo) -} - -// FreeBlocksCount implements BlockGroup.FreeBlocksCount. -func (bg *BlockGroup64Bit) FreeBlocksCount() uint32 { - return (uint32(bg.FreeBlocksCountHi) << 16) | uint32(bg.FreeBlocksCountLo) -} - -// FreeInodesCount implements BlockGroup.FreeInodesCount. -func (bg *BlockGroup64Bit) FreeInodesCount() uint32 { - return (uint32(bg.FreeInodesCountHi) << 16) | uint32(bg.FreeInodesCountLo) -} - -// DirectoryCount implements BlockGroup.DirectoryCount. -func (bg *BlockGroup64Bit) DirectoryCount() uint32 { - return (uint32(bg.UsedDirsCountHi) << 16) | uint32(bg.UsedDirsCountLo) -} - -// UnusedInodeCount implements BlockGroup.UnusedInodeCount. -func (bg *BlockGroup64Bit) UnusedInodeCount() uint32 { - return (uint32(bg.ItableUnusedHi) << 16) | uint32(bg.ItableUnusedLo) -} - -// BlockBitmapChecksum implements BlockGroup.BlockBitmapChecksum. -func (bg *BlockGroup64Bit) BlockBitmapChecksum() uint32 { - return (uint32(bg.BlockBitmapChecksumHi) << 16) | uint32(bg.BlockBitmapChecksumLo) -} - -// InodeBitmapChecksum implements BlockGroup.InodeBitmapChecksum. -func (bg *BlockGroup64Bit) InodeBitmapChecksum() uint32 { - return (uint32(bg.InodeBitmapChecksumHi) << 16) | uint32(bg.InodeBitmapChecksumLo) -} diff --git a/pkg/sentry/fs/ext/disklayout/block_group_test.go b/pkg/sentry/fs/ext/disklayout/block_group_test.go deleted file mode 100644 index 0ef4294c0..000000000 --- a/pkg/sentry/fs/ext/disklayout/block_group_test.go +++ /dev/null @@ -1,26 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "testing" -) - -// TestBlockGroupSize tests that the block group descriptor structs are of the -// correct size. -func TestBlockGroupSize(t *testing.T) { - assertSize(t, BlockGroup32Bit{}, 32) - assertSize(t, BlockGroup64Bit{}, 64) -} diff --git a/pkg/sentry/fs/ext/disklayout/dirent.go b/pkg/sentry/fs/ext/disklayout/dirent.go deleted file mode 100644 index 685bf57b8..000000000 --- a/pkg/sentry/fs/ext/disklayout/dirent.go +++ /dev/null @@ -1,69 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "gvisor.dev/gvisor/pkg/sentry/fs" -) - -const ( - // MaxFileName is the maximum length of an ext fs file's name. - MaxFileName = 255 -) - -var ( - // inodeTypeByFileType maps ext4 file types to vfs inode types. - // - // See https://www.kernel.org/doc/html/latest/filesystems/ext4/dynamic.html#ftype. - inodeTypeByFileType = map[uint8]fs.InodeType{ - 0: fs.Anonymous, - 1: fs.RegularFile, - 2: fs.Directory, - 3: fs.CharacterDevice, - 4: fs.BlockDevice, - 5: fs.Pipe, - 6: fs.Socket, - 7: fs.Symlink, - } -) - -// The Dirent interface should be implemented by structs representing ext -// directory entries. These are for the linear classical directories which -// just store a list of dirent structs. A directory is a series of data blocks -// where is each data block contains a linear array of dirents. The last entry -// of the block has a record size that takes it to the end of the block. The -// end of the directory is when you read dirInode.Size() bytes from the blocks. -// -// See https://www.kernel.org/doc/html/latest/filesystems/ext4/dynamic.html#linear-classic-directories. -type Dirent interface { - // Inode returns the absolute inode number of the underlying inode. - // Inode number 0 signifies an unused dirent. - Inode() uint32 - - // RecordSize returns the record length of this dirent on disk. The next - // dirent in the dirent list should be read after these many bytes from - // the current dirent. Must be a multiple of 4. - RecordSize() uint16 - - // FileName returns the name of the file. Can be at most 255 is length. - FileName() string - - // FileType returns the inode type of the underlying inode. This is a - // performance hack so that we do not have to read the underlying inode struct - // to know the type of inode. This will only work when the SbDirentFileType - // feature is set. If not, the second returned value will be false indicating - // that user code has to use the inode mode to extract the file type. - FileType() (fs.InodeType, bool) -} diff --git a/pkg/sentry/fs/ext/disklayout/dirent_new.go b/pkg/sentry/fs/ext/disklayout/dirent_new.go deleted file mode 100644 index 29ae4a5c2..000000000 --- a/pkg/sentry/fs/ext/disklayout/dirent_new.go +++ /dev/null @@ -1,61 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "fmt" - - "gvisor.dev/gvisor/pkg/sentry/fs" -) - -// DirentNew represents the ext4 directory entry struct. This emulates Linux's -// ext4_dir_entry_2 struct. The FileName can not be more than 255 bytes so we -// only need 8 bits to store the NameLength. As a result, NameLength has been -// shortened and the other 8 bits are used to encode the file type. Use the -// FileTypeRaw field only if the SbDirentFileType feature is set. -// -// Note: This struct can be of variable size on disk. The one described below -// is of maximum size and the FileName beyond NameLength bytes might contain -// garbage. -type DirentNew struct { - InodeNumber uint32 - RecordLength uint16 - NameLength uint8 - FileTypeRaw uint8 - FileNameRaw [MaxFileName]byte -} - -// Compiles only if DirentNew implements Dirent. -var _ Dirent = (*DirentNew)(nil) - -// Inode implements Dirent.Inode. -func (d *DirentNew) Inode() uint32 { return d.InodeNumber } - -// RecordSize implements Dirent.RecordSize. -func (d *DirentNew) RecordSize() uint16 { return d.RecordLength } - -// FileName implements Dirent.FileName. -func (d *DirentNew) FileName() string { - return string(d.FileNameRaw[:d.NameLength]) -} - -// FileType implements Dirent.FileType. -func (d *DirentNew) FileType() (fs.InodeType, bool) { - if inodeType, ok := inodeTypeByFileType[d.FileTypeRaw]; ok { - return inodeType, true - } - - panic(fmt.Sprintf("unknown file type %v", d.FileTypeRaw)) -} diff --git a/pkg/sentry/fs/ext/disklayout/dirent_old.go b/pkg/sentry/fs/ext/disklayout/dirent_old.go deleted file mode 100644 index 6fff12a6e..000000000 --- a/pkg/sentry/fs/ext/disklayout/dirent_old.go +++ /dev/null @@ -1,49 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import "gvisor.dev/gvisor/pkg/sentry/fs" - -// DirentOld represents the old directory entry struct which does not contain -// the file type. This emulates Linux's ext4_dir_entry struct. -// -// Note: This struct can be of variable size on disk. The one described below -// is of maximum size and the FileName beyond NameLength bytes might contain -// garbage. -type DirentOld struct { - InodeNumber uint32 - RecordLength uint16 - NameLength uint16 - FileNameRaw [MaxFileName]byte -} - -// Compiles only if DirentOld implements Dirent. -var _ Dirent = (*DirentOld)(nil) - -// Inode implements Dirent.Inode. -func (d *DirentOld) Inode() uint32 { return d.InodeNumber } - -// RecordSize implements Dirent.RecordSize. -func (d *DirentOld) RecordSize() uint16 { return d.RecordLength } - -// FileName implements Dirent.FileName. -func (d *DirentOld) FileName() string { - return string(d.FileNameRaw[:d.NameLength]) -} - -// FileType implements Dirent.FileType. -func (d *DirentOld) FileType() (fs.InodeType, bool) { - return fs.Anonymous, false -} diff --git a/pkg/sentry/fs/ext/disklayout/dirent_test.go b/pkg/sentry/fs/ext/disklayout/dirent_test.go deleted file mode 100644 index cc6dff2c9..000000000 --- a/pkg/sentry/fs/ext/disklayout/dirent_test.go +++ /dev/null @@ -1,28 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "testing" -) - -// TestDirentSize tests that the dirent structs are of the correct -// size. -func TestDirentSize(t *testing.T) { - want := uintptr(263) - - assertSize(t, DirentOld{}, want) - assertSize(t, DirentNew{}, want) -} diff --git a/pkg/sentry/fs/ext/disklayout/disklayout.go b/pkg/sentry/fs/ext/disklayout/disklayout.go deleted file mode 100644 index bdf4e2132..000000000 --- a/pkg/sentry/fs/ext/disklayout/disklayout.go +++ /dev/null @@ -1,50 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package disklayout provides Linux ext file system's disk level structures -// which can be directly read into from the underlying device. Structs aim to -// emulate structures `exactly` how they are layed out on disk. -// -// This library aims to be compatible with all ext(2/3/4) systems so it -// provides a generic interface for all major structures and various -// implementations (for different versions). The user code is responsible for -// using appropriate implementations based on the underlying device. -// -// Interfacing all major structures here serves a few purposes: -// - Abstracts away the complexity of the underlying structure from client -// code. The client only has to figure out versioning on set up and then -// can use these as black boxes and pass it higher up the stack. -// - Having pointer receivers forces the user to use pointers to these -// heavy structs. Hence, prevents the client code from unintentionally -// copying these by value while passing the interface around. -// - Version-based implementation selection is resolved on set up hence -// avoiding per call overhead of choosing implementation. -// - All interface methods are pretty light weight (do not take in any -// parameters by design). Passing pointer arguments to interface methods -// can lead to heap allocation as the compiler won't be able to perform -// escape analysis on an unknown implementation at compile time. -// -// Notes: -// - All fields in these structs are exported because binary.Read would -// panic otherwise. -// - All structures on disk are in little-endian order. Only jbd2 (journal) -// structures are in big-endian order. -// - All OS dependent fields in these structures will be interpretted using -// the Linux version of that field. -// - The suffix `Lo` in field names stands for lower bits of that field. -// - The suffix `Hi` in field names stands for upper bits of that field. -// - The suffix `Raw` has been added to indicate that the field is not split -// into Lo and Hi fields and also to resolve name collision with the -// respective interface. -package disklayout diff --git a/pkg/sentry/fs/ext/disklayout/extent.go b/pkg/sentry/fs/ext/disklayout/extent.go deleted file mode 100644 index 567523d32..000000000 --- a/pkg/sentry/fs/ext/disklayout/extent.go +++ /dev/null @@ -1,139 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -// Extents were introduced in ext4 and provide huge performance gains in terms -// data locality and reduced metadata block usage. Extents are organized in -// extent trees. The root node is contained in inode.BlocksRaw. -// -// Terminology: -// - Physical Block: -// Filesystem data block which is addressed normally wrt the entire -// filesystem (addressed with 48 bits). -// -// - File Block: -// Data block containing *only* file data and addressed wrt to the file -// with only 32 bits. The (i)th file block contains file data from -// byte (i * sb.BlockSize()) to ((i+1) * sb.BlockSize()). - -const ( - // ExtentStructsSize is the size of all the three extent on-disk structs. - ExtentStructsSize = 12 - - // ExtentMagic is the magic number which must be present in the header. - ExtentMagic = 0xf30a -) - -// ExtentEntryPair couples an in-memory ExtendNode with the ExtentEntry that -// points to it. We want to cache these structs in memory to avoid repeated -// disk reads. -// -// Note: This struct itself does not represent an on-disk struct. -type ExtentEntryPair struct { - // Entry points to the child node on disk. - Entry ExtentEntry - // Node points to child node in memory. Is nil if the current node is a leaf. - Node *ExtentNode -} - -// ExtentNode represents an extent tree node. For internal nodes, all Entries -// will be ExtendIdxs. For leaf nodes, they will all be Extents. -// -// Note: This struct itself does not represent an on-disk struct. -type ExtentNode struct { - Header ExtentHeader - Entries []ExtentEntryPair -} - -// ExtentEntry reprsents an extent tree node entry. The entry can either be -// an ExtentIdx or Extent itself. This exists to simplify navigation logic. -type ExtentEntry interface { - // FileBlock returns the first file block number covered by this entry. - FileBlock() uint32 - - // PhysicalBlock returns the child physical block that this entry points to. - PhysicalBlock() uint64 -} - -// ExtentHeader emulates the ext4_extent_header struct in ext4. Each extent -// tree node begins with this and is followed by `NumEntries` number of: -// - Extent if `Depth` == 0 -// - ExtentIdx otherwise -type ExtentHeader struct { - // Magic in the extent magic number, must be 0xf30a. - Magic uint16 - - // NumEntries indicates the number of valid entries following the header. - NumEntries uint16 - - // MaxEntries that could follow the header. Used while adding entries. - MaxEntries uint16 - - // Height represents the distance of this node from the farthest leaf. Please - // note that Linux incorrectly calls this `Depth` (which means the distance - // of the node from the root). - Height uint16 - _ uint32 -} - -// ExtentIdx emulates the ext4_extent_idx struct in ext4. Only present in -// internal nodes. Sorted in ascending order based on FirstFileBlock since -// Linux does a binary search on this. This points to a block containing the -// child node. -type ExtentIdx struct { - FirstFileBlock uint32 - ChildBlockLo uint32 - ChildBlockHi uint16 - _ uint16 -} - -// Compiles only if ExtentIdx implements ExtentEntry. -var _ ExtentEntry = (*ExtentIdx)(nil) - -// FileBlock implements ExtentEntry.FileBlock. -func (ei *ExtentIdx) FileBlock() uint32 { - return ei.FirstFileBlock -} - -// PhysicalBlock implements ExtentEntry.PhysicalBlock. It returns the -// physical block number of the child block. -func (ei *ExtentIdx) PhysicalBlock() uint64 { - return (uint64(ei.ChildBlockHi) << 32) | uint64(ei.ChildBlockLo) -} - -// Extent represents the ext4_extent struct in ext4. Only present in leaf -// nodes. Sorted in ascending order based on FirstFileBlock since Linux does a -// binary search on this. This points to an array of data blocks containing the -// file data. It covers `Length` data blocks starting from `StartBlock`. -type Extent struct { - FirstFileBlock uint32 - Length uint16 - StartBlockHi uint16 - StartBlockLo uint32 -} - -// Compiles only if Extent implements ExtentEntry. -var _ ExtentEntry = (*Extent)(nil) - -// FileBlock implements ExtentEntry.FileBlock. -func (e *Extent) FileBlock() uint32 { - return e.FirstFileBlock -} - -// PhysicalBlock implements ExtentEntry.PhysicalBlock. It returns the -// physical block number of the first data block this extent covers. -func (e *Extent) PhysicalBlock() uint64 { - return (uint64(e.StartBlockHi) << 32) | uint64(e.StartBlockLo) -} diff --git a/pkg/sentry/fs/ext/disklayout/extent_test.go b/pkg/sentry/fs/ext/disklayout/extent_test.go deleted file mode 100644 index b0fad9b71..000000000 --- a/pkg/sentry/fs/ext/disklayout/extent_test.go +++ /dev/null @@ -1,27 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "testing" -) - -// TestExtentSize tests that the extent structs are of the correct -// size. -func TestExtentSize(t *testing.T) { - assertSize(t, ExtentHeader{}, ExtentStructsSize) - assertSize(t, ExtentIdx{}, ExtentStructsSize) - assertSize(t, Extent{}, ExtentStructsSize) -} diff --git a/pkg/sentry/fs/ext/disklayout/inode.go b/pkg/sentry/fs/ext/disklayout/inode.go deleted file mode 100644 index 88ae913f5..000000000 --- a/pkg/sentry/fs/ext/disklayout/inode.go +++ /dev/null @@ -1,274 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/time" -) - -// Special inodes. See https://www.kernel.org/doc/html/latest/filesystems/ext4/overview.html#special-inodes. -const ( - // RootDirInode is the inode number of the root directory inode. - RootDirInode = 2 -) - -// The Inode interface must be implemented by structs representing ext inodes. -// The inode stores all the metadata pertaining to the file (except for the -// file name which is held by the directory entry). It does NOT expose all -// fields and should be extended if need be. -// -// Some file systems (e.g. FAT) use the directory entry to store all this -// information. Ext file systems do not so that they can support hard links. -// However, ext4 cheats a little bit and duplicates the file type in the -// directory entry for performance gains. -// -// See https://www.kernel.org/doc/html/latest/filesystems/ext4/dynamic.html#index-nodes. -type Inode interface { - // Mode returns the linux file mode which is majorly used to extract - // information like: - // - File permissions (read/write/execute by user/group/others). - // - Sticky, set UID and GID bits. - // - File type. - // - // Masks to extract this information are provided in pkg/abi/linux/file.go. - Mode() linux.FileMode - - // UID returns the owner UID. - UID() auth.KUID - - // GID returns the owner GID. - GID() auth.KGID - - // Size returns the size of the file in bytes. - Size() uint64 - - // InodeSize returns the size of this inode struct in bytes. - // In ext2 and ext3, the inode struct and inode disk record size was fixed at - // 128 bytes. Ext4 makes it possible for the inode struct to be bigger. - // However, accessing any field beyond the 128 bytes marker must be verified - // using this method. - InodeSize() uint16 - - // AccessTime returns the last access time. Shows when the file was last read. - // - // If InExtendedAttr is set, then this should NOT be used because the - // underlying field is used to store the extended attribute value checksum. - AccessTime() time.Time - - // ChangeTime returns the last change time. Shows when the file meta data - // (like permissions) was last changed. - // - // If InExtendedAttr is set, then this should NOT be used because the - // underlying field is used to store the lower 32 bits of the attribute - // value’s reference count. - ChangeTime() time.Time - - // ModificationTime returns the last modification time. Shows when the file - // content was last modified. - // - // If InExtendedAttr is set, then this should NOT be used because - // the underlying field contains the number of the inode that owns the - // extended attribute. - ModificationTime() time.Time - - // DeletionTime returns the deletion time. Inodes are marked as deleted by - // writing to the underlying field. FS tools can restore files until they are - // actually overwritten. - DeletionTime() time.Time - - // LinksCount returns the number of hard links to this inode. - // - // Normally there is an upper limit on the number of hard links: - // - ext2/ext3 = 32,000 - // - ext4 = 65,000 - // - // This implies that an ext4 directory cannot have more than 64,998 - // subdirectories because each subdirectory will have a hard link to the - // directory via the `..` entry. The directory has hard link via the `.` entry - // of its own. And finally the inode is initiated with 1 hard link (itself). - // - // The underlying value is reset to 1 if all the following hold: - // - Inode is a directory. - // - SbDirNlink is enabled. - // - Number of hard links is incremented past 64,999. - // Hard link value of 1 for a directory would indicate that the number of hard - // links is unknown because a directory can have minimum 2 hard links (itself - // and `.` entry). - LinksCount() uint16 - - // Flags returns InodeFlags which represents the inode flags. - Flags() InodeFlags - - // Data returns the underlying inode.i_block array as a slice so it's - // modifiable. This field is special and is used to store various kinds of - // things depending on the filesystem version and inode type. The underlying - // field name in Linux is a little misleading. - // - In ext2/ext3, it contains the block map. - // - In ext4, it contains the extent tree root node. - // - For inline files, it contains the file contents. - // - For symlinks, it contains the link path (if it fits here). - // - // See https://www.kernel.org/doc/html/latest/filesystems/ext4/dynamic.html#the-contents-of-inode-i-block. - Data() []byte -} - -// Inode flags. This is not comprehensive and flags which were not used in -// the Linux kernel have been excluded. -const ( - // InSync indicates that all writes to the file must be synchronous. - InSync = 0x8 - - // InImmutable indicates that this file is immutable. - InImmutable = 0x10 - - // InAppend indicates that this file can only be appended to. - InAppend = 0x20 - - // InNoDump indicates that teh dump(1) utility should not dump this file. - InNoDump = 0x40 - - // InNoAccessTime indicates that the access time of this inode must not be - // updated. - InNoAccessTime = 0x80 - - // InIndex indicates that this directory has hashed indexes. - InIndex = 0x1000 - - // InJournalData indicates that file data must always be written through a - // journal device. - InJournalData = 0x4000 - - // InDirSync indicates that all the directory entiry data must be written - // synchronously. - InDirSync = 0x10000 - - // InTopDir indicates that this inode is at the top of the directory hierarchy. - InTopDir = 0x20000 - - // InHugeFile indicates that this is a huge file. - InHugeFile = 0x40000 - - // InExtents indicates that this inode uses extents. - InExtents = 0x80000 - - // InExtendedAttr indicates that this inode stores a large extended attribute - // value in its data blocks. - InExtendedAttr = 0x200000 - - // InInline indicates that this inode has inline data. - InInline = 0x10000000 - - // InReserved indicates that this inode is reserved for the ext4 library. - InReserved = 0x80000000 -) - -// InodeFlags represents all possible combinations of inode flags. It aims to -// cover the bit masks and provide a more user-friendly interface. -type InodeFlags struct { - Sync bool - Immutable bool - Append bool - NoDump bool - NoAccessTime bool - Index bool - JournalData bool - DirSync bool - TopDir bool - HugeFile bool - Extents bool - ExtendedAttr bool - Inline bool - Reserved bool -} - -// ToInt converts inode flags back to its 32-bit rep. -func (f InodeFlags) ToInt() uint32 { - var res uint32 - - if f.Sync { - res |= InSync - } - if f.Immutable { - res |= InImmutable - } - if f.Append { - res |= InAppend - } - if f.NoDump { - res |= InNoDump - } - if f.NoAccessTime { - res |= InNoAccessTime - } - if f.Index { - res |= InIndex - } - if f.JournalData { - res |= InJournalData - } - if f.DirSync { - res |= InDirSync - } - if f.TopDir { - res |= InTopDir - } - if f.HugeFile { - res |= InHugeFile - } - if f.Extents { - res |= InExtents - } - if f.ExtendedAttr { - res |= InExtendedAttr - } - if f.Inline { - res |= InInline - } - if f.Reserved { - res |= InReserved - } - - return res -} - -// InodeFlagsFromInt converts the integer representation of inode flags to -// a InodeFlags struct. -func InodeFlagsFromInt(f uint32) InodeFlags { - return InodeFlags{ - Sync: f&InSync > 0, - Immutable: f&InImmutable > 0, - Append: f&InAppend > 0, - NoDump: f&InNoDump > 0, - NoAccessTime: f&InNoAccessTime > 0, - Index: f&InIndex > 0, - JournalData: f&InJournalData > 0, - DirSync: f&InDirSync > 0, - TopDir: f&InTopDir > 0, - HugeFile: f&InHugeFile > 0, - Extents: f&InExtents > 0, - ExtendedAttr: f&InExtendedAttr > 0, - Inline: f&InInline > 0, - Reserved: f&InReserved > 0, - } -} - -// These masks define how users can view/modify inode flags. The rest of the -// flags are for internal kernel usage only. -const ( - InUserReadFlagMask = 0x4BDFFF - InUserWriteFlagMask = 0x4B80FF -) diff --git a/pkg/sentry/fs/ext/disklayout/inode_new.go b/pkg/sentry/fs/ext/disklayout/inode_new.go deleted file mode 100644 index 8f9f574ce..000000000 --- a/pkg/sentry/fs/ext/disklayout/inode_new.go +++ /dev/null @@ -1,96 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import "gvisor.dev/gvisor/pkg/sentry/kernel/time" - -// InodeNew represents ext4 inode structure which can be bigger than -// OldInodeSize. The actual size of this struct should be determined using -// inode.ExtraInodeSize. Accessing any field here should be verified with the -// actual size. The extra space between the end of the inode struct and end of -// the inode record can be used to store extended attr. -// -// If the TimeExtra fields are in scope, the lower 2 bits of those are used -// to extend their counter part to be 34 bits wide; the rest (upper) 30 bits -// are used to provide nanoscond precision. Hence, these timestamps will now -// overflow in May 2446. -// See https://www.kernel.org/doc/html/latest/filesystems/ext4/dynamic.html#inode-timestamps. -type InodeNew struct { - InodeOld - - ExtraInodeSize uint16 - ChecksumHi uint16 - ChangeTimeExtra uint32 - ModificationTimeExtra uint32 - AccessTimeExtra uint32 - CreationTime uint32 - CreationTimeExtra uint32 - VersionHi uint32 - ProjectID uint32 -} - -// Compiles only if InodeNew implements Inode. -var _ Inode = (*InodeNew)(nil) - -// fromExtraTime decodes the extra time and constructs the kernel time struct -// with nanosecond precision. -func fromExtraTime(lo int32, extra uint32) time.Time { - // See description above InodeNew for format. - seconds := (int64(extra&0x3) << 32) + int64(lo) - nanoseconds := int64(extra >> 2) - return time.FromUnix(seconds, nanoseconds) -} - -// Only override methods which change due to ext4 specific fields. - -// Size implements Inode.Size. -func (in *InodeNew) Size() uint64 { - return (uint64(in.SizeHi) << 32) | uint64(in.SizeLo) -} - -// InodeSize implements Inode.InodeSize. -func (in *InodeNew) InodeSize() uint16 { - return OldInodeSize + in.ExtraInodeSize -} - -// ChangeTime implements Inode.ChangeTime. -func (in *InodeNew) ChangeTime() time.Time { - // Apply new timestamp logic if inode.ChangeTimeExtra is in scope. - if in.ExtraInodeSize >= 8 { - return fromExtraTime(in.ChangeTimeRaw, in.ChangeTimeExtra) - } - - return in.InodeOld.ChangeTime() -} - -// ModificationTime implements Inode.ModificationTime. -func (in *InodeNew) ModificationTime() time.Time { - // Apply new timestamp logic if inode.ModificationTimeExtra is in scope. - if in.ExtraInodeSize >= 12 { - return fromExtraTime(in.ModificationTimeRaw, in.ModificationTimeExtra) - } - - return in.InodeOld.ModificationTime() -} - -// AccessTime implements Inode.AccessTime. -func (in *InodeNew) AccessTime() time.Time { - // Apply new timestamp logic if inode.AccessTimeExtra is in scope. - if in.ExtraInodeSize >= 16 { - return fromExtraTime(in.AccessTimeRaw, in.AccessTimeExtra) - } - - return in.InodeOld.AccessTime() -} diff --git a/pkg/sentry/fs/ext/disklayout/inode_old.go b/pkg/sentry/fs/ext/disklayout/inode_old.go deleted file mode 100644 index db25b11b6..000000000 --- a/pkg/sentry/fs/ext/disklayout/inode_old.go +++ /dev/null @@ -1,117 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/time" -) - -const ( - // OldInodeSize is the inode size in ext2/ext3. - OldInodeSize = 128 -) - -// InodeOld implements Inode interface. It emulates ext2/ext3 inode struct. -// Inode struct size and record size are both 128 bytes for this. -// -// All fields representing time are in seconds since the epoch. Which means that -// they will overflow in January 2038. -type InodeOld struct { - ModeRaw uint16 - UIDLo uint16 - SizeLo uint32 - - // The time fields are signed integers because they could be negative to - // represent time before the epoch. - AccessTimeRaw int32 - ChangeTimeRaw int32 - ModificationTimeRaw int32 - DeletionTimeRaw int32 - - GIDLo uint16 - LinksCountRaw uint16 - BlocksCountLo uint32 - FlagsRaw uint32 - VersionLo uint32 // This is OS dependent. - DataRaw [60]byte - Generation uint32 - FileACLLo uint32 - SizeHi uint32 - ObsoFaddr uint32 - - // OS dependent fields have been inlined here. - BlocksCountHi uint16 - FileACLHi uint16 - UIDHi uint16 - GIDHi uint16 - ChecksumLo uint16 - _ uint16 -} - -// Compiles only if InodeOld implements Inode. -var _ Inode = (*InodeOld)(nil) - -// Mode implements Inode.Mode. -func (in *InodeOld) Mode() linux.FileMode { return linux.FileMode(in.ModeRaw) } - -// UID implements Inode.UID. -func (in *InodeOld) UID() auth.KUID { - return auth.KUID((uint32(in.UIDHi) << 16) | uint32(in.UIDLo)) -} - -// GID implements Inode.GID. -func (in *InodeOld) GID() auth.KGID { - return auth.KGID((uint32(in.GIDHi) << 16) | uint32(in.GIDLo)) -} - -// Size implements Inode.Size. -func (in *InodeOld) Size() uint64 { - // In ext2/ext3, in.SizeHi did not exist, it was instead named in.DirACL. - return uint64(in.SizeLo) -} - -// InodeSize implements Inode.InodeSize. -func (in *InodeOld) InodeSize() uint16 { return OldInodeSize } - -// AccessTime implements Inode.AccessTime. -func (in *InodeOld) AccessTime() time.Time { - return time.FromUnix(int64(in.AccessTimeRaw), 0) -} - -// ChangeTime implements Inode.ChangeTime. -func (in *InodeOld) ChangeTime() time.Time { - return time.FromUnix(int64(in.ChangeTimeRaw), 0) -} - -// ModificationTime implements Inode.ModificationTime. -func (in *InodeOld) ModificationTime() time.Time { - return time.FromUnix(int64(in.ModificationTimeRaw), 0) -} - -// DeletionTime implements Inode.DeletionTime. -func (in *InodeOld) DeletionTime() time.Time { - return time.FromUnix(int64(in.DeletionTimeRaw), 0) -} - -// LinksCount implements Inode.LinksCount. -func (in *InodeOld) LinksCount() uint16 { return in.LinksCountRaw } - -// Flags implements Inode.Flags. -func (in *InodeOld) Flags() InodeFlags { return InodeFlagsFromInt(in.FlagsRaw) } - -// Data implements Inode.Data. -func (in *InodeOld) Data() []byte { return in.DataRaw[:] } diff --git a/pkg/sentry/fs/ext/disklayout/inode_test.go b/pkg/sentry/fs/ext/disklayout/inode_test.go deleted file mode 100644 index dd03ee50e..000000000 --- a/pkg/sentry/fs/ext/disklayout/inode_test.go +++ /dev/null @@ -1,222 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "fmt" - "strconv" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/kernel/time" -) - -// TestInodeSize tests that the inode structs are of the correct size. -func TestInodeSize(t *testing.T) { - assertSize(t, InodeOld{}, OldInodeSize) - - // This was updated from 156 bytes to 160 bytes in Oct 2015. - assertSize(t, InodeNew{}, 160) -} - -// TestTimestampSeconds tests that the seconds part of [a/c/m] timestamps in -// ext4 inode structs are decoded correctly. -// -// These tests are derived from the table under https://www.kernel.org/doc/html/latest/filesystems/ext4/dynamic.html#inode-timestamps. -func TestTimestampSeconds(t *testing.T) { - type timestampTest struct { - // msbSet tells if the most significant bit of InodeOld.[X]TimeRaw is set. - // If this is set then the 32-bit time is negative. - msbSet bool - - // lowerBound tells if we should take the lowest possible value of - // InodeOld.[X]TimeRaw while satisfying test.msbSet condition. If set to - // false it tells to take the highest possible value. - lowerBound bool - - // extraBits is InodeNew.[X]TimeExtra. - extraBits uint32 - - // want is the kernel time struct that is expected. - want time.Time - } - - tests := []timestampTest{ - // 1901-12-13 - { - msbSet: true, - lowerBound: true, - extraBits: 0, - want: time.FromUnix(int64(-0x80000000), 0), - }, - - // 1969-12-31 - { - msbSet: true, - lowerBound: false, - extraBits: 0, - want: time.FromUnix(int64(-1), 0), - }, - - // 1970-01-01 - { - msbSet: false, - lowerBound: true, - extraBits: 0, - want: time.FromUnix(int64(0), 0), - }, - - // 2038-01-19 - { - msbSet: false, - lowerBound: false, - extraBits: 0, - want: time.FromUnix(int64(0x7fffffff), 0), - }, - - // 2038-01-19 - { - msbSet: true, - lowerBound: true, - extraBits: 1, - want: time.FromUnix(int64(0x80000000), 0), - }, - - // 2106-02-07 - { - msbSet: true, - lowerBound: false, - extraBits: 1, - want: time.FromUnix(int64(0xffffffff), 0), - }, - - // 2106-02-07 - { - msbSet: false, - lowerBound: true, - extraBits: 1, - want: time.FromUnix(int64(0x100000000), 0), - }, - - // 2174-02-25 - { - msbSet: false, - lowerBound: false, - extraBits: 1, - want: time.FromUnix(int64(0x17fffffff), 0), - }, - - // 2174-02-25 - { - msbSet: true, - lowerBound: true, - extraBits: 2, - want: time.FromUnix(int64(0x180000000), 0), - }, - - // 2242-03-16 - { - msbSet: true, - lowerBound: false, - extraBits: 2, - want: time.FromUnix(int64(0x1ffffffff), 0), - }, - - // 2242-03-16 - { - msbSet: false, - lowerBound: true, - extraBits: 2, - want: time.FromUnix(int64(0x200000000), 0), - }, - - // 2310-04-04 - { - msbSet: false, - lowerBound: false, - extraBits: 2, - want: time.FromUnix(int64(0x27fffffff), 0), - }, - - // 2310-04-04 - { - msbSet: true, - lowerBound: true, - extraBits: 3, - want: time.FromUnix(int64(0x280000000), 0), - }, - - // 2378-04-22 - { - msbSet: true, - lowerBound: false, - extraBits: 3, - want: time.FromUnix(int64(0x2ffffffff), 0), - }, - - // 2378-04-22 - { - msbSet: false, - lowerBound: true, - extraBits: 3, - want: time.FromUnix(int64(0x300000000), 0), - }, - - // 2446-05-10 - { - msbSet: false, - lowerBound: false, - extraBits: 3, - want: time.FromUnix(int64(0x37fffffff), 0), - }, - } - - lowerMSB0 := int32(0) // binary: 00000000 00000000 00000000 00000000 - upperMSB0 := int32(0x7fffffff) // binary: 01111111 11111111 11111111 11111111 - lowerMSB1 := int32(-0x80000000) // binary: 10000000 00000000 00000000 00000000 - upperMSB1 := int32(-1) // binary: 11111111 11111111 11111111 11111111 - - get32BitTime := func(test timestampTest) int32 { - if test.msbSet { - if test.lowerBound { - return lowerMSB1 - } - - return upperMSB1 - } - - if test.lowerBound { - return lowerMSB0 - } - - return upperMSB0 - } - - getTestName := func(test timestampTest) string { - return fmt.Sprintf( - "Tests time decoding with epoch bits 0b%s and 32-bit raw time: MSB set=%t, lower bound=%t", - strconv.FormatInt(int64(test.extraBits), 2), - test.msbSet, - test.lowerBound, - ) - } - - for _, test := range tests { - t.Run(getTestName(test), func(t *testing.T) { - if got := fromExtraTime(get32BitTime(test), test.extraBits); got != test.want { - t.Errorf("Expected: %v, Got: %v", test.want, got) - } - }) - } -} diff --git a/pkg/sentry/fs/ext/disklayout/superblock.go b/pkg/sentry/fs/ext/disklayout/superblock.go deleted file mode 100644 index 7a337a5e0..000000000 --- a/pkg/sentry/fs/ext/disklayout/superblock.go +++ /dev/null @@ -1,471 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -const ( - // SbOffset is the absolute offset at which the superblock is placed. - SbOffset = 1024 -) - -// SuperBlock should be implemented by structs representing the ext superblock. -// The superblock holds a lot of information about the enclosing filesystem. -// This interface aims to provide access methods to important information held -// by the superblock. It does NOT expose all fields of the superblock, only the -// ones necessary. This can be expanded when need be. -// -// Location and replication: -// - The superblock is located at offset 1024 in block group 0. -// - Redundant copies of the superblock and group descriptors are kept in -// all groups if SbSparse feature flag is NOT set. If it is set, the -// replicas only exist in groups whose group number is either 0 or a -// power of 3, 5, or 7. -// - There is also a sparse superblock feature v2 in which there are just -// two replicas saved in the block groups pointed by sb.s_backup_bgs. -// -// Replicas should eventually be updated if the superblock is updated. -// -// See https://www.kernel.org/doc/html/latest/filesystems/ext4/globals.html#super-block. -type SuperBlock interface { - // InodesCount returns the total number of inodes in this filesystem. - InodesCount() uint32 - - // BlocksCount returns the total number of data blocks in this filesystem. - BlocksCount() uint64 - - // FreeBlocksCount returns the number of free blocks in this filesystem. - FreeBlocksCount() uint64 - - // FreeInodesCount returns the number of free inodes in this filesystem. - FreeInodesCount() uint32 - - // MountCount returns the number of mounts since the last fsck. - MountCount() uint16 - - // MaxMountCount returns the number of mounts allowed beyond which a fsck is - // needed. - MaxMountCount() uint16 - - // FirstDataBlock returns the absolute block number of the first data block, - // which contains the super block itself. - // - // If the filesystem has 1kb data blocks then this should return 1. For all - // other configurations, this typically returns 0. - FirstDataBlock() uint32 - - // BlockSize returns the size of one data block in this filesystem. - // This can be calculated by 2^(10 + sb.s_log_block_size). This ensures that - // the smallest block size is 1kb. - BlockSize() uint64 - - // BlocksPerGroup returns the number of data blocks in a block group. - BlocksPerGroup() uint32 - - // ClusterSize returns block cluster size (set during mkfs time by admin). - // This can be calculated by 2^(10 + sb.s_log_cluster_size). This ensures that - // the smallest cluster size is 1kb. - // - // sb.s_log_cluster_size must equal sb.s_log_block_size if bigalloc feature - // is NOT set and consequently BlockSize() = ClusterSize() in that case. - ClusterSize() uint64 - - // ClustersPerGroup returns: - // - number of clusters per group if bigalloc is enabled. - // - BlocksPerGroup() otherwise. - ClustersPerGroup() uint32 - - // InodeSize returns the size of the inode disk record size in bytes. Use this - // to iterate over inode arrays on disk. - // - // In ext2 and ext3: - // - Each inode had a disk record of 128 bytes. - // - The inode struct size was fixed at 128 bytes. - // - // In ext4 its possible to allocate larger on-disk inodes: - // - Inode disk record size = sb.s_inode_size (function return value). - // = 256 (default) - // - Inode struct size = 128 + inode.i_extra_isize. - // = 128 + 32 = 160 (default) - InodeSize() uint16 - - // InodesPerGroup returns the number of inodes in a block group. - InodesPerGroup() uint32 - - // BgDescSize returns the size of the block group descriptor struct. - // - // In ext2, ext3, ext4 (without 64-bit feature), the block group descriptor - // is only 32 bytes long. - // In ext4 with 64-bit feature, the block group descriptor expands to AT LEAST - // 64 bytes. It might be bigger than that. - BgDescSize() uint16 - - // CompatibleFeatures returns the CompatFeatures struct which holds all the - // compatible features this fs supports. - CompatibleFeatures() CompatFeatures - - // IncompatibleFeatures returns the CompatFeatures struct which holds all the - // incompatible features this fs supports. - IncompatibleFeatures() IncompatFeatures - - // ReadOnlyCompatibleFeatures returns the CompatFeatures struct which holds all the - // readonly compatible features this fs supports. - ReadOnlyCompatibleFeatures() RoCompatFeatures - - // Magic() returns the magic signature which must be 0xef53. - Magic() uint16 - - // Revision returns the superblock revision. Superblock struct fields from - // offset 0x54 till 0x150 should only be used if superblock has DynamicRev. - Revision() SbRevision -} - -// SbRevision is the type for superblock revisions. -type SbRevision uint32 - -// Super block revisions. -const ( - // OldRev is the good old (original) format. - OldRev SbRevision = 0 - - // DynamicRev is v2 format w/ dynamic inode sizes. - DynamicRev SbRevision = 1 -) - -// Superblock compatible features. -// This is not exhaustive, unused features are not listed. -const ( - // SbDirPrealloc indicates directory preallocation. - SbDirPrealloc = 0x1 - - // SbHasJournal indicates the presence of a journal. jbd2 should only work - // with this being set. - SbHasJournal = 0x4 - - // SbExtAttr indicates extended attributes support. - SbExtAttr = 0x8 - - // SbResizeInode indicates that the fs has reserved GDT blocks (right after - // group descriptors) for fs expansion. - SbResizeInode = 0x10 - - // SbDirIndex indicates that the fs has directory indices. - SbDirIndex = 0x20 - - // SbSparseV2 stands for Sparse superblock version 2. - SbSparseV2 = 0x200 -) - -// CompatFeatures represents a superblock's compatible feature set. If the -// kernel does not understand any of these feature, it can still read/write -// to this fs. -type CompatFeatures struct { - DirPrealloc bool - HasJournal bool - ExtAttr bool - ResizeInode bool - DirIndex bool - SparseV2 bool -} - -// ToInt converts superblock compatible features back to its 32-bit rep. -func (f CompatFeatures) ToInt() uint32 { - var res uint32 - - if f.DirPrealloc { - res |= SbDirPrealloc - } - if f.HasJournal { - res |= SbHasJournal - } - if f.ExtAttr { - res |= SbExtAttr - } - if f.ResizeInode { - res |= SbResizeInode - } - if f.DirIndex { - res |= SbDirIndex - } - if f.SparseV2 { - res |= SbSparseV2 - } - - return res -} - -// CompatFeaturesFromInt converts the integer representation of superblock -// compatible features to CompatFeatures struct. -func CompatFeaturesFromInt(f uint32) CompatFeatures { - return CompatFeatures{ - DirPrealloc: f&SbDirPrealloc > 0, - HasJournal: f&SbHasJournal > 0, - ExtAttr: f&SbExtAttr > 0, - ResizeInode: f&SbResizeInode > 0, - DirIndex: f&SbDirIndex > 0, - SparseV2: f&SbSparseV2 > 0, - } -} - -// Superblock incompatible features. -// This is not exhaustive, unused features are not listed. -const ( - // SbDirentFileType indicates that directory entries record the file type. - // We should use struct ext4_dir_entry_2 for dirents then. - SbDirentFileType = 0x2 - - // SbRecovery indicates that the filesystem needs recovery. - SbRecovery = 0x4 - - // SbJournalDev indicates that the filesystem has a separate journal device. - SbJournalDev = 0x8 - - // SbMetaBG indicates that the filesystem is using Meta block groups. Moves - // the group descriptors from the congested first block group into the first - // group of each metablock group to increase the maximum block groups limit - // and hence support much larger filesystems. - // - // See https://www.kernel.org/doc/html/latest/filesystems/ext4/overview.html#meta-block-groups. - SbMetaBG = 0x10 - - // SbExtents indicates that the filesystem uses extents. Must be set in ext4 - // filesystems. - SbExtents = 0x40 - - // SbIs64Bit indicates that this filesystem addresses blocks with 64-bits. - // Hence can support 2^64 data blocks. - SbIs64Bit = 0x80 - - // SbMMP indicates that this filesystem has multiple mount protection. - // - // See https://www.kernel.org/doc/html/latest/filesystems/ext4/globals.html#multiple-mount-protection. - SbMMP = 0x100 - - // SbFlexBg indicates that this filesystem has flexible block groups. Several - // block groups are tied into one logical block group so that all the metadata - // for the block groups (bitmaps and inode tables) are close together for - // faster loading. Consequently, large files will be continuous on disk. - // However, this does not affect the placement of redundant superblocks and - // group descriptors. - // - // See https://www.kernel.org/doc/html/latest/filesystems/ext4/overview.html#flexible-block-groups. - SbFlexBg = 0x200 - - // SbLargeDir shows that large directory enabled. Directory htree can be 3 - // levels deep. Directory htrees are allowed to be 2 levels deep otherwise. - SbLargeDir = 0x4000 - - // SbInlineData allows inline data in inodes for really small files. - SbInlineData = 0x8000 - - // SbEncrypted indicates that this fs contains encrypted inodes. - SbEncrypted = 0x10000 -) - -// IncompatFeatures represents a superblock's incompatible feature set. If the -// kernel does not understand any of these feature, it should refuse to mount. -type IncompatFeatures struct { - DirentFileType bool - Recovery bool - JournalDev bool - MetaBG bool - Extents bool - Is64Bit bool - MMP bool - FlexBg bool - LargeDir bool - InlineData bool - Encrypted bool -} - -// ToInt converts superblock incompatible features back to its 32-bit rep. -func (f IncompatFeatures) ToInt() uint32 { - var res uint32 - - if f.DirentFileType { - res |= SbDirentFileType - } - if f.Recovery { - res |= SbRecovery - } - if f.JournalDev { - res |= SbJournalDev - } - if f.MetaBG { - res |= SbMetaBG - } - if f.Extents { - res |= SbExtents - } - if f.Is64Bit { - res |= SbIs64Bit - } - if f.MMP { - res |= SbMMP - } - if f.FlexBg { - res |= SbFlexBg - } - if f.LargeDir { - res |= SbLargeDir - } - if f.InlineData { - res |= SbInlineData - } - if f.Encrypted { - res |= SbEncrypted - } - - return res -} - -// IncompatFeaturesFromInt converts the integer representation of superblock -// incompatible features to IncompatFeatures struct. -func IncompatFeaturesFromInt(f uint32) IncompatFeatures { - return IncompatFeatures{ - DirentFileType: f&SbDirentFileType > 0, - Recovery: f&SbRecovery > 0, - JournalDev: f&SbJournalDev > 0, - MetaBG: f&SbMetaBG > 0, - Extents: f&SbExtents > 0, - Is64Bit: f&SbIs64Bit > 0, - MMP: f&SbMMP > 0, - FlexBg: f&SbFlexBg > 0, - LargeDir: f&SbLargeDir > 0, - InlineData: f&SbInlineData > 0, - Encrypted: f&SbEncrypted > 0, - } -} - -// Superblock readonly compatible features. -// This is not exhaustive, unused features are not listed. -const ( - // SbSparse indicates sparse superblocks. Only groups with number either 0 or - // a power of 3, 5, or 7 will have redundant copies of the superblock and - // block descriptors. - SbSparse = 0x1 - - // SbLargeFile indicates that this fs has been used to store a file >= 2GiB. - SbLargeFile = 0x2 - - // SbHugeFile indicates that this fs contains files whose sizes are - // represented in units of logicals blocks, not 512-byte sectors. - SbHugeFile = 0x8 - - // SbGdtCsum indicates that group descriptors have checksums. - SbGdtCsum = 0x10 - - // SbDirNlink indicates that the new subdirectory limit is 64,999. Ext3 has a - // 32,000 subdirectory limit. - SbDirNlink = 0x20 - - // SbExtraIsize indicates that large inodes exist on this filesystem. - SbExtraIsize = 0x40 - - // SbHasSnapshot indicates the existence of a snapshot. - SbHasSnapshot = 0x80 - - // SbQuota enables usage tracking for all quota types. - SbQuota = 0x100 - - // SbBigalloc maps to the bigalloc feature. When set, the minimum allocation - // unit becomes a cluster rather than a data block. Then block bitmaps track - // clusters, not data blocks. - // - // See https://www.kernel.org/doc/html/latest/filesystems/ext4/overview.html#bigalloc. - SbBigalloc = 0x200 - - // SbMetadataCsum indicates that the fs supports metadata checksumming. - SbMetadataCsum = 0x400 - - // SbReadOnly marks this filesystem as readonly. Should refuse to mount in - // read/write mode. - SbReadOnly = 0x1000 -) - -// RoCompatFeatures represents a superblock's readonly compatible feature set. -// If the kernel does not understand any of these feature, it can still mount -// readonly. But if the user wants to mount read/write, the kernel should -// refuse to mount. -type RoCompatFeatures struct { - Sparse bool - LargeFile bool - HugeFile bool - GdtCsum bool - DirNlink bool - ExtraIsize bool - HasSnapshot bool - Quota bool - Bigalloc bool - MetadataCsum bool - ReadOnly bool -} - -// ToInt converts superblock readonly compatible features to its 32-bit rep. -func (f RoCompatFeatures) ToInt() uint32 { - var res uint32 - - if f.Sparse { - res |= SbSparse - } - if f.LargeFile { - res |= SbLargeFile - } - if f.HugeFile { - res |= SbHugeFile - } - if f.GdtCsum { - res |= SbGdtCsum - } - if f.DirNlink { - res |= SbDirNlink - } - if f.ExtraIsize { - res |= SbExtraIsize - } - if f.HasSnapshot { - res |= SbHasSnapshot - } - if f.Quota { - res |= SbQuota - } - if f.Bigalloc { - res |= SbBigalloc - } - if f.MetadataCsum { - res |= SbMetadataCsum - } - if f.ReadOnly { - res |= SbReadOnly - } - - return res -} - -// RoCompatFeaturesFromInt converts the integer representation of superblock -// readonly compatible features to RoCompatFeatures struct. -func RoCompatFeaturesFromInt(f uint32) RoCompatFeatures { - return RoCompatFeatures{ - Sparse: f&SbSparse > 0, - LargeFile: f&SbLargeFile > 0, - HugeFile: f&SbHugeFile > 0, - GdtCsum: f&SbGdtCsum > 0, - DirNlink: f&SbDirNlink > 0, - ExtraIsize: f&SbExtraIsize > 0, - HasSnapshot: f&SbHasSnapshot > 0, - Quota: f&SbQuota > 0, - Bigalloc: f&SbBigalloc > 0, - MetadataCsum: f&SbMetadataCsum > 0, - ReadOnly: f&SbReadOnly > 0, - } -} diff --git a/pkg/sentry/fs/ext/disklayout/superblock_32.go b/pkg/sentry/fs/ext/disklayout/superblock_32.go deleted file mode 100644 index 53e515fd3..000000000 --- a/pkg/sentry/fs/ext/disklayout/superblock_32.go +++ /dev/null @@ -1,76 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -// SuperBlock32Bit implements SuperBlock and represents the 32-bit version of -// the ext4_super_block struct in fs/ext4/ext4.h. Should be used only if -// RevLevel = DynamicRev and 64-bit feature is disabled. -type SuperBlock32Bit struct { - // We embed the old superblock struct here because the 32-bit version is just - // an extension of the old version. - SuperBlockOld - - FirstInode uint32 - InodeSizeRaw uint16 - BlockGroupNumber uint16 - FeatureCompat uint32 - FeatureIncompat uint32 - FeatureRoCompat uint32 - UUID [16]byte - VolumeName [16]byte - LastMounted [64]byte - AlgoUsageBitmap uint32 - PreallocBlocks uint8 - PreallocDirBlocks uint8 - ReservedGdtBlocks uint16 - JournalUUID [16]byte - JournalInum uint32 - JournalDev uint32 - LastOrphan uint32 - HashSeed [4]uint32 - DefaultHashVersion uint8 - JnlBackupType uint8 - BgDescSizeRaw uint16 - DefaultMountOpts uint32 - FirstMetaBg uint32 - MkfsTime uint32 - JnlBlocks [17]uint32 -} - -// Compiles only if SuperBlock32Bit implements SuperBlock. -var _ SuperBlock = (*SuperBlock32Bit)(nil) - -// Only override methods which change based on the additional fields above. -// Not overriding SuperBlock.BgDescSize because it would still return 32 here. - -// InodeSize implements SuperBlock.InodeSize. -func (sb *SuperBlock32Bit) InodeSize() uint16 { - return sb.InodeSizeRaw -} - -// CompatibleFeatures implements SuperBlock.CompatibleFeatures. -func (sb *SuperBlock32Bit) CompatibleFeatures() CompatFeatures { - return CompatFeaturesFromInt(sb.FeatureCompat) -} - -// IncompatibleFeatures implements SuperBlock.IncompatibleFeatures. -func (sb *SuperBlock32Bit) IncompatibleFeatures() IncompatFeatures { - return IncompatFeaturesFromInt(sb.FeatureIncompat) -} - -// ReadOnlyCompatibleFeatures implements SuperBlock.ReadOnlyCompatibleFeatures. -func (sb *SuperBlock32Bit) ReadOnlyCompatibleFeatures() RoCompatFeatures { - return RoCompatFeaturesFromInt(sb.FeatureRoCompat) -} diff --git a/pkg/sentry/fs/ext/disklayout/superblock_64.go b/pkg/sentry/fs/ext/disklayout/superblock_64.go deleted file mode 100644 index 7c1053fb4..000000000 --- a/pkg/sentry/fs/ext/disklayout/superblock_64.go +++ /dev/null @@ -1,95 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -// SuperBlock64Bit implements SuperBlock and represents the 64-bit version of -// the ext4_super_block struct in fs/ext4/ext4.h. This sums up to be exactly -// 1024 bytes (smallest possible block size) and hence the superblock always -// fits in no more than one data block. Should only be used when the 64-bit -// feature is set. -type SuperBlock64Bit struct { - // We embed the 32-bit struct here because 64-bit version is just an extension - // of the 32-bit version. - SuperBlock32Bit - - BlocksCountHi uint32 - ReservedBlocksCountHi uint32 - FreeBlocksCountHi uint32 - MinInodeSize uint16 - WantInodeSize uint16 - Flags uint32 - RaidStride uint16 - MmpInterval uint16 - MmpBlock uint64 - RaidStripeWidth uint32 - LogGroupsPerFlex uint8 - ChecksumType uint8 - _ uint16 - KbytesWritten uint64 - SnapshotInum uint32 - SnapshotID uint32 - SnapshotRsrvBlocksCount uint64 - SnapshotList uint32 - ErrorCount uint32 - FirstErrorTime uint32 - FirstErrorInode uint32 - FirstErrorBlock uint64 - FirstErrorFunction [32]byte - FirstErrorLine uint32 - LastErrorTime uint32 - LastErrorInode uint32 - LastErrorLine uint32 - LastErrorBlock uint64 - LastErrorFunction [32]byte - MountOpts [64]byte - UserQuotaInum uint32 - GroupQuotaInum uint32 - OverheadBlocks uint32 - BackupBgs [2]uint32 - EncryptAlgos [4]uint8 - EncryptPwSalt [16]uint8 - LostFoundInode uint32 - ProjectQuotaInode uint32 - ChecksumSeed uint32 - WtimeHi uint8 - MtimeHi uint8 - MkfsTimeHi uint8 - LastCheckHi uint8 - FirstErrorTimeHi uint8 - LastErrorTimeHi uint8 - _ [2]uint8 - Encoding uint16 - EncodingFlags uint16 - _ [95]uint32 - Checksum uint32 -} - -// Compiles only if SuperBlock64Bit implements SuperBlock. -var _ SuperBlock = (*SuperBlock64Bit)(nil) - -// Only override methods which change based on the 64-bit feature. - -// BlocksCount implements SuperBlock.BlocksCount. -func (sb *SuperBlock64Bit) BlocksCount() uint64 { - return (uint64(sb.BlocksCountHi) << 32) | uint64(sb.BlocksCountLo) -} - -// FreeBlocksCount implements SuperBlock.FreeBlocksCount. -func (sb *SuperBlock64Bit) FreeBlocksCount() uint64 { - return (uint64(sb.FreeBlocksCountHi) << 32) | uint64(sb.FreeBlocksCountLo) -} - -// BgDescSize implements SuperBlock.BgDescSize. -func (sb *SuperBlock64Bit) BgDescSize() uint16 { return sb.BgDescSizeRaw } diff --git a/pkg/sentry/fs/ext/disklayout/superblock_old.go b/pkg/sentry/fs/ext/disklayout/superblock_old.go deleted file mode 100644 index 9221e0251..000000000 --- a/pkg/sentry/fs/ext/disklayout/superblock_old.go +++ /dev/null @@ -1,105 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -// SuperBlockOld implements SuperBlock and represents the old version of the -// superblock struct. Should be used only if RevLevel = OldRev. -type SuperBlockOld struct { - InodesCountRaw uint32 - BlocksCountLo uint32 - ReservedBlocksCount uint32 - FreeBlocksCountLo uint32 - FreeInodesCountRaw uint32 - FirstDataBlockRaw uint32 - LogBlockSize uint32 - LogClusterSize uint32 - BlocksPerGroupRaw uint32 - ClustersPerGroupRaw uint32 - InodesPerGroupRaw uint32 - Mtime uint32 - Wtime uint32 - MountCountRaw uint16 - MaxMountCountRaw uint16 - MagicRaw uint16 - State uint16 - Errors uint16 - MinorRevLevel uint16 - LastCheck uint32 - CheckInterval uint32 - CreatorOS uint32 - RevLevel uint32 - DefResUID uint16 - DefResGID uint16 -} - -// Compiles only if SuperBlockOld implements SuperBlock. -var _ SuperBlock = (*SuperBlockOld)(nil) - -// InodesCount implements SuperBlock.InodesCount. -func (sb *SuperBlockOld) InodesCount() uint32 { return sb.InodesCountRaw } - -// BlocksCount implements SuperBlock.BlocksCount. -func (sb *SuperBlockOld) BlocksCount() uint64 { return uint64(sb.BlocksCountLo) } - -// FreeBlocksCount implements SuperBlock.FreeBlocksCount. -func (sb *SuperBlockOld) FreeBlocksCount() uint64 { return uint64(sb.FreeBlocksCountLo) } - -// FreeInodesCount implements SuperBlock.FreeInodesCount. -func (sb *SuperBlockOld) FreeInodesCount() uint32 { return sb.FreeInodesCountRaw } - -// MountCount implements SuperBlock.MountCount. -func (sb *SuperBlockOld) MountCount() uint16 { return sb.MountCountRaw } - -// MaxMountCount implements SuperBlock.MaxMountCount. -func (sb *SuperBlockOld) MaxMountCount() uint16 { return sb.MaxMountCountRaw } - -// FirstDataBlock implements SuperBlock.FirstDataBlock. -func (sb *SuperBlockOld) FirstDataBlock() uint32 { return sb.FirstDataBlockRaw } - -// BlockSize implements SuperBlock.BlockSize. -func (sb *SuperBlockOld) BlockSize() uint64 { return 1 << (10 + sb.LogBlockSize) } - -// BlocksPerGroup implements SuperBlock.BlocksPerGroup. -func (sb *SuperBlockOld) BlocksPerGroup() uint32 { return sb.BlocksPerGroupRaw } - -// ClusterSize implements SuperBlock.ClusterSize. -func (sb *SuperBlockOld) ClusterSize() uint64 { return 1 << (10 + sb.LogClusterSize) } - -// ClustersPerGroup implements SuperBlock.ClustersPerGroup. -func (sb *SuperBlockOld) ClustersPerGroup() uint32 { return sb.ClustersPerGroupRaw } - -// InodeSize implements SuperBlock.InodeSize. -func (sb *SuperBlockOld) InodeSize() uint16 { return OldInodeSize } - -// InodesPerGroup implements SuperBlock.InodesPerGroup. -func (sb *SuperBlockOld) InodesPerGroup() uint32 { return sb.InodesPerGroupRaw } - -// BgDescSize implements SuperBlock.BgDescSize. -func (sb *SuperBlockOld) BgDescSize() uint16 { return 32 } - -// CompatibleFeatures implements SuperBlock.CompatibleFeatures. -func (sb *SuperBlockOld) CompatibleFeatures() CompatFeatures { return CompatFeatures{} } - -// IncompatibleFeatures implements SuperBlock.IncompatibleFeatures. -func (sb *SuperBlockOld) IncompatibleFeatures() IncompatFeatures { return IncompatFeatures{} } - -// ReadOnlyCompatibleFeatures implements SuperBlock.ReadOnlyCompatibleFeatures. -func (sb *SuperBlockOld) ReadOnlyCompatibleFeatures() RoCompatFeatures { return RoCompatFeatures{} } - -// Magic implements SuperBlock.Magic. -func (sb *SuperBlockOld) Magic() uint16 { return sb.MagicRaw } - -// Revision implements SuperBlock.Revision. -func (sb *SuperBlockOld) Revision() SbRevision { return SbRevision(sb.RevLevel) } diff --git a/pkg/sentry/fs/ext/disklayout/superblock_test.go b/pkg/sentry/fs/ext/disklayout/superblock_test.go deleted file mode 100644 index 463b5ba21..000000000 --- a/pkg/sentry/fs/ext/disklayout/superblock_test.go +++ /dev/null @@ -1,27 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "testing" -) - -// TestSuperBlockSize tests that the superblock structs are of the correct -// size. -func TestSuperBlockSize(t *testing.T) { - assertSize(t, SuperBlockOld{}, 84) - assertSize(t, SuperBlock32Bit{}, 336) - assertSize(t, SuperBlock64Bit{}, 1024) -} diff --git a/pkg/sentry/fs/ext/disklayout/test_utils.go b/pkg/sentry/fs/ext/disklayout/test_utils.go deleted file mode 100644 index 9c63f04c0..000000000 --- a/pkg/sentry/fs/ext/disklayout/test_utils.go +++ /dev/null @@ -1,30 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package disklayout - -import ( - "reflect" - "testing" - - "gvisor.dev/gvisor/pkg/binary" -) - -func assertSize(t *testing.T, v interface{}, want uintptr) { - t.Helper() - - if got := binary.Size(v); got != want { - t.Errorf("struct %s should be exactly %d bytes but is %d bytes", reflect.TypeOf(v).Name(), want, got) - } -} diff --git a/pkg/sentry/fs/ext/ext.go b/pkg/sentry/fs/ext/ext.go deleted file mode 100644 index d303dd122..000000000 --- a/pkg/sentry/fs/ext/ext.go +++ /dev/null @@ -1,96 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package ext implements readonly ext(2/3/4) filesystems. -package ext - -import ( - "errors" - "fmt" - "io" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/fd" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs/ext/disklayout" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// filesystemType implements vfs.FilesystemType. -type filesystemType struct{} - -// Compiles only if filesystemType implements vfs.FilesystemType. -var _ vfs.FilesystemType = (*filesystemType)(nil) - -// getDeviceFd returns an io.ReaderAt to the underlying device. -// Currently there are two ways of mounting an ext(2/3/4) fs: -// 1. Specify a mount with our internal special MountType in the OCI spec. -// 2. Expose the device to the container and mount it from application layer. -func getDeviceFd(source string, opts vfs.NewFilesystemOptions) (io.ReaderAt, error) { - if opts.InternalData == nil { - // User mount call. - // TODO(b/134676337): Open the device specified by `source` and return that. - panic("unimplemented") - } - - // NewFilesystem call originated from within the sentry. - devFd, ok := opts.InternalData.(int) - if !ok { - return nil, errors.New("internal data for ext fs must be an int containing the file descriptor to device") - } - - if devFd < 0 { - return nil, fmt.Errorf("ext device file descriptor is not valid: %d", devFd) - } - - // The fd.ReadWriter returned from fd.NewReadWriter() does not take ownership - // of the file descriptor and hence will not close it when it is garbage - // collected. - return fd.NewReadWriter(devFd), nil -} - -// NewFilesystem implements vfs.FilesystemType.NewFilesystem. -func (fstype filesystemType) NewFilesystem(ctx context.Context, creds *auth.Credentials, source string, opts vfs.NewFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - dev, err := getDeviceFd(source, opts) - if err != nil { - return nil, nil, err - } - - fs := filesystem{dev: dev, inodeCache: make(map[uint32]*inode)} - fs.vfsfs.Init(&fs) - fs.sb, err = readSuperBlock(dev) - if err != nil { - return nil, nil, err - } - - if fs.sb.Magic() != linux.EXT_SUPER_MAGIC { - // mount(2) specifies that EINVAL should be returned if the superblock is - // invalid. - return nil, nil, syserror.EINVAL - } - - fs.bgs, err = readBlockGroups(dev, fs.sb) - if err != nil { - return nil, nil, err - } - - rootInode, err := fs.getOrCreateInode(ctx, disklayout.RootDirInode) - if err != nil { - return nil, nil, err - } - - return &fs.vfsfs, &newDentry(rootInode).vfsd, nil -} diff --git a/pkg/sentry/fs/ext/ext_test.go b/pkg/sentry/fs/ext/ext_test.go deleted file mode 100644 index 18764e92a..000000000 --- a/pkg/sentry/fs/ext/ext_test.go +++ /dev/null @@ -1,407 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "fmt" - "os" - "path" - "testing" - - "github.com/google/go-cmp/cmp" - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs/ext/disklayout" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - - "gvisor.dev/gvisor/runsc/test/testutil" -) - -const ( - assetsDir = "pkg/sentry/fs/ext/assets" -) - -var ( - ext2ImagePath = path.Join(assetsDir, "tiny.ext2") - ext3ImagePath = path.Join(assetsDir, "tiny.ext3") - ext4ImagePath = path.Join(assetsDir, "tiny.ext4") -) - -func beginning(_ uint64) uint64 { - return 0 -} - -func middle(i uint64) uint64 { - return i / 2 -} - -func end(i uint64) uint64 { - return i -} - -// setUp opens imagePath as an ext Filesystem and returns all necessary -// elements required to run tests. If error is non-nil, it also returns a tear -// down function which must be called after the test is run for clean up. -func setUp(t *testing.T, imagePath string) (context.Context, *vfs.Filesystem, *vfs.Dentry, func(), error) { - localImagePath, err := testutil.FindFile(imagePath) - if err != nil { - return nil, nil, nil, nil, fmt.Errorf("failed to open local image at path %s: %v", imagePath, err) - } - - f, err := os.Open(localImagePath) - if err != nil { - return nil, nil, nil, nil, err - } - - // Mount the ext4 fs and retrieve the inode structure for the file. - mockCtx := contexttest.Context(t) - fs, d, err := filesystemType{}.NewFilesystem(mockCtx, nil, localImagePath, vfs.NewFilesystemOptions{InternalData: int(f.Fd())}) - if err != nil { - f.Close() - return nil, nil, nil, nil, err - } - - tearDown := func() { - if err := f.Close(); err != nil { - t.Fatalf("tearDown failed: %v", err) - } - } - return mockCtx, fs, d, tearDown, nil -} - -// TestRootDir tests that the root directory inode is correctly initialized and -// returned from setUp. -func TestRootDir(t *testing.T) { - type inodeProps struct { - Mode linux.FileMode - UID auth.KUID - GID auth.KGID - Size uint64 - InodeSize uint16 - Links uint16 - Flags disklayout.InodeFlags - } - - type rootDirTest struct { - name string - image string - wantInode inodeProps - } - - tests := []rootDirTest{ - { - name: "ext4 root dir", - image: ext4ImagePath, - wantInode: inodeProps{ - Mode: linux.ModeDirectory | 0755, - Size: 0x400, - InodeSize: 0x80, - Links: 3, - Flags: disklayout.InodeFlags{Extents: true}, - }, - }, - { - name: "ext3 root dir", - image: ext3ImagePath, - wantInode: inodeProps{ - Mode: linux.ModeDirectory | 0755, - Size: 0x400, - InodeSize: 0x80, - Links: 3, - }, - }, - { - name: "ext2 root dir", - image: ext2ImagePath, - wantInode: inodeProps{ - Mode: linux.ModeDirectory | 0755, - Size: 0x400, - InodeSize: 0x80, - Links: 3, - }, - }, - } - - for _, test := range tests { - t.Run(test.name, func(t *testing.T) { - _, _, vfsd, tearDown, err := setUp(t, test.image) - if err != nil { - t.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - d, ok := vfsd.Impl().(*dentry) - if !ok { - t.Fatalf("ext dentry of incorrect type: %T", vfsd.Impl()) - } - - // Offload inode contents into local structs for comparison. - gotInode := inodeProps{ - Mode: d.inode.diskInode.Mode(), - UID: d.inode.diskInode.UID(), - GID: d.inode.diskInode.GID(), - Size: d.inode.diskInode.Size(), - InodeSize: d.inode.diskInode.InodeSize(), - Links: d.inode.diskInode.LinksCount(), - Flags: d.inode.diskInode.Flags(), - } - - if diff := cmp.Diff(gotInode, test.wantInode); diff != "" { - t.Errorf("inode mismatch (-want +got):\n%s", diff) - } - }) - } -} - -// TestFilesystemInit tests that the filesystem superblock and block group -// descriptors are correctly read in and initialized. -func TestFilesystemInit(t *testing.T) { - // sb only contains the immutable properties of the superblock. - type sb struct { - InodesCount uint32 - BlocksCount uint64 - MaxMountCount uint16 - FirstDataBlock uint32 - BlockSize uint64 - BlocksPerGroup uint32 - ClusterSize uint64 - ClustersPerGroup uint32 - InodeSize uint16 - InodesPerGroup uint32 - BgDescSize uint16 - Magic uint16 - Revision disklayout.SbRevision - CompatFeatures disklayout.CompatFeatures - IncompatFeatures disklayout.IncompatFeatures - RoCompatFeatures disklayout.RoCompatFeatures - } - - // bg only contains the immutable properties of the block group descriptor. - type bg struct { - InodeTable uint64 - BlockBitmap uint64 - InodeBitmap uint64 - ExclusionBitmap uint64 - Flags disklayout.BGFlags - } - - type fsInitTest struct { - name string - image string - wantSb sb - wantBgs []bg - } - - tests := []fsInitTest{ - { - name: "ext4 filesystem init", - image: ext4ImagePath, - wantSb: sb{ - InodesCount: 0x10, - BlocksCount: 0x40, - MaxMountCount: 0xffff, - FirstDataBlock: 0x1, - BlockSize: 0x400, - BlocksPerGroup: 0x2000, - ClusterSize: 0x400, - ClustersPerGroup: 0x2000, - InodeSize: 0x80, - InodesPerGroup: 0x10, - BgDescSize: 0x40, - Magic: linux.EXT_SUPER_MAGIC, - Revision: disklayout.DynamicRev, - CompatFeatures: disklayout.CompatFeatures{ - ExtAttr: true, - ResizeInode: true, - DirIndex: true, - }, - IncompatFeatures: disklayout.IncompatFeatures{ - DirentFileType: true, - Extents: true, - Is64Bit: true, - FlexBg: true, - }, - RoCompatFeatures: disklayout.RoCompatFeatures{ - Sparse: true, - LargeFile: true, - HugeFile: true, - DirNlink: true, - ExtraIsize: true, - MetadataCsum: true, - }, - }, - wantBgs: []bg{ - { - InodeTable: 0x23, - BlockBitmap: 0x3, - InodeBitmap: 0x13, - Flags: disklayout.BGFlags{ - InodeZeroed: true, - }, - }, - }, - }, - { - name: "ext3 filesystem init", - image: ext3ImagePath, - wantSb: sb{ - InodesCount: 0x10, - BlocksCount: 0x40, - MaxMountCount: 0xffff, - FirstDataBlock: 0x1, - BlockSize: 0x400, - BlocksPerGroup: 0x2000, - ClusterSize: 0x400, - ClustersPerGroup: 0x2000, - InodeSize: 0x80, - InodesPerGroup: 0x10, - BgDescSize: 0x20, - Magic: linux.EXT_SUPER_MAGIC, - Revision: disklayout.DynamicRev, - CompatFeatures: disklayout.CompatFeatures{ - ExtAttr: true, - ResizeInode: true, - DirIndex: true, - }, - IncompatFeatures: disklayout.IncompatFeatures{ - DirentFileType: true, - }, - RoCompatFeatures: disklayout.RoCompatFeatures{ - Sparse: true, - LargeFile: true, - }, - }, - wantBgs: []bg{ - { - InodeTable: 0x5, - BlockBitmap: 0x3, - InodeBitmap: 0x4, - Flags: disklayout.BGFlags{ - InodeZeroed: true, - }, - }, - }, - }, - { - name: "ext2 filesystem init", - image: ext2ImagePath, - wantSb: sb{ - InodesCount: 0x10, - BlocksCount: 0x40, - MaxMountCount: 0xffff, - FirstDataBlock: 0x1, - BlockSize: 0x400, - BlocksPerGroup: 0x2000, - ClusterSize: 0x400, - ClustersPerGroup: 0x2000, - InodeSize: 0x80, - InodesPerGroup: 0x10, - BgDescSize: 0x20, - Magic: linux.EXT_SUPER_MAGIC, - Revision: disklayout.DynamicRev, - CompatFeatures: disklayout.CompatFeatures{ - ExtAttr: true, - ResizeInode: true, - DirIndex: true, - }, - IncompatFeatures: disklayout.IncompatFeatures{ - DirentFileType: true, - }, - RoCompatFeatures: disklayout.RoCompatFeatures{ - Sparse: true, - LargeFile: true, - }, - }, - wantBgs: []bg{ - { - InodeTable: 0x5, - BlockBitmap: 0x3, - InodeBitmap: 0x4, - Flags: disklayout.BGFlags{ - InodeZeroed: true, - }, - }, - }, - }, - } - - for _, test := range tests { - t.Run(test.name, func(t *testing.T) { - _, vfsfs, _, tearDown, err := setUp(t, test.image) - if err != nil { - t.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - fs, ok := vfsfs.Impl().(*filesystem) - if !ok { - t.Fatalf("ext filesystem of incorrect type: %T", vfsfs.Impl()) - } - - // Offload superblock and block group descriptors contents into - // local structs for comparison. - totalFreeInodes := uint32(0) - totalFreeBlocks := uint64(0) - gotSb := sb{ - InodesCount: fs.sb.InodesCount(), - BlocksCount: fs.sb.BlocksCount(), - MaxMountCount: fs.sb.MaxMountCount(), - FirstDataBlock: fs.sb.FirstDataBlock(), - BlockSize: fs.sb.BlockSize(), - BlocksPerGroup: fs.sb.BlocksPerGroup(), - ClusterSize: fs.sb.ClusterSize(), - ClustersPerGroup: fs.sb.ClustersPerGroup(), - InodeSize: fs.sb.InodeSize(), - InodesPerGroup: fs.sb.InodesPerGroup(), - BgDescSize: fs.sb.BgDescSize(), - Magic: fs.sb.Magic(), - Revision: fs.sb.Revision(), - CompatFeatures: fs.sb.CompatibleFeatures(), - IncompatFeatures: fs.sb.IncompatibleFeatures(), - RoCompatFeatures: fs.sb.ReadOnlyCompatibleFeatures(), - } - gotNumBgs := len(fs.bgs) - gotBgs := make([]bg, gotNumBgs) - for i := 0; i < gotNumBgs; i++ { - gotBgs[i].InodeTable = fs.bgs[i].InodeTable() - gotBgs[i].BlockBitmap = fs.bgs[i].BlockBitmap() - gotBgs[i].InodeBitmap = fs.bgs[i].InodeBitmap() - gotBgs[i].ExclusionBitmap = fs.bgs[i].ExclusionBitmap() - gotBgs[i].Flags = fs.bgs[i].Flags() - - totalFreeInodes += fs.bgs[i].FreeInodesCount() - totalFreeBlocks += uint64(fs.bgs[i].FreeBlocksCount()) - } - - if diff := cmp.Diff(gotSb, test.wantSb); diff != "" { - t.Errorf("superblock mismatch (-want +got):\n%s", diff) - } - - if diff := cmp.Diff(gotBgs, test.wantBgs); diff != "" { - t.Errorf("block group descriptors mismatch (-want +got):\n%s", diff) - } - - if diff := cmp.Diff(totalFreeInodes, fs.sb.FreeInodesCount()); diff != "" { - t.Errorf("total free inodes mismatch (-want +got):\n%s", diff) - } - - if diff := cmp.Diff(totalFreeBlocks, fs.sb.FreeBlocksCount()); diff != "" { - t.Errorf("total free blocks mismatch (-want +got):\n%s", diff) - } - }) - } -} diff --git a/pkg/sentry/fs/ext/extent_file.go b/pkg/sentry/fs/ext/extent_file.go deleted file mode 100644 index aa4102dbb..000000000 --- a/pkg/sentry/fs/ext/extent_file.go +++ /dev/null @@ -1,260 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "io" - "sort" - - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/fs/ext/disklayout" - "gvisor.dev/gvisor/pkg/syserror" -) - -// extentFile is a type of regular file which uses extents to store file data. -type extentFile struct { - regFile regularFile - - // root is the root extent node. This lives in the 60 byte diskInode.Data(). - // Immutable. - root disklayout.ExtentNode -} - -// Compiles only if extentFile implements fileReader. -var _ fileReader = (*extentFile)(nil) - -// Read implements fileReader.getFileReader. -func (f *extentFile) getFileReader(dev io.ReaderAt, blkSize uint64, offset uint64) io.Reader { - return &extentReader{ - dev: dev, - file: f, - fileOff: offset, - blkSize: blkSize, - } -} - -// newExtentFile is the extent file constructor. It reads the entire extent -// tree into memory. -// TODO(b/134676337): Build extent tree on demand to reduce memory usage. -func newExtentFile(dev io.ReaderAt, blkSize uint64, regFile regularFile) (*extentFile, error) { - file := &extentFile{regFile: regFile} - file.regFile.impl = file - err := file.buildExtTree(dev, blkSize) - if err != nil { - return nil, err - } - return file, nil -} - -// buildExtTree builds the extent tree by reading it from disk by doing -// running a simple DFS. It first reads the root node from the inode struct in -// memory. Then it recursively builds the rest of the tree by reading it off -// disk. -// -// Precondition: inode flag InExtents must be set. -func (f *extentFile) buildExtTree(dev io.ReaderAt, blkSize uint64) error { - rootNodeData := f.regFile.inode.diskInode.Data() - - binary.Unmarshal(rootNodeData[:disklayout.ExtentStructsSize], binary.LittleEndian, &f.root.Header) - - // Root node can not have more than 4 entries: 60 bytes = 1 header + 4 entries. - if f.root.Header.NumEntries > 4 { - // read(2) specifies that EINVAL should be returned if the file is unsuitable - // for reading. - return syserror.EINVAL - } - - f.root.Entries = make([]disklayout.ExtentEntryPair, f.root.Header.NumEntries) - for i, off := uint16(0), disklayout.ExtentStructsSize; i < f.root.Header.NumEntries; i, off = i+1, off+disklayout.ExtentStructsSize { - var curEntry disklayout.ExtentEntry - if f.root.Header.Height == 0 { - // Leaf node. - curEntry = &disklayout.Extent{} - } else { - // Internal node. - curEntry = &disklayout.ExtentIdx{} - } - binary.Unmarshal(rootNodeData[off:off+disklayout.ExtentStructsSize], binary.LittleEndian, curEntry) - f.root.Entries[i].Entry = curEntry - } - - // If this node is internal, perform DFS. - if f.root.Header.Height > 0 { - for i := uint16(0); i < f.root.Header.NumEntries; i++ { - var err error - if f.root.Entries[i].Node, err = buildExtTreeFromDisk(dev, f.root.Entries[i].Entry, blkSize); err != nil { - return err - } - } - } - - return nil -} - -// buildExtTreeFromDisk reads the extent tree nodes from disk and recursively -// builds the tree. Performs a simple DFS. It returns the ExtentNode pointed to -// by the ExtentEntry. -func buildExtTreeFromDisk(dev io.ReaderAt, entry disklayout.ExtentEntry, blkSize uint64) (*disklayout.ExtentNode, error) { - var header disklayout.ExtentHeader - off := entry.PhysicalBlock() * blkSize - err := readFromDisk(dev, int64(off), &header) - if err != nil { - return nil, err - } - - entries := make([]disklayout.ExtentEntryPair, header.NumEntries) - for i, off := uint16(0), off+disklayout.ExtentStructsSize; i < header.NumEntries; i, off = i+1, off+disklayout.ExtentStructsSize { - var curEntry disklayout.ExtentEntry - if header.Height == 0 { - // Leaf node. - curEntry = &disklayout.Extent{} - } else { - // Internal node. - curEntry = &disklayout.ExtentIdx{} - } - - err := readFromDisk(dev, int64(off), curEntry) - if err != nil { - return nil, err - } - entries[i].Entry = curEntry - } - - // If this node is internal, perform DFS. - if header.Height > 0 { - for i := uint16(0); i < header.NumEntries; i++ { - var err error - entries[i].Node, err = buildExtTreeFromDisk(dev, entries[i].Entry, blkSize) - if err != nil { - return nil, err - } - } - } - - return &disklayout.ExtentNode{header, entries}, nil -} - -// extentReader implements io.Reader which can traverse the extent tree and -// read file data. This is not thread safe. -type extentReader struct { - dev io.ReaderAt - file *extentFile - fileOff uint64 // Represents the current file offset being read from. - blkSize uint64 -} - -// Compiles only if inlineReader implements io.Reader. -var _ io.Reader = (*extentReader)(nil) - -// Read implements io.Reader.Read. -func (r *extentReader) Read(dst []byte) (int, error) { - if len(dst) == 0 { - return 0, nil - } - - if r.fileOff >= r.file.regFile.inode.diskInode.Size() { - return 0, io.EOF - } - - return r.read(&r.file.root, dst) -} - -// read is a helper which traverses the extent tree and reads data. -func (r *extentReader) read(node *disklayout.ExtentNode, dst []byte) (int, error) { - // Perform a binary search for the node covering bytes starting at r.fileOff. - // A highly fragmented filesystem can have upto 340 entries and so linear - // search should be avoided. Finds the first entry which does not cover the - // file block we want and subtracts 1 to get the desired index. - fileBlk := r.fileBlock() - n := len(node.Entries) - found := sort.Search(n, func(i int) bool { - return node.Entries[i].Entry.FileBlock() > fileBlk - }) - 1 - - // We should be in this recursive step only if the data we want exists under - // the current node. - if found < 0 { - panic("searching for a file block in an extent entry which does not cover it") - } - - read := 0 - toRead := len(dst) - var curR int - var err error - for i := found; i < n && read < toRead; i++ { - if node.Header.Height == 0 { - curR, err = r.readFromExtent(node.Entries[i].Entry.(*disklayout.Extent), dst[read:]) - } else { - curR, err = r.read(node.Entries[i].Node, dst[read:]) - } - - read += curR - if err != nil { - return read, err - } - } - - return read, nil -} - -// readFromExtent reads file data from the extent. It takes advantage of the -// sequential nature of extents and reads file data from multiple blocks in one -// call. Also updates the file offset. -// -// A non-nil error indicates that this is a partial read and there is probably -// more to read from this extent. The caller should propagate the error upward -// and not move to the next extent in the tree. -// -// A subsequent call to extentReader.Read should continue reading from where we -// left off as expected. -func (r *extentReader) readFromExtent(ex *disklayout.Extent, dst []byte) (int, error) { - curFileBlk := r.fileBlock() - exFirstFileBlk := ex.FileBlock() - exLastFileBlk := exFirstFileBlk + uint32(ex.Length) // This is exclusive. - - // We should be in this recursive step only if the data we want exists under - // the current extent. - if curFileBlk < exFirstFileBlk || exLastFileBlk <= curFileBlk { - panic("searching for a file block in an extent which does not cover it") - } - - curPhyBlk := uint64(curFileBlk-exFirstFileBlk) + ex.PhysicalBlock() - readStart := curPhyBlk*r.blkSize + r.fileBlockOff() - - endPhyBlk := ex.PhysicalBlock() + uint64(ex.Length) - extentEnd := endPhyBlk * r.blkSize // This is exclusive. - - toRead := int(extentEnd - readStart) - if len(dst) < toRead { - toRead = len(dst) - } - - n, _ := r.dev.ReadAt(dst[:toRead], int64(readStart)) - r.fileOff += uint64(n) - if n < toRead { - return n, syserror.EIO - } - return n, nil -} - -// fileBlock returns the file block number we are currently reading. -func (r *extentReader) fileBlock() uint32 { - return uint32(r.fileOff / r.blkSize) -} - -// fileBlockOff returns the current offset within the current file block. -func (r *extentReader) fileBlockOff() uint64 { - return r.fileOff % r.blkSize -} diff --git a/pkg/sentry/fs/ext/extent_test.go b/pkg/sentry/fs/ext/extent_test.go deleted file mode 100644 index dff401114..000000000 --- a/pkg/sentry/fs/ext/extent_test.go +++ /dev/null @@ -1,292 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "bytes" - "io" - "math/rand" - "testing" - - "github.com/google/go-cmp/cmp" - "github.com/google/go-cmp/cmp/cmpopts" - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/fs/ext/disklayout" -) - -const ( - // mockExtentBlkSize is the mock block size used for testing. - // No block has more than 1 header + 4 entries. - mockExtentBlkSize = uint64(64) -) - -// The tree described below looks like: -// -// 0.{Head}[Idx][Idx] -// / \ -// / \ -// 1.{Head}[Ext][Ext] 2.{Head}[Idx] -// / | \ -// [Phy] [Phy, Phy] 3.{Head}[Ext] -// | -// [Phy, Phy, Phy] -// -// Legend: -// - Head = ExtentHeader -// - Idx = ExtentIdx -// - Ext = Extent -// - Phy = Physical Block -// -// Please note that ext4 might not construct extent trees looking like this. -// This is purely for testing the tree traversal logic. -var ( - node3 = &disklayout.ExtentNode{ - Header: disklayout.ExtentHeader{ - Magic: disklayout.ExtentMagic, - NumEntries: 1, - MaxEntries: 4, - Height: 0, - }, - Entries: []disklayout.ExtentEntryPair{ - { - Entry: &disklayout.Extent{ - FirstFileBlock: 3, - Length: 3, - StartBlockLo: 6, - }, - Node: nil, - }, - }, - } - - node2 = &disklayout.ExtentNode{ - Header: disklayout.ExtentHeader{ - Magic: disklayout.ExtentMagic, - NumEntries: 1, - MaxEntries: 4, - Height: 1, - }, - Entries: []disklayout.ExtentEntryPair{ - { - Entry: &disklayout.ExtentIdx{ - FirstFileBlock: 3, - ChildBlockLo: 2, - }, - Node: node3, - }, - }, - } - - node1 = &disklayout.ExtentNode{ - Header: disklayout.ExtentHeader{ - Magic: disklayout.ExtentMagic, - NumEntries: 2, - MaxEntries: 4, - Height: 0, - }, - Entries: []disklayout.ExtentEntryPair{ - { - Entry: &disklayout.Extent{ - FirstFileBlock: 0, - Length: 1, - StartBlockLo: 3, - }, - Node: nil, - }, - { - Entry: &disklayout.Extent{ - FirstFileBlock: 1, - Length: 2, - StartBlockLo: 4, - }, - Node: nil, - }, - }, - } - - node0 = &disklayout.ExtentNode{ - Header: disklayout.ExtentHeader{ - Magic: disklayout.ExtentMagic, - NumEntries: 2, - MaxEntries: 4, - Height: 2, - }, - Entries: []disklayout.ExtentEntryPair{ - { - Entry: &disklayout.ExtentIdx{ - FirstFileBlock: 0, - ChildBlockLo: 0, - }, - Node: node1, - }, - { - Entry: &disklayout.ExtentIdx{ - FirstFileBlock: 3, - ChildBlockLo: 1, - }, - Node: node2, - }, - }, - } -) - -// TestExtentReader tests extentReader functionality. We should be able to use -// the file reader like any other io.Reader. -func TestExtentReader(t *testing.T) { - type extentReaderTest struct { - name string - from func(uint64) uint64 - to func(uint64) uint64 - } - - tests := []extentReaderTest{ - { - name: "read first half", - from: beginning, - to: middle, - }, - { - name: "read entire file", - from: beginning, - to: end, - }, - { - name: "read second half", - from: middle, - to: end, - }, - } - - dev, mockExtentFile, want := extentTreeSetUp(t, node0) - size := mockExtentFile.regFile.inode.diskInode.Size() - - for _, test := range tests { - from := test.from(size) - to := test.to(size) - fileReader := mockExtentFile.getFileReader(dev, mockExtentBlkSize, from) - - got := make([]byte, to-from) - if _, err := io.ReadFull(fileReader, got); err != nil { - t.Errorf("file read failed: %v", err) - } - - if diff := cmp.Diff(got, want[from:to]); diff != "" { - t.Errorf("file data mismatch (-want +got):\n%s", diff) - } - } -} - -// TestBuildExtentTree tests the extent tree building logic. -func TestBuildExtentTree(t *testing.T) { - _, mockExtentFile, _ := extentTreeSetUp(t, node0) - - opt := cmpopts.IgnoreUnexported(disklayout.ExtentIdx{}, disklayout.ExtentHeader{}) - if diff := cmp.Diff(&mockExtentFile.root, node0, opt); diff != "" { - t.Errorf("extent tree mismatch (-want +got):\n%s", diff) - } -} - -// extentTreeSetUp writes the passed extent tree to a mock disk as an extent -// tree. It also constucts a mock extent file with the same tree built in it. -// It also writes random data file data and returns it. -func extentTreeSetUp(t *testing.T, root *disklayout.ExtentNode) (io.ReaderAt, *extentFile, []byte) { - t.Helper() - - mockDisk := make([]byte, mockExtentBlkSize*10) - mockExtentFile := &extentFile{ - regFile: regularFile{ - inode: inode{ - diskInode: &disklayout.InodeNew{ - InodeOld: disklayout.InodeOld{ - SizeLo: uint32(mockExtentBlkSize) * getNumPhyBlks(root), - }, - }, - }, - }, - } - - fileData := writeTree(&mockExtentFile.regFile.inode, mockDisk, node0, mockExtentBlkSize) - - r := bytes.NewReader(mockDisk) - if err := mockExtentFile.buildExtTree(r, mockExtentBlkSize); err != nil { - t.Fatalf("inode.buildExtTree failed: %v", err) - } - return r, mockExtentFile, fileData -} - -// writeTree writes the tree represented by `root` to the inode and disk. It -// also writes random file data on disk. -func writeTree(in *inode, disk []byte, root *disklayout.ExtentNode, mockExtentBlkSize uint64) []byte { - rootData := binary.Marshal(nil, binary.LittleEndian, root.Header) - for _, ep := range root.Entries { - rootData = binary.Marshal(rootData, binary.LittleEndian, ep.Entry) - } - - copy(in.diskInode.Data(), rootData) - - var fileData []byte - for _, ep := range root.Entries { - if root.Header.Height == 0 { - fileData = append(fileData, writeRandomFileData(disk, ep.Entry.(*disklayout.Extent))...) - } else { - fileData = append(fileData, writeTreeToDisk(disk, ep)...) - } - } - return fileData -} - -// writeTreeToDisk is the recursive step for writeTree which writes the tree -// on the disk only. Also writes random file data on disk. -func writeTreeToDisk(disk []byte, curNode disklayout.ExtentEntryPair) []byte { - nodeData := binary.Marshal(nil, binary.LittleEndian, curNode.Node.Header) - for _, ep := range curNode.Node.Entries { - nodeData = binary.Marshal(nodeData, binary.LittleEndian, ep.Entry) - } - - copy(disk[curNode.Entry.PhysicalBlock()*mockExtentBlkSize:], nodeData) - - var fileData []byte - for _, ep := range curNode.Node.Entries { - if curNode.Node.Header.Height == 0 { - fileData = append(fileData, writeRandomFileData(disk, ep.Entry.(*disklayout.Extent))...) - } else { - fileData = append(fileData, writeTreeToDisk(disk, ep)...) - } - } - return fileData -} - -// writeRandomFileData writes random bytes to the blocks on disk that the -// passed extent points to. -func writeRandomFileData(disk []byte, ex *disklayout.Extent) []byte { - phyExStartBlk := ex.PhysicalBlock() - phyExStartOff := phyExStartBlk * mockExtentBlkSize - phyExEndOff := phyExStartOff + uint64(ex.Length)*mockExtentBlkSize - rand.Read(disk[phyExStartOff:phyExEndOff]) - return disk[phyExStartOff:phyExEndOff] -} - -// getNumPhyBlks returns the number of physical blocks covered under the node. -func getNumPhyBlks(node *disklayout.ExtentNode) uint32 { - var res uint32 - for _, ep := range node.Entries { - if node.Header.Height == 0 { - res += uint32(ep.Entry.(*disklayout.Extent).Length) - } else { - res += getNumPhyBlks(ep.Node) - } - } - return res -} diff --git a/pkg/sentry/fs/ext/filesystem.go b/pkg/sentry/fs/ext/filesystem.go deleted file mode 100644 index 12aeb5dac..000000000 --- a/pkg/sentry/fs/ext/filesystem.go +++ /dev/null @@ -1,131 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "io" - "sync" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs/ext/disklayout" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// filesystem implements vfs.FilesystemImpl. -type filesystem struct { - // TODO(b/134676337): Remove when all methods have been implemented. - vfs.FilesystemImpl - - vfsfs vfs.Filesystem - - // mu serializes changes to the Dentry tree. - mu sync.RWMutex - - // dev is the io.ReaderAt for the underlying fs device. It does not require - // protection because io.ReaderAt permits concurrent read calls to it. It - // translates to the pread syscall which passes on the read request directly - // to the device driver. Device drivers are intelligent in serving multiple - // concurrent read requests in the optimal order (taking locality into - // consideration). - dev io.ReaderAt - - // inodeCache maps absolute inode numbers to the corresponding Inode struct. - // Inodes should be removed from this once their reference count hits 0. - // - // Protected by mu because every addition and removal from this corresponds to - // a change in the dentry tree. - inodeCache map[uint32]*inode - - // sb represents the filesystem superblock. Immutable after initialization. - sb disklayout.SuperBlock - - // bgs represents all the block group descriptors for the filesystem. - // Immutable after initialization. - bgs []disklayout.BlockGroup -} - -// Compiles only if filesystem implements vfs.FilesystemImpl. -var _ vfs.FilesystemImpl = (*filesystem)(nil) - -// getOrCreateInode gets the inode corresponding to the inode number passed in. -// It creates a new one with the given inode number if one does not exist. -// -// Precondition: must be holding fs.mu. -func (fs *filesystem) getOrCreateInode(ctx context.Context, inodeNum uint32) (*inode, error) { - if in, ok := fs.inodeCache[inodeNum]; ok { - return in, nil - } - - in, err := newInode(ctx, fs.dev, fs.sb, fs.bgs, inodeNum) - if err != nil { - return nil, err - } - - fs.inodeCache[inodeNum] = in - return in, nil -} - -// Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { -} - -// Sync implements vfs.FilesystemImpl.Sync. -func (fs *filesystem) Sync(ctx context.Context) error { - // This is a readonly filesystem for now. - return nil -} - -// The vfs.FilesystemImpl functions below return EROFS because their respective -// man pages say that EROFS must be returned if the path resolves to a file on -// a read-only filesystem. - -// TODO(b/134676337): Implement path traversal and return EROFS only if the -// path resolves to a Dentry within ext fs. - -// MkdirAt implements vfs.FilesystemImpl.MkdirAt. -func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { - return syserror.EROFS -} - -// MknodAt implements vfs.FilesystemImpl.MknodAt. -func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { - return syserror.EROFS -} - -// RenameAt implements vfs.FilesystemImpl.RenameAt. -func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry, opts vfs.RenameOptions) error { - return syserror.EROFS -} - -// RmdirAt implements vfs.FilesystemImpl.RmdirAt. -func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { - return syserror.EROFS -} - -// SetStatAt implements vfs.FilesystemImpl.SetStatAt. -func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { - return syserror.EROFS -} - -// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. -func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { - return syserror.EROFS -} - -// UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. -func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { - return syserror.EROFS -} diff --git a/pkg/sentry/fs/ext/inline_file.go b/pkg/sentry/fs/ext/inline_file.go deleted file mode 100644 index b9adfe548..000000000 --- a/pkg/sentry/fs/ext/inline_file.go +++ /dev/null @@ -1,66 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "io" -) - -// inlineFile is a type of regular file. All the data here is stored in the -// inode.Data() array. -type inlineFile struct { - regFile regularFile -} - -// Compiles only if inlineFile implements fileReader. -var _ fileReader = (*inlineFile)(nil) - -// getFileReader implements fileReader.getFileReader. -func (f *inlineFile) getFileReader(_ io.ReaderAt, _ uint64, offset uint64) io.Reader { - diskInode := f.regFile.inode.diskInode - return &inlineReader{offset: offset, data: diskInode.Data()[:diskInode.Size()]} -} - -// newInlineFile is the inlineFile constructor. -func newInlineFile(regFile regularFile) *inlineFile { - file := &inlineFile{regFile: regFile} - file.regFile.impl = file - return file -} - -// inlineReader implements io.Reader which can read the underlying data. This -// is not thread safe. -type inlineReader struct { - offset uint64 - data []byte -} - -// Compiles only if inlineReader implements io.Reader. -var _ io.Reader = (*inlineReader)(nil) - -// Read implements io.Reader.Read. -func (r *inlineReader) Read(dst []byte) (int, error) { - if len(dst) == 0 { - return 0, nil - } - - if int(r.offset) >= len(r.data) { - return 0, io.EOF - } - - n := copy(dst, r.data[r.offset:]) - r.offset += uint64(n) - return n, nil -} diff --git a/pkg/sentry/fs/ext/inode.go b/pkg/sentry/fs/ext/inode.go deleted file mode 100644 index 00e022953..000000000 --- a/pkg/sentry/fs/ext/inode.go +++ /dev/null @@ -1,151 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "io" - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs/ext/disklayout" - "gvisor.dev/gvisor/pkg/syserror" -) - -// inode represents an ext inode. -// -// inode uses the same inheritance pattern that pkg/sentry/vfs structures use. -// This has been done to increase memory locality. -// -// Implementations: -// inode -- -// |-- pipe -// |-- dir -// |-- symlink -// |-- regular-- -// |-- extent file -// |-- block map file -// |-- inline file -type inode struct { - // refs is a reference count. refs is accessed using atomic memory operations. - refs int64 - - // inodeNum is the inode number of this inode on disk. This is used to - // identify inodes within the ext filesystem. - inodeNum uint32 - - // diskInode gives us access to the inode struct on disk. Immutable. - diskInode disklayout.Inode - - // This is immutable. The first field of the implementations must have inode - // as the first field to ensure temporality. - impl interface{} -} - -// incRef increments the inode ref count. -func (in *inode) incRef() { - atomic.AddInt64(&in.refs, 1) -} - -// tryIncRef tries to increment the ref count. Returns true if successful. -func (in *inode) tryIncRef() bool { - for { - refs := atomic.LoadInt64(&in.refs) - if refs == 0 { - return false - } - if atomic.CompareAndSwapInt64(&in.refs, refs, refs+1) { - return true - } - } -} - -// decRef decrements the inode ref count and releases the inode resources if -// the ref count hits 0. -// -// Precondition: Must have locked fs.mu. -func (in *inode) decRef(fs *filesystem) { - if refs := atomic.AddInt64(&in.refs, -1); refs == 0 { - delete(fs.inodeCache, in.inodeNum) - } else if refs < 0 { - panic("ext.inode.decRef() called without holding a reference") - } -} - -// newInode is the inode constructor. Reads the inode off disk. Identifies -// inodes based on the absolute inode number on disk. -func newInode(ctx context.Context, dev io.ReaderAt, sb disklayout.SuperBlock, bgs []disklayout.BlockGroup, inodeNum uint32) (*inode, error) { - if inodeNum == 0 { - panic("inode number 0 on ext filesystems is not possible") - } - - inodeRecordSize := sb.InodeSize() - var diskInode disklayout.Inode - if inodeRecordSize == disklayout.OldInodeSize { - diskInode = &disklayout.InodeOld{} - } else { - diskInode = &disklayout.InodeNew{} - } - - // Calculate where the inode is actually placed. - inodesPerGrp := sb.InodesPerGroup() - blkSize := sb.BlockSize() - inodeTableOff := bgs[getBGNum(inodeNum, inodesPerGrp)].InodeTable() * blkSize - inodeOff := inodeTableOff + uint64(uint32(inodeRecordSize)*getBGOff(inodeNum, inodesPerGrp)) - - if err := readFromDisk(dev, int64(inodeOff), diskInode); err != nil { - return nil, err - } - - // Build the inode based on its type. - inode := inode{ - refs: 1, - inodeNum: inodeNum, - diskInode: diskInode, - } - - switch diskInode.Mode().FileType() { - case linux.ModeSymlink: - f, err := newSymlink(dev, blkSize, inode) - if err != nil { - return nil, err - } - return &f.inode, nil - case linux.ModeRegular: - f, err := newRegularFile(dev, blkSize, inode) - if err != nil { - return nil, err - } - return &f.inode, nil - case linux.ModeDirectory: - return &newDirectroy(inode).inode, nil - case linux.ModeNamedPipe: - return &newNamedPipe(ctx, inode).inode, nil - default: - // TODO(b/134676337): Return appropriate errors for sockets and devices. - return nil, syserror.EINVAL - } -} - -// getBGNum returns the block group number that a given inode belongs to. -func getBGNum(inodeNum uint32, inodesPerGrp uint32) uint32 { - return (inodeNum - 1) / inodesPerGrp -} - -// getBGOff returns the offset at which the given inode lives in the block -// group's inode table, i.e. the index of the inode in the inode table. -func getBGOff(inodeNum uint32, inodesPerGrp uint32) uint32 { - return (inodeNum - 1) % inodesPerGrp -} diff --git a/pkg/sentry/fs/ext/named_pipe.go b/pkg/sentry/fs/ext/named_pipe.go deleted file mode 100644 index 0f3af1b53..000000000 --- a/pkg/sentry/fs/ext/named_pipe.go +++ /dev/null @@ -1,40 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" - "gvisor.dev/gvisor/pkg/sentry/usermem" -) - -// namedPipe represents a named pipe inode. It is currently just a wrapper -// around pkg/sentry/kernel/pipe. -type namedPipe struct { - inode inode - - p *pipe.Pipe - inodeOps fs.InodeOperations -} - -// newNamedPipe is the namedPipe constructor. -func newNamedPipe(ctx context.Context, inode inode) *namedPipe { - file := &namedPipe{inode: inode} - file.inode.impl = file - file.p = pipe.NewPipe(ctx, true /* isNamed */, pipe.DefaultPipeSize, usermem.PageSize) - file.inodeOps = pipe.NewInodeOperations(ctx, fs.FilePermsFromMode(file.inode.diskInode.Mode()), file.p) - return file -} diff --git a/pkg/sentry/fs/ext/regular_file.go b/pkg/sentry/fs/ext/regular_file.go deleted file mode 100644 index b48f61795..000000000 --- a/pkg/sentry/fs/ext/regular_file.go +++ /dev/null @@ -1,85 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "io" -) - -// fileReader is used to abstact away the complexity of how the file data is -// stored under the hood. Provides a method to get a file reader which can be -// used to read file data without worrying about how it is organized on disk. -type fileReader interface { - - // getFileReader returns a Reader implementation which can be used to read a - // file. It abstracts away the complexity of how the file is actually - // organized on disk. The reader is initialized with the passed offset. - // - // This reader is not meant to be retained across Read operations as it needs - // to be reinitialized with the correct offset for every Read. - getFileReader(dev io.ReaderAt, blkSize uint64, offset uint64) io.Reader -} - -// regularFile represents a regular file's inode. This too follows the -// inheritance pattern prevelant in the vfs layer described in -// pkg/sentry/vfs/README.md. -type regularFile struct { - inode inode - - // This is immutable. The first field of fileReader implementations must be - // regularFile to ensure temporality. - impl fileReader -} - -// newRegularFile is the regularFile constructor. It figures out what kind of -// file this is and initializes the fileReader. -func newRegularFile(dev io.ReaderAt, blkSize uint64, inode inode) (*regularFile, error) { - regFile := regularFile{ - inode: inode, - } - - inodeFlags := inode.diskInode.Flags() - - if inodeFlags.Extents { - file, err := newExtentFile(dev, blkSize, regFile) - if err != nil { - return nil, err - } - - file.regFile.inode.impl = &file.regFile - return &file.regFile, nil - } - - if inodeFlags.Inline { - if inode.diskInode.Size() > 60 { - panic("ext fs: inline file larger than 60 bytes") - } - - file := newInlineFile(regFile) - file.regFile.inode.impl = &file.regFile - return &file.regFile, nil - } - - file, err := newBlockMapFile(blkSize, regFile) - if err != nil { - return nil, err - } - file.regFile.inode.impl = &file.regFile - return &file.regFile, nil -} - -func (f *regularFile) blksUsed(blkSize uint64) uint64 { - return (f.inode.diskInode.Size() + blkSize - 1) / blkSize -} diff --git a/pkg/sentry/fs/ext/symlink.go b/pkg/sentry/fs/ext/symlink.go deleted file mode 100644 index 6a55c1a7b..000000000 --- a/pkg/sentry/fs/ext/symlink.go +++ /dev/null @@ -1,57 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "io" - - "gvisor.dev/gvisor/pkg/syserror" -) - -// symlink represents a symlink inode. -type symlink struct { - inode inode - target string // immutable -} - -// newSymlink is the symlink constructor. It reads out the symlink target from -// the inode (however it might have been stored). -func newSymlink(dev io.ReaderAt, blkSize uint64, inode inode) (*symlink, error) { - var file *symlink - var link []byte - - // If the symlink target is lesser than 60 bytes, its stores in inode.Data(). - // Otherwise either extents or block maps will be used to store the link. - size := inode.diskInode.Size() - if size < 60 { - link = inode.diskInode.Data()[:size] - } else { - // Create a regular file out of this inode and read out the target. - regFile, err := newRegularFile(dev, blkSize, inode) - if err != nil { - return nil, err - } - - link = make([]byte, size) - reader := regFile.impl.getFileReader(dev, blkSize, 0) - if _, err := io.ReadFull(reader, link); err != nil { - return nil, syserror.EIO - } - } - - file = &symlink{inode: inode, target: string(link)} - file.inode.impl = file - return file, nil -} diff --git a/pkg/sentry/fs/ext/utils.go b/pkg/sentry/fs/ext/utils.go deleted file mode 100644 index 3d89d664d..000000000 --- a/pkg/sentry/fs/ext/utils.go +++ /dev/null @@ -1,94 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ext - -import ( - "io" - - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/fs/ext/disklayout" - "gvisor.dev/gvisor/pkg/syserror" -) - -// readFromDisk performs a binary read from disk into the given struct from -// the absolute offset provided. -func readFromDisk(dev io.ReaderAt, abOff int64, v interface{}) error { - n := binary.Size(v) - buf := make([]byte, n) - if read, _ := dev.ReadAt(buf, abOff); read < int(n) { - return syserror.EIO - } - - binary.Unmarshal(buf, binary.LittleEndian, v) - return nil -} - -// readSuperBlock reads the SuperBlock from block group 0 in the underlying -// device. There are three versions of the superblock. This function identifies -// and returns the correct version. -func readSuperBlock(dev io.ReaderAt) (disklayout.SuperBlock, error) { - var sb disklayout.SuperBlock = &disklayout.SuperBlockOld{} - if err := readFromDisk(dev, disklayout.SbOffset, sb); err != nil { - return nil, err - } - if sb.Revision() == disklayout.OldRev { - return sb, nil - } - - sb = &disklayout.SuperBlock32Bit{} - if err := readFromDisk(dev, disklayout.SbOffset, sb); err != nil { - return nil, err - } - if !sb.IncompatibleFeatures().Is64Bit { - return sb, nil - } - - sb = &disklayout.SuperBlock64Bit{} - if err := readFromDisk(dev, disklayout.SbOffset, sb); err != nil { - return nil, err - } - return sb, nil -} - -// blockGroupsCount returns the number of block groups in the ext fs. -func blockGroupsCount(sb disklayout.SuperBlock) uint64 { - blocksCount := sb.BlocksCount() - blocksPerGroup := uint64(sb.BlocksPerGroup()) - - // Round up the result. float64 can compromise precision so do it manually. - return (blocksCount + blocksPerGroup - 1) / blocksPerGroup -} - -// readBlockGroups reads the block group descriptor table from block group 0 in -// the underlying device. -func readBlockGroups(dev io.ReaderAt, sb disklayout.SuperBlock) ([]disklayout.BlockGroup, error) { - bgCount := blockGroupsCount(sb) - bgdSize := uint64(sb.BgDescSize()) - is64Bit := sb.IncompatibleFeatures().Is64Bit - bgds := make([]disklayout.BlockGroup, bgCount) - - for i, off := uint64(0), uint64(sb.FirstDataBlock()+1)*sb.BlockSize(); i < bgCount; i, off = i+1, off+bgdSize { - if is64Bit { - bgds[i] = &disklayout.BlockGroup64Bit{} - } else { - bgds[i] = &disklayout.BlockGroup32Bit{} - } - - if err := readFromDisk(dev, int64(off), bgds[i]); err != nil { - return nil, err - } - } - return bgds, nil -} diff --git a/pkg/sentry/fs/fdpipe/BUILD b/pkg/sentry/fs/fdpipe/BUILD deleted file mode 100644 index bf00b9c09..000000000 --- a/pkg/sentry/fs/fdpipe/BUILD +++ /dev/null @@ -1,48 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_stateify:defs.bzl", "go_library", "go_test") - -go_library( - name = "fdpipe", - srcs = [ - "pipe.go", - "pipe_opener.go", - "pipe_state.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/fdpipe", - imports = ["gvisor.dev/gvisor/pkg/sentry/fs"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/fd", - "//pkg/fdnotifier", - "//pkg/log", - "//pkg/secio", - "//pkg/sentry/context", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/safemem", - "//pkg/sentry/usermem", - "//pkg/syserror", - "//pkg/waiter", - ], -) - -go_test( - name = "fdpipe_test", - size = "small", - srcs = [ - "pipe_opener_test.go", - "pipe_test.go", - ], - embed = [":fdpipe"], - deps = [ - "//pkg/fd", - "//pkg/fdnotifier", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", - "//pkg/sentry/fs", - "//pkg/sentry/usermem", - "//pkg/syserror", - "@com_github_google_uuid//:go_default_library", - ], -) diff --git a/pkg/sentry/fs/fdpipe/fdpipe_state_autogen.go b/pkg/sentry/fs/fdpipe/fdpipe_state_autogen.go new file mode 100755 index 000000000..38c1ed916 --- /dev/null +++ b/pkg/sentry/fs/fdpipe/fdpipe_state_autogen.go @@ -0,0 +1,27 @@ +// automatically generated by stateify. + +package fdpipe + +import ( + "gvisor.dev/gvisor/pkg/state" + "gvisor.dev/gvisor/pkg/sentry/fs" +) + +func (x *pipeOperations) save(m state.Map) { + x.beforeSave() + var flags fs.FileFlags = x.saveFlags() + m.SaveValue("flags", flags) + m.Save("opener", &x.opener) + m.Save("readAheadBuffer", &x.readAheadBuffer) +} + +func (x *pipeOperations) load(m state.Map) { + m.LoadWait("opener", &x.opener) + m.Load("readAheadBuffer", &x.readAheadBuffer) + m.LoadValue("flags", new(fs.FileFlags), func(y interface{}) { x.loadFlags(y.(fs.FileFlags)) }) + m.AfterLoad(x.afterLoad) +} + +func init() { + state.Register("fdpipe.pipeOperations", (*pipeOperations)(nil), state.Fns{Save: (*pipeOperations).save, Load: (*pipeOperations).load}) +} diff --git a/pkg/sentry/fs/fdpipe/pipe_opener_test.go b/pkg/sentry/fs/fdpipe/pipe_opener_test.go deleted file mode 100644 index 8e4d839e1..000000000 --- a/pkg/sentry/fs/fdpipe/pipe_opener_test.go +++ /dev/null @@ -1,522 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fdpipe - -import ( - "bytes" - "fmt" - "io" - "os" - "path" - "syscall" - "testing" - "time" - - "github.com/google/uuid" - "gvisor.dev/gvisor/pkg/fd" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/syserror" -) - -type hostOpener struct { - name string -} - -func (h *hostOpener) NonBlockingOpen(_ context.Context, p fs.PermMask) (*fd.FD, error) { - var flags int - switch { - case p.Read && p.Write: - flags = syscall.O_RDWR - case p.Write: - flags = syscall.O_WRONLY - case p.Read: - flags = syscall.O_RDONLY - default: - return nil, syscall.EINVAL - } - f, err := syscall.Open(h.name, flags|syscall.O_NONBLOCK, 0666) - if err != nil { - return nil, err - } - return fd.New(f), nil -} - -func pipename() string { - return fmt.Sprintf(path.Join(os.TempDir(), "test-named-pipe-%s"), uuid.New()) -} - -func mkpipe(name string) error { - return syscall.Mknod(name, syscall.S_IFIFO|0666, 0) -} - -func TestTryOpen(t *testing.T) { - for _, test := range []struct { - // desc is the test's description. - desc string - - // makePipe is true if the test case should create the pipe. - makePipe bool - - // flags are the fs.FileFlags used to open the pipe. - flags fs.FileFlags - - // expectFile is true if a fs.File is expected. - expectFile bool - - // err is the expected error - err error - }{ - { - desc: "FileFlags lacking Read and Write are invalid", - makePipe: false, - flags: fs.FileFlags{}, /* bogus */ - expectFile: false, - err: syscall.EINVAL, - }, - { - desc: "NonBlocking Read only error returns immediately", - makePipe: false, /* causes the error */ - flags: fs.FileFlags{Read: true, NonBlocking: true}, - expectFile: false, - err: syscall.ENOENT, - }, - { - desc: "NonBlocking Read only success returns immediately", - makePipe: true, - flags: fs.FileFlags{Read: true, NonBlocking: true}, - expectFile: true, - err: nil, - }, - { - desc: "NonBlocking Write only error returns immediately", - makePipe: false, /* causes the error */ - flags: fs.FileFlags{Write: true, NonBlocking: true}, - expectFile: false, - err: syscall.ENOENT, - }, - { - desc: "NonBlocking Write only no reader error returns immediately", - makePipe: true, - flags: fs.FileFlags{Write: true, NonBlocking: true}, - expectFile: false, - err: syscall.ENXIO, - }, - { - desc: "ReadWrite error returns immediately", - makePipe: false, /* causes the error */ - flags: fs.FileFlags{Read: true, Write: true}, - expectFile: false, - err: syscall.ENOENT, - }, - { - desc: "ReadWrite returns immediately", - makePipe: true, - flags: fs.FileFlags{Read: true, Write: true}, - expectFile: true, - err: nil, - }, - { - desc: "Blocking Write only returns open error", - makePipe: false, /* causes the error */ - flags: fs.FileFlags{Write: true}, - expectFile: false, - err: syscall.ENOENT, /* from bogus perms */ - }, - { - desc: "Blocking Read only returns open error", - makePipe: false, /* causes the error */ - flags: fs.FileFlags{Read: true}, - expectFile: false, - err: syscall.ENOENT, - }, - { - desc: "Blocking Write only returns with syserror.ErrWouldBlock", - makePipe: true, - flags: fs.FileFlags{Write: true}, - expectFile: false, - err: syserror.ErrWouldBlock, - }, - { - desc: "Blocking Read only returns with syserror.ErrWouldBlock", - makePipe: true, - flags: fs.FileFlags{Read: true}, - expectFile: false, - err: syserror.ErrWouldBlock, - }, - } { - name := pipename() - if test.makePipe { - // Create the pipe. We do this per-test case to keep tests independent. - if err := mkpipe(name); err != nil { - t.Errorf("%s: failed to make host pipe: %v", test.desc, err) - continue - } - defer syscall.Unlink(name) - } - - // Use a host opener to keep things simple. - opener := &hostOpener{name: name} - - pipeOpenState := &pipeOpenState{} - ctx := contexttest.Context(t) - pipeOps, err := pipeOpenState.TryOpen(ctx, opener, test.flags) - if unwrapError(err) != test.err { - t.Errorf("%s: got error %v, want %v", test.desc, err, test.err) - if pipeOps != nil { - // Cleanup the state of the pipe, and remove the fd from the - // fdnotifier. Sadly this needed to maintain the correctness - // of other tests because the fdnotifier is global. - pipeOps.Release() - } - continue - } - if (pipeOps != nil) != test.expectFile { - t.Errorf("%s: got non-nil file %v, want %v", test.desc, pipeOps != nil, test.expectFile) - } - if pipeOps != nil { - // Same as above. - pipeOps.Release() - } - } -} - -func TestPipeOpenUnblocksEventually(t *testing.T) { - for _, test := range []struct { - // desc is the test's description. - desc string - - // partnerIsReader is true if the goroutine opening the same pipe as the test case - // should open the pipe read only. Otherwise write only. This also means that the - // test case will open the pipe in the opposite way. - partnerIsReader bool - - // partnerIsBlocking is true if the goroutine opening the same pipe as the test case - // should do so without the O_NONBLOCK flag, otherwise opens the pipe with O_NONBLOCK - // until ENXIO is not returned. - partnerIsBlocking bool - }{ - { - desc: "Blocking Read with blocking writer partner opens eventually", - partnerIsReader: false, - partnerIsBlocking: true, - }, - { - desc: "Blocking Write with blocking reader partner opens eventually", - partnerIsReader: true, - partnerIsBlocking: true, - }, - { - desc: "Blocking Read with non-blocking writer partner opens eventually", - partnerIsReader: false, - partnerIsBlocking: false, - }, - { - desc: "Blocking Write with non-blocking reader partner opens eventually", - partnerIsReader: true, - partnerIsBlocking: false, - }, - } { - // Create the pipe. We do this per-test case to keep tests independent. - name := pipename() - if err := mkpipe(name); err != nil { - t.Errorf("%s: failed to make host pipe: %v", test.desc, err) - continue - } - defer syscall.Unlink(name) - - // Spawn the partner. - type fderr struct { - fd int - err error - } - errch := make(chan fderr, 1) - go func() { - var flags int - if test.partnerIsReader { - flags = syscall.O_RDONLY - } else { - flags = syscall.O_WRONLY - } - if test.partnerIsBlocking { - fd, err := syscall.Open(name, flags, 0666) - errch <- fderr{fd: fd, err: err} - } else { - var fd int - err := error(syscall.ENXIO) - for err == syscall.ENXIO { - fd, err = syscall.Open(name, flags|syscall.O_NONBLOCK, 0666) - time.Sleep(1 * time.Second) - } - errch <- fderr{fd: fd, err: err} - } - }() - - // Setup file flags for either a read only or write only open. - flags := fs.FileFlags{ - Read: !test.partnerIsReader, - Write: test.partnerIsReader, - } - - // Open the pipe in a blocking way, which should succeed eventually. - opener := &hostOpener{name: name} - ctx := contexttest.Context(t) - pipeOps, err := Open(ctx, opener, flags) - if pipeOps != nil { - // Same as TestTryOpen. - pipeOps.Release() - } - - // Check that the partner opened the file successfully. - e := <-errch - if e.err != nil { - t.Errorf("%s: partner got error %v, wanted nil", test.desc, e.err) - continue - } - // If so, then close the partner fd to avoid leaking an fd. - syscall.Close(e.fd) - - // Check that our blocking open was successful. - if err != nil { - t.Errorf("%s: blocking open got error %v, wanted nil", test.desc, err) - continue - } - if pipeOps == nil { - t.Errorf("%s: blocking open got nil file, wanted non-nil", test.desc) - continue - } - } -} - -func TestCopiedReadAheadBuffer(t *testing.T) { - // Create the pipe. - name := pipename() - if err := mkpipe(name); err != nil { - t.Fatalf("failed to make host pipe: %v", err) - } - defer syscall.Unlink(name) - - // We're taking advantage of the fact that pipes opened read only always return - // success, but internally they are not deemed "opened" until we're sure that - // another writer comes along. This means we can open the same pipe write only - // with no problems + write to it, given that opener.Open already tried to open - // the pipe RDONLY and succeeded, which we know happened if TryOpen returns - // syserror.ErrwouldBlock. - // - // This simulates the open(RDONLY) <-> open(WRONLY)+write race we care about, but - // does not cause our test to be racy (which would be terrible). - opener := &hostOpener{name: name} - pipeOpenState := &pipeOpenState{} - ctx := contexttest.Context(t) - pipeOps, err := pipeOpenState.TryOpen(ctx, opener, fs.FileFlags{Read: true}) - if pipeOps != nil { - pipeOps.Release() - t.Fatalf("open(%s, %o) got file, want nil", name, syscall.O_RDONLY) - } - if err != syserror.ErrWouldBlock { - t.Fatalf("open(%s, %o) got error %v, want %v", name, syscall.O_RDONLY, err, syserror.ErrWouldBlock) - } - - // Then open the same pipe write only and write some bytes to it. The next - // time we try to open the pipe read only again via the pipeOpenState, we should - // succeed and buffer some of the bytes written. - fd, err := syscall.Open(name, syscall.O_WRONLY, 0666) - if err != nil { - t.Fatalf("open(%s, %o) got error %v, want nil", name, syscall.O_WRONLY, err) - } - defer syscall.Close(fd) - - data := []byte("hello") - if n, err := syscall.Write(fd, data); n != len(data) || err != nil { - t.Fatalf("write(%v) got (%d, %v), want (%d, nil)", data, n, err, len(data)) - } - - // Try the read again, knowing that it should succeed this time. - pipeOps, err = pipeOpenState.TryOpen(ctx, opener, fs.FileFlags{Read: true}) - if pipeOps == nil { - t.Fatalf("open(%s, %o) got nil file, want not nil", name, syscall.O_RDONLY) - } - defer pipeOps.Release() - - if err != nil { - t.Fatalf("open(%s, %o) got error %v, want nil", name, syscall.O_RDONLY, err) - } - - inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{ - Type: fs.Pipe, - }) - file := fs.NewFile(ctx, fs.NewDirent(ctx, inode, "pipe"), fs.FileFlags{Read: true}, pipeOps) - - // Check that the file we opened points to a pipe with a non-empty read ahead buffer. - bufsize := len(pipeOps.readAheadBuffer) - if bufsize != 1 { - t.Fatalf("read ahead buffer got %d bytes, want %d", bufsize, 1) - } - - // Now for the final test, try to read everything in, expecting to get back all of - // the bytes that were written at once. Note that in the wild there is no atomic - // read size so expecting to get all bytes from a single writer when there are - // multiple readers is a bad expectation. - buf := make([]byte, len(data)) - ioseq := usermem.BytesIOSequence(buf) - n, err := pipeOps.Read(ctx, file, ioseq, 0) - if err != nil { - t.Fatalf("read request got error %v, want nil", err) - } - if n != int64(len(data)) { - t.Fatalf("read request got %d bytes, want %d", n, len(data)) - } - if !bytes.Equal(buf, data) { - t.Errorf("read request got bytes [%v], want [%v]", buf, data) - } -} - -func TestPipeHangup(t *testing.T) { - for _, test := range []struct { - // desc is the test's description. - desc string - - // flags control how we open our end of the pipe and must be read - // only or write only. They also dicate how a coordinating partner - // fd is opened, which is their inverse (read only -> write only, etc). - flags fs.FileFlags - - // hangupSelf if true causes the test case to close our end of the pipe - // and causes hangup errors to be asserted on our coordinating partner's - // fd. If hangupSelf is false, then our partner's fd is closed and the - // hangup errors are expected on our end of the pipe. - hangupSelf bool - }{ - { - desc: "Read only gets hangup error", - flags: fs.FileFlags{Read: true}, - }, - { - desc: "Write only gets hangup error", - flags: fs.FileFlags{Write: true}, - }, - { - desc: "Read only generates hangup error", - flags: fs.FileFlags{Read: true}, - hangupSelf: true, - }, - { - desc: "Write only generates hangup error", - flags: fs.FileFlags{Write: true}, - hangupSelf: true, - }, - } { - if test.flags.Read == test.flags.Write { - t.Errorf("%s: test requires a single reader or writer", test.desc) - continue - } - - // Create the pipe. We do this per-test case to keep tests independent. - name := pipename() - if err := mkpipe(name); err != nil { - t.Errorf("%s: failed to make host pipe: %v", test.desc, err) - continue - } - defer syscall.Unlink(name) - - // Fire off a partner routine which tries to open the same pipe blocking, - // which will synchronize with us. The channel allows us to get back the - // fd once we expect this partner routine to succeed, so we can manifest - // hangup events more directly. - fdchan := make(chan int, 1) - go func() { - // Be explicit about the flags to protect the test from - // misconfiguration. - var flags int - if test.flags.Read { - flags = syscall.O_WRONLY - } else { - flags = syscall.O_RDONLY - } - fd, err := syscall.Open(name, flags, 0666) - if err != nil { - t.Logf("Open(%q, %o, 0666) partner failed: %v", name, flags, err) - } - fdchan <- fd - }() - - // Open our end in a blocking way to ensure that we coordinate. - opener := &hostOpener{name: name} - ctx := contexttest.Context(t) - pipeOps, err := Open(ctx, opener, test.flags) - if err != nil { - t.Errorf("%s: Open got error %v, want nil", test.desc, err) - continue - } - // Don't defer file.DecRef here because that causes the hangup we're - // trying to test for. - - // Expect the partner routine to have coordinated with us and get back - // its open fd. - f := <-fdchan - if f < 0 { - t.Errorf("%s: partner routine got fd %d, want > 0", test.desc, f) - pipeOps.Release() - continue - } - - if test.hangupSelf { - // Hangup self and assert that our partner got the expected hangup - // error. - pipeOps.Release() - - if test.flags.Read { - // Partner is writer. - assertWriterHungup(t, test.desc, fd.NewReadWriter(f)) - } else { - // Partner is reader. - assertReaderHungup(t, test.desc, fd.NewReadWriter(f)) - } - } else { - // Hangup our partner and expect us to get the hangup error. - syscall.Close(f) - defer pipeOps.Release() - - if test.flags.Read { - assertReaderHungup(t, test.desc, pipeOps.(*pipeOperations).file) - } else { - assertWriterHungup(t, test.desc, pipeOps.(*pipeOperations).file) - } - } - } -} - -func assertReaderHungup(t *testing.T, desc string, reader io.Reader) bool { - // Drain the pipe completely, it might have crap in it, but expect EOF eventually. - var err error - for err == nil { - _, err = reader.Read(make([]byte, 10)) - } - if err != io.EOF { - t.Errorf("%s: read from self after hangup got error %v, want %v", desc, err, io.EOF) - return false - } - return true -} - -func assertWriterHungup(t *testing.T, desc string, writer io.Writer) bool { - if _, err := writer.Write([]byte("hello")); unwrapError(err) != syscall.EPIPE { - t.Errorf("%s: write to self after hangup got error %v, want %v", desc, err, syscall.EPIPE) - return false - } - return true -} diff --git a/pkg/sentry/fs/fdpipe/pipe_test.go b/pkg/sentry/fs/fdpipe/pipe_test.go deleted file mode 100644 index 69abc1e71..000000000 --- a/pkg/sentry/fs/fdpipe/pipe_test.go +++ /dev/null @@ -1,505 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fdpipe - -import ( - "bytes" - "io" - "os" - "syscall" - "testing" - - "gvisor.dev/gvisor/pkg/fd" - "gvisor.dev/gvisor/pkg/fdnotifier" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/syserror" -) - -func singlePipeFD() (int, error) { - fds := make([]int, 2) - if err := syscall.Pipe(fds); err != nil { - return -1, err - } - syscall.Close(fds[1]) - return fds[0], nil -} - -func singleDirFD() (int, error) { - return syscall.Open(os.TempDir(), syscall.O_RDONLY, 0666) -} - -func mockPipeDirent(t *testing.T) *fs.Dirent { - ctx := contexttest.Context(t) - node := fs.NewMockInodeOperations(ctx) - node.UAttr = fs.UnstableAttr{ - Perms: fs.FilePermissions{ - User: fs.PermMask{Read: true, Write: true}, - }, - } - inode := fs.NewInode(ctx, node, fs.NewMockMountSource(nil), fs.StableAttr{ - Type: fs.Pipe, - BlockSize: usermem.PageSize, - }) - return fs.NewDirent(ctx, inode, "") -} - -func TestNewPipe(t *testing.T) { - for _, test := range []struct { - // desc is the test's description. - desc string - - // getfd generates the fd to pass to newPipeOperations. - getfd func() (int, error) - - // flags are the fs.FileFlags passed to newPipeOperations. - flags fs.FileFlags - - // readAheadBuffer is the buffer passed to newPipeOperations. - readAheadBuffer []byte - - // err is the expected error. - err error - }{ - { - desc: "Cannot make new pipe from bad fd", - getfd: func() (int, error) { return -1, nil }, - err: syscall.EINVAL, - }, - { - desc: "Cannot make new pipe from non-pipe fd", - getfd: singleDirFD, - err: syscall.EINVAL, - }, - { - desc: "Can make new pipe from pipe fd", - getfd: singlePipeFD, - flags: fs.FileFlags{Read: true}, - readAheadBuffer: []byte("hello"), - }, - } { - gfd, err := test.getfd() - if err != nil { - t.Errorf("%s: getfd got (%d, %v), want (fd, nil)", test.desc, gfd, err) - continue - } - f := fd.New(gfd) - - p, err := newPipeOperations(contexttest.Context(t), nil, test.flags, f, test.readAheadBuffer) - if p != nil { - // This is necessary to remove the fd from the global fd notifier. - defer p.Release() - } else { - // If there is no p to DecRef on, because newPipeOperations failed, then the - // file still needs to be closed. - defer f.Close() - } - - if err != test.err { - t.Errorf("%s: got error %v, want %v", test.desc, err, test.err) - continue - } - // Check the state of the pipe given that it was successfully opened. - if err == nil { - if p == nil { - t.Errorf("%s: got nil pipe and nil error, want (pipe, nil)", test.desc) - continue - } - if flags := p.flags; test.flags != flags { - t.Errorf("%s: got file flags %s, want %s", test.desc, flags, test.flags) - continue - } - if len(test.readAheadBuffer) != len(p.readAheadBuffer) { - t.Errorf("%s: got read ahead buffer length %d, want %d", test.desc, len(p.readAheadBuffer), len(test.readAheadBuffer)) - continue - } - fileFlags, _, errno := syscall.Syscall(syscall.SYS_FCNTL, uintptr(p.file.FD()), syscall.F_GETFL, 0) - if errno != 0 { - t.Errorf("%s: failed to get file flags for fd %d, got %v, want 0", test.desc, p.file.FD(), errno) - continue - } - if fileFlags&syscall.O_NONBLOCK == 0 { - t.Errorf("%s: pipe is blocking, expected non-blocking", test.desc) - continue - } - if !fdnotifier.HasFD(int32(f.FD())) { - t.Errorf("%s: pipe fd %d is not registered for events", test.desc, f.FD) - } - } - } -} - -func TestPipeDestruction(t *testing.T) { - fds := make([]int, 2) - if err := syscall.Pipe(fds); err != nil { - t.Fatalf("failed to create pipes: got %v, want nil", err) - } - f := fd.New(fds[0]) - - // We don't care about the other end, just use the read end. - syscall.Close(fds[1]) - - // Test the read end, but it doesn't really matter which. - p, err := newPipeOperations(contexttest.Context(t), nil, fs.FileFlags{Read: true}, f, nil) - if err != nil { - f.Close() - t.Fatalf("newPipeOperations got error %v, want nil", err) - } - // Drop our only reference, which should trigger the destructor. - p.Release() - - if fdnotifier.HasFD(int32(fds[0])) { - t.Fatalf("after DecRef fdnotifier has fd %d, want no longer registered", fds[0]) - } - if p.file != nil { - t.Errorf("after DecRef got file, want nil") - } -} - -type Seek struct{} - -type ReadDir struct{} - -type Writev struct { - Src usermem.IOSequence -} - -type Readv struct { - Dst usermem.IOSequence -} - -type Fsync struct{} - -func TestPipeRequest(t *testing.T) { - for _, test := range []struct { - // desc is the test's description. - desc string - - // request to execute. - context interface{} - - // flags determines whether to use the read or write end - // of the pipe, for this test it can only be Read or Write. - flags fs.FileFlags - - // keepOpenPartner if false closes the other end of the pipe, - // otherwise this is delayed until the end of the test. - keepOpenPartner bool - - // expected error - err error - }{ - { - desc: "ReadDir on pipe returns ENOTDIR", - context: &ReadDir{}, - err: syscall.ENOTDIR, - }, - { - desc: "Fsync on pipe returns EINVAL", - context: &Fsync{}, - err: syscall.EINVAL, - }, - { - desc: "Seek on pipe returns ESPIPE", - context: &Seek{}, - err: syscall.ESPIPE, - }, - { - desc: "Readv on pipe from empty buffer returns nil", - context: &Readv{Dst: usermem.BytesIOSequence(nil)}, - flags: fs.FileFlags{Read: true}, - }, - { - desc: "Readv on pipe from non-empty buffer and closed partner returns EOF", - context: &Readv{Dst: usermem.BytesIOSequence(make([]byte, 10))}, - flags: fs.FileFlags{Read: true}, - err: io.EOF, - }, - { - desc: "Readv on pipe from non-empty buffer and open partner returns EWOULDBLOCK", - context: &Readv{Dst: usermem.BytesIOSequence(make([]byte, 10))}, - flags: fs.FileFlags{Read: true}, - keepOpenPartner: true, - err: syserror.ErrWouldBlock, - }, - { - desc: "Writev on pipe from empty buffer returns nil", - context: &Writev{Src: usermem.BytesIOSequence(nil)}, - flags: fs.FileFlags{Write: true}, - }, - { - desc: "Writev on pipe from non-empty buffer and closed partner returns EPIPE", - context: &Writev{Src: usermem.BytesIOSequence([]byte("hello"))}, - flags: fs.FileFlags{Write: true}, - err: syscall.EPIPE, - }, - { - desc: "Writev on pipe from non-empty buffer and open partner succeeds", - context: &Writev{Src: usermem.BytesIOSequence([]byte("hello"))}, - flags: fs.FileFlags{Write: true}, - keepOpenPartner: true, - }, - } { - if test.flags.Read && test.flags.Write { - panic("both read and write not supported for this test") - } - - fds := make([]int, 2) - if err := syscall.Pipe(fds); err != nil { - t.Errorf("%s: failed to create pipes: got %v, want nil", test.desc, err) - continue - } - - // Configure the fd and partner fd based on the file flags. - testFd, partnerFd := fds[0], fds[1] - if test.flags.Write { - testFd, partnerFd = fds[1], fds[0] - } - - // Configure closing the fds. - if test.keepOpenPartner { - defer syscall.Close(partnerFd) - } else { - syscall.Close(partnerFd) - } - - // Create the pipe. - ctx := contexttest.Context(t) - p, err := newPipeOperations(ctx, nil, test.flags, fd.New(testFd), nil) - if err != nil { - t.Fatalf("%s: newPipeOperations got error %v, want nil", test.desc, err) - } - defer p.Release() - - inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{Type: fs.Pipe}) - file := fs.NewFile(ctx, fs.NewDirent(ctx, inode, "pipe"), fs.FileFlags{Read: true}, p) - - // Issue request via the appropriate function. - switch c := test.context.(type) { - case *Seek: - _, err = p.Seek(ctx, file, 0, 0) - case *ReadDir: - _, err = p.Readdir(ctx, file, nil) - case *Readv: - _, err = p.Read(ctx, file, c.Dst, 0) - case *Writev: - _, err = p.Write(ctx, file, c.Src, 0) - case *Fsync: - err = p.Fsync(ctx, file, 0, fs.FileMaxOffset, fs.SyncAll) - default: - t.Errorf("%s: unknown request type %T", test.desc, test.context) - } - - if unwrapError(err) != test.err { - t.Errorf("%s: got error %v, want %v", test.desc, err, test.err) - } - } -} - -func TestPipeReadAheadBuffer(t *testing.T) { - fds := make([]int, 2) - if err := syscall.Pipe(fds); err != nil { - t.Fatalf("failed to create pipes: got %v, want nil", err) - } - rfile := fd.New(fds[0]) - - // Eventually close the write end, which is not wrapped in a pipe object. - defer syscall.Close(fds[1]) - - // Write some bytes to this end. - data := []byte("world") - if n, err := syscall.Write(fds[1], data); n != len(data) || err != nil { - rfile.Close() - t.Fatalf("write to pipe got (%d, %v), want (%d, nil)", n, err, len(data)) - } - // Close the write end immediately, we don't care about it. - - buffered := []byte("hello ") - ctx := contexttest.Context(t) - p, err := newPipeOperations(ctx, nil, fs.FileFlags{Read: true}, rfile, buffered) - if err != nil { - rfile.Close() - t.Fatalf("newPipeOperations got error %v, want nil", err) - } - defer p.Release() - - inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{ - Type: fs.Pipe, - }) - file := fs.NewFile(ctx, fs.NewDirent(ctx, inode, "pipe"), fs.FileFlags{Read: true}, p) - - // In total we expect to read data + buffered. - total := append(buffered, data...) - - buf := make([]byte, len(total)) - iov := usermem.BytesIOSequence(buf) - n, err := p.Read(contexttest.Context(t), file, iov, 0) - if err != nil { - t.Fatalf("read request got error %v, want nil", err) - } - if n != int64(len(total)) { - t.Fatalf("read request got %d bytes, want %d", n, len(total)) - } - if !bytes.Equal(buf, total) { - t.Errorf("read request got bytes [%v], want [%v]", buf, total) - } -} - -// This is very important for pipes in general because they can return -// EWOULDBLOCK and for those that block they must continue until they have read -// all of the data (and report it as such). -func TestPipeReadsAccumulate(t *testing.T) { - fds := make([]int, 2) - if err := syscall.Pipe(fds); err != nil { - t.Fatalf("failed to create pipes: got %v, want nil", err) - } - rfile := fd.New(fds[0]) - - // Eventually close the write end, it doesn't depend on a pipe object. - defer syscall.Close(fds[1]) - - // Get a new read only pipe reference. - ctx := contexttest.Context(t) - p, err := newPipeOperations(ctx, nil, fs.FileFlags{Read: true}, rfile, nil) - if err != nil { - rfile.Close() - t.Fatalf("newPipeOperations got error %v, want nil", err) - } - // Don't forget to remove the fd from the fd notifier. Otherwise other tests will - // likely be borked, because it's global :( - defer p.Release() - - inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{ - Type: fs.Pipe, - }) - file := fs.NewFile(ctx, fs.NewDirent(ctx, inode, "pipe"), fs.FileFlags{Read: true}, p) - - // Write some some bytes to the pipe. - data := []byte("some message") - if n, err := syscall.Write(fds[1], data); n != len(data) || err != nil { - t.Fatalf("write to pipe got (%d, %v), want (%d, nil)", n, err, len(data)) - } - - // Construct a segment vec that is a bit more than we have written so we - // trigger an EWOULDBLOCK. - wantBytes := len(data) + 1 - readBuffer := make([]byte, wantBytes) - iov := usermem.BytesIOSequence(readBuffer) - n, err := p.Read(ctx, file, iov, 0) - total := n - iov = iov.DropFirst64(n) - if err != syserror.ErrWouldBlock { - t.Fatalf("Readv got error %v, want %v", err, syserror.ErrWouldBlock) - } - - // Write a few more bytes to allow us to read more/accumulate. - extra := []byte("extra") - if n, err := syscall.Write(fds[1], extra); n != len(extra) || err != nil { - t.Fatalf("write to pipe got (%d, %v), want (%d, nil)", n, err, len(extra)) - } - - // This time, using the same request, we should not block. - n, err = p.Read(ctx, file, iov, 0) - total += n - if err != nil { - t.Fatalf("Readv got error %v, want nil", err) - } - - // Assert that the result we got back is cumulative. - if total != int64(wantBytes) { - t.Fatalf("Readv sequence got %d bytes, want %d", total, wantBytes) - } - - if want := append(data, extra[0]); !bytes.Equal(readBuffer, want) { - t.Errorf("Readv sequence got %v, want %v", readBuffer, want) - } -} - -// Same as TestReadsAccumulate. -func TestPipeWritesAccumulate(t *testing.T) { - fds := make([]int, 2) - if err := syscall.Pipe(fds); err != nil { - t.Fatalf("failed to create pipes: got %v, want nil", err) - } - wfile := fd.New(fds[1]) - - // Eventually close the read end, it doesn't depend on a pipe object. - defer syscall.Close(fds[0]) - - // Get a new write only pipe reference. - ctx := contexttest.Context(t) - p, err := newPipeOperations(ctx, nil, fs.FileFlags{Write: true}, wfile, nil) - if err != nil { - wfile.Close() - t.Fatalf("newPipeOperations got error %v, want nil", err) - } - // Don't forget to remove the fd from the fd notifier. Otherwise other tests - // will likely be borked, because it's global :( - defer p.Release() - - inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{ - Type: fs.Pipe, - }) - file := fs.NewFile(ctx, fs.NewDirent(ctx, inode, "pipe"), fs.FileFlags{Read: true}, p) - - pipeSize, _, errno := syscall.Syscall(syscall.SYS_FCNTL, uintptr(wfile.FD()), syscall.F_GETPIPE_SZ, 0) - if errno != 0 { - t.Fatalf("fcntl(F_GETPIPE_SZ) failed: %v", errno) - } - t.Logf("Pipe buffer size: %d", pipeSize) - - // Construct a segment vec that is larger than the pipe size to trigger an - // EWOULDBLOCK. - wantBytes := int(pipeSize) * 2 - writeBuffer := make([]byte, wantBytes) - for i := 0; i < wantBytes; i++ { - writeBuffer[i] = 'a' - } - iov := usermem.BytesIOSequence(writeBuffer) - n, err := p.Write(ctx, file, iov, 0) - if err != syserror.ErrWouldBlock { - t.Fatalf("Writev got error %v, want %v", err, syserror.ErrWouldBlock) - } - if n != int64(pipeSize) { - t.Fatalf("Writev partial write, got: %v, want %v", n, pipeSize) - } - total := n - iov = iov.DropFirst64(n) - - // Read the entire pipe buf size to make space for the second half. - readBuffer := make([]byte, n) - if n, err := syscall.Read(fds[0], readBuffer); n != len(readBuffer) || err != nil { - t.Fatalf("write to pipe got (%d, %v), want (%d, nil)", n, err, len(readBuffer)) - } - if !bytes.Equal(readBuffer, writeBuffer[:len(readBuffer)]) { - t.Fatalf("wrong data read from pipe, got: %v, want: %v", readBuffer, writeBuffer) - } - - // This time we should not block. - n, err = p.Write(ctx, file, iov, 0) - if err != nil { - t.Fatalf("Writev got error %v, want nil", err) - } - if n != int64(pipeSize) { - t.Fatalf("Writev partial write, got: %v, want %v", n, pipeSize) - } - total += n - - // Assert that the result we got back is cumulative. - if total != int64(wantBytes) { - t.Fatalf("Writev sequence got %d bytes, want %d", total, wantBytes) - } -} diff --git a/pkg/sentry/fs/file_overlay_test.go b/pkg/sentry/fs/file_overlay_test.go deleted file mode 100644 index 2fb824d5c..000000000 --- a/pkg/sentry/fs/file_overlay_test.go +++ /dev/null @@ -1,275 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fs_test - -import ( - "reflect" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" - "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" -) - -func TestReaddir(t *testing.T) { - ctx := contexttest.Context(t) - ctx = &rootContext{ - Context: ctx, - root: fs.NewDirent(ctx, newTestRamfsDir(ctx, nil, nil), "root"), - } - for _, test := range []struct { - // Test description. - desc string - - // Lookup parameters. - dir *fs.Inode - - // Want from lookup. - err error - names []string - }{ - { - desc: "no upper, lower has entries", - dir: fs.NewTestOverlayDir(ctx, - nil, /* upper */ - newTestRamfsDir(ctx, []dirContent{ - {name: "a"}, - {name: "b"}, - }, nil), /* lower */ - false /* revalidate */), - names: []string{".", "..", "a", "b"}, - }, - { - desc: "upper has entries, no lower", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - {name: "a"}, - {name: "b"}, - }, nil), /* upper */ - nil, /* lower */ - false /* revalidate */), - names: []string{".", "..", "a", "b"}, - }, - { - desc: "upper and lower, entries combine", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - {name: "a"}, - }, nil), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - {name: "b"}, - }, nil), /* lower */ - false /* revalidate */), - names: []string{".", "..", "a", "b"}, - }, - { - desc: "upper and lower, entries combine, none are masked", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - {name: "a"}, - }, []string{"b"}), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - {name: "c"}, - }, nil), /* lower */ - false /* revalidate */), - names: []string{".", "..", "a", "c"}, - }, - { - desc: "upper and lower, entries combine, upper masks some of lower", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - {name: "a"}, - }, []string{"b"}), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - {name: "b"}, /* will be masked */ - {name: "c"}, - }, nil), /* lower */ - false /* revalidate */), - names: []string{".", "..", "a", "c"}, - }, - } { - t.Run(test.desc, func(t *testing.T) { - openDir, err := test.dir.GetFile(ctx, fs.NewDirent(ctx, test.dir, "stub"), fs.FileFlags{Read: true}) - if err != nil { - t.Fatalf("GetFile got error %v, want nil", err) - } - stubSerializer := &fs.CollectEntriesSerializer{} - err = openDir.Readdir(ctx, stubSerializer) - if err != test.err { - t.Fatalf("Readdir got error %v, want nil", err) - } - if err != nil { - return - } - if !reflect.DeepEqual(stubSerializer.Order, test.names) { - t.Errorf("Readdir got names %v, want %v", stubSerializer.Order, test.names) - } - }) - } -} - -func TestReaddirRevalidation(t *testing.T) { - ctx := contexttest.Context(t) - ctx = &rootContext{ - Context: ctx, - root: fs.NewDirent(ctx, newTestRamfsDir(ctx, nil, nil), "root"), - } - - // Create an overlay with two directories, each with one file. - upper := newTestRamfsDir(ctx, []dirContent{{name: "a"}}, nil) - lower := newTestRamfsDir(ctx, []dirContent{{name: "b"}}, nil) - overlay := fs.NewTestOverlayDir(ctx, upper, lower, true /* revalidate */) - - // Get a handle to the dirent in the upper filesystem so that we can - // modify it without going through the dirent. - upperDir := upper.InodeOperations.(*dir).InodeOperations.(*ramfs.Dir) - - // Check that overlay returns the files from both upper and lower. - openDir, err := overlay.GetFile(ctx, fs.NewDirent(ctx, overlay, "stub"), fs.FileFlags{Read: true}) - if err != nil { - t.Fatalf("GetFile got error %v, want nil", err) - } - ser := &fs.CollectEntriesSerializer{} - if err := openDir.Readdir(ctx, ser); err != nil { - t.Fatalf("Readdir got error %v, want nil", err) - } - got, want := ser.Order, []string{".", "..", "a", "b"} - if !reflect.DeepEqual(got, want) { - t.Errorf("Readdir got names %v, want %v", got, want) - } - - // Remove "a" from the upper and add "c". - if err := upperDir.Remove(ctx, upper, "a"); err != nil { - t.Fatalf("error removing child: %v", err) - } - upperDir.AddChild(ctx, "c", fs.NewInode(ctx, fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermissions{}, 0), - upper.MountSource, fs.StableAttr{Type: fs.RegularFile})) - - // Seek to beginning of the directory and do the readdir again. - if _, err := openDir.Seek(ctx, fs.SeekSet, 0); err != nil { - t.Fatalf("error seeking to beginning of dir: %v", err) - } - ser = &fs.CollectEntriesSerializer{} - if err := openDir.Readdir(ctx, ser); err != nil { - t.Fatalf("Readdir got error %v, want nil", err) - } - - // Readdir should return the updated children. - got, want = ser.Order, []string{".", "..", "b", "c"} - if !reflect.DeepEqual(got, want) { - t.Errorf("Readdir got names %v, want %v", got, want) - } -} - -// TestReaddirOverlayFrozen tests that calling Readdir on an overlay file with -// a frozen dirent tree does not make Readdir calls to the underlying files. -func TestReaddirOverlayFrozen(t *testing.T) { - ctx := contexttest.Context(t) - - // Create an overlay with two directories, each with two files. - upper := newTestRamfsDir(ctx, []dirContent{{name: "upper-file1"}, {name: "upper-file2"}}, nil) - lower := newTestRamfsDir(ctx, []dirContent{{name: "lower-file1"}, {name: "lower-file2"}}, nil) - overlayInode := fs.NewTestOverlayDir(ctx, upper, lower, false) - - // Set that overlay as the root. - root := fs.NewDirent(ctx, overlayInode, "root") - ctx = &rootContext{ - Context: ctx, - root: root, - } - - // Check that calling Readdir on the root now returns all 4 files (2 - // from each layer in the overlay). - rootFile, err := root.Inode.GetFile(ctx, root, fs.FileFlags{Read: true}) - if err != nil { - t.Fatalf("root.Inode.GetFile failed: %v", err) - } - defer rootFile.DecRef() - ser := &fs.CollectEntriesSerializer{} - if err := rootFile.Readdir(ctx, ser); err != nil { - t.Fatalf("rootFile.Readdir failed: %v", err) - } - if got, want := ser.Order, []string{".", "..", "lower-file1", "lower-file2", "upper-file1", "upper-file2"}; !reflect.DeepEqual(got, want) { - t.Errorf("Readdir got names %v, want %v", got, want) - } - - // Readdir should have been called on upper and lower. - upperDir := upper.InodeOperations.(*dir) - lowerDir := lower.InodeOperations.(*dir) - if !upperDir.ReaddirCalled { - t.Errorf("upperDir.ReaddirCalled got %v, want true", upperDir.ReaddirCalled) - } - if !lowerDir.ReaddirCalled { - t.Errorf("lowerDir.ReaddirCalled got %v, want true", lowerDir.ReaddirCalled) - } - - // Reset. - upperDir.ReaddirCalled = false - lowerDir.ReaddirCalled = false - - // Take references on "upper-file1" and "lower-file1", pinning them in - // the dirent tree. - for _, name := range []string{"upper-file1", "lower-file1"} { - if _, err := root.Walk(ctx, root, name); err != nil { - t.Fatalf("root.Walk(%q) failed: %v", name, err) - } - // Don't drop a reference on the returned dirent so that it - // will stay in the tree. - } - - // Freeze the dirent tree. - root.Freeze() - - // Seek back to the beginning of the file. - if _, err := rootFile.Seek(ctx, fs.SeekSet, 0); err != nil { - t.Fatalf("error seeking to beginning of directory: %v", err) - } - - // Calling Readdir on the root now will return only the pinned - // children. - ser = &fs.CollectEntriesSerializer{} - if err := rootFile.Readdir(ctx, ser); err != nil { - t.Fatalf("rootFile.Readdir failed: %v", err) - } - if got, want := ser.Order, []string{".", "..", "lower-file1", "upper-file1"}; !reflect.DeepEqual(got, want) { - t.Errorf("Readdir got names %v, want %v", got, want) - } - - // Readdir should NOT have been called on upper or lower. - if upperDir.ReaddirCalled { - t.Errorf("upperDir.ReaddirCalled got %v, want false", upperDir.ReaddirCalled) - } - if lowerDir.ReaddirCalled { - t.Errorf("lowerDir.ReaddirCalled got %v, want false", lowerDir.ReaddirCalled) - } -} - -type rootContext struct { - context.Context - root *fs.Dirent -} - -// Value implements context.Context. -func (r *rootContext) Value(key interface{}) interface{} { - switch key { - case fs.CtxRoot: - r.root.IncRef() - return r.root - default: - return r.Context.Value(key) - } -} diff --git a/pkg/sentry/fs/filetest/BUILD b/pkg/sentry/fs/filetest/BUILD deleted file mode 100644 index a9d6d9301..000000000 --- a/pkg/sentry/fs/filetest/BUILD +++ /dev/null @@ -1,20 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_stateify:defs.bzl", "go_library") - -go_library( - name = "filetest", - testonly = 1, - srcs = ["filetest.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/filetest", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", - "//pkg/sentry/fs", - "//pkg/sentry/fs/anon", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/usermem", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/fs/filetest/filetest.go b/pkg/sentry/fs/filetest/filetest.go deleted file mode 100644 index 22270a494..000000000 --- a/pkg/sentry/fs/filetest/filetest.go +++ /dev/null @@ -1,61 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package filetest provides a test implementation of an fs.File. -package filetest - -import ( - "fmt" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/anon" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/waiter" -) - -// TestFileOperations is an implementation of the File interface. It provides all -// required methods. -type TestFileOperations struct { - fsutil.FileNoopRelease `state:"nosave"` - fsutil.FilePipeSeek `state:"nosave"` - fsutil.FileNotDirReaddir `state:"nosave"` - fsutil.FileNoFsync `state:"nosave"` - fsutil.FileNoopFlush `state:"nosave"` - fsutil.FileNoMMap `state:"nosave"` - fsutil.FileNoIoctl `state:"nosave"` - fsutil.FileNoSplice `state:"nosave"` - fsutil.FileUseInodeUnstableAttr `state:"nosave"` - waiter.AlwaysReady `state:"nosave"` -} - -// NewTestFile creates and initializes a new test file. -func NewTestFile(tb testing.TB) *fs.File { - ctx := contexttest.Context(tb) - dirent := fs.NewDirent(ctx, anon.NewInode(ctx), "test") - return fs.NewFile(ctx, dirent, fs.FileFlags{}, &TestFileOperations{}) -} - -// Read just fails the request. -func (*TestFileOperations) Read(context.Context, *fs.File, usermem.IOSequence, int64) (int64, error) { - return 0, fmt.Errorf("Readv not implemented") -} - -// Write just fails the request. -func (*TestFileOperations) Write(context.Context, *fs.File, usermem.IOSequence, int64) (int64, error) { - return 0, fmt.Errorf("Writev not implemented") -} diff --git a/pkg/sentry/fs/fs_state_autogen.go b/pkg/sentry/fs/fs_state_autogen.go new file mode 100755 index 000000000..5ea2669e6 --- /dev/null +++ b/pkg/sentry/fs/fs_state_autogen.go @@ -0,0 +1,632 @@ +// automatically generated by stateify. + +package fs + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *StableAttr) beforeSave() {} +func (x *StableAttr) save(m state.Map) { + x.beforeSave() + m.Save("Type", &x.Type) + m.Save("DeviceID", &x.DeviceID) + m.Save("InodeID", &x.InodeID) + m.Save("BlockSize", &x.BlockSize) + m.Save("DeviceFileMajor", &x.DeviceFileMajor) + m.Save("DeviceFileMinor", &x.DeviceFileMinor) +} + +func (x *StableAttr) afterLoad() {} +func (x *StableAttr) load(m state.Map) { + m.Load("Type", &x.Type) + m.Load("DeviceID", &x.DeviceID) + m.Load("InodeID", &x.InodeID) + m.Load("BlockSize", &x.BlockSize) + m.Load("DeviceFileMajor", &x.DeviceFileMajor) + m.Load("DeviceFileMinor", &x.DeviceFileMinor) +} + +func (x *UnstableAttr) beforeSave() {} +func (x *UnstableAttr) save(m state.Map) { + x.beforeSave() + m.Save("Size", &x.Size) + m.Save("Usage", &x.Usage) + m.Save("Perms", &x.Perms) + m.Save("Owner", &x.Owner) + m.Save("AccessTime", &x.AccessTime) + m.Save("ModificationTime", &x.ModificationTime) + m.Save("StatusChangeTime", &x.StatusChangeTime) + m.Save("Links", &x.Links) +} + +func (x *UnstableAttr) afterLoad() {} +func (x *UnstableAttr) load(m state.Map) { + m.Load("Size", &x.Size) + m.Load("Usage", &x.Usage) + m.Load("Perms", &x.Perms) + m.Load("Owner", &x.Owner) + m.Load("AccessTime", &x.AccessTime) + m.Load("ModificationTime", &x.ModificationTime) + m.Load("StatusChangeTime", &x.StatusChangeTime) + m.Load("Links", &x.Links) +} + +func (x *AttrMask) beforeSave() {} +func (x *AttrMask) save(m state.Map) { + x.beforeSave() + m.Save("Type", &x.Type) + m.Save("DeviceID", &x.DeviceID) + m.Save("InodeID", &x.InodeID) + m.Save("BlockSize", &x.BlockSize) + m.Save("Size", &x.Size) + m.Save("Usage", &x.Usage) + m.Save("Perms", &x.Perms) + m.Save("UID", &x.UID) + m.Save("GID", &x.GID) + m.Save("AccessTime", &x.AccessTime) + m.Save("ModificationTime", &x.ModificationTime) + m.Save("StatusChangeTime", &x.StatusChangeTime) + m.Save("Links", &x.Links) +} + +func (x *AttrMask) afterLoad() {} +func (x *AttrMask) load(m state.Map) { + m.Load("Type", &x.Type) + m.Load("DeviceID", &x.DeviceID) + m.Load("InodeID", &x.InodeID) + m.Load("BlockSize", &x.BlockSize) + m.Load("Size", &x.Size) + m.Load("Usage", &x.Usage) + m.Load("Perms", &x.Perms) + m.Load("UID", &x.UID) + m.Load("GID", &x.GID) + m.Load("AccessTime", &x.AccessTime) + m.Load("ModificationTime", &x.ModificationTime) + m.Load("StatusChangeTime", &x.StatusChangeTime) + m.Load("Links", &x.Links) +} + +func (x *PermMask) beforeSave() {} +func (x *PermMask) save(m state.Map) { + x.beforeSave() + m.Save("Read", &x.Read) + m.Save("Write", &x.Write) + m.Save("Execute", &x.Execute) +} + +func (x *PermMask) afterLoad() {} +func (x *PermMask) load(m state.Map) { + m.Load("Read", &x.Read) + m.Load("Write", &x.Write) + m.Load("Execute", &x.Execute) +} + +func (x *FilePermissions) beforeSave() {} +func (x *FilePermissions) save(m state.Map) { + x.beforeSave() + m.Save("User", &x.User) + m.Save("Group", &x.Group) + m.Save("Other", &x.Other) + m.Save("Sticky", &x.Sticky) + m.Save("SetUID", &x.SetUID) + m.Save("SetGID", &x.SetGID) +} + +func (x *FilePermissions) afterLoad() {} +func (x *FilePermissions) load(m state.Map) { + m.Load("User", &x.User) + m.Load("Group", &x.Group) + m.Load("Other", &x.Other) + m.Load("Sticky", &x.Sticky) + m.Load("SetUID", &x.SetUID) + m.Load("SetGID", &x.SetGID) +} + +func (x *FileOwner) beforeSave() {} +func (x *FileOwner) save(m state.Map) { + x.beforeSave() + m.Save("UID", &x.UID) + m.Save("GID", &x.GID) +} + +func (x *FileOwner) afterLoad() {} +func (x *FileOwner) load(m state.Map) { + m.Load("UID", &x.UID) + m.Load("GID", &x.GID) +} + +func (x *DentAttr) beforeSave() {} +func (x *DentAttr) save(m state.Map) { + x.beforeSave() + m.Save("Type", &x.Type) + m.Save("InodeID", &x.InodeID) +} + +func (x *DentAttr) afterLoad() {} +func (x *DentAttr) load(m state.Map) { + m.Load("Type", &x.Type) + m.Load("InodeID", &x.InodeID) +} + +func (x *SortedDentryMap) beforeSave() {} +func (x *SortedDentryMap) save(m state.Map) { + x.beforeSave() + m.Save("names", &x.names) + m.Save("entries", &x.entries) +} + +func (x *SortedDentryMap) afterLoad() {} +func (x *SortedDentryMap) load(m state.Map) { + m.Load("names", &x.names) + m.Load("entries", &x.entries) +} + +func (x *Dirent) save(m state.Map) { + x.beforeSave() + var children map[string]*Dirent = x.saveChildren() + m.SaveValue("children", children) + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("userVisible", &x.userVisible) + m.Save("Inode", &x.Inode) + m.Save("name", &x.name) + m.Save("parent", &x.parent) + m.Save("deleted", &x.deleted) + m.Save("frozen", &x.frozen) + m.Save("mounted", &x.mounted) +} + +func (x *Dirent) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("userVisible", &x.userVisible) + m.Load("Inode", &x.Inode) + m.Load("name", &x.name) + m.Load("parent", &x.parent) + m.Load("deleted", &x.deleted) + m.Load("frozen", &x.frozen) + m.Load("mounted", &x.mounted) + m.LoadValue("children", new(map[string]*Dirent), func(y interface{}) { x.loadChildren(y.(map[string]*Dirent)) }) + m.AfterLoad(x.afterLoad) +} + +func (x *DirentCache) beforeSave() {} +func (x *DirentCache) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.currentSize) { m.Failf("currentSize is %v, expected zero", x.currentSize) } + if !state.IsZeroValue(x.list) { m.Failf("list is %v, expected zero", x.list) } + m.Save("maxSize", &x.maxSize) + m.Save("limit", &x.limit) +} + +func (x *DirentCache) afterLoad() {} +func (x *DirentCache) load(m state.Map) { + m.Load("maxSize", &x.maxSize) + m.Load("limit", &x.limit) +} + +func (x *DirentCacheLimiter) beforeSave() {} +func (x *DirentCacheLimiter) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.count) { m.Failf("count is %v, expected zero", x.count) } + m.Save("max", &x.max) +} + +func (x *DirentCacheLimiter) afterLoad() {} +func (x *DirentCacheLimiter) load(m state.Map) { + m.Load("max", &x.max) +} + +func (x *direntList) beforeSave() {} +func (x *direntList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *direntList) afterLoad() {} +func (x *direntList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *direntEntry) beforeSave() {} +func (x *direntEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *direntEntry) afterLoad() {} +func (x *direntEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func (x *eventList) beforeSave() {} +func (x *eventList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *eventList) afterLoad() {} +func (x *eventList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *eventEntry) beforeSave() {} +func (x *eventEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *eventEntry) afterLoad() {} +func (x *eventEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func (x *File) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("UniqueID", &x.UniqueID) + m.Save("Dirent", &x.Dirent) + m.Save("flags", &x.flags) + m.Save("async", &x.async) + m.Save("FileOperations", &x.FileOperations) + m.Save("offset", &x.offset) +} + +func (x *File) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("UniqueID", &x.UniqueID) + m.Load("Dirent", &x.Dirent) + m.Load("flags", &x.flags) + m.Load("async", &x.async) + m.LoadWait("FileOperations", &x.FileOperations) + m.Load("offset", &x.offset) + m.AfterLoad(x.afterLoad) +} + +func (x *overlayFileOperations) beforeSave() {} +func (x *overlayFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("upper", &x.upper) + m.Save("lower", &x.lower) + m.Save("dirCursor", &x.dirCursor) +} + +func (x *overlayFileOperations) afterLoad() {} +func (x *overlayFileOperations) load(m state.Map) { + m.Load("upper", &x.upper) + m.Load("lower", &x.lower) + m.Load("dirCursor", &x.dirCursor) +} + +func (x *overlayMappingIdentity) beforeSave() {} +func (x *overlayMappingIdentity) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("id", &x.id) + m.Save("overlayFile", &x.overlayFile) +} + +func (x *overlayMappingIdentity) afterLoad() {} +func (x *overlayMappingIdentity) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("id", &x.id) + m.Load("overlayFile", &x.overlayFile) +} + +func (x *MountSourceFlags) beforeSave() {} +func (x *MountSourceFlags) save(m state.Map) { + x.beforeSave() + m.Save("ReadOnly", &x.ReadOnly) + m.Save("NoAtime", &x.NoAtime) + m.Save("ForcePageCache", &x.ForcePageCache) + m.Save("NoExec", &x.NoExec) +} + +func (x *MountSourceFlags) afterLoad() {} +func (x *MountSourceFlags) load(m state.Map) { + m.Load("ReadOnly", &x.ReadOnly) + m.Load("NoAtime", &x.NoAtime) + m.Load("ForcePageCache", &x.ForcePageCache) + m.Load("NoExec", &x.NoExec) +} + +func (x *FileFlags) beforeSave() {} +func (x *FileFlags) save(m state.Map) { + x.beforeSave() + m.Save("Direct", &x.Direct) + m.Save("NonBlocking", &x.NonBlocking) + m.Save("DSync", &x.DSync) + m.Save("Sync", &x.Sync) + m.Save("Append", &x.Append) + m.Save("Read", &x.Read) + m.Save("Write", &x.Write) + m.Save("Pread", &x.Pread) + m.Save("Pwrite", &x.Pwrite) + m.Save("Directory", &x.Directory) + m.Save("Async", &x.Async) + m.Save("LargeFile", &x.LargeFile) + m.Save("NonSeekable", &x.NonSeekable) +} + +func (x *FileFlags) afterLoad() {} +func (x *FileFlags) load(m state.Map) { + m.Load("Direct", &x.Direct) + m.Load("NonBlocking", &x.NonBlocking) + m.Load("DSync", &x.DSync) + m.Load("Sync", &x.Sync) + m.Load("Append", &x.Append) + m.Load("Read", &x.Read) + m.Load("Write", &x.Write) + m.Load("Pread", &x.Pread) + m.Load("Pwrite", &x.Pwrite) + m.Load("Directory", &x.Directory) + m.Load("Async", &x.Async) + m.Load("LargeFile", &x.LargeFile) + m.Load("NonSeekable", &x.NonSeekable) +} + +func (x *Inode) beforeSave() {} +func (x *Inode) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("InodeOperations", &x.InodeOperations) + m.Save("StableAttr", &x.StableAttr) + m.Save("LockCtx", &x.LockCtx) + m.Save("Watches", &x.Watches) + m.Save("MountSource", &x.MountSource) + m.Save("overlay", &x.overlay) +} + +func (x *Inode) afterLoad() {} +func (x *Inode) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("InodeOperations", &x.InodeOperations) + m.Load("StableAttr", &x.StableAttr) + m.Load("LockCtx", &x.LockCtx) + m.Load("Watches", &x.Watches) + m.Load("MountSource", &x.MountSource) + m.Load("overlay", &x.overlay) +} + +func (x *LockCtx) beforeSave() {} +func (x *LockCtx) save(m state.Map) { + x.beforeSave() + m.Save("Posix", &x.Posix) + m.Save("BSD", &x.BSD) +} + +func (x *LockCtx) afterLoad() {} +func (x *LockCtx) load(m state.Map) { + m.Load("Posix", &x.Posix) + m.Load("BSD", &x.BSD) +} + +func (x *Watches) beforeSave() {} +func (x *Watches) save(m state.Map) { + x.beforeSave() + m.Save("ws", &x.ws) + m.Save("unlinked", &x.unlinked) +} + +func (x *Watches) afterLoad() {} +func (x *Watches) load(m state.Map) { + m.Load("ws", &x.ws) + m.Load("unlinked", &x.unlinked) +} + +func (x *Inotify) beforeSave() {} +func (x *Inotify) save(m state.Map) { + x.beforeSave() + m.Save("id", &x.id) + m.Save("events", &x.events) + m.Save("scratch", &x.scratch) + m.Save("nextWatch", &x.nextWatch) + m.Save("watches", &x.watches) +} + +func (x *Inotify) afterLoad() {} +func (x *Inotify) load(m state.Map) { + m.Load("id", &x.id) + m.Load("events", &x.events) + m.Load("scratch", &x.scratch) + m.Load("nextWatch", &x.nextWatch) + m.Load("watches", &x.watches) +} + +func (x *Event) beforeSave() {} +func (x *Event) save(m state.Map) { + x.beforeSave() + m.Save("eventEntry", &x.eventEntry) + m.Save("wd", &x.wd) + m.Save("mask", &x.mask) + m.Save("cookie", &x.cookie) + m.Save("len", &x.len) + m.Save("name", &x.name) +} + +func (x *Event) afterLoad() {} +func (x *Event) load(m state.Map) { + m.Load("eventEntry", &x.eventEntry) + m.Load("wd", &x.wd) + m.Load("mask", &x.mask) + m.Load("cookie", &x.cookie) + m.Load("len", &x.len) + m.Load("name", &x.name) +} + +func (x *Watch) beforeSave() {} +func (x *Watch) save(m state.Map) { + x.beforeSave() + m.Save("owner", &x.owner) + m.Save("wd", &x.wd) + m.Save("target", &x.target) + m.Save("unpinned", &x.unpinned) + m.Save("mask", &x.mask) + m.Save("pins", &x.pins) +} + +func (x *Watch) afterLoad() {} +func (x *Watch) load(m state.Map) { + m.Load("owner", &x.owner) + m.Load("wd", &x.wd) + m.Load("target", &x.target) + m.Load("unpinned", &x.unpinned) + m.Load("mask", &x.mask) + m.Load("pins", &x.pins) +} + +func (x *MountSource) beforeSave() {} +func (x *MountSource) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("MountSourceOperations", &x.MountSourceOperations) + m.Save("FilesystemType", &x.FilesystemType) + m.Save("Flags", &x.Flags) + m.Save("fscache", &x.fscache) + m.Save("direntRefs", &x.direntRefs) +} + +func (x *MountSource) afterLoad() {} +func (x *MountSource) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("MountSourceOperations", &x.MountSourceOperations) + m.Load("FilesystemType", &x.FilesystemType) + m.Load("Flags", &x.Flags) + m.Load("fscache", &x.fscache) + m.Load("direntRefs", &x.direntRefs) +} + +func (x *SimpleMountSourceOperations) beforeSave() {} +func (x *SimpleMountSourceOperations) save(m state.Map) { + x.beforeSave() + m.Save("keep", &x.keep) + m.Save("revalidate", &x.revalidate) + m.Save("cacheReaddir", &x.cacheReaddir) +} + +func (x *SimpleMountSourceOperations) afterLoad() {} +func (x *SimpleMountSourceOperations) load(m state.Map) { + m.Load("keep", &x.keep) + m.Load("revalidate", &x.revalidate) + m.Load("cacheReaddir", &x.cacheReaddir) +} + +func (x *overlayMountSourceOperations) beforeSave() {} +func (x *overlayMountSourceOperations) save(m state.Map) { + x.beforeSave() + m.Save("upper", &x.upper) + m.Save("lower", &x.lower) +} + +func (x *overlayMountSourceOperations) afterLoad() {} +func (x *overlayMountSourceOperations) load(m state.Map) { + m.Load("upper", &x.upper) + m.Load("lower", &x.lower) +} + +func (x *overlayFilesystem) beforeSave() {} +func (x *overlayFilesystem) save(m state.Map) { + x.beforeSave() +} + +func (x *overlayFilesystem) afterLoad() {} +func (x *overlayFilesystem) load(m state.Map) { +} + +func (x *Mount) beforeSave() {} +func (x *Mount) save(m state.Map) { + x.beforeSave() + m.Save("ID", &x.ID) + m.Save("ParentID", &x.ParentID) + m.Save("root", &x.root) + m.Save("previous", &x.previous) +} + +func (x *Mount) afterLoad() {} +func (x *Mount) load(m state.Map) { + m.Load("ID", &x.ID) + m.Load("ParentID", &x.ParentID) + m.Load("root", &x.root) + m.Load("previous", &x.previous) +} + +func (x *MountNamespace) beforeSave() {} +func (x *MountNamespace) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("userns", &x.userns) + m.Save("root", &x.root) + m.Save("mounts", &x.mounts) + m.Save("mountID", &x.mountID) +} + +func (x *MountNamespace) afterLoad() {} +func (x *MountNamespace) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("userns", &x.userns) + m.Load("root", &x.root) + m.Load("mounts", &x.mounts) + m.Load("mountID", &x.mountID) +} + +func (x *overlayEntry) beforeSave() {} +func (x *overlayEntry) save(m state.Map) { + x.beforeSave() + m.Save("lowerExists", &x.lowerExists) + m.Save("lower", &x.lower) + m.Save("mappings", &x.mappings) + m.Save("upper", &x.upper) + m.Save("dirCache", &x.dirCache) +} + +func (x *overlayEntry) afterLoad() {} +func (x *overlayEntry) load(m state.Map) { + m.Load("lowerExists", &x.lowerExists) + m.Load("lower", &x.lower) + m.Load("mappings", &x.mappings) + m.Load("upper", &x.upper) + m.Load("dirCache", &x.dirCache) +} + +func init() { + state.Register("fs.StableAttr", (*StableAttr)(nil), state.Fns{Save: (*StableAttr).save, Load: (*StableAttr).load}) + state.Register("fs.UnstableAttr", (*UnstableAttr)(nil), state.Fns{Save: (*UnstableAttr).save, Load: (*UnstableAttr).load}) + state.Register("fs.AttrMask", (*AttrMask)(nil), state.Fns{Save: (*AttrMask).save, Load: (*AttrMask).load}) + state.Register("fs.PermMask", (*PermMask)(nil), state.Fns{Save: (*PermMask).save, Load: (*PermMask).load}) + state.Register("fs.FilePermissions", (*FilePermissions)(nil), state.Fns{Save: (*FilePermissions).save, Load: (*FilePermissions).load}) + state.Register("fs.FileOwner", (*FileOwner)(nil), state.Fns{Save: (*FileOwner).save, Load: (*FileOwner).load}) + state.Register("fs.DentAttr", (*DentAttr)(nil), state.Fns{Save: (*DentAttr).save, Load: (*DentAttr).load}) + state.Register("fs.SortedDentryMap", (*SortedDentryMap)(nil), state.Fns{Save: (*SortedDentryMap).save, Load: (*SortedDentryMap).load}) + state.Register("fs.Dirent", (*Dirent)(nil), state.Fns{Save: (*Dirent).save, Load: (*Dirent).load}) + state.Register("fs.DirentCache", (*DirentCache)(nil), state.Fns{Save: (*DirentCache).save, Load: (*DirentCache).load}) + state.Register("fs.DirentCacheLimiter", (*DirentCacheLimiter)(nil), state.Fns{Save: (*DirentCacheLimiter).save, Load: (*DirentCacheLimiter).load}) + state.Register("fs.direntList", (*direntList)(nil), state.Fns{Save: (*direntList).save, Load: (*direntList).load}) + state.Register("fs.direntEntry", (*direntEntry)(nil), state.Fns{Save: (*direntEntry).save, Load: (*direntEntry).load}) + state.Register("fs.eventList", (*eventList)(nil), state.Fns{Save: (*eventList).save, Load: (*eventList).load}) + state.Register("fs.eventEntry", (*eventEntry)(nil), state.Fns{Save: (*eventEntry).save, Load: (*eventEntry).load}) + state.Register("fs.File", (*File)(nil), state.Fns{Save: (*File).save, Load: (*File).load}) + state.Register("fs.overlayFileOperations", (*overlayFileOperations)(nil), state.Fns{Save: (*overlayFileOperations).save, Load: (*overlayFileOperations).load}) + state.Register("fs.overlayMappingIdentity", (*overlayMappingIdentity)(nil), state.Fns{Save: (*overlayMappingIdentity).save, Load: (*overlayMappingIdentity).load}) + state.Register("fs.MountSourceFlags", (*MountSourceFlags)(nil), state.Fns{Save: (*MountSourceFlags).save, Load: (*MountSourceFlags).load}) + state.Register("fs.FileFlags", (*FileFlags)(nil), state.Fns{Save: (*FileFlags).save, Load: (*FileFlags).load}) + state.Register("fs.Inode", (*Inode)(nil), state.Fns{Save: (*Inode).save, Load: (*Inode).load}) + state.Register("fs.LockCtx", (*LockCtx)(nil), state.Fns{Save: (*LockCtx).save, Load: (*LockCtx).load}) + state.Register("fs.Watches", (*Watches)(nil), state.Fns{Save: (*Watches).save, Load: (*Watches).load}) + state.Register("fs.Inotify", (*Inotify)(nil), state.Fns{Save: (*Inotify).save, Load: (*Inotify).load}) + state.Register("fs.Event", (*Event)(nil), state.Fns{Save: (*Event).save, Load: (*Event).load}) + state.Register("fs.Watch", (*Watch)(nil), state.Fns{Save: (*Watch).save, Load: (*Watch).load}) + state.Register("fs.MountSource", (*MountSource)(nil), state.Fns{Save: (*MountSource).save, Load: (*MountSource).load}) + state.Register("fs.SimpleMountSourceOperations", (*SimpleMountSourceOperations)(nil), state.Fns{Save: (*SimpleMountSourceOperations).save, Load: (*SimpleMountSourceOperations).load}) + state.Register("fs.overlayMountSourceOperations", (*overlayMountSourceOperations)(nil), state.Fns{Save: (*overlayMountSourceOperations).save, Load: (*overlayMountSourceOperations).load}) + state.Register("fs.overlayFilesystem", (*overlayFilesystem)(nil), state.Fns{Save: (*overlayFilesystem).save, Load: (*overlayFilesystem).load}) + state.Register("fs.Mount", (*Mount)(nil), state.Fns{Save: (*Mount).save, Load: (*Mount).load}) + state.Register("fs.MountNamespace", (*MountNamespace)(nil), state.Fns{Save: (*MountNamespace).save, Load: (*MountNamespace).load}) + state.Register("fs.overlayEntry", (*overlayEntry)(nil), state.Fns{Save: (*overlayEntry).save, Load: (*overlayEntry).load}) +} diff --git a/pkg/sentry/fs/fsutil/BUILD b/pkg/sentry/fs/fsutil/BUILD deleted file mode 100644 index 6499f87ac..000000000 --- a/pkg/sentry/fs/fsutil/BUILD +++ /dev/null @@ -1,118 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library", "go_test") - -go_template_instance( - name = "dirty_set_impl", - out = "dirty_set_impl.go", - imports = { - "memmap": "gvisor.dev/gvisor/pkg/sentry/memmap", - "platform": "gvisor.dev/gvisor/pkg/sentry/platform", - }, - package = "fsutil", - prefix = "Dirty", - template = "//pkg/segment:generic_set", - types = { - "Key": "uint64", - "Range": "memmap.MappableRange", - "Value": "DirtyInfo", - "Functions": "dirtySetFunctions", - }, -) - -go_template_instance( - name = "frame_ref_set_impl", - out = "frame_ref_set_impl.go", - imports = { - "platform": "gvisor.dev/gvisor/pkg/sentry/platform", - }, - package = "fsutil", - prefix = "frameRef", - template = "//pkg/segment:generic_set", - types = { - "Key": "uint64", - "Range": "platform.FileRange", - "Value": "uint64", - "Functions": "frameRefSetFunctions", - }, -) - -go_template_instance( - name = "file_range_set_impl", - out = "file_range_set_impl.go", - imports = { - "memmap": "gvisor.dev/gvisor/pkg/sentry/memmap", - "platform": "gvisor.dev/gvisor/pkg/sentry/platform", - }, - package = "fsutil", - prefix = "FileRange", - template = "//pkg/segment:generic_set", - types = { - "Key": "uint64", - "Range": "memmap.MappableRange", - "Value": "uint64", - "Functions": "fileRangeSetFunctions", - }, -) - -go_library( - name = "fsutil", - srcs = [ - "dirty_set.go", - "dirty_set_impl.go", - "file.go", - "file_range_set.go", - "file_range_set_impl.go", - "frame_ref_set.go", - "frame_ref_set_impl.go", - "fsutil.go", - "host_file_mapper.go", - "host_file_mapper_state.go", - "host_file_mapper_unsafe.go", - "host_mappable.go", - "inode.go", - "inode_cached.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/fsutil", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/log", - "//pkg/sentry/arch", - "//pkg/sentry/context", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/kernel/time", - "//pkg/sentry/memmap", - "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", - "//pkg/sentry/safemem", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usage", - "//pkg/sentry/usermem", - "//pkg/state", - "//pkg/syserror", - "//pkg/waiter", - ], -) - -go_test( - name = "fsutil_test", - size = "small", - srcs = [ - "dirty_set_test.go", - "inode_cached_test.go", - ], - embed = [":fsutil"], - deps = [ - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", - "//pkg/sentry/fs", - "//pkg/sentry/kernel/time", - "//pkg/sentry/memmap", - "//pkg/sentry/safemem", - "//pkg/sentry/usermem", - "//pkg/syserror", - ], -) diff --git a/pkg/sentry/fs/fsutil/README.md b/pkg/sentry/fs/fsutil/README.md deleted file mode 100644 index 8be367334..000000000 --- a/pkg/sentry/fs/fsutil/README.md +++ /dev/null @@ -1,207 +0,0 @@ -This package provides utilities for implementing virtual filesystem objects. - -[TOC] - -## Page cache - -`CachingInodeOperations` implements a page cache for files that cannot use the -host page cache. Normally these are files that store their data in a remote -filesystem. This also applies to files that are accessed on a platform that does -not support directly memory mapping host file descriptors (e.g. the ptrace -platform). - -An `CachingInodeOperations` buffers regions of a single file into memory. It is -owned by an `fs.Inode`, the in-memory representation of a file (all open file -descriptors are backed by an `fs.Inode`). The `fs.Inode` provides operations for -reading memory into an `CachingInodeOperations`, to represent the contents of -the file in-memory, and for writing memory out, to relieve memory pressure on -the kernel and to synchronize in-memory changes to filesystems. - -An `CachingInodeOperations` enables readable and/or writable memory access to -file content. Files can be mapped shared or private, see mmap(2). When a file is -mapped shared, changes to the file via write(2) and truncate(2) are reflected in -the shared memory region. Conversely, when the shared memory region is modified, -changes to the file are visible via read(2). Multiple shared mappings of the -same file are coherent with each other. This is consistent with Linux. - -When a file is mapped private, updates to the mapped memory are not visible to -other memory mappings. Updates to the mapped memory are also not reflected in -the file content as seen by read(2). If the file is changed after a private -mapping is created, for instance by write(2), the change to the file may or may -not be reflected in the private mapping. This is consistent with Linux. - -An `CachingInodeOperations` keeps track of ranges of memory that were modified -(or "dirtied"). When the file is explicitly synced via fsync(2), only the dirty -ranges are written out to the filesystem. Any error returned indicates a failure -to write all dirty memory of an `CachingInodeOperations` to the filesystem. In -this case the filesystem may be in an inconsistent state. The same operation can -be performed on the shared memory itself using msync(2). If neither fsync(2) nor -msync(2) is performed, then the dirty memory is written out in accordance with -the `CachingInodeOperations` eviction strategy (see below) and there is no -guarantee that memory will be written out successfully in full. - -### Memory allocation and eviction - -An `CachingInodeOperations` implements the following allocation and eviction -strategy: - -- Memory is allocated and brought up to date with the contents of a file when - a region of mapped memory is accessed (or "faulted on"). - -- Dirty memory is written out to filesystems when an fsync(2) or msync(2) - operation is performed on a memory mapped file, for all memory mapped files - when saved, and/or when there are no longer any memory mappings of a range - of a file, see munmap(2). As the latter implies, in the absence of a panic - or SIGKILL, dirty memory is written out for all memory mapped files when an - application exits. - -- Memory is freed when there are no longer any memory mappings of a range of a - file (e.g. when an application exits). This behavior is consistent with - Linux for shared memory that has been locked via mlock(2). - -Notably, memory is not allocated for read(2) or write(2) operations. This means -that reads and writes to the file are only accelerated by an -`CachingInodeOperations` if the file being read or written has been memory -mapped *and* if the shared memory has been accessed at the region being read or -written. This diverges from Linux which buffers memory into a page cache on -read(2) proactively (i.e. readahead) and delays writing it out to filesystems on -write(2) (i.e. writeback). The absence of these optimizations is not visible to -applications beyond less than optimal performance when repeatedly reading and/or -writing to same region of a file. See [Future Work](#future-work) for plans to -implement these optimizations. - -Additionally, memory held by `CachingInodeOperationss` is currently unbounded in -size. An `CachingInodeOperations` does not write out dirty memory and free it -under system memory pressure. This can cause pathological memory usage. - -When memory is written back, an `CachingInodeOperations` may write regions of -shared memory that were never modified. This is due to the strategy of -minimizing page faults (see below) and handling only a subset of memory write -faults. In the absence of an application or sentry crash, it is guaranteed that -if a region of shared memory was written to, it is written back to a filesystem. - -### Life of a shared memory mapping - -A file is memory mapped via mmap(2). For example, if `A` is an address, an -application may execute: - -``` -mmap(A, 0x1000, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0); -``` - -This creates a shared mapping of fd that reflects 4k of the contents of fd -starting at offset 0, accessible at address `A`. This in turn creates a virtual -memory area region ("vma") which indicates that [`A`, `A`+0x1000) is now a valid -address range for this application to access. - -At this point, memory has not been allocated in the file's -`CachingInodeOperations`. It is also the case that the address range [`A`, -`A`+0x1000) has not been mapped on the host on behalf of the application. If the -application then tries to modify 8 bytes of the shared memory: - -``` -char buffer[] = "aaaaaaaa"; -memcpy(A, buffer, 8); -``` - -The host then sends a `SIGSEGV` to the sentry because the address range [`A`, -`A`+8) is not mapped on the host. The `SIGSEGV` indicates that the memory was -accessed writable. The sentry looks up the vma associated with [`A`, `A`+8), -finds the file that was mapped and its `CachingInodeOperations`. It then calls -`CachingInodeOperations.Translate` which allocates memory to back [`A`, `A`+8). -It may choose to allocate more memory (i.e. do "readahead") to minimize -subsequent faults. - -Memory that is allocated comes from a host tmpfs file (see -`pgalloc.MemoryFile`). The host tmpfs file memory is brought up to date with the -contents of the mapped file on its filesystem. The region of the host tmpfs file -that reflects the mapped file is then mapped into the host address space of the -application so that subsequent memory accesses do not repeatedly generate a -`SIGSEGV`. - -The range that was allocated, including any extra memory allocation to minimize -faults, is marked dirty due to the write fault. This overcounts dirty memory if -the extra memory allocated is never modified. - -To make the scenario more interesting, imagine that this application spawns -another process and maps the same file in the exact same way: - -``` -mmap(A, 0x1000, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0); -``` - -Imagine that this process then tries to modify the file again but with only 4 -bytes: - -``` -char buffer[] = "bbbb"; -memcpy(A, buffer, 4); -``` - -Since the first process has already mapped and accessed the same region of the -file writable, `CachingInodeOperations.Translate` is called but returns the -memory that has already been allocated rather than allocating new memory. The -address range [`A`, `A`+0x1000) reflects the same cached view of the file as the -first process sees. For example, reading 8 bytes from the file from either -process via read(2) starting at offset 0 returns a consistent "bbbbaaaa". - -When this process no longer needs the shared memory, it may do: - -``` -munmap(A, 0x1000); -``` - -At this point, the modified memory cached by the `CachingInodeOperations` is not -written back to the file because it is still in use by the first process that -mapped it. When the first process also does: - -``` -munmap(A, 0x1000); -``` - -Then the last memory mapping of the file at the range [0, 0x1000) is gone. The -file's `CachingInodeOperations` then starts writing back memory marked dirty to -the file on its filesystem. Once writing completes, regardless of whether it was -successful, the `CachingInodeOperations` frees the memory cached at the range -[0, 0x1000). - -Subsequent read(2) or write(2) operations on the file go directly to the -filesystem since there no longer exists memory for it in its -`CachingInodeOperations`. - -## Future Work - -### Page cache - -The sentry does not yet implement the readahead and writeback optimizations for -read(2) and write(2) respectively. To do so, on read(2) and/or write(2) the -sentry must ensure that memory is allocated in a page cache to read or write -into. However, the sentry cannot boundlessly allocate memory. If it did, the -host would eventually OOM-kill the sentry+application process. This means that -the sentry must implement a page cache memory allocation strategy that is -bounded by a global user or container imposed limit. When this limit is -approached, the sentry must decide from which page cache memory should be freed -so that it can allocate more memory. If it makes a poor decision, the sentry may -end up freeing and re-allocating memory to back regions of files that are -frequently used, nullifying the optimization (and in some cases causing worse -performance due to the overhead of memory allocation and general management). -This is a form of "cache thrashing". - -In Linux, much research has been done to select and implement a lightweight but -optimal page cache eviction algorithm. Linux makes use of hardware page bits to -keep track of whether memory has been accessed. The sentry does not have direct -access to hardware. Implementing a similarly lightweight and optimal page cache -eviction algorithm will need to either introduce a kernel interface to obtain -these page bits or find a suitable alternative proxy for access events. - -In Linux, readahead happens by default but is not always ideal. For instance, -for files that are not read sequentially, it would be more ideal to simply read -from only those regions of the file rather than to optimistically cache some -number of bytes ahead of the read (up to 2MB in Linux) if the bytes cached won't -be accessed. Linux implements the fadvise64(2) system call for applications to -specify that a range of a file will not be accessed sequentially. The advice bit -FADV_RANDOM turns off the readahead optimization for the given range in the -given file. However fadvise64 is rarely used by applications so Linux implements -a readahead backoff strategy if reads are not sequential. To ensure that -application performance is not degraded, the sentry must implement a similar -backoff strategy. diff --git a/pkg/sentry/fs/fsutil/dirty_set_impl.go b/pkg/sentry/fs/fsutil/dirty_set_impl.go new file mode 100755 index 000000000..2510b81b3 --- /dev/null +++ b/pkg/sentry/fs/fsutil/dirty_set_impl.go @@ -0,0 +1,1274 @@ +package fsutil + +import ( + __generics_imported0 "gvisor.dev/gvisor/pkg/sentry/memmap" +) + +import ( + "bytes" + "fmt" +) + +const ( + // minDegree is the minimum degree of an internal node in a Set B-tree. + // + // - Any non-root node has at least minDegree-1 segments. + // + // - Any non-root internal (non-leaf) node has at least minDegree children. + // + // - The root node may have fewer than minDegree-1 segments, but it may + // only have 0 segments if the tree is empty. + // + // Our implementation requires minDegree >= 3. Higher values of minDegree + // usually improve performance, but increase memory usage for small sets. + DirtyminDegree = 3 + + DirtymaxDegree = 2 * DirtyminDegree +) + +// A Set is a mapping of segments with non-overlapping Range keys. The zero +// value for a Set is an empty set. Set values are not safely movable nor +// copyable. Set is thread-compatible. +// +// +stateify savable +type DirtySet struct { + root Dirtynode `state:".(*DirtySegmentDataSlices)"` +} + +// IsEmpty returns true if the set contains no segments. +func (s *DirtySet) IsEmpty() bool { + return s.root.nrSegments == 0 +} + +// IsEmptyRange returns true iff no segments in the set overlap the given +// range. This is semantically equivalent to s.SpanRange(r) == 0, but may be +// more efficient. +func (s *DirtySet) IsEmptyRange(r __generics_imported0.MappableRange) bool { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return true + } + _, gap := s.Find(r.Start) + if !gap.Ok() { + return false + } + return r.End <= gap.End() +} + +// Span returns the total size of all segments in the set. +func (s *DirtySet) Span() uint64 { + var sz uint64 + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sz += seg.Range().Length() + } + return sz +} + +// SpanRange returns the total size of the intersection of segments in the set +// with the given range. +func (s *DirtySet) SpanRange(r __generics_imported0.MappableRange) uint64 { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return 0 + } + var sz uint64 + for seg := s.LowerBoundSegment(r.Start); seg.Ok() && seg.Start() < r.End; seg = seg.NextSegment() { + sz += seg.Range().Intersect(r).Length() + } + return sz +} + +// FirstSegment returns the first segment in the set. If the set is empty, +// FirstSegment returns a terminal iterator. +func (s *DirtySet) FirstSegment() DirtyIterator { + if s.root.nrSegments == 0 { + return DirtyIterator{} + } + return s.root.firstSegment() +} + +// LastSegment returns the last segment in the set. If the set is empty, +// LastSegment returns a terminal iterator. +func (s *DirtySet) LastSegment() DirtyIterator { + if s.root.nrSegments == 0 { + return DirtyIterator{} + } + return s.root.lastSegment() +} + +// FirstGap returns the first gap in the set. +func (s *DirtySet) FirstGap() DirtyGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[0] + } + return DirtyGapIterator{n, 0} +} + +// LastGap returns the last gap in the set. +func (s *DirtySet) LastGap() DirtyGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[n.nrSegments] + } + return DirtyGapIterator{n, n.nrSegments} +} + +// Find returns the segment or gap whose range contains the given key. If a +// segment is found, the returned Iterator is non-terminal and the +// returned GapIterator is terminal. Otherwise, the returned Iterator is +// terminal and the returned GapIterator is non-terminal. +func (s *DirtySet) Find(key uint64) (DirtyIterator, DirtyGapIterator) { + n := &s.root + for { + + lower := 0 + upper := n.nrSegments + for lower < upper { + i := lower + (upper-lower)/2 + if r := n.keys[i]; key < r.End { + if key >= r.Start { + return DirtyIterator{n, i}, DirtyGapIterator{} + } + upper = i + } else { + lower = i + 1 + } + } + i := lower + if !n.hasChildren { + return DirtyIterator{}, DirtyGapIterator{n, i} + } + n = n.children[i] + } +} + +// FindSegment returns the segment whose range contains the given key. If no +// such segment exists, FindSegment returns a terminal iterator. +func (s *DirtySet) FindSegment(key uint64) DirtyIterator { + seg, _ := s.Find(key) + return seg +} + +// LowerBoundSegment returns the segment with the lowest range that contains a +// key greater than or equal to min. If no such segment exists, +// LowerBoundSegment returns a terminal iterator. +func (s *DirtySet) LowerBoundSegment(min uint64) DirtyIterator { + seg, gap := s.Find(min) + if seg.Ok() { + return seg + } + return gap.NextSegment() +} + +// UpperBoundSegment returns the segment with the highest range that contains a +// key less than or equal to max. If no such segment exists, UpperBoundSegment +// returns a terminal iterator. +func (s *DirtySet) UpperBoundSegment(max uint64) DirtyIterator { + seg, gap := s.Find(max) + if seg.Ok() { + return seg + } + return gap.PrevSegment() +} + +// FindGap returns the gap containing the given key. If no such gap exists +// (i.e. the set contains a segment containing that key), FindGap returns a +// terminal iterator. +func (s *DirtySet) FindGap(key uint64) DirtyGapIterator { + _, gap := s.Find(key) + return gap +} + +// LowerBoundGap returns the gap with the lowest range that is greater than or +// equal to min. +func (s *DirtySet) LowerBoundGap(min uint64) DirtyGapIterator { + seg, gap := s.Find(min) + if gap.Ok() { + return gap + } + return seg.NextGap() +} + +// UpperBoundGap returns the gap with the highest range that is less than or +// equal to max. +func (s *DirtySet) UpperBoundGap(max uint64) DirtyGapIterator { + seg, gap := s.Find(max) + if gap.Ok() { + return gap + } + return seg.PrevGap() +} + +// Add inserts the given segment into the set and returns true. If the new +// segment can be merged with adjacent segments, Add will do so. If the new +// segment would overlap an existing segment, Add returns false. If Add +// succeeds, all existing iterators are invalidated. +func (s *DirtySet) Add(r __generics_imported0.MappableRange, val DirtyInfo) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.Insert(gap, r, val) + return true +} + +// AddWithoutMerging inserts the given segment into the set and returns true. +// If it would overlap an existing segment, AddWithoutMerging does nothing and +// returns false. If AddWithoutMerging succeeds, all existing iterators are +// invalidated. +func (s *DirtySet) AddWithoutMerging(r __generics_imported0.MappableRange, val DirtyInfo) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.InsertWithoutMergingUnchecked(gap, r, val) + return true +} + +// Insert inserts the given segment into the given gap. If the new segment can +// be merged with adjacent segments, Insert will do so. Insert returns an +// iterator to the segment containing the inserted value (which may have been +// merged with other values). All existing iterators (including gap, but not +// including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, Insert panics. +// +// Insert is semantically equivalent to a InsertWithoutMerging followed by a +// Merge, but may be more efficient. Note that there is no unchecked variant of +// Insert since Insert must retrieve and inspect gap's predecessor and +// successor segments regardless. +func (s *DirtySet) Insert(gap DirtyGapIterator, r __generics_imported0.MappableRange, val DirtyInfo) DirtyIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + prev, next := gap.PrevSegment(), gap.NextSegment() + if prev.Ok() && prev.End() > r.Start { + panic(fmt.Sprintf("new segment %v overlaps predecessor %v", r, prev.Range())) + } + if next.Ok() && next.Start() < r.End { + panic(fmt.Sprintf("new segment %v overlaps successor %v", r, next.Range())) + } + if prev.Ok() && prev.End() == r.Start { + if mval, ok := (dirtySetFunctions{}).Merge(prev.Range(), prev.Value(), r, val); ok { + prev.SetEndUnchecked(r.End) + prev.SetValue(mval) + if next.Ok() && next.Start() == r.End { + val = mval + if mval, ok := (dirtySetFunctions{}).Merge(prev.Range(), val, next.Range(), next.Value()); ok { + prev.SetEndUnchecked(next.End()) + prev.SetValue(mval) + return s.Remove(next).PrevSegment() + } + } + return prev + } + } + if next.Ok() && next.Start() == r.End { + if mval, ok := (dirtySetFunctions{}).Merge(r, val, next.Range(), next.Value()); ok { + next.SetStartUnchecked(r.Start) + next.SetValue(mval) + return next + } + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMerging inserts the given segment into the given gap and +// returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, +// InsertWithoutMerging panics. +func (s *DirtySet) InsertWithoutMerging(gap DirtyGapIterator, r __generics_imported0.MappableRange, val DirtyInfo) DirtyIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if gr := gap.Range(); !gr.IsSupersetOf(r) { + panic(fmt.Sprintf("cannot insert segment range %v into gap range %v", r, gr)) + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMergingUnchecked inserts the given segment into the given gap +// and returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// Preconditions: r.Start >= gap.Start(); r.End <= gap.End(). +func (s *DirtySet) InsertWithoutMergingUnchecked(gap DirtyGapIterator, r __generics_imported0.MappableRange, val DirtyInfo) DirtyIterator { + gap = gap.node.rebalanceBeforeInsert(gap) + copy(gap.node.keys[gap.index+1:], gap.node.keys[gap.index:gap.node.nrSegments]) + copy(gap.node.values[gap.index+1:], gap.node.values[gap.index:gap.node.nrSegments]) + gap.node.keys[gap.index] = r + gap.node.values[gap.index] = val + gap.node.nrSegments++ + return DirtyIterator{gap.node, gap.index} +} + +// Remove removes the given segment and returns an iterator to the vacated gap. +// All existing iterators (including seg, but not including the returned +// iterator) are invalidated. +func (s *DirtySet) Remove(seg DirtyIterator) DirtyGapIterator { + + if seg.node.hasChildren { + + victim := seg.PrevSegment() + + seg.SetRangeUnchecked(victim.Range()) + seg.SetValue(victim.Value()) + return s.Remove(victim).NextGap() + } + copy(seg.node.keys[seg.index:], seg.node.keys[seg.index+1:seg.node.nrSegments]) + copy(seg.node.values[seg.index:], seg.node.values[seg.index+1:seg.node.nrSegments]) + dirtySetFunctions{}.ClearValue(&seg.node.values[seg.node.nrSegments-1]) + seg.node.nrSegments-- + return seg.node.rebalanceAfterRemove(DirtyGapIterator{seg.node, seg.index}) +} + +// RemoveAll removes all segments from the set. All existing iterators are +// invalidated. +func (s *DirtySet) RemoveAll() { + s.root = Dirtynode{} +} + +// RemoveRange removes all segments in the given range. An iterator to the +// newly formed gap is returned, and all existing iterators are invalidated. +func (s *DirtySet) RemoveRange(r __generics_imported0.MappableRange) DirtyGapIterator { + seg, gap := s.Find(r.Start) + if seg.Ok() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + for seg = gap.NextSegment(); seg.Ok() && seg.Start() < r.End; seg = gap.NextSegment() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + return gap +} + +// Merge attempts to merge two neighboring segments. If successful, Merge +// returns an iterator to the merged segment, and all existing iterators are +// invalidated. Otherwise, Merge returns a terminal iterator. +// +// If first is not the predecessor of second, Merge panics. +func (s *DirtySet) Merge(first, second DirtyIterator) DirtyIterator { + if first.NextSegment() != second { + panic(fmt.Sprintf("attempt to merge non-neighboring segments %v, %v", first.Range(), second.Range())) + } + return s.MergeUnchecked(first, second) +} + +// MergeUnchecked attempts to merge two neighboring segments. If successful, +// MergeUnchecked returns an iterator to the merged segment, and all existing +// iterators are invalidated. Otherwise, MergeUnchecked returns a terminal +// iterator. +// +// Precondition: first is the predecessor of second: first.NextSegment() == +// second, first == second.PrevSegment(). +func (s *DirtySet) MergeUnchecked(first, second DirtyIterator) DirtyIterator { + if first.End() == second.Start() { + if mval, ok := (dirtySetFunctions{}).Merge(first.Range(), first.Value(), second.Range(), second.Value()); ok { + + first.SetEndUnchecked(second.End()) + first.SetValue(mval) + return s.Remove(second).PrevSegment() + } + } + return DirtyIterator{} +} + +// MergeAll attempts to merge all adjacent segments in the set. All existing +// iterators are invalidated. +func (s *DirtySet) MergeAll() { + seg := s.FirstSegment() + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeRange attempts to merge all adjacent segments that contain a key in the +// specific range. All existing iterators are invalidated. +func (s *DirtySet) MergeRange(r __generics_imported0.MappableRange) { + seg := s.LowerBoundSegment(r.Start) + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() && next.Range().Start < r.End { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeAdjacent attempts to merge the segment containing r.Start with its +// predecessor, and the segment containing r.End-1 with its successor. +func (s *DirtySet) MergeAdjacent(r __generics_imported0.MappableRange) { + first := s.FindSegment(r.Start) + if first.Ok() { + if prev := first.PrevSegment(); prev.Ok() { + s.Merge(prev, first) + } + } + last := s.FindSegment(r.End - 1) + if last.Ok() { + if next := last.NextSegment(); next.Ok() { + s.Merge(last, next) + } + } +} + +// Split splits the given segment at the given key and returns iterators to the +// two resulting segments. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +// +// If the segment cannot be split at split (because split is at the start or +// end of the segment's range, so splitting would produce a segment with zero +// length, or because split falls outside the segment's range altogether), +// Split panics. +func (s *DirtySet) Split(seg DirtyIterator, split uint64) (DirtyIterator, DirtyIterator) { + if !seg.Range().CanSplitAt(split) { + panic(fmt.Sprintf("can't split %v at %v", seg.Range(), split)) + } + return s.SplitUnchecked(seg, split) +} + +// SplitUnchecked splits the given segment at the given key and returns +// iterators to the two resulting segments. All existing iterators (including +// seg, but not including the returned iterators) are invalidated. +// +// Preconditions: seg.Start() < key < seg.End(). +func (s *DirtySet) SplitUnchecked(seg DirtyIterator, split uint64) (DirtyIterator, DirtyIterator) { + val1, val2 := (dirtySetFunctions{}).Split(seg.Range(), seg.Value(), split) + end2 := seg.End() + seg.SetEndUnchecked(split) + seg.SetValue(val1) + seg2 := s.InsertWithoutMergingUnchecked(seg.NextGap(), __generics_imported0.MappableRange{split, end2}, val2) + + return seg2.PrevSegment(), seg2 +} + +// SplitAt splits the segment straddling split, if one exists. SplitAt returns +// true if a segment was split and false otherwise. If SplitAt splits a +// segment, all existing iterators are invalidated. +func (s *DirtySet) SplitAt(split uint64) bool { + if seg := s.FindSegment(split); seg.Ok() && seg.Range().CanSplitAt(split) { + s.SplitUnchecked(seg, split) + return true + } + return false +} + +// Isolate ensures that the given segment's range does not escape r by +// splitting at r.Start and r.End if necessary, and returns an updated iterator +// to the bounded segment. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +func (s *DirtySet) Isolate(seg DirtyIterator, r __generics_imported0.MappableRange) DirtyIterator { + if seg.Range().CanSplitAt(r.Start) { + _, seg = s.SplitUnchecked(seg, r.Start) + } + if seg.Range().CanSplitAt(r.End) { + seg, _ = s.SplitUnchecked(seg, r.End) + } + return seg +} + +// ApplyContiguous applies a function to a contiguous range of segments, +// splitting if necessary. The function is applied until the first gap is +// encountered, at which point the gap is returned. If the function is applied +// across the entire range, a terminal gap is returned. All existing iterators +// are invalidated. +// +// N.B. The Iterator must not be invalidated by the function. +func (s *DirtySet) ApplyContiguous(r __generics_imported0.MappableRange, fn func(seg DirtyIterator)) DirtyGapIterator { + seg, gap := s.Find(r.Start) + if !seg.Ok() { + return gap + } + for { + seg = s.Isolate(seg, r) + fn(seg) + if seg.End() >= r.End { + return DirtyGapIterator{} + } + gap = seg.NextGap() + if !gap.IsEmpty() { + return gap + } + seg = gap.NextSegment() + if !seg.Ok() { + + return DirtyGapIterator{} + } + } +} + +// +stateify savable +type Dirtynode struct { + // An internal binary tree node looks like: + // + // K + // / \ + // Cl Cr + // + // where all keys in the subtree rooted by Cl (the left subtree) are less + // than K (the key of the parent node), and all keys in the subtree rooted + // by Cr (the right subtree) are greater than K. + // + // An internal B-tree node's indexes work out to look like: + // + // K0 K1 K2 ... Kn-1 + // / \/ \/ \ ... / \ + // C0 C1 C2 C3 ... Cn-1 Cn + // + // where n is nrSegments. + nrSegments int + + // parent is a pointer to this node's parent. If this node is root, parent + // is nil. + parent *Dirtynode + + // parentIndex is the index of this node in parent.children. + parentIndex int + + // Flag for internal nodes that is technically redundant with "children[0] + // != nil", but is stored in the first cache line. "hasChildren" rather + // than "isLeaf" because false must be the correct value for an empty root. + hasChildren bool + + // Nodes store keys and values in separate arrays to maximize locality in + // the common case (scanning keys for lookup). + keys [DirtymaxDegree - 1]__generics_imported0.MappableRange + values [DirtymaxDegree - 1]DirtyInfo + children [DirtymaxDegree]*Dirtynode +} + +// firstSegment returns the first segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *Dirtynode) firstSegment() DirtyIterator { + for n.hasChildren { + n = n.children[0] + } + return DirtyIterator{n, 0} +} + +// lastSegment returns the last segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *Dirtynode) lastSegment() DirtyIterator { + for n.hasChildren { + n = n.children[n.nrSegments] + } + return DirtyIterator{n, n.nrSegments - 1} +} + +func (n *Dirtynode) prevSibling() *Dirtynode { + if n.parent == nil || n.parentIndex == 0 { + return nil + } + return n.parent.children[n.parentIndex-1] +} + +func (n *Dirtynode) nextSibling() *Dirtynode { + if n.parent == nil || n.parentIndex == n.parent.nrSegments { + return nil + } + return n.parent.children[n.parentIndex+1] +} + +// rebalanceBeforeInsert splits n and its ancestors if they are full, as +// required for insertion, and returns an updated iterator to the position +// represented by gap. +func (n *Dirtynode) rebalanceBeforeInsert(gap DirtyGapIterator) DirtyGapIterator { + if n.parent != nil { + gap = n.parent.rebalanceBeforeInsert(gap) + } + if n.nrSegments < DirtymaxDegree-1 { + return gap + } + if n.parent == nil { + + left := &Dirtynode{ + nrSegments: DirtyminDegree - 1, + parent: n, + parentIndex: 0, + hasChildren: n.hasChildren, + } + right := &Dirtynode{ + nrSegments: DirtyminDegree - 1, + parent: n, + parentIndex: 1, + hasChildren: n.hasChildren, + } + copy(left.keys[:DirtyminDegree-1], n.keys[:DirtyminDegree-1]) + copy(left.values[:DirtyminDegree-1], n.values[:DirtyminDegree-1]) + copy(right.keys[:DirtyminDegree-1], n.keys[DirtyminDegree:]) + copy(right.values[:DirtyminDegree-1], n.values[DirtyminDegree:]) + n.keys[0], n.values[0] = n.keys[DirtyminDegree-1], n.values[DirtyminDegree-1] + DirtyzeroValueSlice(n.values[1:]) + if n.hasChildren { + copy(left.children[:DirtyminDegree], n.children[:DirtyminDegree]) + copy(right.children[:DirtyminDegree], n.children[DirtyminDegree:]) + DirtyzeroNodeSlice(n.children[2:]) + for i := 0; i < DirtyminDegree; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + right.children[i].parent = right + right.children[i].parentIndex = i + } + } + n.nrSegments = 1 + n.hasChildren = true + n.children[0] = left + n.children[1] = right + if gap.node != n { + return gap + } + if gap.index < DirtyminDegree { + return DirtyGapIterator{left, gap.index} + } + return DirtyGapIterator{right, gap.index - DirtyminDegree} + } + + copy(n.parent.keys[n.parentIndex+1:], n.parent.keys[n.parentIndex:n.parent.nrSegments]) + copy(n.parent.values[n.parentIndex+1:], n.parent.values[n.parentIndex:n.parent.nrSegments]) + n.parent.keys[n.parentIndex], n.parent.values[n.parentIndex] = n.keys[DirtyminDegree-1], n.values[DirtyminDegree-1] + copy(n.parent.children[n.parentIndex+2:], n.parent.children[n.parentIndex+1:n.parent.nrSegments+1]) + for i := n.parentIndex + 2; i < n.parent.nrSegments+2; i++ { + n.parent.children[i].parentIndex = i + } + sibling := &Dirtynode{ + nrSegments: DirtyminDegree - 1, + parent: n.parent, + parentIndex: n.parentIndex + 1, + hasChildren: n.hasChildren, + } + n.parent.children[n.parentIndex+1] = sibling + n.parent.nrSegments++ + copy(sibling.keys[:DirtyminDegree-1], n.keys[DirtyminDegree:]) + copy(sibling.values[:DirtyminDegree-1], n.values[DirtyminDegree:]) + DirtyzeroValueSlice(n.values[DirtyminDegree-1:]) + if n.hasChildren { + copy(sibling.children[:DirtyminDegree], n.children[DirtyminDegree:]) + DirtyzeroNodeSlice(n.children[DirtyminDegree:]) + for i := 0; i < DirtyminDegree; i++ { + sibling.children[i].parent = sibling + sibling.children[i].parentIndex = i + } + } + n.nrSegments = DirtyminDegree - 1 + + if gap.node != n { + return gap + } + if gap.index < DirtyminDegree { + return gap + } + return DirtyGapIterator{sibling, gap.index - DirtyminDegree} +} + +// rebalanceAfterRemove "unsplits" n and its ancestors if they are deficient +// (contain fewer segments than required by B-tree invariants), as required for +// removal, and returns an updated iterator to the position represented by gap. +// +// Precondition: n is the only node in the tree that may currently violate a +// B-tree invariant. +func (n *Dirtynode) rebalanceAfterRemove(gap DirtyGapIterator) DirtyGapIterator { + for { + if n.nrSegments >= DirtyminDegree-1 { + return gap + } + if n.parent == nil { + + return gap + } + + if sibling := n.prevSibling(); sibling != nil && sibling.nrSegments >= DirtyminDegree { + copy(n.keys[1:], n.keys[:n.nrSegments]) + copy(n.values[1:], n.values[:n.nrSegments]) + n.keys[0] = n.parent.keys[n.parentIndex-1] + n.values[0] = n.parent.values[n.parentIndex-1] + n.parent.keys[n.parentIndex-1] = sibling.keys[sibling.nrSegments-1] + n.parent.values[n.parentIndex-1] = sibling.values[sibling.nrSegments-1] + dirtySetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + copy(n.children[1:], n.children[:n.nrSegments+1]) + n.children[0] = sibling.children[sibling.nrSegments] + sibling.children[sibling.nrSegments] = nil + n.children[0].parent = n + n.children[0].parentIndex = 0 + for i := 1; i < n.nrSegments+2; i++ { + n.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling && gap.index == sibling.nrSegments { + return DirtyGapIterator{n, 0} + } + if gap.node == n { + return DirtyGapIterator{n, gap.index + 1} + } + return gap + } + if sibling := n.nextSibling(); sibling != nil && sibling.nrSegments >= DirtyminDegree { + n.keys[n.nrSegments] = n.parent.keys[n.parentIndex] + n.values[n.nrSegments] = n.parent.values[n.parentIndex] + n.parent.keys[n.parentIndex] = sibling.keys[0] + n.parent.values[n.parentIndex] = sibling.values[0] + copy(sibling.keys[:sibling.nrSegments-1], sibling.keys[1:]) + copy(sibling.values[:sibling.nrSegments-1], sibling.values[1:]) + dirtySetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + n.children[n.nrSegments+1] = sibling.children[0] + copy(sibling.children[:sibling.nrSegments], sibling.children[1:]) + sibling.children[sibling.nrSegments] = nil + n.children[n.nrSegments+1].parent = n + n.children[n.nrSegments+1].parentIndex = n.nrSegments + 1 + for i := 0; i < sibling.nrSegments; i++ { + sibling.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling { + if gap.index == 0 { + return DirtyGapIterator{n, n.nrSegments} + } + return DirtyGapIterator{sibling, gap.index - 1} + } + return gap + } + + p := n.parent + if p.nrSegments == 1 { + + left, right := p.children[0], p.children[1] + p.nrSegments = left.nrSegments + right.nrSegments + 1 + p.hasChildren = left.hasChildren + p.keys[left.nrSegments] = p.keys[0] + p.values[left.nrSegments] = p.values[0] + copy(p.keys[:left.nrSegments], left.keys[:left.nrSegments]) + copy(p.values[:left.nrSegments], left.values[:left.nrSegments]) + copy(p.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(p.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(p.children[:left.nrSegments+1], left.children[:left.nrSegments+1]) + copy(p.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := 0; i < p.nrSegments+1; i++ { + p.children[i].parent = p + p.children[i].parentIndex = i + } + } else { + p.children[0] = nil + p.children[1] = nil + } + if gap.node == left { + return DirtyGapIterator{p, gap.index} + } + if gap.node == right { + return DirtyGapIterator{p, gap.index + left.nrSegments + 1} + } + return gap + } + // Merge n and either sibling, along with the segment separating the + // two, into whichever of the two nodes comes first. This is the + // reverse of the non-root splitting case in + // node.rebalanceBeforeInsert. + var left, right *Dirtynode + if n.parentIndex > 0 { + left = n.prevSibling() + right = n + } else { + left = n + right = n.nextSibling() + } + + if gap.node == right { + gap = DirtyGapIterator{left, gap.index + left.nrSegments + 1} + } + left.keys[left.nrSegments] = p.keys[left.parentIndex] + left.values[left.nrSegments] = p.values[left.parentIndex] + copy(left.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(left.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(left.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := left.nrSegments + 1; i < left.nrSegments+right.nrSegments+2; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + } + } + left.nrSegments += right.nrSegments + 1 + copy(p.keys[left.parentIndex:], p.keys[left.parentIndex+1:p.nrSegments]) + copy(p.values[left.parentIndex:], p.values[left.parentIndex+1:p.nrSegments]) + dirtySetFunctions{}.ClearValue(&p.values[p.nrSegments-1]) + copy(p.children[left.parentIndex+1:], p.children[left.parentIndex+2:p.nrSegments+1]) + for i := 0; i < p.nrSegments; i++ { + p.children[i].parentIndex = i + } + p.children[p.nrSegments] = nil + p.nrSegments-- + + n = p + } +} + +// A Iterator is conceptually one of: +// +// - A pointer to a segment in a set; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Iterators are copyable values and are meaningfully equality-comparable. The +// zero value of Iterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type DirtyIterator struct { + // node is the node containing the iterated segment. If the iterator is + // terminal, node is nil. + node *Dirtynode + + // index is the index of the segment in node.keys/values. + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (seg DirtyIterator) Ok() bool { + return seg.node != nil +} + +// Range returns the iterated segment's range key. +func (seg DirtyIterator) Range() __generics_imported0.MappableRange { + return seg.node.keys[seg.index] +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (seg DirtyIterator) Start() uint64 { + return seg.node.keys[seg.index].Start +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (seg DirtyIterator) End() uint64 { + return seg.node.keys[seg.index].End +} + +// SetRangeUnchecked mutates the iterated segment's range key. This operation +// does not invalidate any iterators. +// +// Preconditions: +// +// - r.Length() > 0. +// +// - The new range must not overlap an existing one: If seg.NextSegment().Ok(), +// then r.end <= seg.NextSegment().Start(); if seg.PrevSegment().Ok(), then +// r.start >= seg.PrevSegment().End(). +func (seg DirtyIterator) SetRangeUnchecked(r __generics_imported0.MappableRange) { + seg.node.keys[seg.index] = r +} + +// SetRange mutates the iterated segment's range key. If the new range would +// cause the iterated segment to overlap another segment, or if the new range +// is invalid, SetRange panics. This operation does not invalidate any +// iterators. +func (seg DirtyIterator) SetRange(r __generics_imported0.MappableRange) { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if prev := seg.PrevSegment(); prev.Ok() && r.Start < prev.End() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, prev.Range())) + } + if next := seg.NextSegment(); next.Ok() && r.End > next.Start() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, next.Range())) + } + seg.SetRangeUnchecked(r) +} + +// SetStartUnchecked mutates the iterated segment's start. This operation does +// not invalidate any iterators. +// +// Preconditions: The new start must be valid: start < seg.End(); if +// seg.PrevSegment().Ok(), then start >= seg.PrevSegment().End(). +func (seg DirtyIterator) SetStartUnchecked(start uint64) { + seg.node.keys[seg.index].Start = start +} + +// SetStart mutates the iterated segment's start. If the new start value would +// cause the iterated segment to overlap another segment, or would result in an +// invalid range, SetStart panics. This operation does not invalidate any +// iterators. +func (seg DirtyIterator) SetStart(start uint64) { + if start >= seg.End() { + panic(fmt.Sprintf("new start %v would invalidate segment range %v", start, seg.Range())) + } + if prev := seg.PrevSegment(); prev.Ok() && start < prev.End() { + panic(fmt.Sprintf("new start %v would cause segment range %v to overlap segment range %v", start, seg.Range(), prev.Range())) + } + seg.SetStartUnchecked(start) +} + +// SetEndUnchecked mutates the iterated segment's end. This operation does not +// invalidate any iterators. +// +// Preconditions: The new end must be valid: end > seg.Start(); if +// seg.NextSegment().Ok(), then end <= seg.NextSegment().Start(). +func (seg DirtyIterator) SetEndUnchecked(end uint64) { + seg.node.keys[seg.index].End = end +} + +// SetEnd mutates the iterated segment's end. If the new end value would cause +// the iterated segment to overlap another segment, or would result in an +// invalid range, SetEnd panics. This operation does not invalidate any +// iterators. +func (seg DirtyIterator) SetEnd(end uint64) { + if end <= seg.Start() { + panic(fmt.Sprintf("new end %v would invalidate segment range %v", end, seg.Range())) + } + if next := seg.NextSegment(); next.Ok() && end > next.Start() { + panic(fmt.Sprintf("new end %v would cause segment range %v to overlap segment range %v", end, seg.Range(), next.Range())) + } + seg.SetEndUnchecked(end) +} + +// Value returns a copy of the iterated segment's value. +func (seg DirtyIterator) Value() DirtyInfo { + return seg.node.values[seg.index] +} + +// ValuePtr returns a pointer to the iterated segment's value. The pointer is +// invalidated if the iterator is invalidated. This operation does not +// invalidate any iterators. +func (seg DirtyIterator) ValuePtr() *DirtyInfo { + return &seg.node.values[seg.index] +} + +// SetValue mutates the iterated segment's value. This operation does not +// invalidate any iterators. +func (seg DirtyIterator) SetValue(val DirtyInfo) { + seg.node.values[seg.index] = val +} + +// PrevSegment returns the iterated segment's predecessor. If there is no +// preceding segment, PrevSegment returns a terminal iterator. +func (seg DirtyIterator) PrevSegment() DirtyIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index].lastSegment() + } + if seg.index > 0 { + return DirtyIterator{seg.node, seg.index - 1} + } + if seg.node.parent == nil { + return DirtyIterator{} + } + return DirtysegmentBeforePosition(seg.node.parent, seg.node.parentIndex) +} + +// NextSegment returns the iterated segment's successor. If there is no +// succeeding segment, NextSegment returns a terminal iterator. +func (seg DirtyIterator) NextSegment() DirtyIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment() + } + if seg.index < seg.node.nrSegments-1 { + return DirtyIterator{seg.node, seg.index + 1} + } + if seg.node.parent == nil { + return DirtyIterator{} + } + return DirtysegmentAfterPosition(seg.node.parent, seg.node.parentIndex) +} + +// PrevGap returns the gap immediately before the iterated segment. +func (seg DirtyIterator) PrevGap() DirtyGapIterator { + if seg.node.hasChildren { + + return seg.node.children[seg.index].lastSegment().NextGap() + } + return DirtyGapIterator{seg.node, seg.index} +} + +// NextGap returns the gap immediately after the iterated segment. +func (seg DirtyIterator) NextGap() DirtyGapIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment().PrevGap() + } + return DirtyGapIterator{seg.node, seg.index + 1} +} + +// PrevNonEmpty returns the iterated segment's predecessor if it is adjacent, +// or the gap before the iterated segment otherwise. If seg.Start() == +// Functions.MinKey(), PrevNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by PrevNonEmpty will be +// non-terminal. +func (seg DirtyIterator) PrevNonEmpty() (DirtyIterator, DirtyGapIterator) { + gap := seg.PrevGap() + if gap.Range().Length() != 0 { + return DirtyIterator{}, gap + } + return gap.PrevSegment(), DirtyGapIterator{} +} + +// NextNonEmpty returns the iterated segment's successor if it is adjacent, or +// the gap after the iterated segment otherwise. If seg.End() == +// Functions.MaxKey(), NextNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by NextNonEmpty will be +// non-terminal. +func (seg DirtyIterator) NextNonEmpty() (DirtyIterator, DirtyGapIterator) { + gap := seg.NextGap() + if gap.Range().Length() != 0 { + return DirtyIterator{}, gap + } + return gap.NextSegment(), DirtyGapIterator{} +} + +// A GapIterator is conceptually one of: +// +// - A pointer to a position between two segments, before the first segment, or +// after the last segment in a set, called a *gap*; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Note that the gap between two adjacent segments exists (iterators to it are +// non-terminal), but has a length of zero. GapIterator.IsEmpty returns true +// for such gaps. An empty set contains a single gap, spanning the entire range +// of the set's keys. +// +// GapIterators are copyable values and are meaningfully equality-comparable. +// The zero value of GapIterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type DirtyGapIterator struct { + // The representation of a GapIterator is identical to that of an Iterator, + // except that index corresponds to positions between segments in the same + // way as for node.children (see comment for node.nrSegments). + node *Dirtynode + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (gap DirtyGapIterator) Ok() bool { + return gap.node != nil +} + +// Range returns the range spanned by the iterated gap. +func (gap DirtyGapIterator) Range() __generics_imported0.MappableRange { + return __generics_imported0.MappableRange{gap.Start(), gap.End()} +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (gap DirtyGapIterator) Start() uint64 { + if ps := gap.PrevSegment(); ps.Ok() { + return ps.End() + } + return dirtySetFunctions{}.MinKey() +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (gap DirtyGapIterator) End() uint64 { + if ns := gap.NextSegment(); ns.Ok() { + return ns.Start() + } + return dirtySetFunctions{}.MaxKey() +} + +// IsEmpty returns true if the iterated gap is empty (that is, the "gap" is +// between two adjacent segments.) +func (gap DirtyGapIterator) IsEmpty() bool { + return gap.Range().Length() == 0 +} + +// PrevSegment returns the segment immediately before the iterated gap. If no +// such segment exists, PrevSegment returns a terminal iterator. +func (gap DirtyGapIterator) PrevSegment() DirtyIterator { + return DirtysegmentBeforePosition(gap.node, gap.index) +} + +// NextSegment returns the segment immediately after the iterated gap. If no +// such segment exists, NextSegment returns a terminal iterator. +func (gap DirtyGapIterator) NextSegment() DirtyIterator { + return DirtysegmentAfterPosition(gap.node, gap.index) +} + +// PrevGap returns the iterated gap's predecessor. If no such gap exists, +// PrevGap returns a terminal iterator. +func (gap DirtyGapIterator) PrevGap() DirtyGapIterator { + seg := gap.PrevSegment() + if !seg.Ok() { + return DirtyGapIterator{} + } + return seg.PrevGap() +} + +// NextGap returns the iterated gap's successor. If no such gap exists, NextGap +// returns a terminal iterator. +func (gap DirtyGapIterator) NextGap() DirtyGapIterator { + seg := gap.NextSegment() + if !seg.Ok() { + return DirtyGapIterator{} + } + return seg.NextGap() +} + +// segmentBeforePosition returns the predecessor segment of the position given +// by n.children[i], which may or may not contain a child. If no such segment +// exists, segmentBeforePosition returns a terminal iterator. +func DirtysegmentBeforePosition(n *Dirtynode, i int) DirtyIterator { + for i == 0 { + if n.parent == nil { + return DirtyIterator{} + } + n, i = n.parent, n.parentIndex + } + return DirtyIterator{n, i - 1} +} + +// segmentAfterPosition returns the successor segment of the position given by +// n.children[i], which may or may not contain a child. If no such segment +// exists, segmentAfterPosition returns a terminal iterator. +func DirtysegmentAfterPosition(n *Dirtynode, i int) DirtyIterator { + for i == n.nrSegments { + if n.parent == nil { + return DirtyIterator{} + } + n, i = n.parent, n.parentIndex + } + return DirtyIterator{n, i} +} + +func DirtyzeroValueSlice(slice []DirtyInfo) { + + for i := range slice { + dirtySetFunctions{}.ClearValue(&slice[i]) + } +} + +func DirtyzeroNodeSlice(slice []*Dirtynode) { + for i := range slice { + slice[i] = nil + } +} + +// String stringifies a Set for debugging. +func (s *DirtySet) String() string { + return s.root.String() +} + +// String stringifies a node (and all of its children) for debugging. +func (n *Dirtynode) String() string { + var buf bytes.Buffer + n.writeDebugString(&buf, "") + return buf.String() +} + +func (n *Dirtynode) writeDebugString(buf *bytes.Buffer, prefix string) { + if n.hasChildren != (n.nrSegments > 0 && n.children[0] != nil) { + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent value of hasChildren: got %v, want %v\n", n.hasChildren, !n.hasChildren)) + } + for i := 0; i < n.nrSegments; i++ { + if child := n.children[i]; child != nil { + cprefix := fmt.Sprintf("%s- % 3d ", prefix, i) + if child.parent != n || child.parentIndex != i { + buf.WriteString(cprefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent linkage to parent: got (%p, %d), want (%p, %d)\n", child.parent, child.parentIndex, n, i)) + } + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, i)) + } + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + } + if child := n.children[n.nrSegments]; child != nil { + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, n.nrSegments)) + } +} + +// SegmentDataSlices represents segments from a set as slices of start, end, and +// values. SegmentDataSlices is primarily used as an intermediate representation +// for save/restore and the layout here is optimized for that. +// +// +stateify savable +type DirtySegmentDataSlices struct { + Start []uint64 + End []uint64 + Values []DirtyInfo +} + +// ExportSortedSlice returns a copy of all segments in the given set, in ascending +// key order. +func (s *DirtySet) ExportSortedSlices() *DirtySegmentDataSlices { + var sds DirtySegmentDataSlices + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sds.Start = append(sds.Start, seg.Start()) + sds.End = append(sds.End, seg.End()) + sds.Values = append(sds.Values, seg.Value()) + } + sds.Start = sds.Start[:len(sds.Start):len(sds.Start)] + sds.End = sds.End[:len(sds.End):len(sds.End)] + sds.Values = sds.Values[:len(sds.Values):len(sds.Values)] + return &sds +} + +// ImportSortedSlice initializes the given set from the given slice. +// +// Preconditions: s must be empty. sds must represent a valid set (the segments +// in sds must have valid lengths that do not overlap). The segments in sds +// must be sorted in ascending key order. +func (s *DirtySet) ImportSortedSlices(sds *DirtySegmentDataSlices) error { + if !s.IsEmpty() { + return fmt.Errorf("cannot import into non-empty set %v", s) + } + gap := s.FirstGap() + for i := range sds.Start { + r := __generics_imported0.MappableRange{sds.Start[i], sds.End[i]} + if !gap.Range().IsSupersetOf(r) { + return fmt.Errorf("segment overlaps a preceding segment or is incorrectly sorted: [%d, %d) => %v", sds.Start[i], sds.End[i], sds.Values[i]) + } + gap = s.InsertWithoutMerging(gap, r, sds.Values[i]).NextGap() + } + return nil +} +func (s *DirtySet) saveRoot() *DirtySegmentDataSlices { + return s.ExportSortedSlices() +} + +func (s *DirtySet) loadRoot(sds *DirtySegmentDataSlices) { + if err := s.ImportSortedSlices(sds); err != nil { + panic(err) + } +} diff --git a/pkg/sentry/fs/fsutil/dirty_set_test.go b/pkg/sentry/fs/fsutil/dirty_set_test.go deleted file mode 100644 index 75575d994..000000000 --- a/pkg/sentry/fs/fsutil/dirty_set_test.go +++ /dev/null @@ -1,38 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fsutil - -import ( - "reflect" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" -) - -func TestDirtySet(t *testing.T) { - var set DirtySet - set.MarkDirty(memmap.MappableRange{0, 2 * usermem.PageSize}) - set.KeepDirty(memmap.MappableRange{usermem.PageSize, 2 * usermem.PageSize}) - set.MarkClean(memmap.MappableRange{0, 2 * usermem.PageSize}) - want := &DirtySegmentDataSlices{ - Start: []uint64{usermem.PageSize}, - End: []uint64{2 * usermem.PageSize}, - Values: []DirtyInfo{{Keep: true}}, - } - if got := set.ExportSortedSlices(); !reflect.DeepEqual(got, want) { - t.Errorf("set:\n\tgot %v,\n\twant %v", got, want) - } -} diff --git a/pkg/sentry/fs/fsutil/file_range_set_impl.go b/pkg/sentry/fs/fsutil/file_range_set_impl.go new file mode 100755 index 000000000..0548bba08 --- /dev/null +++ b/pkg/sentry/fs/fsutil/file_range_set_impl.go @@ -0,0 +1,1274 @@ +package fsutil + +import ( + __generics_imported0 "gvisor.dev/gvisor/pkg/sentry/memmap" +) + +import ( + "bytes" + "fmt" +) + +const ( + // minDegree is the minimum degree of an internal node in a Set B-tree. + // + // - Any non-root node has at least minDegree-1 segments. + // + // - Any non-root internal (non-leaf) node has at least minDegree children. + // + // - The root node may have fewer than minDegree-1 segments, but it may + // only have 0 segments if the tree is empty. + // + // Our implementation requires minDegree >= 3. Higher values of minDegree + // usually improve performance, but increase memory usage for small sets. + FileRangeminDegree = 3 + + FileRangemaxDegree = 2 * FileRangeminDegree +) + +// A Set is a mapping of segments with non-overlapping Range keys. The zero +// value for a Set is an empty set. Set values are not safely movable nor +// copyable. Set is thread-compatible. +// +// +stateify savable +type FileRangeSet struct { + root FileRangenode `state:".(*FileRangeSegmentDataSlices)"` +} + +// IsEmpty returns true if the set contains no segments. +func (s *FileRangeSet) IsEmpty() bool { + return s.root.nrSegments == 0 +} + +// IsEmptyRange returns true iff no segments in the set overlap the given +// range. This is semantically equivalent to s.SpanRange(r) == 0, but may be +// more efficient. +func (s *FileRangeSet) IsEmptyRange(r __generics_imported0.MappableRange) bool { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return true + } + _, gap := s.Find(r.Start) + if !gap.Ok() { + return false + } + return r.End <= gap.End() +} + +// Span returns the total size of all segments in the set. +func (s *FileRangeSet) Span() uint64 { + var sz uint64 + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sz += seg.Range().Length() + } + return sz +} + +// SpanRange returns the total size of the intersection of segments in the set +// with the given range. +func (s *FileRangeSet) SpanRange(r __generics_imported0.MappableRange) uint64 { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return 0 + } + var sz uint64 + for seg := s.LowerBoundSegment(r.Start); seg.Ok() && seg.Start() < r.End; seg = seg.NextSegment() { + sz += seg.Range().Intersect(r).Length() + } + return sz +} + +// FirstSegment returns the first segment in the set. If the set is empty, +// FirstSegment returns a terminal iterator. +func (s *FileRangeSet) FirstSegment() FileRangeIterator { + if s.root.nrSegments == 0 { + return FileRangeIterator{} + } + return s.root.firstSegment() +} + +// LastSegment returns the last segment in the set. If the set is empty, +// LastSegment returns a terminal iterator. +func (s *FileRangeSet) LastSegment() FileRangeIterator { + if s.root.nrSegments == 0 { + return FileRangeIterator{} + } + return s.root.lastSegment() +} + +// FirstGap returns the first gap in the set. +func (s *FileRangeSet) FirstGap() FileRangeGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[0] + } + return FileRangeGapIterator{n, 0} +} + +// LastGap returns the last gap in the set. +func (s *FileRangeSet) LastGap() FileRangeGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[n.nrSegments] + } + return FileRangeGapIterator{n, n.nrSegments} +} + +// Find returns the segment or gap whose range contains the given key. If a +// segment is found, the returned Iterator is non-terminal and the +// returned GapIterator is terminal. Otherwise, the returned Iterator is +// terminal and the returned GapIterator is non-terminal. +func (s *FileRangeSet) Find(key uint64) (FileRangeIterator, FileRangeGapIterator) { + n := &s.root + for { + + lower := 0 + upper := n.nrSegments + for lower < upper { + i := lower + (upper-lower)/2 + if r := n.keys[i]; key < r.End { + if key >= r.Start { + return FileRangeIterator{n, i}, FileRangeGapIterator{} + } + upper = i + } else { + lower = i + 1 + } + } + i := lower + if !n.hasChildren { + return FileRangeIterator{}, FileRangeGapIterator{n, i} + } + n = n.children[i] + } +} + +// FindSegment returns the segment whose range contains the given key. If no +// such segment exists, FindSegment returns a terminal iterator. +func (s *FileRangeSet) FindSegment(key uint64) FileRangeIterator { + seg, _ := s.Find(key) + return seg +} + +// LowerBoundSegment returns the segment with the lowest range that contains a +// key greater than or equal to min. If no such segment exists, +// LowerBoundSegment returns a terminal iterator. +func (s *FileRangeSet) LowerBoundSegment(min uint64) FileRangeIterator { + seg, gap := s.Find(min) + if seg.Ok() { + return seg + } + return gap.NextSegment() +} + +// UpperBoundSegment returns the segment with the highest range that contains a +// key less than or equal to max. If no such segment exists, UpperBoundSegment +// returns a terminal iterator. +func (s *FileRangeSet) UpperBoundSegment(max uint64) FileRangeIterator { + seg, gap := s.Find(max) + if seg.Ok() { + return seg + } + return gap.PrevSegment() +} + +// FindGap returns the gap containing the given key. If no such gap exists +// (i.e. the set contains a segment containing that key), FindGap returns a +// terminal iterator. +func (s *FileRangeSet) FindGap(key uint64) FileRangeGapIterator { + _, gap := s.Find(key) + return gap +} + +// LowerBoundGap returns the gap with the lowest range that is greater than or +// equal to min. +func (s *FileRangeSet) LowerBoundGap(min uint64) FileRangeGapIterator { + seg, gap := s.Find(min) + if gap.Ok() { + return gap + } + return seg.NextGap() +} + +// UpperBoundGap returns the gap with the highest range that is less than or +// equal to max. +func (s *FileRangeSet) UpperBoundGap(max uint64) FileRangeGapIterator { + seg, gap := s.Find(max) + if gap.Ok() { + return gap + } + return seg.PrevGap() +} + +// Add inserts the given segment into the set and returns true. If the new +// segment can be merged with adjacent segments, Add will do so. If the new +// segment would overlap an existing segment, Add returns false. If Add +// succeeds, all existing iterators are invalidated. +func (s *FileRangeSet) Add(r __generics_imported0.MappableRange, val uint64) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.Insert(gap, r, val) + return true +} + +// AddWithoutMerging inserts the given segment into the set and returns true. +// If it would overlap an existing segment, AddWithoutMerging does nothing and +// returns false. If AddWithoutMerging succeeds, all existing iterators are +// invalidated. +func (s *FileRangeSet) AddWithoutMerging(r __generics_imported0.MappableRange, val uint64) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.InsertWithoutMergingUnchecked(gap, r, val) + return true +} + +// Insert inserts the given segment into the given gap. If the new segment can +// be merged with adjacent segments, Insert will do so. Insert returns an +// iterator to the segment containing the inserted value (which may have been +// merged with other values). All existing iterators (including gap, but not +// including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, Insert panics. +// +// Insert is semantically equivalent to a InsertWithoutMerging followed by a +// Merge, but may be more efficient. Note that there is no unchecked variant of +// Insert since Insert must retrieve and inspect gap's predecessor and +// successor segments regardless. +func (s *FileRangeSet) Insert(gap FileRangeGapIterator, r __generics_imported0.MappableRange, val uint64) FileRangeIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + prev, next := gap.PrevSegment(), gap.NextSegment() + if prev.Ok() && prev.End() > r.Start { + panic(fmt.Sprintf("new segment %v overlaps predecessor %v", r, prev.Range())) + } + if next.Ok() && next.Start() < r.End { + panic(fmt.Sprintf("new segment %v overlaps successor %v", r, next.Range())) + } + if prev.Ok() && prev.End() == r.Start { + if mval, ok := (fileRangeSetFunctions{}).Merge(prev.Range(), prev.Value(), r, val); ok { + prev.SetEndUnchecked(r.End) + prev.SetValue(mval) + if next.Ok() && next.Start() == r.End { + val = mval + if mval, ok := (fileRangeSetFunctions{}).Merge(prev.Range(), val, next.Range(), next.Value()); ok { + prev.SetEndUnchecked(next.End()) + prev.SetValue(mval) + return s.Remove(next).PrevSegment() + } + } + return prev + } + } + if next.Ok() && next.Start() == r.End { + if mval, ok := (fileRangeSetFunctions{}).Merge(r, val, next.Range(), next.Value()); ok { + next.SetStartUnchecked(r.Start) + next.SetValue(mval) + return next + } + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMerging inserts the given segment into the given gap and +// returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, +// InsertWithoutMerging panics. +func (s *FileRangeSet) InsertWithoutMerging(gap FileRangeGapIterator, r __generics_imported0.MappableRange, val uint64) FileRangeIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if gr := gap.Range(); !gr.IsSupersetOf(r) { + panic(fmt.Sprintf("cannot insert segment range %v into gap range %v", r, gr)) + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMergingUnchecked inserts the given segment into the given gap +// and returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// Preconditions: r.Start >= gap.Start(); r.End <= gap.End(). +func (s *FileRangeSet) InsertWithoutMergingUnchecked(gap FileRangeGapIterator, r __generics_imported0.MappableRange, val uint64) FileRangeIterator { + gap = gap.node.rebalanceBeforeInsert(gap) + copy(gap.node.keys[gap.index+1:], gap.node.keys[gap.index:gap.node.nrSegments]) + copy(gap.node.values[gap.index+1:], gap.node.values[gap.index:gap.node.nrSegments]) + gap.node.keys[gap.index] = r + gap.node.values[gap.index] = val + gap.node.nrSegments++ + return FileRangeIterator{gap.node, gap.index} +} + +// Remove removes the given segment and returns an iterator to the vacated gap. +// All existing iterators (including seg, but not including the returned +// iterator) are invalidated. +func (s *FileRangeSet) Remove(seg FileRangeIterator) FileRangeGapIterator { + + if seg.node.hasChildren { + + victim := seg.PrevSegment() + + seg.SetRangeUnchecked(victim.Range()) + seg.SetValue(victim.Value()) + return s.Remove(victim).NextGap() + } + copy(seg.node.keys[seg.index:], seg.node.keys[seg.index+1:seg.node.nrSegments]) + copy(seg.node.values[seg.index:], seg.node.values[seg.index+1:seg.node.nrSegments]) + fileRangeSetFunctions{}.ClearValue(&seg.node.values[seg.node.nrSegments-1]) + seg.node.nrSegments-- + return seg.node.rebalanceAfterRemove(FileRangeGapIterator{seg.node, seg.index}) +} + +// RemoveAll removes all segments from the set. All existing iterators are +// invalidated. +func (s *FileRangeSet) RemoveAll() { + s.root = FileRangenode{} +} + +// RemoveRange removes all segments in the given range. An iterator to the +// newly formed gap is returned, and all existing iterators are invalidated. +func (s *FileRangeSet) RemoveRange(r __generics_imported0.MappableRange) FileRangeGapIterator { + seg, gap := s.Find(r.Start) + if seg.Ok() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + for seg = gap.NextSegment(); seg.Ok() && seg.Start() < r.End; seg = gap.NextSegment() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + return gap +} + +// Merge attempts to merge two neighboring segments. If successful, Merge +// returns an iterator to the merged segment, and all existing iterators are +// invalidated. Otherwise, Merge returns a terminal iterator. +// +// If first is not the predecessor of second, Merge panics. +func (s *FileRangeSet) Merge(first, second FileRangeIterator) FileRangeIterator { + if first.NextSegment() != second { + panic(fmt.Sprintf("attempt to merge non-neighboring segments %v, %v", first.Range(), second.Range())) + } + return s.MergeUnchecked(first, second) +} + +// MergeUnchecked attempts to merge two neighboring segments. If successful, +// MergeUnchecked returns an iterator to the merged segment, and all existing +// iterators are invalidated. Otherwise, MergeUnchecked returns a terminal +// iterator. +// +// Precondition: first is the predecessor of second: first.NextSegment() == +// second, first == second.PrevSegment(). +func (s *FileRangeSet) MergeUnchecked(first, second FileRangeIterator) FileRangeIterator { + if first.End() == second.Start() { + if mval, ok := (fileRangeSetFunctions{}).Merge(first.Range(), first.Value(), second.Range(), second.Value()); ok { + + first.SetEndUnchecked(second.End()) + first.SetValue(mval) + return s.Remove(second).PrevSegment() + } + } + return FileRangeIterator{} +} + +// MergeAll attempts to merge all adjacent segments in the set. All existing +// iterators are invalidated. +func (s *FileRangeSet) MergeAll() { + seg := s.FirstSegment() + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeRange attempts to merge all adjacent segments that contain a key in the +// specific range. All existing iterators are invalidated. +func (s *FileRangeSet) MergeRange(r __generics_imported0.MappableRange) { + seg := s.LowerBoundSegment(r.Start) + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() && next.Range().Start < r.End { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeAdjacent attempts to merge the segment containing r.Start with its +// predecessor, and the segment containing r.End-1 with its successor. +func (s *FileRangeSet) MergeAdjacent(r __generics_imported0.MappableRange) { + first := s.FindSegment(r.Start) + if first.Ok() { + if prev := first.PrevSegment(); prev.Ok() { + s.Merge(prev, first) + } + } + last := s.FindSegment(r.End - 1) + if last.Ok() { + if next := last.NextSegment(); next.Ok() { + s.Merge(last, next) + } + } +} + +// Split splits the given segment at the given key and returns iterators to the +// two resulting segments. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +// +// If the segment cannot be split at split (because split is at the start or +// end of the segment's range, so splitting would produce a segment with zero +// length, or because split falls outside the segment's range altogether), +// Split panics. +func (s *FileRangeSet) Split(seg FileRangeIterator, split uint64) (FileRangeIterator, FileRangeIterator) { + if !seg.Range().CanSplitAt(split) { + panic(fmt.Sprintf("can't split %v at %v", seg.Range(), split)) + } + return s.SplitUnchecked(seg, split) +} + +// SplitUnchecked splits the given segment at the given key and returns +// iterators to the two resulting segments. All existing iterators (including +// seg, but not including the returned iterators) are invalidated. +// +// Preconditions: seg.Start() < key < seg.End(). +func (s *FileRangeSet) SplitUnchecked(seg FileRangeIterator, split uint64) (FileRangeIterator, FileRangeIterator) { + val1, val2 := (fileRangeSetFunctions{}).Split(seg.Range(), seg.Value(), split) + end2 := seg.End() + seg.SetEndUnchecked(split) + seg.SetValue(val1) + seg2 := s.InsertWithoutMergingUnchecked(seg.NextGap(), __generics_imported0.MappableRange{split, end2}, val2) + + return seg2.PrevSegment(), seg2 +} + +// SplitAt splits the segment straddling split, if one exists. SplitAt returns +// true if a segment was split and false otherwise. If SplitAt splits a +// segment, all existing iterators are invalidated. +func (s *FileRangeSet) SplitAt(split uint64) bool { + if seg := s.FindSegment(split); seg.Ok() && seg.Range().CanSplitAt(split) { + s.SplitUnchecked(seg, split) + return true + } + return false +} + +// Isolate ensures that the given segment's range does not escape r by +// splitting at r.Start and r.End if necessary, and returns an updated iterator +// to the bounded segment. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +func (s *FileRangeSet) Isolate(seg FileRangeIterator, r __generics_imported0.MappableRange) FileRangeIterator { + if seg.Range().CanSplitAt(r.Start) { + _, seg = s.SplitUnchecked(seg, r.Start) + } + if seg.Range().CanSplitAt(r.End) { + seg, _ = s.SplitUnchecked(seg, r.End) + } + return seg +} + +// ApplyContiguous applies a function to a contiguous range of segments, +// splitting if necessary. The function is applied until the first gap is +// encountered, at which point the gap is returned. If the function is applied +// across the entire range, a terminal gap is returned. All existing iterators +// are invalidated. +// +// N.B. The Iterator must not be invalidated by the function. +func (s *FileRangeSet) ApplyContiguous(r __generics_imported0.MappableRange, fn func(seg FileRangeIterator)) FileRangeGapIterator { + seg, gap := s.Find(r.Start) + if !seg.Ok() { + return gap + } + for { + seg = s.Isolate(seg, r) + fn(seg) + if seg.End() >= r.End { + return FileRangeGapIterator{} + } + gap = seg.NextGap() + if !gap.IsEmpty() { + return gap + } + seg = gap.NextSegment() + if !seg.Ok() { + + return FileRangeGapIterator{} + } + } +} + +// +stateify savable +type FileRangenode struct { + // An internal binary tree node looks like: + // + // K + // / \ + // Cl Cr + // + // where all keys in the subtree rooted by Cl (the left subtree) are less + // than K (the key of the parent node), and all keys in the subtree rooted + // by Cr (the right subtree) are greater than K. + // + // An internal B-tree node's indexes work out to look like: + // + // K0 K1 K2 ... Kn-1 + // / \/ \/ \ ... / \ + // C0 C1 C2 C3 ... Cn-1 Cn + // + // where n is nrSegments. + nrSegments int + + // parent is a pointer to this node's parent. If this node is root, parent + // is nil. + parent *FileRangenode + + // parentIndex is the index of this node in parent.children. + parentIndex int + + // Flag for internal nodes that is technically redundant with "children[0] + // != nil", but is stored in the first cache line. "hasChildren" rather + // than "isLeaf" because false must be the correct value for an empty root. + hasChildren bool + + // Nodes store keys and values in separate arrays to maximize locality in + // the common case (scanning keys for lookup). + keys [FileRangemaxDegree - 1]__generics_imported0.MappableRange + values [FileRangemaxDegree - 1]uint64 + children [FileRangemaxDegree]*FileRangenode +} + +// firstSegment returns the first segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *FileRangenode) firstSegment() FileRangeIterator { + for n.hasChildren { + n = n.children[0] + } + return FileRangeIterator{n, 0} +} + +// lastSegment returns the last segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *FileRangenode) lastSegment() FileRangeIterator { + for n.hasChildren { + n = n.children[n.nrSegments] + } + return FileRangeIterator{n, n.nrSegments - 1} +} + +func (n *FileRangenode) prevSibling() *FileRangenode { + if n.parent == nil || n.parentIndex == 0 { + return nil + } + return n.parent.children[n.parentIndex-1] +} + +func (n *FileRangenode) nextSibling() *FileRangenode { + if n.parent == nil || n.parentIndex == n.parent.nrSegments { + return nil + } + return n.parent.children[n.parentIndex+1] +} + +// rebalanceBeforeInsert splits n and its ancestors if they are full, as +// required for insertion, and returns an updated iterator to the position +// represented by gap. +func (n *FileRangenode) rebalanceBeforeInsert(gap FileRangeGapIterator) FileRangeGapIterator { + if n.parent != nil { + gap = n.parent.rebalanceBeforeInsert(gap) + } + if n.nrSegments < FileRangemaxDegree-1 { + return gap + } + if n.parent == nil { + + left := &FileRangenode{ + nrSegments: FileRangeminDegree - 1, + parent: n, + parentIndex: 0, + hasChildren: n.hasChildren, + } + right := &FileRangenode{ + nrSegments: FileRangeminDegree - 1, + parent: n, + parentIndex: 1, + hasChildren: n.hasChildren, + } + copy(left.keys[:FileRangeminDegree-1], n.keys[:FileRangeminDegree-1]) + copy(left.values[:FileRangeminDegree-1], n.values[:FileRangeminDegree-1]) + copy(right.keys[:FileRangeminDegree-1], n.keys[FileRangeminDegree:]) + copy(right.values[:FileRangeminDegree-1], n.values[FileRangeminDegree:]) + n.keys[0], n.values[0] = n.keys[FileRangeminDegree-1], n.values[FileRangeminDegree-1] + FileRangezeroValueSlice(n.values[1:]) + if n.hasChildren { + copy(left.children[:FileRangeminDegree], n.children[:FileRangeminDegree]) + copy(right.children[:FileRangeminDegree], n.children[FileRangeminDegree:]) + FileRangezeroNodeSlice(n.children[2:]) + for i := 0; i < FileRangeminDegree; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + right.children[i].parent = right + right.children[i].parentIndex = i + } + } + n.nrSegments = 1 + n.hasChildren = true + n.children[0] = left + n.children[1] = right + if gap.node != n { + return gap + } + if gap.index < FileRangeminDegree { + return FileRangeGapIterator{left, gap.index} + } + return FileRangeGapIterator{right, gap.index - FileRangeminDegree} + } + + copy(n.parent.keys[n.parentIndex+1:], n.parent.keys[n.parentIndex:n.parent.nrSegments]) + copy(n.parent.values[n.parentIndex+1:], n.parent.values[n.parentIndex:n.parent.nrSegments]) + n.parent.keys[n.parentIndex], n.parent.values[n.parentIndex] = n.keys[FileRangeminDegree-1], n.values[FileRangeminDegree-1] + copy(n.parent.children[n.parentIndex+2:], n.parent.children[n.parentIndex+1:n.parent.nrSegments+1]) + for i := n.parentIndex + 2; i < n.parent.nrSegments+2; i++ { + n.parent.children[i].parentIndex = i + } + sibling := &FileRangenode{ + nrSegments: FileRangeminDegree - 1, + parent: n.parent, + parentIndex: n.parentIndex + 1, + hasChildren: n.hasChildren, + } + n.parent.children[n.parentIndex+1] = sibling + n.parent.nrSegments++ + copy(sibling.keys[:FileRangeminDegree-1], n.keys[FileRangeminDegree:]) + copy(sibling.values[:FileRangeminDegree-1], n.values[FileRangeminDegree:]) + FileRangezeroValueSlice(n.values[FileRangeminDegree-1:]) + if n.hasChildren { + copy(sibling.children[:FileRangeminDegree], n.children[FileRangeminDegree:]) + FileRangezeroNodeSlice(n.children[FileRangeminDegree:]) + for i := 0; i < FileRangeminDegree; i++ { + sibling.children[i].parent = sibling + sibling.children[i].parentIndex = i + } + } + n.nrSegments = FileRangeminDegree - 1 + + if gap.node != n { + return gap + } + if gap.index < FileRangeminDegree { + return gap + } + return FileRangeGapIterator{sibling, gap.index - FileRangeminDegree} +} + +// rebalanceAfterRemove "unsplits" n and its ancestors if they are deficient +// (contain fewer segments than required by B-tree invariants), as required for +// removal, and returns an updated iterator to the position represented by gap. +// +// Precondition: n is the only node in the tree that may currently violate a +// B-tree invariant. +func (n *FileRangenode) rebalanceAfterRemove(gap FileRangeGapIterator) FileRangeGapIterator { + for { + if n.nrSegments >= FileRangeminDegree-1 { + return gap + } + if n.parent == nil { + + return gap + } + + if sibling := n.prevSibling(); sibling != nil && sibling.nrSegments >= FileRangeminDegree { + copy(n.keys[1:], n.keys[:n.nrSegments]) + copy(n.values[1:], n.values[:n.nrSegments]) + n.keys[0] = n.parent.keys[n.parentIndex-1] + n.values[0] = n.parent.values[n.parentIndex-1] + n.parent.keys[n.parentIndex-1] = sibling.keys[sibling.nrSegments-1] + n.parent.values[n.parentIndex-1] = sibling.values[sibling.nrSegments-1] + fileRangeSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + copy(n.children[1:], n.children[:n.nrSegments+1]) + n.children[0] = sibling.children[sibling.nrSegments] + sibling.children[sibling.nrSegments] = nil + n.children[0].parent = n + n.children[0].parentIndex = 0 + for i := 1; i < n.nrSegments+2; i++ { + n.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling && gap.index == sibling.nrSegments { + return FileRangeGapIterator{n, 0} + } + if gap.node == n { + return FileRangeGapIterator{n, gap.index + 1} + } + return gap + } + if sibling := n.nextSibling(); sibling != nil && sibling.nrSegments >= FileRangeminDegree { + n.keys[n.nrSegments] = n.parent.keys[n.parentIndex] + n.values[n.nrSegments] = n.parent.values[n.parentIndex] + n.parent.keys[n.parentIndex] = sibling.keys[0] + n.parent.values[n.parentIndex] = sibling.values[0] + copy(sibling.keys[:sibling.nrSegments-1], sibling.keys[1:]) + copy(sibling.values[:sibling.nrSegments-1], sibling.values[1:]) + fileRangeSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + n.children[n.nrSegments+1] = sibling.children[0] + copy(sibling.children[:sibling.nrSegments], sibling.children[1:]) + sibling.children[sibling.nrSegments] = nil + n.children[n.nrSegments+1].parent = n + n.children[n.nrSegments+1].parentIndex = n.nrSegments + 1 + for i := 0; i < sibling.nrSegments; i++ { + sibling.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling { + if gap.index == 0 { + return FileRangeGapIterator{n, n.nrSegments} + } + return FileRangeGapIterator{sibling, gap.index - 1} + } + return gap + } + + p := n.parent + if p.nrSegments == 1 { + + left, right := p.children[0], p.children[1] + p.nrSegments = left.nrSegments + right.nrSegments + 1 + p.hasChildren = left.hasChildren + p.keys[left.nrSegments] = p.keys[0] + p.values[left.nrSegments] = p.values[0] + copy(p.keys[:left.nrSegments], left.keys[:left.nrSegments]) + copy(p.values[:left.nrSegments], left.values[:left.nrSegments]) + copy(p.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(p.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(p.children[:left.nrSegments+1], left.children[:left.nrSegments+1]) + copy(p.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := 0; i < p.nrSegments+1; i++ { + p.children[i].parent = p + p.children[i].parentIndex = i + } + } else { + p.children[0] = nil + p.children[1] = nil + } + if gap.node == left { + return FileRangeGapIterator{p, gap.index} + } + if gap.node == right { + return FileRangeGapIterator{p, gap.index + left.nrSegments + 1} + } + return gap + } + // Merge n and either sibling, along with the segment separating the + // two, into whichever of the two nodes comes first. This is the + // reverse of the non-root splitting case in + // node.rebalanceBeforeInsert. + var left, right *FileRangenode + if n.parentIndex > 0 { + left = n.prevSibling() + right = n + } else { + left = n + right = n.nextSibling() + } + + if gap.node == right { + gap = FileRangeGapIterator{left, gap.index + left.nrSegments + 1} + } + left.keys[left.nrSegments] = p.keys[left.parentIndex] + left.values[left.nrSegments] = p.values[left.parentIndex] + copy(left.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(left.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(left.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := left.nrSegments + 1; i < left.nrSegments+right.nrSegments+2; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + } + } + left.nrSegments += right.nrSegments + 1 + copy(p.keys[left.parentIndex:], p.keys[left.parentIndex+1:p.nrSegments]) + copy(p.values[left.parentIndex:], p.values[left.parentIndex+1:p.nrSegments]) + fileRangeSetFunctions{}.ClearValue(&p.values[p.nrSegments-1]) + copy(p.children[left.parentIndex+1:], p.children[left.parentIndex+2:p.nrSegments+1]) + for i := 0; i < p.nrSegments; i++ { + p.children[i].parentIndex = i + } + p.children[p.nrSegments] = nil + p.nrSegments-- + + n = p + } +} + +// A Iterator is conceptually one of: +// +// - A pointer to a segment in a set; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Iterators are copyable values and are meaningfully equality-comparable. The +// zero value of Iterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type FileRangeIterator struct { + // node is the node containing the iterated segment. If the iterator is + // terminal, node is nil. + node *FileRangenode + + // index is the index of the segment in node.keys/values. + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (seg FileRangeIterator) Ok() bool { + return seg.node != nil +} + +// Range returns the iterated segment's range key. +func (seg FileRangeIterator) Range() __generics_imported0.MappableRange { + return seg.node.keys[seg.index] +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (seg FileRangeIterator) Start() uint64 { + return seg.node.keys[seg.index].Start +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (seg FileRangeIterator) End() uint64 { + return seg.node.keys[seg.index].End +} + +// SetRangeUnchecked mutates the iterated segment's range key. This operation +// does not invalidate any iterators. +// +// Preconditions: +// +// - r.Length() > 0. +// +// - The new range must not overlap an existing one: If seg.NextSegment().Ok(), +// then r.end <= seg.NextSegment().Start(); if seg.PrevSegment().Ok(), then +// r.start >= seg.PrevSegment().End(). +func (seg FileRangeIterator) SetRangeUnchecked(r __generics_imported0.MappableRange) { + seg.node.keys[seg.index] = r +} + +// SetRange mutates the iterated segment's range key. If the new range would +// cause the iterated segment to overlap another segment, or if the new range +// is invalid, SetRange panics. This operation does not invalidate any +// iterators. +func (seg FileRangeIterator) SetRange(r __generics_imported0.MappableRange) { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if prev := seg.PrevSegment(); prev.Ok() && r.Start < prev.End() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, prev.Range())) + } + if next := seg.NextSegment(); next.Ok() && r.End > next.Start() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, next.Range())) + } + seg.SetRangeUnchecked(r) +} + +// SetStartUnchecked mutates the iterated segment's start. This operation does +// not invalidate any iterators. +// +// Preconditions: The new start must be valid: start < seg.End(); if +// seg.PrevSegment().Ok(), then start >= seg.PrevSegment().End(). +func (seg FileRangeIterator) SetStartUnchecked(start uint64) { + seg.node.keys[seg.index].Start = start +} + +// SetStart mutates the iterated segment's start. If the new start value would +// cause the iterated segment to overlap another segment, or would result in an +// invalid range, SetStart panics. This operation does not invalidate any +// iterators. +func (seg FileRangeIterator) SetStart(start uint64) { + if start >= seg.End() { + panic(fmt.Sprintf("new start %v would invalidate segment range %v", start, seg.Range())) + } + if prev := seg.PrevSegment(); prev.Ok() && start < prev.End() { + panic(fmt.Sprintf("new start %v would cause segment range %v to overlap segment range %v", start, seg.Range(), prev.Range())) + } + seg.SetStartUnchecked(start) +} + +// SetEndUnchecked mutates the iterated segment's end. This operation does not +// invalidate any iterators. +// +// Preconditions: The new end must be valid: end > seg.Start(); if +// seg.NextSegment().Ok(), then end <= seg.NextSegment().Start(). +func (seg FileRangeIterator) SetEndUnchecked(end uint64) { + seg.node.keys[seg.index].End = end +} + +// SetEnd mutates the iterated segment's end. If the new end value would cause +// the iterated segment to overlap another segment, or would result in an +// invalid range, SetEnd panics. This operation does not invalidate any +// iterators. +func (seg FileRangeIterator) SetEnd(end uint64) { + if end <= seg.Start() { + panic(fmt.Sprintf("new end %v would invalidate segment range %v", end, seg.Range())) + } + if next := seg.NextSegment(); next.Ok() && end > next.Start() { + panic(fmt.Sprintf("new end %v would cause segment range %v to overlap segment range %v", end, seg.Range(), next.Range())) + } + seg.SetEndUnchecked(end) +} + +// Value returns a copy of the iterated segment's value. +func (seg FileRangeIterator) Value() uint64 { + return seg.node.values[seg.index] +} + +// ValuePtr returns a pointer to the iterated segment's value. The pointer is +// invalidated if the iterator is invalidated. This operation does not +// invalidate any iterators. +func (seg FileRangeIterator) ValuePtr() *uint64 { + return &seg.node.values[seg.index] +} + +// SetValue mutates the iterated segment's value. This operation does not +// invalidate any iterators. +func (seg FileRangeIterator) SetValue(val uint64) { + seg.node.values[seg.index] = val +} + +// PrevSegment returns the iterated segment's predecessor. If there is no +// preceding segment, PrevSegment returns a terminal iterator. +func (seg FileRangeIterator) PrevSegment() FileRangeIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index].lastSegment() + } + if seg.index > 0 { + return FileRangeIterator{seg.node, seg.index - 1} + } + if seg.node.parent == nil { + return FileRangeIterator{} + } + return FileRangesegmentBeforePosition(seg.node.parent, seg.node.parentIndex) +} + +// NextSegment returns the iterated segment's successor. If there is no +// succeeding segment, NextSegment returns a terminal iterator. +func (seg FileRangeIterator) NextSegment() FileRangeIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment() + } + if seg.index < seg.node.nrSegments-1 { + return FileRangeIterator{seg.node, seg.index + 1} + } + if seg.node.parent == nil { + return FileRangeIterator{} + } + return FileRangesegmentAfterPosition(seg.node.parent, seg.node.parentIndex) +} + +// PrevGap returns the gap immediately before the iterated segment. +func (seg FileRangeIterator) PrevGap() FileRangeGapIterator { + if seg.node.hasChildren { + + return seg.node.children[seg.index].lastSegment().NextGap() + } + return FileRangeGapIterator{seg.node, seg.index} +} + +// NextGap returns the gap immediately after the iterated segment. +func (seg FileRangeIterator) NextGap() FileRangeGapIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment().PrevGap() + } + return FileRangeGapIterator{seg.node, seg.index + 1} +} + +// PrevNonEmpty returns the iterated segment's predecessor if it is adjacent, +// or the gap before the iterated segment otherwise. If seg.Start() == +// Functions.MinKey(), PrevNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by PrevNonEmpty will be +// non-terminal. +func (seg FileRangeIterator) PrevNonEmpty() (FileRangeIterator, FileRangeGapIterator) { + gap := seg.PrevGap() + if gap.Range().Length() != 0 { + return FileRangeIterator{}, gap + } + return gap.PrevSegment(), FileRangeGapIterator{} +} + +// NextNonEmpty returns the iterated segment's successor if it is adjacent, or +// the gap after the iterated segment otherwise. If seg.End() == +// Functions.MaxKey(), NextNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by NextNonEmpty will be +// non-terminal. +func (seg FileRangeIterator) NextNonEmpty() (FileRangeIterator, FileRangeGapIterator) { + gap := seg.NextGap() + if gap.Range().Length() != 0 { + return FileRangeIterator{}, gap + } + return gap.NextSegment(), FileRangeGapIterator{} +} + +// A GapIterator is conceptually one of: +// +// - A pointer to a position between two segments, before the first segment, or +// after the last segment in a set, called a *gap*; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Note that the gap between two adjacent segments exists (iterators to it are +// non-terminal), but has a length of zero. GapIterator.IsEmpty returns true +// for such gaps. An empty set contains a single gap, spanning the entire range +// of the set's keys. +// +// GapIterators are copyable values and are meaningfully equality-comparable. +// The zero value of GapIterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type FileRangeGapIterator struct { + // The representation of a GapIterator is identical to that of an Iterator, + // except that index corresponds to positions between segments in the same + // way as for node.children (see comment for node.nrSegments). + node *FileRangenode + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (gap FileRangeGapIterator) Ok() bool { + return gap.node != nil +} + +// Range returns the range spanned by the iterated gap. +func (gap FileRangeGapIterator) Range() __generics_imported0.MappableRange { + return __generics_imported0.MappableRange{gap.Start(), gap.End()} +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (gap FileRangeGapIterator) Start() uint64 { + if ps := gap.PrevSegment(); ps.Ok() { + return ps.End() + } + return fileRangeSetFunctions{}.MinKey() +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (gap FileRangeGapIterator) End() uint64 { + if ns := gap.NextSegment(); ns.Ok() { + return ns.Start() + } + return fileRangeSetFunctions{}.MaxKey() +} + +// IsEmpty returns true if the iterated gap is empty (that is, the "gap" is +// between two adjacent segments.) +func (gap FileRangeGapIterator) IsEmpty() bool { + return gap.Range().Length() == 0 +} + +// PrevSegment returns the segment immediately before the iterated gap. If no +// such segment exists, PrevSegment returns a terminal iterator. +func (gap FileRangeGapIterator) PrevSegment() FileRangeIterator { + return FileRangesegmentBeforePosition(gap.node, gap.index) +} + +// NextSegment returns the segment immediately after the iterated gap. If no +// such segment exists, NextSegment returns a terminal iterator. +func (gap FileRangeGapIterator) NextSegment() FileRangeIterator { + return FileRangesegmentAfterPosition(gap.node, gap.index) +} + +// PrevGap returns the iterated gap's predecessor. If no such gap exists, +// PrevGap returns a terminal iterator. +func (gap FileRangeGapIterator) PrevGap() FileRangeGapIterator { + seg := gap.PrevSegment() + if !seg.Ok() { + return FileRangeGapIterator{} + } + return seg.PrevGap() +} + +// NextGap returns the iterated gap's successor. If no such gap exists, NextGap +// returns a terminal iterator. +func (gap FileRangeGapIterator) NextGap() FileRangeGapIterator { + seg := gap.NextSegment() + if !seg.Ok() { + return FileRangeGapIterator{} + } + return seg.NextGap() +} + +// segmentBeforePosition returns the predecessor segment of the position given +// by n.children[i], which may or may not contain a child. If no such segment +// exists, segmentBeforePosition returns a terminal iterator. +func FileRangesegmentBeforePosition(n *FileRangenode, i int) FileRangeIterator { + for i == 0 { + if n.parent == nil { + return FileRangeIterator{} + } + n, i = n.parent, n.parentIndex + } + return FileRangeIterator{n, i - 1} +} + +// segmentAfterPosition returns the successor segment of the position given by +// n.children[i], which may or may not contain a child. If no such segment +// exists, segmentAfterPosition returns a terminal iterator. +func FileRangesegmentAfterPosition(n *FileRangenode, i int) FileRangeIterator { + for i == n.nrSegments { + if n.parent == nil { + return FileRangeIterator{} + } + n, i = n.parent, n.parentIndex + } + return FileRangeIterator{n, i} +} + +func FileRangezeroValueSlice(slice []uint64) { + + for i := range slice { + fileRangeSetFunctions{}.ClearValue(&slice[i]) + } +} + +func FileRangezeroNodeSlice(slice []*FileRangenode) { + for i := range slice { + slice[i] = nil + } +} + +// String stringifies a Set for debugging. +func (s *FileRangeSet) String() string { + return s.root.String() +} + +// String stringifies a node (and all of its children) for debugging. +func (n *FileRangenode) String() string { + var buf bytes.Buffer + n.writeDebugString(&buf, "") + return buf.String() +} + +func (n *FileRangenode) writeDebugString(buf *bytes.Buffer, prefix string) { + if n.hasChildren != (n.nrSegments > 0 && n.children[0] != nil) { + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent value of hasChildren: got %v, want %v\n", n.hasChildren, !n.hasChildren)) + } + for i := 0; i < n.nrSegments; i++ { + if child := n.children[i]; child != nil { + cprefix := fmt.Sprintf("%s- % 3d ", prefix, i) + if child.parent != n || child.parentIndex != i { + buf.WriteString(cprefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent linkage to parent: got (%p, %d), want (%p, %d)\n", child.parent, child.parentIndex, n, i)) + } + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, i)) + } + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + } + if child := n.children[n.nrSegments]; child != nil { + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, n.nrSegments)) + } +} + +// SegmentDataSlices represents segments from a set as slices of start, end, and +// values. SegmentDataSlices is primarily used as an intermediate representation +// for save/restore and the layout here is optimized for that. +// +// +stateify savable +type FileRangeSegmentDataSlices struct { + Start []uint64 + End []uint64 + Values []uint64 +} + +// ExportSortedSlice returns a copy of all segments in the given set, in ascending +// key order. +func (s *FileRangeSet) ExportSortedSlices() *FileRangeSegmentDataSlices { + var sds FileRangeSegmentDataSlices + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sds.Start = append(sds.Start, seg.Start()) + sds.End = append(sds.End, seg.End()) + sds.Values = append(sds.Values, seg.Value()) + } + sds.Start = sds.Start[:len(sds.Start):len(sds.Start)] + sds.End = sds.End[:len(sds.End):len(sds.End)] + sds.Values = sds.Values[:len(sds.Values):len(sds.Values)] + return &sds +} + +// ImportSortedSlice initializes the given set from the given slice. +// +// Preconditions: s must be empty. sds must represent a valid set (the segments +// in sds must have valid lengths that do not overlap). The segments in sds +// must be sorted in ascending key order. +func (s *FileRangeSet) ImportSortedSlices(sds *FileRangeSegmentDataSlices) error { + if !s.IsEmpty() { + return fmt.Errorf("cannot import into non-empty set %v", s) + } + gap := s.FirstGap() + for i := range sds.Start { + r := __generics_imported0.MappableRange{sds.Start[i], sds.End[i]} + if !gap.Range().IsSupersetOf(r) { + return fmt.Errorf("segment overlaps a preceding segment or is incorrectly sorted: [%d, %d) => %v", sds.Start[i], sds.End[i], sds.Values[i]) + } + gap = s.InsertWithoutMerging(gap, r, sds.Values[i]).NextGap() + } + return nil +} +func (s *FileRangeSet) saveRoot() *FileRangeSegmentDataSlices { + return s.ExportSortedSlices() +} + +func (s *FileRangeSet) loadRoot(sds *FileRangeSegmentDataSlices) { + if err := s.ImportSortedSlices(sds); err != nil { + panic(err) + } +} diff --git a/pkg/sentry/fs/fsutil/frame_ref_set_impl.go b/pkg/sentry/fs/fsutil/frame_ref_set_impl.go new file mode 100755 index 000000000..d4601bffa --- /dev/null +++ b/pkg/sentry/fs/fsutil/frame_ref_set_impl.go @@ -0,0 +1,1274 @@ +package fsutil + +import ( + __generics_imported0 "gvisor.dev/gvisor/pkg/sentry/platform" +) + +import ( + "bytes" + "fmt" +) + +const ( + // minDegree is the minimum degree of an internal node in a Set B-tree. + // + // - Any non-root node has at least minDegree-1 segments. + // + // - Any non-root internal (non-leaf) node has at least minDegree children. + // + // - The root node may have fewer than minDegree-1 segments, but it may + // only have 0 segments if the tree is empty. + // + // Our implementation requires minDegree >= 3. Higher values of minDegree + // usually improve performance, but increase memory usage for small sets. + frameRefminDegree = 3 + + frameRefmaxDegree = 2 * frameRefminDegree +) + +// A Set is a mapping of segments with non-overlapping Range keys. The zero +// value for a Set is an empty set. Set values are not safely movable nor +// copyable. Set is thread-compatible. +// +// +stateify savable +type frameRefSet struct { + root frameRefnode `state:".(*frameRefSegmentDataSlices)"` +} + +// IsEmpty returns true if the set contains no segments. +func (s *frameRefSet) IsEmpty() bool { + return s.root.nrSegments == 0 +} + +// IsEmptyRange returns true iff no segments in the set overlap the given +// range. This is semantically equivalent to s.SpanRange(r) == 0, but may be +// more efficient. +func (s *frameRefSet) IsEmptyRange(r __generics_imported0.FileRange) bool { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return true + } + _, gap := s.Find(r.Start) + if !gap.Ok() { + return false + } + return r.End <= gap.End() +} + +// Span returns the total size of all segments in the set. +func (s *frameRefSet) Span() uint64 { + var sz uint64 + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sz += seg.Range().Length() + } + return sz +} + +// SpanRange returns the total size of the intersection of segments in the set +// with the given range. +func (s *frameRefSet) SpanRange(r __generics_imported0.FileRange) uint64 { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return 0 + } + var sz uint64 + for seg := s.LowerBoundSegment(r.Start); seg.Ok() && seg.Start() < r.End; seg = seg.NextSegment() { + sz += seg.Range().Intersect(r).Length() + } + return sz +} + +// FirstSegment returns the first segment in the set. If the set is empty, +// FirstSegment returns a terminal iterator. +func (s *frameRefSet) FirstSegment() frameRefIterator { + if s.root.nrSegments == 0 { + return frameRefIterator{} + } + return s.root.firstSegment() +} + +// LastSegment returns the last segment in the set. If the set is empty, +// LastSegment returns a terminal iterator. +func (s *frameRefSet) LastSegment() frameRefIterator { + if s.root.nrSegments == 0 { + return frameRefIterator{} + } + return s.root.lastSegment() +} + +// FirstGap returns the first gap in the set. +func (s *frameRefSet) FirstGap() frameRefGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[0] + } + return frameRefGapIterator{n, 0} +} + +// LastGap returns the last gap in the set. +func (s *frameRefSet) LastGap() frameRefGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[n.nrSegments] + } + return frameRefGapIterator{n, n.nrSegments} +} + +// Find returns the segment or gap whose range contains the given key. If a +// segment is found, the returned Iterator is non-terminal and the +// returned GapIterator is terminal. Otherwise, the returned Iterator is +// terminal and the returned GapIterator is non-terminal. +func (s *frameRefSet) Find(key uint64) (frameRefIterator, frameRefGapIterator) { + n := &s.root + for { + + lower := 0 + upper := n.nrSegments + for lower < upper { + i := lower + (upper-lower)/2 + if r := n.keys[i]; key < r.End { + if key >= r.Start { + return frameRefIterator{n, i}, frameRefGapIterator{} + } + upper = i + } else { + lower = i + 1 + } + } + i := lower + if !n.hasChildren { + return frameRefIterator{}, frameRefGapIterator{n, i} + } + n = n.children[i] + } +} + +// FindSegment returns the segment whose range contains the given key. If no +// such segment exists, FindSegment returns a terminal iterator. +func (s *frameRefSet) FindSegment(key uint64) frameRefIterator { + seg, _ := s.Find(key) + return seg +} + +// LowerBoundSegment returns the segment with the lowest range that contains a +// key greater than or equal to min. If no such segment exists, +// LowerBoundSegment returns a terminal iterator. +func (s *frameRefSet) LowerBoundSegment(min uint64) frameRefIterator { + seg, gap := s.Find(min) + if seg.Ok() { + return seg + } + return gap.NextSegment() +} + +// UpperBoundSegment returns the segment with the highest range that contains a +// key less than or equal to max. If no such segment exists, UpperBoundSegment +// returns a terminal iterator. +func (s *frameRefSet) UpperBoundSegment(max uint64) frameRefIterator { + seg, gap := s.Find(max) + if seg.Ok() { + return seg + } + return gap.PrevSegment() +} + +// FindGap returns the gap containing the given key. If no such gap exists +// (i.e. the set contains a segment containing that key), FindGap returns a +// terminal iterator. +func (s *frameRefSet) FindGap(key uint64) frameRefGapIterator { + _, gap := s.Find(key) + return gap +} + +// LowerBoundGap returns the gap with the lowest range that is greater than or +// equal to min. +func (s *frameRefSet) LowerBoundGap(min uint64) frameRefGapIterator { + seg, gap := s.Find(min) + if gap.Ok() { + return gap + } + return seg.NextGap() +} + +// UpperBoundGap returns the gap with the highest range that is less than or +// equal to max. +func (s *frameRefSet) UpperBoundGap(max uint64) frameRefGapIterator { + seg, gap := s.Find(max) + if gap.Ok() { + return gap + } + return seg.PrevGap() +} + +// Add inserts the given segment into the set and returns true. If the new +// segment can be merged with adjacent segments, Add will do so. If the new +// segment would overlap an existing segment, Add returns false. If Add +// succeeds, all existing iterators are invalidated. +func (s *frameRefSet) Add(r __generics_imported0.FileRange, val uint64) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.Insert(gap, r, val) + return true +} + +// AddWithoutMerging inserts the given segment into the set and returns true. +// If it would overlap an existing segment, AddWithoutMerging does nothing and +// returns false. If AddWithoutMerging succeeds, all existing iterators are +// invalidated. +func (s *frameRefSet) AddWithoutMerging(r __generics_imported0.FileRange, val uint64) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.InsertWithoutMergingUnchecked(gap, r, val) + return true +} + +// Insert inserts the given segment into the given gap. If the new segment can +// be merged with adjacent segments, Insert will do so. Insert returns an +// iterator to the segment containing the inserted value (which may have been +// merged with other values). All existing iterators (including gap, but not +// including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, Insert panics. +// +// Insert is semantically equivalent to a InsertWithoutMerging followed by a +// Merge, but may be more efficient. Note that there is no unchecked variant of +// Insert since Insert must retrieve and inspect gap's predecessor and +// successor segments regardless. +func (s *frameRefSet) Insert(gap frameRefGapIterator, r __generics_imported0.FileRange, val uint64) frameRefIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + prev, next := gap.PrevSegment(), gap.NextSegment() + if prev.Ok() && prev.End() > r.Start { + panic(fmt.Sprintf("new segment %v overlaps predecessor %v", r, prev.Range())) + } + if next.Ok() && next.Start() < r.End { + panic(fmt.Sprintf("new segment %v overlaps successor %v", r, next.Range())) + } + if prev.Ok() && prev.End() == r.Start { + if mval, ok := (frameRefSetFunctions{}).Merge(prev.Range(), prev.Value(), r, val); ok { + prev.SetEndUnchecked(r.End) + prev.SetValue(mval) + if next.Ok() && next.Start() == r.End { + val = mval + if mval, ok := (frameRefSetFunctions{}).Merge(prev.Range(), val, next.Range(), next.Value()); ok { + prev.SetEndUnchecked(next.End()) + prev.SetValue(mval) + return s.Remove(next).PrevSegment() + } + } + return prev + } + } + if next.Ok() && next.Start() == r.End { + if mval, ok := (frameRefSetFunctions{}).Merge(r, val, next.Range(), next.Value()); ok { + next.SetStartUnchecked(r.Start) + next.SetValue(mval) + return next + } + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMerging inserts the given segment into the given gap and +// returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, +// InsertWithoutMerging panics. +func (s *frameRefSet) InsertWithoutMerging(gap frameRefGapIterator, r __generics_imported0.FileRange, val uint64) frameRefIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if gr := gap.Range(); !gr.IsSupersetOf(r) { + panic(fmt.Sprintf("cannot insert segment range %v into gap range %v", r, gr)) + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMergingUnchecked inserts the given segment into the given gap +// and returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// Preconditions: r.Start >= gap.Start(); r.End <= gap.End(). +func (s *frameRefSet) InsertWithoutMergingUnchecked(gap frameRefGapIterator, r __generics_imported0.FileRange, val uint64) frameRefIterator { + gap = gap.node.rebalanceBeforeInsert(gap) + copy(gap.node.keys[gap.index+1:], gap.node.keys[gap.index:gap.node.nrSegments]) + copy(gap.node.values[gap.index+1:], gap.node.values[gap.index:gap.node.nrSegments]) + gap.node.keys[gap.index] = r + gap.node.values[gap.index] = val + gap.node.nrSegments++ + return frameRefIterator{gap.node, gap.index} +} + +// Remove removes the given segment and returns an iterator to the vacated gap. +// All existing iterators (including seg, but not including the returned +// iterator) are invalidated. +func (s *frameRefSet) Remove(seg frameRefIterator) frameRefGapIterator { + + if seg.node.hasChildren { + + victim := seg.PrevSegment() + + seg.SetRangeUnchecked(victim.Range()) + seg.SetValue(victim.Value()) + return s.Remove(victim).NextGap() + } + copy(seg.node.keys[seg.index:], seg.node.keys[seg.index+1:seg.node.nrSegments]) + copy(seg.node.values[seg.index:], seg.node.values[seg.index+1:seg.node.nrSegments]) + frameRefSetFunctions{}.ClearValue(&seg.node.values[seg.node.nrSegments-1]) + seg.node.nrSegments-- + return seg.node.rebalanceAfterRemove(frameRefGapIterator{seg.node, seg.index}) +} + +// RemoveAll removes all segments from the set. All existing iterators are +// invalidated. +func (s *frameRefSet) RemoveAll() { + s.root = frameRefnode{} +} + +// RemoveRange removes all segments in the given range. An iterator to the +// newly formed gap is returned, and all existing iterators are invalidated. +func (s *frameRefSet) RemoveRange(r __generics_imported0.FileRange) frameRefGapIterator { + seg, gap := s.Find(r.Start) + if seg.Ok() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + for seg = gap.NextSegment(); seg.Ok() && seg.Start() < r.End; seg = gap.NextSegment() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + return gap +} + +// Merge attempts to merge two neighboring segments. If successful, Merge +// returns an iterator to the merged segment, and all existing iterators are +// invalidated. Otherwise, Merge returns a terminal iterator. +// +// If first is not the predecessor of second, Merge panics. +func (s *frameRefSet) Merge(first, second frameRefIterator) frameRefIterator { + if first.NextSegment() != second { + panic(fmt.Sprintf("attempt to merge non-neighboring segments %v, %v", first.Range(), second.Range())) + } + return s.MergeUnchecked(first, second) +} + +// MergeUnchecked attempts to merge two neighboring segments. If successful, +// MergeUnchecked returns an iterator to the merged segment, and all existing +// iterators are invalidated. Otherwise, MergeUnchecked returns a terminal +// iterator. +// +// Precondition: first is the predecessor of second: first.NextSegment() == +// second, first == second.PrevSegment(). +func (s *frameRefSet) MergeUnchecked(first, second frameRefIterator) frameRefIterator { + if first.End() == second.Start() { + if mval, ok := (frameRefSetFunctions{}).Merge(first.Range(), first.Value(), second.Range(), second.Value()); ok { + + first.SetEndUnchecked(second.End()) + first.SetValue(mval) + return s.Remove(second).PrevSegment() + } + } + return frameRefIterator{} +} + +// MergeAll attempts to merge all adjacent segments in the set. All existing +// iterators are invalidated. +func (s *frameRefSet) MergeAll() { + seg := s.FirstSegment() + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeRange attempts to merge all adjacent segments that contain a key in the +// specific range. All existing iterators are invalidated. +func (s *frameRefSet) MergeRange(r __generics_imported0.FileRange) { + seg := s.LowerBoundSegment(r.Start) + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() && next.Range().Start < r.End { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeAdjacent attempts to merge the segment containing r.Start with its +// predecessor, and the segment containing r.End-1 with its successor. +func (s *frameRefSet) MergeAdjacent(r __generics_imported0.FileRange) { + first := s.FindSegment(r.Start) + if first.Ok() { + if prev := first.PrevSegment(); prev.Ok() { + s.Merge(prev, first) + } + } + last := s.FindSegment(r.End - 1) + if last.Ok() { + if next := last.NextSegment(); next.Ok() { + s.Merge(last, next) + } + } +} + +// Split splits the given segment at the given key and returns iterators to the +// two resulting segments. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +// +// If the segment cannot be split at split (because split is at the start or +// end of the segment's range, so splitting would produce a segment with zero +// length, or because split falls outside the segment's range altogether), +// Split panics. +func (s *frameRefSet) Split(seg frameRefIterator, split uint64) (frameRefIterator, frameRefIterator) { + if !seg.Range().CanSplitAt(split) { + panic(fmt.Sprintf("can't split %v at %v", seg.Range(), split)) + } + return s.SplitUnchecked(seg, split) +} + +// SplitUnchecked splits the given segment at the given key and returns +// iterators to the two resulting segments. All existing iterators (including +// seg, but not including the returned iterators) are invalidated. +// +// Preconditions: seg.Start() < key < seg.End(). +func (s *frameRefSet) SplitUnchecked(seg frameRefIterator, split uint64) (frameRefIterator, frameRefIterator) { + val1, val2 := (frameRefSetFunctions{}).Split(seg.Range(), seg.Value(), split) + end2 := seg.End() + seg.SetEndUnchecked(split) + seg.SetValue(val1) + seg2 := s.InsertWithoutMergingUnchecked(seg.NextGap(), __generics_imported0.FileRange{split, end2}, val2) + + return seg2.PrevSegment(), seg2 +} + +// SplitAt splits the segment straddling split, if one exists. SplitAt returns +// true if a segment was split and false otherwise. If SplitAt splits a +// segment, all existing iterators are invalidated. +func (s *frameRefSet) SplitAt(split uint64) bool { + if seg := s.FindSegment(split); seg.Ok() && seg.Range().CanSplitAt(split) { + s.SplitUnchecked(seg, split) + return true + } + return false +} + +// Isolate ensures that the given segment's range does not escape r by +// splitting at r.Start and r.End if necessary, and returns an updated iterator +// to the bounded segment. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +func (s *frameRefSet) Isolate(seg frameRefIterator, r __generics_imported0.FileRange) frameRefIterator { + if seg.Range().CanSplitAt(r.Start) { + _, seg = s.SplitUnchecked(seg, r.Start) + } + if seg.Range().CanSplitAt(r.End) { + seg, _ = s.SplitUnchecked(seg, r.End) + } + return seg +} + +// ApplyContiguous applies a function to a contiguous range of segments, +// splitting if necessary. The function is applied until the first gap is +// encountered, at which point the gap is returned. If the function is applied +// across the entire range, a terminal gap is returned. All existing iterators +// are invalidated. +// +// N.B. The Iterator must not be invalidated by the function. +func (s *frameRefSet) ApplyContiguous(r __generics_imported0.FileRange, fn func(seg frameRefIterator)) frameRefGapIterator { + seg, gap := s.Find(r.Start) + if !seg.Ok() { + return gap + } + for { + seg = s.Isolate(seg, r) + fn(seg) + if seg.End() >= r.End { + return frameRefGapIterator{} + } + gap = seg.NextGap() + if !gap.IsEmpty() { + return gap + } + seg = gap.NextSegment() + if !seg.Ok() { + + return frameRefGapIterator{} + } + } +} + +// +stateify savable +type frameRefnode struct { + // An internal binary tree node looks like: + // + // K + // / \ + // Cl Cr + // + // where all keys in the subtree rooted by Cl (the left subtree) are less + // than K (the key of the parent node), and all keys in the subtree rooted + // by Cr (the right subtree) are greater than K. + // + // An internal B-tree node's indexes work out to look like: + // + // K0 K1 K2 ... Kn-1 + // / \/ \/ \ ... / \ + // C0 C1 C2 C3 ... Cn-1 Cn + // + // where n is nrSegments. + nrSegments int + + // parent is a pointer to this node's parent. If this node is root, parent + // is nil. + parent *frameRefnode + + // parentIndex is the index of this node in parent.children. + parentIndex int + + // Flag for internal nodes that is technically redundant with "children[0] + // != nil", but is stored in the first cache line. "hasChildren" rather + // than "isLeaf" because false must be the correct value for an empty root. + hasChildren bool + + // Nodes store keys and values in separate arrays to maximize locality in + // the common case (scanning keys for lookup). + keys [frameRefmaxDegree - 1]__generics_imported0.FileRange + values [frameRefmaxDegree - 1]uint64 + children [frameRefmaxDegree]*frameRefnode +} + +// firstSegment returns the first segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *frameRefnode) firstSegment() frameRefIterator { + for n.hasChildren { + n = n.children[0] + } + return frameRefIterator{n, 0} +} + +// lastSegment returns the last segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *frameRefnode) lastSegment() frameRefIterator { + for n.hasChildren { + n = n.children[n.nrSegments] + } + return frameRefIterator{n, n.nrSegments - 1} +} + +func (n *frameRefnode) prevSibling() *frameRefnode { + if n.parent == nil || n.parentIndex == 0 { + return nil + } + return n.parent.children[n.parentIndex-1] +} + +func (n *frameRefnode) nextSibling() *frameRefnode { + if n.parent == nil || n.parentIndex == n.parent.nrSegments { + return nil + } + return n.parent.children[n.parentIndex+1] +} + +// rebalanceBeforeInsert splits n and its ancestors if they are full, as +// required for insertion, and returns an updated iterator to the position +// represented by gap. +func (n *frameRefnode) rebalanceBeforeInsert(gap frameRefGapIterator) frameRefGapIterator { + if n.parent != nil { + gap = n.parent.rebalanceBeforeInsert(gap) + } + if n.nrSegments < frameRefmaxDegree-1 { + return gap + } + if n.parent == nil { + + left := &frameRefnode{ + nrSegments: frameRefminDegree - 1, + parent: n, + parentIndex: 0, + hasChildren: n.hasChildren, + } + right := &frameRefnode{ + nrSegments: frameRefminDegree - 1, + parent: n, + parentIndex: 1, + hasChildren: n.hasChildren, + } + copy(left.keys[:frameRefminDegree-1], n.keys[:frameRefminDegree-1]) + copy(left.values[:frameRefminDegree-1], n.values[:frameRefminDegree-1]) + copy(right.keys[:frameRefminDegree-1], n.keys[frameRefminDegree:]) + copy(right.values[:frameRefminDegree-1], n.values[frameRefminDegree:]) + n.keys[0], n.values[0] = n.keys[frameRefminDegree-1], n.values[frameRefminDegree-1] + frameRefzeroValueSlice(n.values[1:]) + if n.hasChildren { + copy(left.children[:frameRefminDegree], n.children[:frameRefminDegree]) + copy(right.children[:frameRefminDegree], n.children[frameRefminDegree:]) + frameRefzeroNodeSlice(n.children[2:]) + for i := 0; i < frameRefminDegree; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + right.children[i].parent = right + right.children[i].parentIndex = i + } + } + n.nrSegments = 1 + n.hasChildren = true + n.children[0] = left + n.children[1] = right + if gap.node != n { + return gap + } + if gap.index < frameRefminDegree { + return frameRefGapIterator{left, gap.index} + } + return frameRefGapIterator{right, gap.index - frameRefminDegree} + } + + copy(n.parent.keys[n.parentIndex+1:], n.parent.keys[n.parentIndex:n.parent.nrSegments]) + copy(n.parent.values[n.parentIndex+1:], n.parent.values[n.parentIndex:n.parent.nrSegments]) + n.parent.keys[n.parentIndex], n.parent.values[n.parentIndex] = n.keys[frameRefminDegree-1], n.values[frameRefminDegree-1] + copy(n.parent.children[n.parentIndex+2:], n.parent.children[n.parentIndex+1:n.parent.nrSegments+1]) + for i := n.parentIndex + 2; i < n.parent.nrSegments+2; i++ { + n.parent.children[i].parentIndex = i + } + sibling := &frameRefnode{ + nrSegments: frameRefminDegree - 1, + parent: n.parent, + parentIndex: n.parentIndex + 1, + hasChildren: n.hasChildren, + } + n.parent.children[n.parentIndex+1] = sibling + n.parent.nrSegments++ + copy(sibling.keys[:frameRefminDegree-1], n.keys[frameRefminDegree:]) + copy(sibling.values[:frameRefminDegree-1], n.values[frameRefminDegree:]) + frameRefzeroValueSlice(n.values[frameRefminDegree-1:]) + if n.hasChildren { + copy(sibling.children[:frameRefminDegree], n.children[frameRefminDegree:]) + frameRefzeroNodeSlice(n.children[frameRefminDegree:]) + for i := 0; i < frameRefminDegree; i++ { + sibling.children[i].parent = sibling + sibling.children[i].parentIndex = i + } + } + n.nrSegments = frameRefminDegree - 1 + + if gap.node != n { + return gap + } + if gap.index < frameRefminDegree { + return gap + } + return frameRefGapIterator{sibling, gap.index - frameRefminDegree} +} + +// rebalanceAfterRemove "unsplits" n and its ancestors if they are deficient +// (contain fewer segments than required by B-tree invariants), as required for +// removal, and returns an updated iterator to the position represented by gap. +// +// Precondition: n is the only node in the tree that may currently violate a +// B-tree invariant. +func (n *frameRefnode) rebalanceAfterRemove(gap frameRefGapIterator) frameRefGapIterator { + for { + if n.nrSegments >= frameRefminDegree-1 { + return gap + } + if n.parent == nil { + + return gap + } + + if sibling := n.prevSibling(); sibling != nil && sibling.nrSegments >= frameRefminDegree { + copy(n.keys[1:], n.keys[:n.nrSegments]) + copy(n.values[1:], n.values[:n.nrSegments]) + n.keys[0] = n.parent.keys[n.parentIndex-1] + n.values[0] = n.parent.values[n.parentIndex-1] + n.parent.keys[n.parentIndex-1] = sibling.keys[sibling.nrSegments-1] + n.parent.values[n.parentIndex-1] = sibling.values[sibling.nrSegments-1] + frameRefSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + copy(n.children[1:], n.children[:n.nrSegments+1]) + n.children[0] = sibling.children[sibling.nrSegments] + sibling.children[sibling.nrSegments] = nil + n.children[0].parent = n + n.children[0].parentIndex = 0 + for i := 1; i < n.nrSegments+2; i++ { + n.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling && gap.index == sibling.nrSegments { + return frameRefGapIterator{n, 0} + } + if gap.node == n { + return frameRefGapIterator{n, gap.index + 1} + } + return gap + } + if sibling := n.nextSibling(); sibling != nil && sibling.nrSegments >= frameRefminDegree { + n.keys[n.nrSegments] = n.parent.keys[n.parentIndex] + n.values[n.nrSegments] = n.parent.values[n.parentIndex] + n.parent.keys[n.parentIndex] = sibling.keys[0] + n.parent.values[n.parentIndex] = sibling.values[0] + copy(sibling.keys[:sibling.nrSegments-1], sibling.keys[1:]) + copy(sibling.values[:sibling.nrSegments-1], sibling.values[1:]) + frameRefSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + n.children[n.nrSegments+1] = sibling.children[0] + copy(sibling.children[:sibling.nrSegments], sibling.children[1:]) + sibling.children[sibling.nrSegments] = nil + n.children[n.nrSegments+1].parent = n + n.children[n.nrSegments+1].parentIndex = n.nrSegments + 1 + for i := 0; i < sibling.nrSegments; i++ { + sibling.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling { + if gap.index == 0 { + return frameRefGapIterator{n, n.nrSegments} + } + return frameRefGapIterator{sibling, gap.index - 1} + } + return gap + } + + p := n.parent + if p.nrSegments == 1 { + + left, right := p.children[0], p.children[1] + p.nrSegments = left.nrSegments + right.nrSegments + 1 + p.hasChildren = left.hasChildren + p.keys[left.nrSegments] = p.keys[0] + p.values[left.nrSegments] = p.values[0] + copy(p.keys[:left.nrSegments], left.keys[:left.nrSegments]) + copy(p.values[:left.nrSegments], left.values[:left.nrSegments]) + copy(p.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(p.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(p.children[:left.nrSegments+1], left.children[:left.nrSegments+1]) + copy(p.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := 0; i < p.nrSegments+1; i++ { + p.children[i].parent = p + p.children[i].parentIndex = i + } + } else { + p.children[0] = nil + p.children[1] = nil + } + if gap.node == left { + return frameRefGapIterator{p, gap.index} + } + if gap.node == right { + return frameRefGapIterator{p, gap.index + left.nrSegments + 1} + } + return gap + } + // Merge n and either sibling, along with the segment separating the + // two, into whichever of the two nodes comes first. This is the + // reverse of the non-root splitting case in + // node.rebalanceBeforeInsert. + var left, right *frameRefnode + if n.parentIndex > 0 { + left = n.prevSibling() + right = n + } else { + left = n + right = n.nextSibling() + } + + if gap.node == right { + gap = frameRefGapIterator{left, gap.index + left.nrSegments + 1} + } + left.keys[left.nrSegments] = p.keys[left.parentIndex] + left.values[left.nrSegments] = p.values[left.parentIndex] + copy(left.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(left.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(left.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := left.nrSegments + 1; i < left.nrSegments+right.nrSegments+2; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + } + } + left.nrSegments += right.nrSegments + 1 + copy(p.keys[left.parentIndex:], p.keys[left.parentIndex+1:p.nrSegments]) + copy(p.values[left.parentIndex:], p.values[left.parentIndex+1:p.nrSegments]) + frameRefSetFunctions{}.ClearValue(&p.values[p.nrSegments-1]) + copy(p.children[left.parentIndex+1:], p.children[left.parentIndex+2:p.nrSegments+1]) + for i := 0; i < p.nrSegments; i++ { + p.children[i].parentIndex = i + } + p.children[p.nrSegments] = nil + p.nrSegments-- + + n = p + } +} + +// A Iterator is conceptually one of: +// +// - A pointer to a segment in a set; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Iterators are copyable values and are meaningfully equality-comparable. The +// zero value of Iterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type frameRefIterator struct { + // node is the node containing the iterated segment. If the iterator is + // terminal, node is nil. + node *frameRefnode + + // index is the index of the segment in node.keys/values. + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (seg frameRefIterator) Ok() bool { + return seg.node != nil +} + +// Range returns the iterated segment's range key. +func (seg frameRefIterator) Range() __generics_imported0.FileRange { + return seg.node.keys[seg.index] +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (seg frameRefIterator) Start() uint64 { + return seg.node.keys[seg.index].Start +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (seg frameRefIterator) End() uint64 { + return seg.node.keys[seg.index].End +} + +// SetRangeUnchecked mutates the iterated segment's range key. This operation +// does not invalidate any iterators. +// +// Preconditions: +// +// - r.Length() > 0. +// +// - The new range must not overlap an existing one: If seg.NextSegment().Ok(), +// then r.end <= seg.NextSegment().Start(); if seg.PrevSegment().Ok(), then +// r.start >= seg.PrevSegment().End(). +func (seg frameRefIterator) SetRangeUnchecked(r __generics_imported0.FileRange) { + seg.node.keys[seg.index] = r +} + +// SetRange mutates the iterated segment's range key. If the new range would +// cause the iterated segment to overlap another segment, or if the new range +// is invalid, SetRange panics. This operation does not invalidate any +// iterators. +func (seg frameRefIterator) SetRange(r __generics_imported0.FileRange) { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if prev := seg.PrevSegment(); prev.Ok() && r.Start < prev.End() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, prev.Range())) + } + if next := seg.NextSegment(); next.Ok() && r.End > next.Start() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, next.Range())) + } + seg.SetRangeUnchecked(r) +} + +// SetStartUnchecked mutates the iterated segment's start. This operation does +// not invalidate any iterators. +// +// Preconditions: The new start must be valid: start < seg.End(); if +// seg.PrevSegment().Ok(), then start >= seg.PrevSegment().End(). +func (seg frameRefIterator) SetStartUnchecked(start uint64) { + seg.node.keys[seg.index].Start = start +} + +// SetStart mutates the iterated segment's start. If the new start value would +// cause the iterated segment to overlap another segment, or would result in an +// invalid range, SetStart panics. This operation does not invalidate any +// iterators. +func (seg frameRefIterator) SetStart(start uint64) { + if start >= seg.End() { + panic(fmt.Sprintf("new start %v would invalidate segment range %v", start, seg.Range())) + } + if prev := seg.PrevSegment(); prev.Ok() && start < prev.End() { + panic(fmt.Sprintf("new start %v would cause segment range %v to overlap segment range %v", start, seg.Range(), prev.Range())) + } + seg.SetStartUnchecked(start) +} + +// SetEndUnchecked mutates the iterated segment's end. This operation does not +// invalidate any iterators. +// +// Preconditions: The new end must be valid: end > seg.Start(); if +// seg.NextSegment().Ok(), then end <= seg.NextSegment().Start(). +func (seg frameRefIterator) SetEndUnchecked(end uint64) { + seg.node.keys[seg.index].End = end +} + +// SetEnd mutates the iterated segment's end. If the new end value would cause +// the iterated segment to overlap another segment, or would result in an +// invalid range, SetEnd panics. This operation does not invalidate any +// iterators. +func (seg frameRefIterator) SetEnd(end uint64) { + if end <= seg.Start() { + panic(fmt.Sprintf("new end %v would invalidate segment range %v", end, seg.Range())) + } + if next := seg.NextSegment(); next.Ok() && end > next.Start() { + panic(fmt.Sprintf("new end %v would cause segment range %v to overlap segment range %v", end, seg.Range(), next.Range())) + } + seg.SetEndUnchecked(end) +} + +// Value returns a copy of the iterated segment's value. +func (seg frameRefIterator) Value() uint64 { + return seg.node.values[seg.index] +} + +// ValuePtr returns a pointer to the iterated segment's value. The pointer is +// invalidated if the iterator is invalidated. This operation does not +// invalidate any iterators. +func (seg frameRefIterator) ValuePtr() *uint64 { + return &seg.node.values[seg.index] +} + +// SetValue mutates the iterated segment's value. This operation does not +// invalidate any iterators. +func (seg frameRefIterator) SetValue(val uint64) { + seg.node.values[seg.index] = val +} + +// PrevSegment returns the iterated segment's predecessor. If there is no +// preceding segment, PrevSegment returns a terminal iterator. +func (seg frameRefIterator) PrevSegment() frameRefIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index].lastSegment() + } + if seg.index > 0 { + return frameRefIterator{seg.node, seg.index - 1} + } + if seg.node.parent == nil { + return frameRefIterator{} + } + return frameRefsegmentBeforePosition(seg.node.parent, seg.node.parentIndex) +} + +// NextSegment returns the iterated segment's successor. If there is no +// succeeding segment, NextSegment returns a terminal iterator. +func (seg frameRefIterator) NextSegment() frameRefIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment() + } + if seg.index < seg.node.nrSegments-1 { + return frameRefIterator{seg.node, seg.index + 1} + } + if seg.node.parent == nil { + return frameRefIterator{} + } + return frameRefsegmentAfterPosition(seg.node.parent, seg.node.parentIndex) +} + +// PrevGap returns the gap immediately before the iterated segment. +func (seg frameRefIterator) PrevGap() frameRefGapIterator { + if seg.node.hasChildren { + + return seg.node.children[seg.index].lastSegment().NextGap() + } + return frameRefGapIterator{seg.node, seg.index} +} + +// NextGap returns the gap immediately after the iterated segment. +func (seg frameRefIterator) NextGap() frameRefGapIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment().PrevGap() + } + return frameRefGapIterator{seg.node, seg.index + 1} +} + +// PrevNonEmpty returns the iterated segment's predecessor if it is adjacent, +// or the gap before the iterated segment otherwise. If seg.Start() == +// Functions.MinKey(), PrevNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by PrevNonEmpty will be +// non-terminal. +func (seg frameRefIterator) PrevNonEmpty() (frameRefIterator, frameRefGapIterator) { + gap := seg.PrevGap() + if gap.Range().Length() != 0 { + return frameRefIterator{}, gap + } + return gap.PrevSegment(), frameRefGapIterator{} +} + +// NextNonEmpty returns the iterated segment's successor if it is adjacent, or +// the gap after the iterated segment otherwise. If seg.End() == +// Functions.MaxKey(), NextNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by NextNonEmpty will be +// non-terminal. +func (seg frameRefIterator) NextNonEmpty() (frameRefIterator, frameRefGapIterator) { + gap := seg.NextGap() + if gap.Range().Length() != 0 { + return frameRefIterator{}, gap + } + return gap.NextSegment(), frameRefGapIterator{} +} + +// A GapIterator is conceptually one of: +// +// - A pointer to a position between two segments, before the first segment, or +// after the last segment in a set, called a *gap*; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Note that the gap between two adjacent segments exists (iterators to it are +// non-terminal), but has a length of zero. GapIterator.IsEmpty returns true +// for such gaps. An empty set contains a single gap, spanning the entire range +// of the set's keys. +// +// GapIterators are copyable values and are meaningfully equality-comparable. +// The zero value of GapIterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type frameRefGapIterator struct { + // The representation of a GapIterator is identical to that of an Iterator, + // except that index corresponds to positions between segments in the same + // way as for node.children (see comment for node.nrSegments). + node *frameRefnode + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (gap frameRefGapIterator) Ok() bool { + return gap.node != nil +} + +// Range returns the range spanned by the iterated gap. +func (gap frameRefGapIterator) Range() __generics_imported0.FileRange { + return __generics_imported0.FileRange{gap.Start(), gap.End()} +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (gap frameRefGapIterator) Start() uint64 { + if ps := gap.PrevSegment(); ps.Ok() { + return ps.End() + } + return frameRefSetFunctions{}.MinKey() +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (gap frameRefGapIterator) End() uint64 { + if ns := gap.NextSegment(); ns.Ok() { + return ns.Start() + } + return frameRefSetFunctions{}.MaxKey() +} + +// IsEmpty returns true if the iterated gap is empty (that is, the "gap" is +// between two adjacent segments.) +func (gap frameRefGapIterator) IsEmpty() bool { + return gap.Range().Length() == 0 +} + +// PrevSegment returns the segment immediately before the iterated gap. If no +// such segment exists, PrevSegment returns a terminal iterator. +func (gap frameRefGapIterator) PrevSegment() frameRefIterator { + return frameRefsegmentBeforePosition(gap.node, gap.index) +} + +// NextSegment returns the segment immediately after the iterated gap. If no +// such segment exists, NextSegment returns a terminal iterator. +func (gap frameRefGapIterator) NextSegment() frameRefIterator { + return frameRefsegmentAfterPosition(gap.node, gap.index) +} + +// PrevGap returns the iterated gap's predecessor. If no such gap exists, +// PrevGap returns a terminal iterator. +func (gap frameRefGapIterator) PrevGap() frameRefGapIterator { + seg := gap.PrevSegment() + if !seg.Ok() { + return frameRefGapIterator{} + } + return seg.PrevGap() +} + +// NextGap returns the iterated gap's successor. If no such gap exists, NextGap +// returns a terminal iterator. +func (gap frameRefGapIterator) NextGap() frameRefGapIterator { + seg := gap.NextSegment() + if !seg.Ok() { + return frameRefGapIterator{} + } + return seg.NextGap() +} + +// segmentBeforePosition returns the predecessor segment of the position given +// by n.children[i], which may or may not contain a child. If no such segment +// exists, segmentBeforePosition returns a terminal iterator. +func frameRefsegmentBeforePosition(n *frameRefnode, i int) frameRefIterator { + for i == 0 { + if n.parent == nil { + return frameRefIterator{} + } + n, i = n.parent, n.parentIndex + } + return frameRefIterator{n, i - 1} +} + +// segmentAfterPosition returns the successor segment of the position given by +// n.children[i], which may or may not contain a child. If no such segment +// exists, segmentAfterPosition returns a terminal iterator. +func frameRefsegmentAfterPosition(n *frameRefnode, i int) frameRefIterator { + for i == n.nrSegments { + if n.parent == nil { + return frameRefIterator{} + } + n, i = n.parent, n.parentIndex + } + return frameRefIterator{n, i} +} + +func frameRefzeroValueSlice(slice []uint64) { + + for i := range slice { + frameRefSetFunctions{}.ClearValue(&slice[i]) + } +} + +func frameRefzeroNodeSlice(slice []*frameRefnode) { + for i := range slice { + slice[i] = nil + } +} + +// String stringifies a Set for debugging. +func (s *frameRefSet) String() string { + return s.root.String() +} + +// String stringifies a node (and all of its children) for debugging. +func (n *frameRefnode) String() string { + var buf bytes.Buffer + n.writeDebugString(&buf, "") + return buf.String() +} + +func (n *frameRefnode) writeDebugString(buf *bytes.Buffer, prefix string) { + if n.hasChildren != (n.nrSegments > 0 && n.children[0] != nil) { + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent value of hasChildren: got %v, want %v\n", n.hasChildren, !n.hasChildren)) + } + for i := 0; i < n.nrSegments; i++ { + if child := n.children[i]; child != nil { + cprefix := fmt.Sprintf("%s- % 3d ", prefix, i) + if child.parent != n || child.parentIndex != i { + buf.WriteString(cprefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent linkage to parent: got (%p, %d), want (%p, %d)\n", child.parent, child.parentIndex, n, i)) + } + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, i)) + } + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + } + if child := n.children[n.nrSegments]; child != nil { + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, n.nrSegments)) + } +} + +// SegmentDataSlices represents segments from a set as slices of start, end, and +// values. SegmentDataSlices is primarily used as an intermediate representation +// for save/restore and the layout here is optimized for that. +// +// +stateify savable +type frameRefSegmentDataSlices struct { + Start []uint64 + End []uint64 + Values []uint64 +} + +// ExportSortedSlice returns a copy of all segments in the given set, in ascending +// key order. +func (s *frameRefSet) ExportSortedSlices() *frameRefSegmentDataSlices { + var sds frameRefSegmentDataSlices + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sds.Start = append(sds.Start, seg.Start()) + sds.End = append(sds.End, seg.End()) + sds.Values = append(sds.Values, seg.Value()) + } + sds.Start = sds.Start[:len(sds.Start):len(sds.Start)] + sds.End = sds.End[:len(sds.End):len(sds.End)] + sds.Values = sds.Values[:len(sds.Values):len(sds.Values)] + return &sds +} + +// ImportSortedSlice initializes the given set from the given slice. +// +// Preconditions: s must be empty. sds must represent a valid set (the segments +// in sds must have valid lengths that do not overlap). The segments in sds +// must be sorted in ascending key order. +func (s *frameRefSet) ImportSortedSlices(sds *frameRefSegmentDataSlices) error { + if !s.IsEmpty() { + return fmt.Errorf("cannot import into non-empty set %v", s) + } + gap := s.FirstGap() + for i := range sds.Start { + r := __generics_imported0.FileRange{sds.Start[i], sds.End[i]} + if !gap.Range().IsSupersetOf(r) { + return fmt.Errorf("segment overlaps a preceding segment or is incorrectly sorted: [%d, %d) => %v", sds.Start[i], sds.End[i], sds.Values[i]) + } + gap = s.InsertWithoutMerging(gap, r, sds.Values[i]).NextGap() + } + return nil +} +func (s *frameRefSet) saveRoot() *frameRefSegmentDataSlices { + return s.ExportSortedSlices() +} + +func (s *frameRefSet) loadRoot(sds *frameRefSegmentDataSlices) { + if err := s.ImportSortedSlices(sds); err != nil { + panic(err) + } +} diff --git a/pkg/sentry/fs/fsutil/fsutil_state_autogen.go b/pkg/sentry/fs/fsutil/fsutil_state_autogen.go new file mode 100755 index 000000000..6371a66a5 --- /dev/null +++ b/pkg/sentry/fs/fsutil/fsutil_state_autogen.go @@ -0,0 +1,349 @@ +// automatically generated by stateify. + +package fsutil + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *DirtyInfo) beforeSave() {} +func (x *DirtyInfo) save(m state.Map) { + x.beforeSave() + m.Save("Keep", &x.Keep) +} + +func (x *DirtyInfo) afterLoad() {} +func (x *DirtyInfo) load(m state.Map) { + m.Load("Keep", &x.Keep) +} + +func (x *DirtySet) beforeSave() {} +func (x *DirtySet) save(m state.Map) { + x.beforeSave() + var root *DirtySegmentDataSlices = x.saveRoot() + m.SaveValue("root", root) +} + +func (x *DirtySet) afterLoad() {} +func (x *DirtySet) load(m state.Map) { + m.LoadValue("root", new(*DirtySegmentDataSlices), func(y interface{}) { x.loadRoot(y.(*DirtySegmentDataSlices)) }) +} + +func (x *Dirtynode) beforeSave() {} +func (x *Dirtynode) save(m state.Map) { + x.beforeSave() + m.Save("nrSegments", &x.nrSegments) + m.Save("parent", &x.parent) + m.Save("parentIndex", &x.parentIndex) + m.Save("hasChildren", &x.hasChildren) + m.Save("keys", &x.keys) + m.Save("values", &x.values) + m.Save("children", &x.children) +} + +func (x *Dirtynode) afterLoad() {} +func (x *Dirtynode) load(m state.Map) { + m.Load("nrSegments", &x.nrSegments) + m.Load("parent", &x.parent) + m.Load("parentIndex", &x.parentIndex) + m.Load("hasChildren", &x.hasChildren) + m.Load("keys", &x.keys) + m.Load("values", &x.values) + m.Load("children", &x.children) +} + +func (x *DirtySegmentDataSlices) beforeSave() {} +func (x *DirtySegmentDataSlices) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) + m.Save("Values", &x.Values) +} + +func (x *DirtySegmentDataSlices) afterLoad() {} +func (x *DirtySegmentDataSlices) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) + m.Load("Values", &x.Values) +} + +func (x *StaticDirFileOperations) beforeSave() {} +func (x *StaticDirFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("dentryMap", &x.dentryMap) + m.Save("dirCursor", &x.dirCursor) +} + +func (x *StaticDirFileOperations) afterLoad() {} +func (x *StaticDirFileOperations) load(m state.Map) { + m.Load("dentryMap", &x.dentryMap) + m.Load("dirCursor", &x.dirCursor) +} + +func (x *NoReadWriteFile) beforeSave() {} +func (x *NoReadWriteFile) save(m state.Map) { + x.beforeSave() +} + +func (x *NoReadWriteFile) afterLoad() {} +func (x *NoReadWriteFile) load(m state.Map) { +} + +func (x *FileStaticContentReader) beforeSave() {} +func (x *FileStaticContentReader) save(m state.Map) { + x.beforeSave() + m.Save("content", &x.content) +} + +func (x *FileStaticContentReader) afterLoad() {} +func (x *FileStaticContentReader) load(m state.Map) { + m.Load("content", &x.content) +} + +func (x *FileRangeSet) beforeSave() {} +func (x *FileRangeSet) save(m state.Map) { + x.beforeSave() + var root *FileRangeSegmentDataSlices = x.saveRoot() + m.SaveValue("root", root) +} + +func (x *FileRangeSet) afterLoad() {} +func (x *FileRangeSet) load(m state.Map) { + m.LoadValue("root", new(*FileRangeSegmentDataSlices), func(y interface{}) { x.loadRoot(y.(*FileRangeSegmentDataSlices)) }) +} + +func (x *FileRangenode) beforeSave() {} +func (x *FileRangenode) save(m state.Map) { + x.beforeSave() + m.Save("nrSegments", &x.nrSegments) + m.Save("parent", &x.parent) + m.Save("parentIndex", &x.parentIndex) + m.Save("hasChildren", &x.hasChildren) + m.Save("keys", &x.keys) + m.Save("values", &x.values) + m.Save("children", &x.children) +} + +func (x *FileRangenode) afterLoad() {} +func (x *FileRangenode) load(m state.Map) { + m.Load("nrSegments", &x.nrSegments) + m.Load("parent", &x.parent) + m.Load("parentIndex", &x.parentIndex) + m.Load("hasChildren", &x.hasChildren) + m.Load("keys", &x.keys) + m.Load("values", &x.values) + m.Load("children", &x.children) +} + +func (x *FileRangeSegmentDataSlices) beforeSave() {} +func (x *FileRangeSegmentDataSlices) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) + m.Save("Values", &x.Values) +} + +func (x *FileRangeSegmentDataSlices) afterLoad() {} +func (x *FileRangeSegmentDataSlices) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) + m.Load("Values", &x.Values) +} + +func (x *frameRefSet) beforeSave() {} +func (x *frameRefSet) save(m state.Map) { + x.beforeSave() + var root *frameRefSegmentDataSlices = x.saveRoot() + m.SaveValue("root", root) +} + +func (x *frameRefSet) afterLoad() {} +func (x *frameRefSet) load(m state.Map) { + m.LoadValue("root", new(*frameRefSegmentDataSlices), func(y interface{}) { x.loadRoot(y.(*frameRefSegmentDataSlices)) }) +} + +func (x *frameRefnode) beforeSave() {} +func (x *frameRefnode) save(m state.Map) { + x.beforeSave() + m.Save("nrSegments", &x.nrSegments) + m.Save("parent", &x.parent) + m.Save("parentIndex", &x.parentIndex) + m.Save("hasChildren", &x.hasChildren) + m.Save("keys", &x.keys) + m.Save("values", &x.values) + m.Save("children", &x.children) +} + +func (x *frameRefnode) afterLoad() {} +func (x *frameRefnode) load(m state.Map) { + m.Load("nrSegments", &x.nrSegments) + m.Load("parent", &x.parent) + m.Load("parentIndex", &x.parentIndex) + m.Load("hasChildren", &x.hasChildren) + m.Load("keys", &x.keys) + m.Load("values", &x.values) + m.Load("children", &x.children) +} + +func (x *frameRefSegmentDataSlices) beforeSave() {} +func (x *frameRefSegmentDataSlices) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) + m.Save("Values", &x.Values) +} + +func (x *frameRefSegmentDataSlices) afterLoad() {} +func (x *frameRefSegmentDataSlices) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) + m.Load("Values", &x.Values) +} + +func (x *HostFileMapper) beforeSave() {} +func (x *HostFileMapper) save(m state.Map) { + x.beforeSave() + m.Save("refs", &x.refs) +} + +func (x *HostFileMapper) load(m state.Map) { + m.Load("refs", &x.refs) + m.AfterLoad(x.afterLoad) +} + +func (x *HostMappable) beforeSave() {} +func (x *HostMappable) save(m state.Map) { + x.beforeSave() + m.Save("hostFileMapper", &x.hostFileMapper) + m.Save("backingFile", &x.backingFile) + m.Save("mappings", &x.mappings) +} + +func (x *HostMappable) afterLoad() {} +func (x *HostMappable) load(m state.Map) { + m.Load("hostFileMapper", &x.hostFileMapper) + m.Load("backingFile", &x.backingFile) + m.Load("mappings", &x.mappings) +} + +func (x *SimpleFileInode) beforeSave() {} +func (x *SimpleFileInode) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *SimpleFileInode) afterLoad() {} +func (x *SimpleFileInode) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *NoReadWriteFileInode) beforeSave() {} +func (x *NoReadWriteFileInode) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *NoReadWriteFileInode) afterLoad() {} +func (x *NoReadWriteFileInode) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) +} + +func (x *InodeSimpleAttributes) beforeSave() {} +func (x *InodeSimpleAttributes) save(m state.Map) { + x.beforeSave() + m.Save("fsType", &x.fsType) + m.Save("unstable", &x.unstable) +} + +func (x *InodeSimpleAttributes) afterLoad() {} +func (x *InodeSimpleAttributes) load(m state.Map) { + m.Load("fsType", &x.fsType) + m.Load("unstable", &x.unstable) +} + +func (x *InodeSimpleExtendedAttributes) beforeSave() {} +func (x *InodeSimpleExtendedAttributes) save(m state.Map) { + x.beforeSave() + m.Save("xattrs", &x.xattrs) +} + +func (x *InodeSimpleExtendedAttributes) afterLoad() {} +func (x *InodeSimpleExtendedAttributes) load(m state.Map) { + m.Load("xattrs", &x.xattrs) +} + +func (x *staticFile) beforeSave() {} +func (x *staticFile) save(m state.Map) { + x.beforeSave() + m.Save("FileStaticContentReader", &x.FileStaticContentReader) +} + +func (x *staticFile) afterLoad() {} +func (x *staticFile) load(m state.Map) { + m.Load("FileStaticContentReader", &x.FileStaticContentReader) +} + +func (x *InodeStaticFileGetter) beforeSave() {} +func (x *InodeStaticFileGetter) save(m state.Map) { + x.beforeSave() + m.Save("Contents", &x.Contents) +} + +func (x *InodeStaticFileGetter) afterLoad() {} +func (x *InodeStaticFileGetter) load(m state.Map) { + m.Load("Contents", &x.Contents) +} + +func (x *CachingInodeOperations) beforeSave() {} +func (x *CachingInodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("backingFile", &x.backingFile) + m.Save("mfp", &x.mfp) + m.Save("forcePageCache", &x.forcePageCache) + m.Save("attr", &x.attr) + m.Save("dirtyAttr", &x.dirtyAttr) + m.Save("mappings", &x.mappings) + m.Save("cache", &x.cache) + m.Save("dirty", &x.dirty) + m.Save("hostFileMapper", &x.hostFileMapper) + m.Save("refs", &x.refs) +} + +func (x *CachingInodeOperations) afterLoad() {} +func (x *CachingInodeOperations) load(m state.Map) { + m.Load("backingFile", &x.backingFile) + m.Load("mfp", &x.mfp) + m.Load("forcePageCache", &x.forcePageCache) + m.Load("attr", &x.attr) + m.Load("dirtyAttr", &x.dirtyAttr) + m.Load("mappings", &x.mappings) + m.Load("cache", &x.cache) + m.Load("dirty", &x.dirty) + m.Load("hostFileMapper", &x.hostFileMapper) + m.Load("refs", &x.refs) +} + +func init() { + state.Register("fsutil.DirtyInfo", (*DirtyInfo)(nil), state.Fns{Save: (*DirtyInfo).save, Load: (*DirtyInfo).load}) + state.Register("fsutil.DirtySet", (*DirtySet)(nil), state.Fns{Save: (*DirtySet).save, Load: (*DirtySet).load}) + state.Register("fsutil.Dirtynode", (*Dirtynode)(nil), state.Fns{Save: (*Dirtynode).save, Load: (*Dirtynode).load}) + state.Register("fsutil.DirtySegmentDataSlices", (*DirtySegmentDataSlices)(nil), state.Fns{Save: (*DirtySegmentDataSlices).save, Load: (*DirtySegmentDataSlices).load}) + state.Register("fsutil.StaticDirFileOperations", (*StaticDirFileOperations)(nil), state.Fns{Save: (*StaticDirFileOperations).save, Load: (*StaticDirFileOperations).load}) + state.Register("fsutil.NoReadWriteFile", (*NoReadWriteFile)(nil), state.Fns{Save: (*NoReadWriteFile).save, Load: (*NoReadWriteFile).load}) + state.Register("fsutil.FileStaticContentReader", (*FileStaticContentReader)(nil), state.Fns{Save: (*FileStaticContentReader).save, Load: (*FileStaticContentReader).load}) + state.Register("fsutil.FileRangeSet", (*FileRangeSet)(nil), state.Fns{Save: (*FileRangeSet).save, Load: (*FileRangeSet).load}) + state.Register("fsutil.FileRangenode", (*FileRangenode)(nil), state.Fns{Save: (*FileRangenode).save, Load: (*FileRangenode).load}) + state.Register("fsutil.FileRangeSegmentDataSlices", (*FileRangeSegmentDataSlices)(nil), state.Fns{Save: (*FileRangeSegmentDataSlices).save, Load: (*FileRangeSegmentDataSlices).load}) + state.Register("fsutil.frameRefSet", (*frameRefSet)(nil), state.Fns{Save: (*frameRefSet).save, Load: (*frameRefSet).load}) + state.Register("fsutil.frameRefnode", (*frameRefnode)(nil), state.Fns{Save: (*frameRefnode).save, Load: (*frameRefnode).load}) + state.Register("fsutil.frameRefSegmentDataSlices", (*frameRefSegmentDataSlices)(nil), state.Fns{Save: (*frameRefSegmentDataSlices).save, Load: (*frameRefSegmentDataSlices).load}) + state.Register("fsutil.HostFileMapper", (*HostFileMapper)(nil), state.Fns{Save: (*HostFileMapper).save, Load: (*HostFileMapper).load}) + state.Register("fsutil.HostMappable", (*HostMappable)(nil), state.Fns{Save: (*HostMappable).save, Load: (*HostMappable).load}) + state.Register("fsutil.SimpleFileInode", (*SimpleFileInode)(nil), state.Fns{Save: (*SimpleFileInode).save, Load: (*SimpleFileInode).load}) + state.Register("fsutil.NoReadWriteFileInode", (*NoReadWriteFileInode)(nil), state.Fns{Save: (*NoReadWriteFileInode).save, Load: (*NoReadWriteFileInode).load}) + state.Register("fsutil.InodeSimpleAttributes", (*InodeSimpleAttributes)(nil), state.Fns{Save: (*InodeSimpleAttributes).save, Load: (*InodeSimpleAttributes).load}) + state.Register("fsutil.InodeSimpleExtendedAttributes", (*InodeSimpleExtendedAttributes)(nil), state.Fns{Save: (*InodeSimpleExtendedAttributes).save, Load: (*InodeSimpleExtendedAttributes).load}) + state.Register("fsutil.staticFile", (*staticFile)(nil), state.Fns{Save: (*staticFile).save, Load: (*staticFile).load}) + state.Register("fsutil.InodeStaticFileGetter", (*InodeStaticFileGetter)(nil), state.Fns{Save: (*InodeStaticFileGetter).save, Load: (*InodeStaticFileGetter).load}) + state.Register("fsutil.CachingInodeOperations", (*CachingInodeOperations)(nil), state.Fns{Save: (*CachingInodeOperations).save, Load: (*CachingInodeOperations).load}) +} diff --git a/pkg/sentry/fs/fsutil/inode_cached_test.go b/pkg/sentry/fs/fsutil/inode_cached_test.go deleted file mode 100644 index dc19255ed..000000000 --- a/pkg/sentry/fs/fsutil/inode_cached_test.go +++ /dev/null @@ -1,389 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fsutil - -import ( - "bytes" - "io" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/syserror" -) - -type noopBackingFile struct{} - -func (noopBackingFile) ReadToBlocksAt(ctx context.Context, dsts safemem.BlockSeq, offset uint64) (uint64, error) { - return dsts.NumBytes(), nil -} - -func (noopBackingFile) WriteFromBlocksAt(ctx context.Context, srcs safemem.BlockSeq, offset uint64) (uint64, error) { - return srcs.NumBytes(), nil -} - -func (noopBackingFile) SetMaskedAttributes(context.Context, fs.AttrMask, fs.UnstableAttr) error { - return nil -} - -func (noopBackingFile) Sync(context.Context) error { - return nil -} - -func (noopBackingFile) FD() int { - return -1 -} - -func (noopBackingFile) Allocate(ctx context.Context, offset int64, length int64) error { - return nil -} - -func TestSetPermissions(t *testing.T) { - ctx := contexttest.Context(t) - - uattr := fs.WithCurrentTime(ctx, fs.UnstableAttr{ - Perms: fs.FilePermsFromMode(0444), - }) - iops := NewCachingInodeOperations(ctx, noopBackingFile{}, uattr, false /*forcePageCache*/) - defer iops.Release() - - perms := fs.FilePermsFromMode(0777) - if !iops.SetPermissions(ctx, nil, perms) { - t.Fatalf("SetPermissions failed, want success") - } - - // Did permissions change? - if iops.attr.Perms != perms { - t.Fatalf("got perms +%v, want +%v", iops.attr.Perms, perms) - } - - // Did status change time change? - if !iops.dirtyAttr.StatusChangeTime { - t.Fatalf("got status change time not dirty, want dirty") - } - if iops.attr.StatusChangeTime.Equal(uattr.StatusChangeTime) { - t.Fatalf("got status change time unchanged") - } -} - -func TestSetTimestamps(t *testing.T) { - ctx := contexttest.Context(t) - for _, test := range []struct { - desc string - ts fs.TimeSpec - wantChanged fs.AttrMask - }{ - { - desc: "noop", - ts: fs.TimeSpec{ - ATimeOmit: true, - MTimeOmit: true, - }, - wantChanged: fs.AttrMask{}, - }, - { - desc: "access time only", - ts: fs.TimeSpec{ - ATime: ktime.NowFromContext(ctx), - MTimeOmit: true, - }, - wantChanged: fs.AttrMask{ - AccessTime: true, - }, - }, - { - desc: "modification time only", - ts: fs.TimeSpec{ - ATimeOmit: true, - MTime: ktime.NowFromContext(ctx), - }, - wantChanged: fs.AttrMask{ - ModificationTime: true, - }, - }, - { - desc: "access and modification time", - ts: fs.TimeSpec{ - ATime: ktime.NowFromContext(ctx), - MTime: ktime.NowFromContext(ctx), - }, - wantChanged: fs.AttrMask{ - AccessTime: true, - ModificationTime: true, - }, - }, - { - desc: "system time access and modification time", - ts: fs.TimeSpec{ - ATimeSetSystemTime: true, - MTimeSetSystemTime: true, - }, - wantChanged: fs.AttrMask{ - AccessTime: true, - ModificationTime: true, - }, - }, - } { - t.Run(test.desc, func(t *testing.T) { - ctx := contexttest.Context(t) - - epoch := ktime.ZeroTime - uattr := fs.UnstableAttr{ - AccessTime: epoch, - ModificationTime: epoch, - StatusChangeTime: epoch, - } - iops := NewCachingInodeOperations(ctx, noopBackingFile{}, uattr, false /*forcePageCache*/) - defer iops.Release() - - if err := iops.SetTimestamps(ctx, nil, test.ts); err != nil { - t.Fatalf("SetTimestamps got error %v, want nil", err) - } - if test.wantChanged.AccessTime { - if !iops.attr.AccessTime.After(uattr.AccessTime) { - t.Fatalf("diritied access time did not advance, want %v > %v", iops.attr.AccessTime, uattr.AccessTime) - } - if !iops.dirtyAttr.StatusChangeTime { - t.Fatalf("dirty access time requires dirty status change time") - } - if !iops.attr.StatusChangeTime.After(uattr.StatusChangeTime) { - t.Fatalf("dirtied status change time did not advance") - } - } - if test.wantChanged.ModificationTime { - if !iops.attr.ModificationTime.After(uattr.ModificationTime) { - t.Fatalf("diritied modification time did not advance") - } - if !iops.dirtyAttr.StatusChangeTime { - t.Fatalf("dirty modification time requires dirty status change time") - } - if !iops.attr.StatusChangeTime.After(uattr.StatusChangeTime) { - t.Fatalf("dirtied status change time did not advance") - } - } - }) - } -} - -func TestTruncate(t *testing.T) { - ctx := contexttest.Context(t) - - uattr := fs.UnstableAttr{ - Size: 0, - } - iops := NewCachingInodeOperations(ctx, noopBackingFile{}, uattr, false /*forcePageCache*/) - defer iops.Release() - - if err := iops.Truncate(ctx, nil, uattr.Size); err != nil { - t.Fatalf("Truncate got error %v, want nil", err) - } - var size int64 = 4096 - if err := iops.Truncate(ctx, nil, size); err != nil { - t.Fatalf("Truncate got error %v, want nil", err) - } - if iops.attr.Size != size { - t.Fatalf("Truncate got %d, want %d", iops.attr.Size, size) - } - if !iops.dirtyAttr.ModificationTime || !iops.dirtyAttr.StatusChangeTime { - t.Fatalf("Truncate did not dirty modification and status change time") - } - if !iops.attr.ModificationTime.After(uattr.ModificationTime) { - t.Fatalf("dirtied modification time did not change") - } - if !iops.attr.StatusChangeTime.After(uattr.StatusChangeTime) { - t.Fatalf("dirtied status change time did not change") - } -} - -type sliceBackingFile struct { - data []byte -} - -func newSliceBackingFile(data []byte) *sliceBackingFile { - return &sliceBackingFile{data} -} - -func (f *sliceBackingFile) ReadToBlocksAt(ctx context.Context, dsts safemem.BlockSeq, offset uint64) (uint64, error) { - r := safemem.BlockSeqReader{safemem.BlockSeqOf(safemem.BlockFromSafeSlice(f.data)).DropFirst64(offset)} - return r.ReadToBlocks(dsts) -} - -func (f *sliceBackingFile) WriteFromBlocksAt(ctx context.Context, srcs safemem.BlockSeq, offset uint64) (uint64, error) { - w := safemem.BlockSeqWriter{safemem.BlockSeqOf(safemem.BlockFromSafeSlice(f.data)).DropFirst64(offset)} - return w.WriteFromBlocks(srcs) -} - -func (*sliceBackingFile) SetMaskedAttributes(context.Context, fs.AttrMask, fs.UnstableAttr) error { - return nil -} - -func (*sliceBackingFile) Sync(context.Context) error { - return nil -} - -func (*sliceBackingFile) FD() int { - return -1 -} - -func (f *sliceBackingFile) Allocate(ctx context.Context, offset int64, length int64) error { - return syserror.EOPNOTSUPP -} - -type noopMappingSpace struct{} - -// Invalidate implements memmap.MappingSpace.Invalidate. -func (noopMappingSpace) Invalidate(ar usermem.AddrRange, opts memmap.InvalidateOpts) { -} - -func anonInode(ctx context.Context) *fs.Inode { - return fs.NewInode(ctx, &SimpleFileInode{ - InodeSimpleAttributes: NewInodeSimpleAttributes(ctx, fs.FileOwnerFromContext(ctx), fs.FilePermissions{ - User: fs.PermMask{Read: true, Write: true}, - }, 0), - }, fs.NewPseudoMountSource(ctx), fs.StableAttr{ - Type: fs.Anonymous, - BlockSize: usermem.PageSize, - }) -} - -func pagesOf(bs ...byte) []byte { - buf := make([]byte, 0, len(bs)*usermem.PageSize) - for _, b := range bs { - buf = append(buf, bytes.Repeat([]byte{b}, usermem.PageSize)...) - } - return buf -} - -func TestRead(t *testing.T) { - ctx := contexttest.Context(t) - - // Construct a 3-page file. - buf := pagesOf('a', 'b', 'c') - file := fs.NewFile(ctx, fs.NewDirent(ctx, anonInode(ctx), "anon"), fs.FileFlags{}, nil) - uattr := fs.UnstableAttr{ - Size: int64(len(buf)), - } - iops := NewCachingInodeOperations(ctx, newSliceBackingFile(buf), uattr, false /*forcePageCache*/) - defer iops.Release() - - // Expect the cache to be initially empty. - if cached := iops.cache.Span(); cached != 0 { - t.Errorf("Span got %d, want 0", cached) - } - - // Create a memory mapping of the second page (as CachingInodeOperations - // expects to only cache mapped pages), then call Translate to force it to - // be cached. - var ms noopMappingSpace - ar := usermem.AddrRange{usermem.PageSize, 2 * usermem.PageSize} - if err := iops.AddMapping(ctx, ms, ar, usermem.PageSize, true); err != nil { - t.Fatalf("AddMapping got %v, want nil", err) - } - mr := memmap.MappableRange{usermem.PageSize, 2 * usermem.PageSize} - if _, err := iops.Translate(ctx, mr, mr, usermem.Read); err != nil { - t.Fatalf("Translate got %v, want nil", err) - } - if cached := iops.cache.Span(); cached != usermem.PageSize { - t.Errorf("SpanRange got %d, want %d", cached, usermem.PageSize) - } - - // Try to read 4 pages. The first and third pages should be read directly - // from the "file", the second page should be read from the cache, and only - // 3 pages (the size of the file) should be readable. - rbuf := make([]byte, 4*usermem.PageSize) - dst := usermem.BytesIOSequence(rbuf) - n, err := iops.Read(ctx, file, dst, 0) - if n != 3*usermem.PageSize || (err != nil && err != io.EOF) { - t.Fatalf("Read got (%d, %v), want (%d, nil or EOF)", n, err, 3*usermem.PageSize) - } - rbuf = rbuf[:3*usermem.PageSize] - - // Did we get the bytes we expect? - if !bytes.Equal(rbuf, buf) { - t.Errorf("Read back bytes %v, want %v", rbuf, buf) - } - - // Delete the memory mapping before iops.Release(). The cached page will - // either be evicted by ctx's pgalloc.MemoryFile, or dropped by - // iops.Release(). - iops.RemoveMapping(ctx, ms, ar, usermem.PageSize, true) -} - -func TestWrite(t *testing.T) { - ctx := contexttest.Context(t) - - // Construct a 4-page file. - buf := pagesOf('a', 'b', 'c', 'd') - orig := append([]byte(nil), buf...) - inode := anonInode(ctx) - uattr := fs.UnstableAttr{ - Size: int64(len(buf)), - } - iops := NewCachingInodeOperations(ctx, newSliceBackingFile(buf), uattr, false /*forcePageCache*/) - defer iops.Release() - - // Expect the cache to be initially empty. - if cached := iops.cache.Span(); cached != 0 { - t.Errorf("Span got %d, want 0", cached) - } - - // Create a memory mapping of the second and third pages (as - // CachingInodeOperations expects to only cache mapped pages), then call - // Translate to force them to be cached. - var ms noopMappingSpace - ar := usermem.AddrRange{usermem.PageSize, 3 * usermem.PageSize} - if err := iops.AddMapping(ctx, ms, ar, usermem.PageSize, true); err != nil { - t.Fatalf("AddMapping got %v, want nil", err) - } - defer iops.RemoveMapping(ctx, ms, ar, usermem.PageSize, true) - mr := memmap.MappableRange{usermem.PageSize, 3 * usermem.PageSize} - if _, err := iops.Translate(ctx, mr, mr, usermem.Read); err != nil { - t.Fatalf("Translate got %v, want nil", err) - } - if cached := iops.cache.Span(); cached != 2*usermem.PageSize { - t.Errorf("SpanRange got %d, want %d", cached, 2*usermem.PageSize) - } - - // Write to the first 2 pages. - wbuf := pagesOf('e', 'f') - src := usermem.BytesIOSequence(wbuf) - n, err := iops.Write(ctx, src, 0) - if n != 2*usermem.PageSize || err != nil { - t.Fatalf("Write got (%d, %v), want (%d, nil)", n, err, 2*usermem.PageSize) - } - - // The first page should have been written directly, since it was not cached. - want := append([]byte(nil), orig...) - copy(want, pagesOf('e')) - if !bytes.Equal(buf, want) { - t.Errorf("File contents are %v, want %v", buf, want) - } - - // Sync back to the "backing file". - if err := iops.WriteOut(ctx, inode); err != nil { - t.Errorf("Sync got %v, want nil", err) - } - - // Now the second page should have been written as well. - copy(want[usermem.PageSize:], pagesOf('f')) - if !bytes.Equal(buf, want) { - t.Errorf("File contents are %v, want %v", buf, want) - } -} diff --git a/pkg/sentry/fs/g3doc/inotify.md b/pkg/sentry/fs/g3doc/inotify.md deleted file mode 100644 index 71a577d9d..000000000 --- a/pkg/sentry/fs/g3doc/inotify.md +++ /dev/null @@ -1,122 +0,0 @@ -# Inotify - -Inotify implements the like-named filesystem event notification system for the -sentry, see `inotify(7)`. - -## Architecture - -For the most part, the sentry implementation of inotify mirrors the Linux -architecture. Inotify instances (i.e. the fd returned by inotify_init(2)) are -backed by a pseudo-filesystem. Events are generated from various places in the -sentry, including the [syscall layer][syscall_dir], the [vfs layer][dirent] and -the [process fd table][fd_table]. Watches are stored in inodes and generated -events are queued to the inotify instance owning the watches for delivery to the -user. - -## Objects - -Here is a brief description of the existing and new objects involved in the -sentry inotify mechanism, and how they interact: - -### [`fs.Inotify`][inotify] - -- An inotify instances, created by inotify_init(2)/inotify_init1(2). -- The inotify fd has a `fs.Dirent`, supports filesystem syscalls to read - events. -- Has multiple `fs.Watch`es, with at most one watch per target inode, per - inotify instance. -- Has an instance `id` which is globally unique. This is *not* the fd number - for this instance, since the fd can be duped. This `id` is not externally - visible. - -### [`fs.Watch`][watch] - -- An inotify watch, created/deleted by - inotify_add_watch(2)/inotify_rm_watch(2). -- Owned by an `fs.Inotify` instance, each watch keeps a pointer to the - `owner`. -- Associated with a single `fs.Inode`, which is the watch `target`. While the - watch is active, it indirectly pins `target` to memory. See the "Reference - Model" section for a detailed explanation. -- Filesystem operations on `target` generate `fs.Event`s. - -### [`fs.Event`][event] - -- A simple struct encapsulating all the fields for an inotify event. -- Generated by `fs.Watch`es and forwarded to the watches' `owner`s. -- Serialized to the user during read(2) syscalls on the associated - `fs.Inotify`'s fd. - -### [`fs.Dirent`][dirent] - -- Many inotify events are generated inside dirent methods. Events are - generated in the dirent methods rather than `fs.Inode` methods because some - events carry the name of the subject node, and node names are generally - unavailable in an `fs.Inode`. -- Dirents do not directly contain state for any watches. Instead, they forward - notifications to the underlying `fs.Inode`. - -### [`fs.Inode`][inode] - -- Interacts with inotify through `fs.Watch`es. -- Inodes contain a map of all active `fs.Watch`es on them. -- An `fs.Inotify` instance can have at most one `fs.Watch` per inode. - `fs.Watch`es on an inode are indexed by their `owner`'s `id`. -- All inotify logic is encapsulated in the [`Watches`][inode_watches] struct - in an inode. Logically, `Watches` is the set of inotify watches on the - inode. - -## Reference Model - -The sentry inotify implementation has a complex reference model. An inotify -watch observes a single inode. For efficient lookup, the state for a watch is -stored directly on the target inode. This state needs to be persistent for the -lifetime of watch. Unlike usual filesystem metadata, the watch state has no -"on-disk" representation, so they cannot be reconstructed by the filesystem if -the inode is flushed from memory. This effectively means we need to keep any -inodes with actives watches pinned to memory. - -We can't just hold an extra ref on the inode to pin it to memory because some -filesystems (such as gofer-based filesystems) don't have persistent inodes. In -such a filesystem, if we just pin the inode, nothing prevents the enclosing -dirent from being GCed. Once the dirent is GCed, the pinned inode is -unreachable -- these filesystems generate a new inode by re-reading the node -state on the next walk. Incidentally, hardlinks also don't work on these -filesystems for this reason. - -To prevent the above scenario, when a new watch is added on an inode, we *pin* -the dirent we used to reach the inode. Note that due to hardlinks, this dirent -may not be the only dirent pointing to the inode. Attempting to set an inotify -watch via multiple hardlinks to the same file results in the same watch being -returned for both links. However, for each new dirent we use to reach the same -inode, we add a new pin. We need a new pin for each new dirent used to reach the -inode because we have no guarantees about the deletion order of the different -links to the inode. - -## Lock Ordering - -There are 4 locks related to the inotify implementation: - -- `Inotify.mu`: the inotify instance lock. -- `Inotify.evMu`: the inotify event queue lock. -- `Watch.mu`: the watch lock, used to protect pins. -- `fs.Watches.mu`: the inode watch set mu, used to protect the collection of - watches on the inode. - -The correct lock ordering for inotify code is: - -`Inotify.mu` -> `fs.Watches.mu` -> `Watch.mu` -> `Inotify.evMu`. - -We need a distinct lock for the event queue because by the time a goroutine -attempts to queue a new event, it is already holding `fs.Watches.mu`. If we used -`Inotify.mu` to also protect the event queue, this would violate the above lock -ordering. - -[dirent]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/dirent.go -[event]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/inotify_event.go -[fd_table]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/kernel/fd_table.go -[inode]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/inode.go -[inode_watches]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/inode_inotify.go -[inotify]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/inotify.go -[syscall_dir]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/syscalls/linux/ -[watch]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/fs/inotify_watch.go diff --git a/pkg/sentry/fs/gofer/BUILD b/pkg/sentry/fs/gofer/BUILD deleted file mode 100644 index 6b993928c..000000000 --- a/pkg/sentry/fs/gofer/BUILD +++ /dev/null @@ -1,65 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_stateify:defs.bzl", "go_library", "go_test") - -go_library( - name = "gofer", - srcs = [ - "attr.go", - "cache_policy.go", - "context_file.go", - "device.go", - "file.go", - "file_state.go", - "fs.go", - "handles.go", - "inode.go", - "inode_state.go", - "path.go", - "session.go", - "session_state.go", - "socket.go", - "util.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/gofer", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/fd", - "//pkg/log", - "//pkg/metric", - "//pkg/p9", - "//pkg/refs", - "//pkg/secio", - "//pkg/sentry/context", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fdpipe", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fs/host", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sentry/memmap", - "//pkg/sentry/safemem", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usermem", - "//pkg/syserr", - "//pkg/syserror", - "//pkg/unet", - "//pkg/waiter", - ], -) - -go_test( - name = "gofer_test", - size = "small", - srcs = ["gofer_test.go"], - embed = [":gofer"], - deps = [ - "//pkg/p9", - "//pkg/p9/p9test", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", - "//pkg/sentry/fs", - ], -) diff --git a/pkg/sentry/fs/gofer/gofer_state_autogen.go b/pkg/sentry/fs/gofer/gofer_state_autogen.go new file mode 100755 index 000000000..e05895fab --- /dev/null +++ b/pkg/sentry/fs/gofer/gofer_state_autogen.go @@ -0,0 +1,113 @@ +// automatically generated by stateify. + +package gofer + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *fileOperations) beforeSave() {} +func (x *fileOperations) save(m state.Map) { + x.beforeSave() + m.Save("inodeOperations", &x.inodeOperations) + m.Save("dirCursor", &x.dirCursor) + m.Save("flags", &x.flags) +} + +func (x *fileOperations) load(m state.Map) { + m.LoadWait("inodeOperations", &x.inodeOperations) + m.Load("dirCursor", &x.dirCursor) + m.LoadWait("flags", &x.flags) + m.AfterLoad(x.afterLoad) +} + +func (x *filesystem) beforeSave() {} +func (x *filesystem) save(m state.Map) { + x.beforeSave() +} + +func (x *filesystem) afterLoad() {} +func (x *filesystem) load(m state.Map) { +} + +func (x *inodeOperations) beforeSave() {} +func (x *inodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("fileState", &x.fileState) + m.Save("cachingInodeOps", &x.cachingInodeOps) +} + +func (x *inodeOperations) afterLoad() {} +func (x *inodeOperations) load(m state.Map) { + m.LoadWait("fileState", &x.fileState) + m.Load("cachingInodeOps", &x.cachingInodeOps) +} + +func (x *inodeFileState) save(m state.Map) { + x.beforeSave() + var loading struct{} = x.saveLoading() + m.SaveValue("loading", loading) + m.Save("s", &x.s) + m.Save("sattr", &x.sattr) + m.Save("savedUAttr", &x.savedUAttr) + m.Save("hostMappable", &x.hostMappable) +} + +func (x *inodeFileState) load(m state.Map) { + m.LoadWait("s", &x.s) + m.LoadWait("sattr", &x.sattr) + m.Load("savedUAttr", &x.savedUAttr) + m.Load("hostMappable", &x.hostMappable) + m.LoadValue("loading", new(struct{}), func(y interface{}) { x.loadLoading(y.(struct{})) }) + m.AfterLoad(x.afterLoad) +} + +func (x *endpointMaps) beforeSave() {} +func (x *endpointMaps) save(m state.Map) { + x.beforeSave() + m.Save("direntMap", &x.direntMap) + m.Save("pathMap", &x.pathMap) +} + +func (x *endpointMaps) afterLoad() {} +func (x *endpointMaps) load(m state.Map) { + m.Load("direntMap", &x.direntMap) + m.Load("pathMap", &x.pathMap) +} + +func (x *session) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("msize", &x.msize) + m.Save("version", &x.version) + m.Save("cachePolicy", &x.cachePolicy) + m.Save("aname", &x.aname) + m.Save("superBlockFlags", &x.superBlockFlags) + m.Save("connID", &x.connID) + m.Save("inodeMappings", &x.inodeMappings) + m.Save("mounter", &x.mounter) + m.Save("endpoints", &x.endpoints) +} + +func (x *session) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.LoadWait("msize", &x.msize) + m.LoadWait("version", &x.version) + m.LoadWait("cachePolicy", &x.cachePolicy) + m.LoadWait("aname", &x.aname) + m.LoadWait("superBlockFlags", &x.superBlockFlags) + m.LoadWait("connID", &x.connID) + m.LoadWait("inodeMappings", &x.inodeMappings) + m.LoadWait("mounter", &x.mounter) + m.LoadWait("endpoints", &x.endpoints) + m.AfterLoad(x.afterLoad) +} + +func init() { + state.Register("gofer.fileOperations", (*fileOperations)(nil), state.Fns{Save: (*fileOperations).save, Load: (*fileOperations).load}) + state.Register("gofer.filesystem", (*filesystem)(nil), state.Fns{Save: (*filesystem).save, Load: (*filesystem).load}) + state.Register("gofer.inodeOperations", (*inodeOperations)(nil), state.Fns{Save: (*inodeOperations).save, Load: (*inodeOperations).load}) + state.Register("gofer.inodeFileState", (*inodeFileState)(nil), state.Fns{Save: (*inodeFileState).save, Load: (*inodeFileState).load}) + state.Register("gofer.endpointMaps", (*endpointMaps)(nil), state.Fns{Save: (*endpointMaps).save, Load: (*endpointMaps).load}) + state.Register("gofer.session", (*session)(nil), state.Fns{Save: (*session).save, Load: (*session).load}) +} diff --git a/pkg/sentry/fs/gofer/gofer_test.go b/pkg/sentry/fs/gofer/gofer_test.go deleted file mode 100644 index 7fc3c32ae..000000000 --- a/pkg/sentry/fs/gofer/gofer_test.go +++ /dev/null @@ -1,310 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package gofer - -import ( - "fmt" - "syscall" - "testing" - "time" - - "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/p9/p9test" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" -) - -// rootTest runs a test with a p9 mock and an fs.InodeOperations created from -// the attached root directory. The root file will be closed and client -// disconnected, but additional files must be closed manually. -func rootTest(t *testing.T, name string, cp cachePolicy, fn func(context.Context, *p9test.Harness, *p9test.Mock, *fs.Inode)) { - t.Run(name, func(t *testing.T) { - h, c := p9test.NewHarness(t) - defer h.Finish() - - // Create a new root. Note that we pass an empty, but non-nil - // map here. This allows tests to extend the root children - // dynamically. - root := h.NewDirectory(map[string]p9test.Generator{})(nil) - - // Return this as the root. - h.Attacher.EXPECT().Attach().Return(root, nil).Times(1) - - // ... and open via the client. - rootFile, err := c.Attach("/") - if err != nil { - t.Fatalf("unable to attach: %v", err) - } - defer rootFile.Close() - - // Wrap an a session. - s := &session{ - mounter: fs.RootOwner, - cachePolicy: cp, - client: c, - } - - // ... and an INode, with only the mode being explicitly valid for now. - ctx := contexttest.Context(t) - sattr, rootInodeOperations := newInodeOperations(ctx, s, contextFile{ - file: rootFile, - }, root.QID, p9.AttrMaskAll(), root.Attr, false /* socket */) - m := fs.NewMountSource(ctx, s, &filesystem{}, fs.MountSourceFlags{}) - rootInode := fs.NewInode(ctx, rootInodeOperations, m, sattr) - - // Ensure that the cache is fully invalidated, so that any - // close actions actually take place before the full harness is - // torn down. - defer func() { - m.FlushDirentRefs() - - // Wait for all resources to be released, otherwise the - // operations may fail after we close the rootFile. - fs.AsyncBarrier() - }() - - // Execute the test. - fn(ctx, h, root, rootInode) - }) -} - -func TestLookup(t *testing.T) { - type lookupTest struct { - // Name of the test. - name string - - // Expected return value. - want error - } - - tests := []lookupTest{ - { - name: "mock Walk passes (function succeeds)", - want: nil, - }, - { - name: "mock Walk fails (function fails)", - want: syscall.ENOENT, - }, - } - - const file = "file" // The walked target file. - - for _, test := range tests { - rootTest(t, test.name, cacheNone, func(ctx context.Context, h *p9test.Harness, rootFile *p9test.Mock, rootInode *fs.Inode) { - // Setup the appropriate result. - rootFile.WalkCallback = func() error { - return test.want - } - if test.want == nil { - // Set the contents of the root. We expect a - // normal file generator for ppp above. This is - // overriden by setting WalkErr in the mock. - rootFile.AddChild(file, h.NewFile()) - } - - // Call function. - dirent, err := rootInode.Lookup(ctx, file) - - // Unwrap the InodeOperations. - var newInodeOperations fs.InodeOperations - if dirent != nil { - if dirent.IsNegative() { - err = syscall.ENOENT - } else { - newInodeOperations = dirent.Inode.InodeOperations - } - } - - // Check return values. - if err != test.want { - t.Errorf("Lookup got err %v, want %v", err, test.want) - } - if err == nil && newInodeOperations == nil { - t.Errorf("Lookup got non-nil err and non-nil node, wanted at least one non-nil") - } - }) - } -} - -func TestRevalidation(t *testing.T) { - type revalidationTest struct { - cachePolicy cachePolicy - - // Whether dirent should be reloaded before any modifications. - preModificationWantReload bool - - // Whether dirent should be reloaded after updating an unstable - // attribute on the remote fs. - postModificationWantReload bool - - // Whether dirent unstable attributes should be updated after - // updating an attribute on the remote fs. - postModificationWantUpdatedAttrs bool - - // Whether dirent should be reloaded after the remote has - // removed the file. - postRemovalWantReload bool - } - - tests := []revalidationTest{ - { - // Policy cacheNone causes Revalidate to always return - // true. - cachePolicy: cacheNone, - preModificationWantReload: true, - postModificationWantReload: true, - postModificationWantUpdatedAttrs: true, - postRemovalWantReload: true, - }, - { - // Policy cacheAll causes Revalidate to always return - // false. - cachePolicy: cacheAll, - preModificationWantReload: false, - postModificationWantReload: false, - postModificationWantUpdatedAttrs: false, - postRemovalWantReload: false, - }, - { - // Policy cacheAllWritethrough causes Revalidate to - // always return false. - cachePolicy: cacheAllWritethrough, - preModificationWantReload: false, - postModificationWantReload: false, - postModificationWantUpdatedAttrs: false, - postRemovalWantReload: false, - }, - { - // Policy cacheRemoteRevalidating causes Revalidate to - // return update cached unstable attrs, and returns - // true only when the remote inode itself has been - // removed or replaced. - cachePolicy: cacheRemoteRevalidating, - preModificationWantReload: false, - postModificationWantReload: false, - postModificationWantUpdatedAttrs: true, - postRemovalWantReload: true, - }, - } - - const file = "file" // The file walked below. - - for _, test := range tests { - name := fmt.Sprintf("cachepolicy=%s", test.cachePolicy) - rootTest(t, name, test.cachePolicy, func(ctx context.Context, h *p9test.Harness, rootFile *p9test.Mock, rootInode *fs.Inode) { - // Wrap in a dirent object. - rootDir := fs.NewDirent(ctx, rootInode, "root") - - // Create a mock file a child of the root. We save when - // this is generated, so that when the time changed, we - // can update the original entry. - var origMocks []*p9test.Mock - rootFile.AddChild(file, func(parent *p9test.Mock) *p9test.Mock { - // Regular a regular file that has a consistent - // path number. This might be used by - // validation so we don't change it. - m := h.NewMock(parent, 0, p9.Attr{ - Mode: p9.ModeRegular, - }) - origMocks = append(origMocks, m) - return m - }) - - // Do the walk. - dirent, err := rootDir.Walk(ctx, rootDir, file) - if err != nil { - t.Fatalf("Lookup failed: %v", err) - } - - // We must release the dirent, of the test will fail - // with a reference leak. This is tracked by p9test. - defer dirent.DecRef() - - // Walk again. Depending on the cache policy, we may - // get a new dirent. - newDirent, err := rootDir.Walk(ctx, rootDir, file) - if err != nil { - t.Fatalf("Lookup failed: %v", err) - } - if test.preModificationWantReload && dirent == newDirent { - t.Errorf("Lookup with cachePolicy=%s got old dirent %+v, wanted a new dirent", test.cachePolicy, dirent) - } - if !test.preModificationWantReload && dirent != newDirent { - t.Errorf("Lookup with cachePolicy=%s got new dirent %+v, wanted old dirent %+v", test.cachePolicy, newDirent, dirent) - } - newDirent.DecRef() // See above. - - // Modify the underlying mocked file's modification - // time for the next walk that occurs. - nowSeconds := time.Now().Unix() - rootFile.AddChild(file, func(parent *p9test.Mock) *p9test.Mock { - // Ensure that the path is the same as above, - // but we change only the modification time of - // the file. - return h.NewMock(parent, 0, p9.Attr{ - Mode: p9.ModeRegular, - MTimeSeconds: uint64(nowSeconds), - }) - }) - - // We also modify the original time, so that GetAttr - // behaves as expected for the caching case. - for _, m := range origMocks { - m.Attr.MTimeSeconds = uint64(nowSeconds) - } - - // Walk again. Depending on the cache policy, we may - // get a new dirent. - newDirent, err = rootDir.Walk(ctx, rootDir, file) - if err != nil { - t.Fatalf("Lookup failed: %v", err) - } - if test.postModificationWantReload && dirent == newDirent { - t.Errorf("Lookup with cachePolicy=%s got old dirent, wanted a new dirent", test.cachePolicy) - } - if !test.postModificationWantReload && dirent != newDirent { - t.Errorf("Lookup with cachePolicy=%s got new dirent, wanted old dirent", test.cachePolicy) - } - uattrs, err := newDirent.Inode.UnstableAttr(ctx) - if err != nil { - t.Fatalf("Error getting unstable attrs: %v", err) - } - gotModTimeSeconds := uattrs.ModificationTime.Seconds() - if test.postModificationWantUpdatedAttrs && gotModTimeSeconds != nowSeconds { - t.Fatalf("Lookup with cachePolicy=%s got new modification time %v, wanted %v", test.cachePolicy, gotModTimeSeconds, nowSeconds) - } - newDirent.DecRef() // See above. - - // Remove the file from the remote fs, subsequent walks - // should now fail to find anything. - rootFile.RemoveChild(file) - - // Walk again. Depending on the cache policy, we may - // get ENOENT. - newDirent, err = rootDir.Walk(ctx, rootDir, file) - if test.postRemovalWantReload && err == nil { - t.Errorf("Lookup with cachePolicy=%s got nil error, wanted ENOENT", test.cachePolicy) - } - if !test.postRemovalWantReload && (err != nil || dirent != newDirent) { - t.Errorf("Lookup with cachePolicy=%s got new dirent and error %v, wanted old dirent and nil error", test.cachePolicy, err) - } - if err == nil { - newDirent.DecRef() // See above. - } - }) - } -} diff --git a/pkg/sentry/fs/host/BUILD b/pkg/sentry/fs/host/BUILD deleted file mode 100644 index b1080fb1a..000000000 --- a/pkg/sentry/fs/host/BUILD +++ /dev/null @@ -1,83 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_stateify:defs.bzl", "go_library", "go_test") - -go_library( - name = "host", - srcs = [ - "control.go", - "descriptor.go", - "descriptor_state.go", - "device.go", - "file.go", - "fs.go", - "inode.go", - "inode_state.go", - "ioctl_unsafe.go", - "socket.go", - "socket_iovec.go", - "socket_state.go", - "socket_unsafe.go", - "tty.go", - "util.go", - "util_unsafe.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/host", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/fd", - "//pkg/fdnotifier", - "//pkg/log", - "//pkg/refs", - "//pkg/secio", - "//pkg/sentry/arch", - "//pkg/sentry/context", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sentry/memmap", - "//pkg/sentry/safemem", - "//pkg/sentry/socket/control", - "//pkg/sentry/socket/unix", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/unimpl", - "//pkg/sentry/uniqueid", - "//pkg/sentry/usermem", - "//pkg/syserr", - "//pkg/syserror", - "//pkg/tcpip", - "//pkg/unet", - "//pkg/waiter", - ], -) - -go_test( - name = "host_test", - size = "small", - srcs = [ - "descriptor_test.go", - "fs_test.go", - "inode_test.go", - "socket_test.go", - "wait_test.go", - ], - embed = [":host"], - deps = [ - "//pkg/fd", - "//pkg/fdnotifier", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", - "//pkg/sentry/fs", - "//pkg/sentry/kernel/time", - "//pkg/sentry/socket", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usermem", - "//pkg/syserr", - "//pkg/tcpip", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/fs/host/descriptor_test.go b/pkg/sentry/fs/host/descriptor_test.go deleted file mode 100644 index 4205981f5..000000000 --- a/pkg/sentry/fs/host/descriptor_test.go +++ /dev/null @@ -1,78 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package host - -import ( - "io/ioutil" - "path/filepath" - "syscall" - "testing" - - "gvisor.dev/gvisor/pkg/fdnotifier" - "gvisor.dev/gvisor/pkg/waiter" -) - -func TestDescriptorRelease(t *testing.T) { - for _, tc := range []struct { - name string - saveable bool - wouldBlock bool - }{ - {name: "all false"}, - {name: "saveable", saveable: true}, - {name: "wouldBlock", wouldBlock: true}, - } { - t.Run(tc.name, func(t *testing.T) { - dir, err := ioutil.TempDir("", "descriptor_test") - if err != nil { - t.Fatal("ioutil.TempDir() failed:", err) - } - - fd, err := syscall.Open(filepath.Join(dir, "file"), syscall.O_RDWR|syscall.O_CREAT, 0666) - if err != nil { - t.Fatal("failed to open temp file:", err) - } - - // FD ownership is transferred to the descritor. - queue := &waiter.Queue{} - d, err := newDescriptor(fd, false /* donated*/, tc.saveable, tc.wouldBlock, queue) - if err != nil { - syscall.Close(fd) - t.Fatalf("newDescriptor(%d, %t, false, %t, queue) failed, err: %v", fd, tc.saveable, tc.wouldBlock, err) - } - if tc.saveable { - if d.origFD < 0 { - t.Errorf("saveable descriptor must preserve origFD, desc: %+v", d) - } - } - if tc.wouldBlock { - if !fdnotifier.HasFD(int32(d.value)) { - t.Errorf("FD not registered with notifier, desc: %+v", d) - } - } - - oldVal := d.value - d.Release() - if d.value != -1 { - t.Errorf("d.value want: -1, got: %d", d.value) - } - if tc.wouldBlock { - if fdnotifier.HasFD(int32(oldVal)) { - t.Errorf("FD not unregistered with notifier, desc: %+v", d) - } - } - }) - } -} diff --git a/pkg/sentry/fs/host/fs_test.go b/pkg/sentry/fs/host/fs_test.go deleted file mode 100644 index c6852ee30..000000000 --- a/pkg/sentry/fs/host/fs_test.go +++ /dev/null @@ -1,380 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package host - -import ( - "fmt" - "io/ioutil" - "os" - "path" - "reflect" - "sort" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" -) - -// newTestMountNamespace creates a MountNamespace with a ramfs root. -// It returns the host folder created, which should be removed when done. -func newTestMountNamespace(t *testing.T) (*fs.MountNamespace, string, error) { - p, err := ioutil.TempDir("", "root") - if err != nil { - return nil, "", err - } - - fd, err := open(nil, p) - if err != nil { - os.RemoveAll(p) - return nil, "", err - } - ctx := contexttest.Context(t) - root, err := newInode(ctx, newMountSource(ctx, p, fs.RootOwner, &Filesystem{}, fs.MountSourceFlags{}, false), fd, false, false) - if err != nil { - os.RemoveAll(p) - return nil, "", err - } - mm, err := fs.NewMountNamespace(ctx, root) - if err != nil { - os.RemoveAll(p) - return nil, "", err - } - return mm, p, nil -} - -// createTestDirs populates the root with some test files and directories. -// /a/a1.txt -// /a/a2.txt -// /b/b1.txt -// /b/c/c1.txt -// /symlinks/normal.txt -// /symlinks/to_normal.txt -> /symlinks/normal.txt -// /symlinks/recursive -> /symlinks -func createTestDirs(ctx context.Context, t *testing.T, m *fs.MountNamespace) error { - r := m.Root() - defer r.DecRef() - - if err := r.CreateDirectory(ctx, r, "a", fs.FilePermsFromMode(0777)); err != nil { - return err - } - - a, err := r.Walk(ctx, r, "a") - if err != nil { - return err - } - defer a.DecRef() - - a1, err := a.Create(ctx, r, "a1.txt", fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0666)) - if err != nil { - return err - } - a1.DecRef() - - a2, err := a.Create(ctx, r, "a2.txt", fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0666)) - if err != nil { - return err - } - a2.DecRef() - - if err := r.CreateDirectory(ctx, r, "b", fs.FilePermsFromMode(0777)); err != nil { - return err - } - - b, err := r.Walk(ctx, r, "b") - if err != nil { - return err - } - defer b.DecRef() - - b1, err := b.Create(ctx, r, "b1.txt", fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0666)) - if err != nil { - return err - } - b1.DecRef() - - if err := b.CreateDirectory(ctx, r, "c", fs.FilePermsFromMode(0777)); err != nil { - return err - } - - c, err := b.Walk(ctx, r, "c") - if err != nil { - return err - } - defer c.DecRef() - - c1, err := c.Create(ctx, r, "c1.txt", fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0666)) - if err != nil { - return err - } - c1.DecRef() - - if err := r.CreateDirectory(ctx, r, "symlinks", fs.FilePermsFromMode(0777)); err != nil { - return err - } - - symlinks, err := r.Walk(ctx, r, "symlinks") - if err != nil { - return err - } - defer symlinks.DecRef() - - normal, err := symlinks.Create(ctx, r, "normal.txt", fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0666)) - if err != nil { - return err - } - normal.DecRef() - - if err := symlinks.CreateLink(ctx, r, "/symlinks/normal.txt", "to_normal.txt"); err != nil { - return err - } - - return symlinks.CreateLink(ctx, r, "/symlinks", "recursive") -} - -// allPaths returns a slice of all paths of entries visible in the rootfs. -func allPaths(ctx context.Context, t *testing.T, m *fs.MountNamespace, base string) ([]string, error) { - var paths []string - root := m.Root() - defer root.DecRef() - - maxTraversals := uint(1) - d, err := m.FindLink(ctx, root, nil, base, &maxTraversals) - if err != nil { - t.Logf("FindLink failed for %q", base) - return paths, err - } - defer d.DecRef() - - if fs.IsDir(d.Inode.StableAttr) { - dir, err := d.Inode.GetFile(ctx, d, fs.FileFlags{Read: true}) - if err != nil { - return nil, fmt.Errorf("failed to open directory %q: %v", base, err) - } - iter, ok := dir.FileOperations.(fs.DirIterator) - if !ok { - return nil, fmt.Errorf("cannot directly iterate on host directory %q", base) - } - dirCtx := &fs.DirCtx{ - Serializer: noopDentrySerializer{}, - } - if _, err := fs.DirentReaddir(ctx, d, iter, root, dirCtx, 0); err != nil { - return nil, err - } - for name := range dirCtx.DentAttrs() { - if name == "." || name == ".." { - continue - } - - fullName := path.Join(base, name) - paths = append(paths, fullName) - - // Recurse. - subpaths, err := allPaths(ctx, t, m, fullName) - if err != nil { - return paths, err - } - paths = append(paths, subpaths...) - } - } - - return paths, nil -} - -type noopDentrySerializer struct{} - -func (noopDentrySerializer) CopyOut(string, fs.DentAttr) error { - return nil -} -func (noopDentrySerializer) Written() int { - return 4096 -} - -// pathsEqual returns true if the two string slices contain the same entries. -func pathsEqual(got, want []string) bool { - sort.Strings(got) - sort.Strings(want) - - if len(got) != len(want) { - return false - } - - for i := range got { - if got[i] != want[i] { - return false - } - } - - return true -} - -func TestWhitelist(t *testing.T) { - for _, test := range []struct { - // description of the test. - desc string - // paths are the paths to whitelist - paths []string - // want are all of the directory entries that should be - // visible (nothing beyond this set should be visible). - want []string - }{ - { - desc: "root", - paths: []string{"/"}, - want: []string{"/a", "/a/a1.txt", "/a/a2.txt", "/b", "/b/b1.txt", "/b/c", "/b/c/c1.txt", "/symlinks", "/symlinks/normal.txt", "/symlinks/to_normal.txt", "/symlinks/recursive"}, - }, - { - desc: "top-level directories", - paths: []string{"/a", "/b"}, - want: []string{"/a", "/a/a1.txt", "/a/a2.txt", "/b", "/b/b1.txt", "/b/c", "/b/c/c1.txt"}, - }, - { - desc: "nested directories (1/2)", - paths: []string{"/b", "/b/c"}, - want: []string{"/b", "/b/b1.txt", "/b/c", "/b/c/c1.txt"}, - }, - { - desc: "nested directories (2/2)", - paths: []string{"/b/c", "/b"}, - want: []string{"/b", "/b/b1.txt", "/b/c", "/b/c/c1.txt"}, - }, - { - desc: "single file", - paths: []string{"/b/c/c1.txt"}, - want: []string{"/b", "/b/c", "/b/c/c1.txt"}, - }, - { - desc: "single file and directory", - paths: []string{"/a/a1.txt", "/b/c"}, - want: []string{"/a", "/a/a1.txt", "/b", "/b/c", "/b/c/c1.txt"}, - }, - { - desc: "symlink", - paths: []string{"/symlinks/to_normal.txt"}, - want: []string{"/symlinks", "/symlinks/normal.txt", "/symlinks/to_normal.txt"}, - }, - { - desc: "recursive symlink", - paths: []string{"/symlinks/recursive/normal.txt"}, - want: []string{"/symlinks", "/symlinks/normal.txt", "/symlinks/recursive"}, - }, - } { - t.Run(test.desc, func(t *testing.T) { - m, p, err := newTestMountNamespace(t) - if err != nil { - t.Errorf("Failed to create MountNamespace: %v", err) - } - defer os.RemoveAll(p) - - ctx := withRoot(contexttest.RootContext(t), m.Root()) - if err := createTestDirs(ctx, t, m); err != nil { - t.Errorf("Failed to create test dirs: %v", err) - } - - if err := installWhitelist(ctx, m, test.paths); err != nil { - t.Errorf("installWhitelist(%v) err got %v want nil", test.paths, err) - } - - got, err := allPaths(ctx, t, m, "/") - if err != nil { - t.Fatalf("Failed to lookup paths (whitelisted: %v): %v", test.paths, err) - } - - if !pathsEqual(got, test.want) { - t.Errorf("For paths %v got %v want %v", test.paths, got, test.want) - } - }) - } -} - -func TestRootPath(t *testing.T) { - // Create a temp dir, which will be the root of our mounted fs. - rootPath, err := ioutil.TempDir(os.TempDir(), "root") - if err != nil { - t.Fatalf("TempDir failed: %v", err) - } - defer os.RemoveAll(rootPath) - - // Create two files inside the new root, one which will be whitelisted - // and one not. - whitelisted, err := ioutil.TempFile(rootPath, "white") - if err != nil { - t.Fatalf("TempFile failed: %v", err) - } - if _, err := ioutil.TempFile(rootPath, "black"); err != nil { - t.Fatalf("TempFile failed: %v", err) - } - - // Create a mount with a root path and single whitelisted file. - hostFS := &Filesystem{} - ctx := contexttest.Context(t) - data := fmt.Sprintf("%s=%s,%s=%s", rootPathKey, rootPath, whitelistKey, whitelisted.Name()) - inode, err := hostFS.Mount(ctx, "", fs.MountSourceFlags{}, data, nil) - if err != nil { - t.Fatalf("Mount failed: %v", err) - } - mm, err := fs.NewMountNamespace(ctx, inode) - if err != nil { - t.Fatalf("NewMountNamespace failed: %v", err) - } - if err := hostFS.InstallWhitelist(ctx, mm); err != nil { - t.Fatalf("InstallWhitelist failed: %v", err) - } - - // Get the contents of the root directory. - rootDir := mm.Root() - rctx := withRoot(ctx, rootDir) - f, err := rootDir.Inode.GetFile(rctx, rootDir, fs.FileFlags{}) - if err != nil { - t.Fatalf("GetFile failed: %v", err) - } - c := &fs.CollectEntriesSerializer{} - if err := f.Readdir(rctx, c); err != nil { - t.Fatalf("Readdir failed: %v", err) - } - - // We should have only our whitelisted file, plus the dots. - want := []string{path.Base(whitelisted.Name()), ".", ".."} - got := c.Order - sort.Strings(want) - sort.Strings(got) - if !reflect.DeepEqual(got, want) { - t.Errorf("Readdir got %v, wanted %v", got, want) - } -} - -type rootContext struct { - context.Context - root *fs.Dirent -} - -// withRoot returns a copy of ctx with the given root. -func withRoot(ctx context.Context, root *fs.Dirent) context.Context { - return &rootContext{ - Context: ctx, - root: root, - } -} - -// Value implements Context.Value. -func (rc rootContext) Value(key interface{}) interface{} { - switch key { - case fs.CtxRoot: - rc.root.IncRef() - return rc.root - default: - return rc.Context.Value(key) - } -} diff --git a/pkg/sentry/fs/host/host_state_autogen.go b/pkg/sentry/fs/host/host_state_autogen.go new file mode 100755 index 000000000..f0e1c4b88 --- /dev/null +++ b/pkg/sentry/fs/host/host_state_autogen.go @@ -0,0 +1,138 @@ +// automatically generated by stateify. + +package host + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *descriptor) save(m state.Map) { + x.beforeSave() + m.Save("donated", &x.donated) + m.Save("origFD", &x.origFD) + m.Save("wouldBlock", &x.wouldBlock) +} + +func (x *descriptor) load(m state.Map) { + m.Load("donated", &x.donated) + m.Load("origFD", &x.origFD) + m.Load("wouldBlock", &x.wouldBlock) + m.AfterLoad(x.afterLoad) +} + +func (x *fileOperations) beforeSave() {} +func (x *fileOperations) save(m state.Map) { + x.beforeSave() + m.Save("iops", &x.iops) + m.Save("dirCursor", &x.dirCursor) +} + +func (x *fileOperations) afterLoad() {} +func (x *fileOperations) load(m state.Map) { + m.LoadWait("iops", &x.iops) + m.Load("dirCursor", &x.dirCursor) +} + +func (x *Filesystem) beforeSave() {} +func (x *Filesystem) save(m state.Map) { + x.beforeSave() + m.Save("paths", &x.paths) +} + +func (x *Filesystem) afterLoad() {} +func (x *Filesystem) load(m state.Map) { + m.Load("paths", &x.paths) +} + +func (x *superOperations) beforeSave() {} +func (x *superOperations) save(m state.Map) { + x.beforeSave() + m.Save("SimpleMountSourceOperations", &x.SimpleMountSourceOperations) + m.Save("root", &x.root) + m.Save("inodeMappings", &x.inodeMappings) + m.Save("mounter", &x.mounter) + m.Save("dontTranslateOwnership", &x.dontTranslateOwnership) +} + +func (x *superOperations) afterLoad() {} +func (x *superOperations) load(m state.Map) { + m.Load("SimpleMountSourceOperations", &x.SimpleMountSourceOperations) + m.Load("root", &x.root) + m.Load("inodeMappings", &x.inodeMappings) + m.Load("mounter", &x.mounter) + m.Load("dontTranslateOwnership", &x.dontTranslateOwnership) +} + +func (x *inodeOperations) beforeSave() {} +func (x *inodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("fileState", &x.fileState) + m.Save("cachingInodeOps", &x.cachingInodeOps) +} + +func (x *inodeOperations) afterLoad() {} +func (x *inodeOperations) load(m state.Map) { + m.LoadWait("fileState", &x.fileState) + m.Load("cachingInodeOps", &x.cachingInodeOps) +} + +func (x *inodeFileState) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.queue) { m.Failf("queue is %v, expected zero", x.queue) } + m.Save("mops", &x.mops) + m.Save("descriptor", &x.descriptor) + m.Save("sattr", &x.sattr) + m.Save("savedUAttr", &x.savedUAttr) +} + +func (x *inodeFileState) load(m state.Map) { + m.LoadWait("mops", &x.mops) + m.LoadWait("descriptor", &x.descriptor) + m.LoadWait("sattr", &x.sattr) + m.Load("savedUAttr", &x.savedUAttr) + m.AfterLoad(x.afterLoad) +} + +func (x *ConnectedEndpoint) save(m state.Map) { + x.beforeSave() + m.Save("ref", &x.ref) + m.Save("queue", &x.queue) + m.Save("path", &x.path) + m.Save("srfd", &x.srfd) + m.Save("stype", &x.stype) +} + +func (x *ConnectedEndpoint) load(m state.Map) { + m.Load("ref", &x.ref) + m.Load("queue", &x.queue) + m.Load("path", &x.path) + m.LoadWait("srfd", &x.srfd) + m.Load("stype", &x.stype) + m.AfterLoad(x.afterLoad) +} + +func (x *TTYFileOperations) beforeSave() {} +func (x *TTYFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("fileOperations", &x.fileOperations) + m.Save("session", &x.session) + m.Save("fgProcessGroup", &x.fgProcessGroup) +} + +func (x *TTYFileOperations) afterLoad() {} +func (x *TTYFileOperations) load(m state.Map) { + m.Load("fileOperations", &x.fileOperations) + m.Load("session", &x.session) + m.Load("fgProcessGroup", &x.fgProcessGroup) +} + +func init() { + state.Register("host.descriptor", (*descriptor)(nil), state.Fns{Save: (*descriptor).save, Load: (*descriptor).load}) + state.Register("host.fileOperations", (*fileOperations)(nil), state.Fns{Save: (*fileOperations).save, Load: (*fileOperations).load}) + state.Register("host.Filesystem", (*Filesystem)(nil), state.Fns{Save: (*Filesystem).save, Load: (*Filesystem).load}) + state.Register("host.superOperations", (*superOperations)(nil), state.Fns{Save: (*superOperations).save, Load: (*superOperations).load}) + state.Register("host.inodeOperations", (*inodeOperations)(nil), state.Fns{Save: (*inodeOperations).save, Load: (*inodeOperations).load}) + state.Register("host.inodeFileState", (*inodeFileState)(nil), state.Fns{Save: (*inodeFileState).save, Load: (*inodeFileState).load}) + state.Register("host.ConnectedEndpoint", (*ConnectedEndpoint)(nil), state.Fns{Save: (*ConnectedEndpoint).save, Load: (*ConnectedEndpoint).load}) + state.Register("host.TTYFileOperations", (*TTYFileOperations)(nil), state.Fns{Save: (*TTYFileOperations).save, Load: (*TTYFileOperations).load}) +} diff --git a/pkg/sentry/fs/host/inode_test.go b/pkg/sentry/fs/host/inode_test.go deleted file mode 100644 index 2d959f10d..000000000 --- a/pkg/sentry/fs/host/inode_test.go +++ /dev/null @@ -1,112 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package host - -import ( - "io/ioutil" - "os" - "path" - "syscall" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" -) - -// TestMultipleReaddir verifies that multiple Readdir calls return the same -// thing if they use different dir contexts. -func TestMultipleReaddir(t *testing.T) { - p, err := ioutil.TempDir("", "readdir") - if err != nil { - t.Fatalf("Failed to create test dir: %v", err) - } - defer os.RemoveAll(p) - - f, err := os.Create(path.Join(p, "a.txt")) - if err != nil { - t.Fatalf("Failed to create a.txt: %v", err) - } - f.Close() - - f, err = os.Create(path.Join(p, "b.txt")) - if err != nil { - t.Fatalf("Failed to create b.txt: %v", err) - } - f.Close() - - fd, err := open(nil, p) - if err != nil { - t.Fatalf("Failed to open %q: %v", p, err) - } - ctx := contexttest.Context(t) - n, err := newInode(ctx, newMountSource(ctx, p, fs.RootOwner, &Filesystem{}, fs.MountSourceFlags{}, false), fd, false, false) - if err != nil { - t.Fatalf("Failed to create inode: %v", err) - } - - dirent := fs.NewDirent(ctx, n, "readdir") - openFile, err := n.GetFile(ctx, dirent, fs.FileFlags{Read: true}) - if err != nil { - t.Fatalf("Failed to get file: %v", err) - } - defer openFile.DecRef() - - c1 := &fs.DirCtx{DirCursor: new(string)} - if _, err := openFile.FileOperations.(*fileOperations).IterateDir(ctx, dirent, c1, 0); err != nil { - t.Fatalf("First Readdir failed: %v", err) - } - - c2 := &fs.DirCtx{DirCursor: new(string)} - if _, err := openFile.FileOperations.(*fileOperations).IterateDir(ctx, dirent, c2, 0); err != nil { - t.Errorf("Second Readdir failed: %v", err) - } - - if _, ok := c1.DentAttrs()["a.txt"]; !ok { - t.Errorf("want a.txt in first Readdir, got %v", c1.DentAttrs()) - } - if _, ok := c1.DentAttrs()["b.txt"]; !ok { - t.Errorf("want b.txt in first Readdir, got %v", c1.DentAttrs()) - } - - if _, ok := c2.DentAttrs()["a.txt"]; !ok { - t.Errorf("want a.txt in second Readdir, got %v", c2.DentAttrs()) - } - if _, ok := c2.DentAttrs()["b.txt"]; !ok { - t.Errorf("want b.txt in second Readdir, got %v", c2.DentAttrs()) - } -} - -// TestCloseFD verifies fds will be closed. -func TestCloseFD(t *testing.T) { - var p [2]int - if err := syscall.Pipe(p[0:]); err != nil { - t.Fatalf("Failed to create pipe %v", err) - } - defer syscall.Close(p[0]) - defer syscall.Close(p[1]) - - // Use the write-end because we will detect if it's closed on the read end. - ctx := contexttest.Context(t) - file, err := NewFile(ctx, p[1], fs.RootOwner) - if err != nil { - t.Fatalf("Failed to create File: %v", err) - } - file.DecRef() - - s := make([]byte, 10) - if c, err := syscall.Read(p[0], s); c != 0 || err != nil { - t.Errorf("want 0, nil (EOF) from read end, got %v, %v", c, err) - } -} diff --git a/pkg/sentry/fs/host/socket_test.go b/pkg/sentry/fs/host/socket_test.go deleted file mode 100644 index 68b38fd1c..000000000 --- a/pkg/sentry/fs/host/socket_test.go +++ /dev/null @@ -1,246 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package host - -import ( - "reflect" - "syscall" - "testing" - - "gvisor.dev/gvisor/pkg/fd" - "gvisor.dev/gvisor/pkg/fdnotifier" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/socket" - "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/syserr" - "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/waiter" -) - -var ( - // Make sure that ConnectedEndpoint implements transport.ConnectedEndpoint. - _ = transport.ConnectedEndpoint(new(ConnectedEndpoint)) - - // Make sure that ConnectedEndpoint implements transport.Receiver. - _ = transport.Receiver(new(ConnectedEndpoint)) -) - -func getFl(fd int) (uint32, error) { - fl, _, err := syscall.RawSyscall(syscall.SYS_FCNTL, uintptr(fd), syscall.F_GETFL, 0) - if err == 0 { - return uint32(fl), nil - } - return 0, err -} - -func TestSocketIsBlocking(t *testing.T) { - // Using socketpair here because it's already connected. - pair, err := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) - if err != nil { - t.Fatalf("host socket creation failed: %v", err) - } - - fl, err := getFl(pair[0]) - if err != nil { - t.Fatalf("getFl: fcntl(%v, GETFL) => %v", pair[0], err) - } - if fl&syscall.O_NONBLOCK == syscall.O_NONBLOCK { - t.Fatalf("Expected socket %v to be blocking", pair[0]) - } - if fl, err = getFl(pair[1]); err != nil { - t.Fatalf("getFl: fcntl(%v, GETFL) => %v", pair[1], err) - } - if fl&syscall.O_NONBLOCK == syscall.O_NONBLOCK { - t.Fatalf("Expected socket %v to be blocking", pair[1]) - } - sock, err := newSocket(contexttest.Context(t), pair[0], false) - if err != nil { - t.Fatalf("newSocket(%v) failed => %v", pair[0], err) - } - defer sock.DecRef() - // Test that the socket now is non-blocking. - if fl, err = getFl(pair[0]); err != nil { - t.Fatalf("getFl: fcntl(%v, GETFL) => %v", pair[0], err) - } - if fl&syscall.O_NONBLOCK != syscall.O_NONBLOCK { - t.Errorf("Expected socket %v to have become non-blocking", pair[0]) - } - if fl, err = getFl(pair[1]); err != nil { - t.Fatalf("getFl: fcntl(%v, GETFL) => %v", pair[1], err) - } - if fl&syscall.O_NONBLOCK == syscall.O_NONBLOCK { - t.Errorf("Did not expect socket %v to become non-blocking", pair[1]) - } -} - -func TestSocketWritev(t *testing.T) { - // Using socketpair here because it's already connected. - pair, err := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) - if err != nil { - t.Fatalf("host socket creation failed: %v", err) - } - socket, err := newSocket(contexttest.Context(t), pair[0], false) - if err != nil { - t.Fatalf("newSocket(%v) => %v", pair[0], err) - } - defer socket.DecRef() - buf := []byte("hello world\n") - n, err := socket.Writev(contexttest.Context(t), usermem.BytesIOSequence(buf)) - if err != nil { - t.Fatalf("socket writev failed: %v", err) - } - - if n != int64(len(buf)) { - t.Fatalf("socket writev wrote incorrect bytes: %d", n) - } -} - -func TestSocketWritevLen0(t *testing.T) { - // Using socketpair here because it's already connected. - pair, err := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) - if err != nil { - t.Fatalf("host socket creation failed: %v", err) - } - socket, err := newSocket(contexttest.Context(t), pair[0], false) - if err != nil { - t.Fatalf("newSocket(%v) => %v", pair[0], err) - } - defer socket.DecRef() - n, err := socket.Writev(contexttest.Context(t), usermem.BytesIOSequence(nil)) - if err != nil { - t.Fatalf("socket writev failed: %v", err) - } - - if n != 0 { - t.Fatalf("socket writev wrote incorrect bytes: %d", n) - } -} - -func TestSocketSendMsgLen0(t *testing.T) { - // Using socketpair here because it's already connected. - pair, err := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) - if err != nil { - t.Fatalf("host socket creation failed: %v", err) - } - sfile, err := newSocket(contexttest.Context(t), pair[0], false) - if err != nil { - t.Fatalf("newSocket(%v) => %v", pair[0], err) - } - defer sfile.DecRef() - - s := sfile.FileOperations.(socket.Socket) - n, terr := s.SendMsg(nil, usermem.BytesIOSequence(nil), []byte{}, 0, false, ktime.Time{}, socket.ControlMessages{}) - if n != 0 { - t.Fatalf("socket sendmsg() failed: %v wrote: %d", terr, n) - } - - if terr != nil { - t.Fatalf("socket sendmsg() failed: %v", terr) - } -} - -func TestListen(t *testing.T) { - pair, err := syscall.Socketpair(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) - if err != nil { - t.Fatalf("syscall.Socket(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) => %v", err) - } - sfile1, err := newSocket(contexttest.Context(t), pair[0], false) - if err != nil { - t.Fatalf("newSocket(%v) => %v", pair[0], err) - } - defer sfile1.DecRef() - socket1 := sfile1.FileOperations.(socket.Socket) - - sfile2, err := newSocket(contexttest.Context(t), pair[1], false) - if err != nil { - t.Fatalf("newSocket(%v) => %v", pair[1], err) - } - defer sfile2.DecRef() - socket2 := sfile2.FileOperations.(socket.Socket) - - // Socketpairs can not be listened to. - if err := socket1.Listen(nil, 64); err != syserr.ErrInvalidEndpointState { - t.Fatalf("socket1.Listen(nil, 64) => %v, want syserr.ErrInvalidEndpointState", err) - } - if err := socket2.Listen(nil, 64); err != syserr.ErrInvalidEndpointState { - t.Fatalf("socket2.Listen(nil, 64) => %v, want syserr.ErrInvalidEndpointState", err) - } - - // Create a Unix socket, do not bind it. - sock, err := syscall.Socket(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) - if err != nil { - t.Fatalf("syscall.Socket(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) => %v", err) - } - sfile3, err := newSocket(contexttest.Context(t), sock, false) - if err != nil { - t.Fatalf("newSocket(%v) => %v", sock, err) - } - defer sfile3.DecRef() - socket3 := sfile3.FileOperations.(socket.Socket) - - // This socket is not bound so we can't listen on it. - if err := socket3.Listen(nil, 64); err != syserr.ErrInvalidEndpointState { - t.Fatalf("socket3.Listen(nil, 64) => %v, want syserr.ErrInvalidEndpointState", err) - } -} - -func TestPasscred(t *testing.T) { - e := ConnectedEndpoint{} - if got, want := e.Passcred(), false; got != want { - t.Errorf("Got %#v.Passcred() = %t, want = %t", e, got, want) - } -} - -func TestGetLocalAddress(t *testing.T) { - e := ConnectedEndpoint{path: "foo"} - want := tcpip.FullAddress{Addr: tcpip.Address("foo")} - if got, err := e.GetLocalAddress(); err != nil || got != want { - t.Errorf("Got %#v.GetLocalAddress() = %#v, %v, want = %#v, %v", e, got, err, want, nil) - } -} - -func TestQueuedSize(t *testing.T) { - e := ConnectedEndpoint{} - tests := []struct { - name string - f func() int64 - }{ - {"SendQueuedSize", e.SendQueuedSize}, - {"RecvQueuedSize", e.RecvQueuedSize}, - } - - for _, test := range tests { - if got, want := test.f(), int64(-1); got != want { - t.Errorf("Got %#v.%s() = %d, want = %d", e, test.name, got, want) - } - } -} - -func TestRelease(t *testing.T) { - f, err := syscall.Socket(syscall.AF_UNIX, syscall.SOCK_STREAM|syscall.SOCK_NONBLOCK|syscall.SOCK_CLOEXEC, 0) - if err != nil { - t.Fatal("Creating socket:", err) - } - c := &ConnectedEndpoint{queue: &waiter.Queue{}, file: fd.New(f)} - want := &ConnectedEndpoint{queue: c.queue} - want.ref.DecRef() - fdnotifier.AddFD(int32(c.file.FD()), nil) - c.Release() - if !reflect.DeepEqual(c, want) { - t.Errorf("got = %#v, want = %#v", c, want) - } -} diff --git a/pkg/sentry/fs/host/wait_test.go b/pkg/sentry/fs/host/wait_test.go deleted file mode 100644 index 88d24d693..000000000 --- a/pkg/sentry/fs/host/wait_test.go +++ /dev/null @@ -1,70 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package host - -import ( - "syscall" - "testing" - "time" - - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/waiter" -) - -func TestWait(t *testing.T) { - var fds [2]int - err := syscall.Pipe(fds[:]) - if err != nil { - t.Fatalf("Unable to create pipe: %v", err) - } - - defer syscall.Close(fds[1]) - - ctx := contexttest.Context(t) - file, err := NewFile(ctx, fds[0], fs.RootOwner) - if err != nil { - syscall.Close(fds[0]) - t.Fatalf("NewFile failed: %v", err) - } - - defer file.DecRef() - - r := file.Readiness(waiter.EventIn) - if r != 0 { - t.Fatalf("File is ready for read when it shouldn't be.") - } - - e, ch := waiter.NewChannelEntry(nil) - file.EventRegister(&e, waiter.EventIn) - defer file.EventUnregister(&e) - - // Check that there are no notifications yet. - if len(ch) != 0 { - t.Fatalf("Channel is non-empty") - } - - // Write to the pipe, so it should be writable now. - syscall.Write(fds[1], []byte{1}) - - // Check that we get a notification. We need to yield the current thread - // so that the fdnotifier can deliver notifications, so we use a - // 1-second timeout instead of just checking the length of the channel. - select { - case <-ch: - case <-time.After(1 * time.Second): - t.Fatalf("Channel not notified") - } -} diff --git a/pkg/sentry/fs/inode_overlay_test.go b/pkg/sentry/fs/inode_overlay_test.go deleted file mode 100644 index 8935aad65..000000000 --- a/pkg/sentry/fs/inode_overlay_test.go +++ /dev/null @@ -1,470 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fs_test - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" - "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" - "gvisor.dev/gvisor/pkg/syserror" -) - -func TestLookup(t *testing.T) { - ctx := contexttest.Context(t) - for _, test := range []struct { - // Test description. - desc string - - // Lookup parameters. - dir *fs.Inode - name string - - // Want from lookup. - found bool - hasUpper bool - hasLower bool - }{ - { - desc: "no upper, lower has name", - dir: fs.NewTestOverlayDir(ctx, - nil, /* upper */ - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: false, - }, - }, nil), /* lower */ - false /* revalidate */), - name: "a", - found: true, - hasUpper: false, - hasLower: true, - }, - { - desc: "no lower, upper has name", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: false, - }, - }, nil), /* upper */ - nil, /* lower */ - false /* revalidate */), - name: "a", - found: true, - hasUpper: true, - hasLower: false, - }, - { - desc: "upper and lower, only lower has name", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - { - name: "b", - dir: false, - }, - }, nil), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: false, - }, - }, nil), /* lower */ - false /* revalidate */), - name: "a", - found: true, - hasUpper: false, - hasLower: true, - }, - { - desc: "upper and lower, only upper has name", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: false, - }, - }, nil), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - { - name: "b", - dir: false, - }, - }, nil), /* lower */ - false /* revalidate */), - name: "a", - found: true, - hasUpper: true, - hasLower: false, - }, - { - desc: "upper and lower, both have file", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: false, - }, - }, nil), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: false, - }, - }, nil), /* lower */ - false /* revalidate */), - name: "a", - found: true, - hasUpper: true, - hasLower: false, - }, - { - desc: "upper and lower, both have directory", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: true, - }, - }, nil), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: true, - }, - }, nil), /* lower */ - false /* revalidate */), - name: "a", - found: true, - hasUpper: true, - hasLower: true, - }, - { - desc: "upper and lower, upper negative masks lower file", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, nil, []string{"a"}), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: false, - }, - }, nil), /* lower */ - false /* revalidate */), - name: "a", - found: false, - hasUpper: false, - hasLower: false, - }, - { - desc: "upper and lower, upper negative does not mask lower file", - dir: fs.NewTestOverlayDir(ctx, - newTestRamfsDir(ctx, nil, []string{"b"}), /* upper */ - newTestRamfsDir(ctx, []dirContent{ - { - name: "a", - dir: false, - }, - }, nil), /* lower */ - false /* revalidate */), - name: "a", - found: true, - hasUpper: false, - hasLower: true, - }, - } { - t.Run(test.desc, func(t *testing.T) { - dirent, err := test.dir.Lookup(ctx, test.name) - if test.found && (err == syserror.ENOENT || dirent.IsNegative()) { - t.Fatalf("lookup %q expected to find positive dirent, got dirent %v err %v", test.name, dirent, err) - } - if !test.found { - if err != syserror.ENOENT && !dirent.IsNegative() { - t.Errorf("lookup %q expected to return ENOENT or negative dirent, got dirent %v err %v", test.name, dirent, err) - } - // Nothing more to check. - return - } - if hasUpper := dirent.Inode.TestHasUpperFS(); hasUpper != test.hasUpper { - t.Fatalf("lookup got upper filesystem %v, want %v", hasUpper, test.hasUpper) - } - if hasLower := dirent.Inode.TestHasLowerFS(); hasLower != test.hasLower { - t.Errorf("lookup got lower filesystem %v, want %v", hasLower, test.hasLower) - } - }) - } -} - -func TestLookupRevalidation(t *testing.T) { - // File name used in the tests. - fileName := "foofile" - ctx := contexttest.Context(t) - for _, tc := range []struct { - // Test description. - desc string - - // Upper and lower fs for the overlay. - upper *fs.Inode - lower *fs.Inode - - // Whether the upper requires revalidation. - revalidate bool - - // Whether we should get the same dirent on second lookup. - wantSame bool - }{ - { - desc: "file from upper with no revalidation", - upper: newTestRamfsDir(ctx, []dirContent{{name: fileName}}, nil), - lower: newTestRamfsDir(ctx, nil, nil), - revalidate: false, - wantSame: true, - }, - { - desc: "file from upper with revalidation", - upper: newTestRamfsDir(ctx, []dirContent{{name: fileName}}, nil), - lower: newTestRamfsDir(ctx, nil, nil), - revalidate: true, - wantSame: false, - }, - { - desc: "file from lower with no revalidation", - upper: newTestRamfsDir(ctx, nil, nil), - lower: newTestRamfsDir(ctx, []dirContent{{name: fileName}}, nil), - revalidate: false, - wantSame: true, - }, - { - desc: "file from lower with revalidation", - upper: newTestRamfsDir(ctx, nil, nil), - lower: newTestRamfsDir(ctx, []dirContent{{name: fileName}}, nil), - revalidate: true, - // The file does not exist in the upper, so we do not - // need to revalidate it. - wantSame: true, - }, - { - desc: "file from upper and lower with no revalidation", - upper: newTestRamfsDir(ctx, []dirContent{{name: fileName}}, nil), - lower: newTestRamfsDir(ctx, []dirContent{{name: fileName}}, nil), - revalidate: false, - wantSame: true, - }, - { - desc: "file from upper and lower with revalidation", - upper: newTestRamfsDir(ctx, []dirContent{{name: fileName}}, nil), - lower: newTestRamfsDir(ctx, []dirContent{{name: fileName}}, nil), - revalidate: true, - wantSame: false, - }, - } { - t.Run(tc.desc, func(t *testing.T) { - root := fs.NewDirent(ctx, newTestRamfsDir(ctx, nil, nil), "root") - ctx = &rootContext{ - Context: ctx, - root: root, - } - overlay := fs.NewDirent(ctx, fs.NewTestOverlayDir(ctx, tc.upper, tc.lower, tc.revalidate), "overlay") - // Lookup the file twice through the overlay. - first, err := overlay.Walk(ctx, root, fileName) - if err != nil { - t.Fatalf("overlay.Walk(%q) failed: %v", fileName, err) - } - second, err := overlay.Walk(ctx, root, fileName) - if err != nil { - t.Fatalf("overlay.Walk(%q) failed: %v", fileName, err) - } - - if tc.wantSame && first != second { - t.Errorf("dirent lookup got different dirents, wanted same\nfirst=%+v\nsecond=%+v", first, second) - } else if !tc.wantSame && first == second { - t.Errorf("dirent lookup got the same dirent, wanted different: %+v", first) - } - }) - } -} - -func TestCacheFlush(t *testing.T) { - ctx := contexttest.Context(t) - - // Upper and lower each have a file. - upperFileName := "file-from-upper" - lowerFileName := "file-from-lower" - upper := newTestRamfsDir(ctx, []dirContent{{name: upperFileName}}, nil) - lower := newTestRamfsDir(ctx, []dirContent{{name: lowerFileName}}, nil) - - overlay := fs.NewTestOverlayDir(ctx, upper, lower, true /* revalidate */) - - mns, err := fs.NewMountNamespace(ctx, overlay) - if err != nil { - t.Fatalf("NewMountNamespace failed: %v", err) - } - root := mns.Root() - defer root.DecRef() - - ctx = &rootContext{ - Context: ctx, - root: root, - } - - for _, fileName := range []string{upperFileName, lowerFileName} { - // Walk to the file. - maxTraversals := uint(0) - dirent, err := mns.FindInode(ctx, root, nil, fileName, &maxTraversals) - if err != nil { - t.Fatalf("FindInode(%q) failed: %v", fileName, err) - } - - // Get a file from the dirent. - file, err := dirent.Inode.GetFile(ctx, dirent, fs.FileFlags{Read: true}) - if err != nil { - t.Fatalf("GetFile() failed: %v", err) - } - - // The dirent should have 3 refs, one from us, one from the - // file, and one from the dirent cache. - // dirent cache. - if got, want := dirent.ReadRefs(), 3; int(got) != want { - t.Errorf("dirent.ReadRefs() got %d want %d", got, want) - } - - // Drop the file reference. - file.DecRef() - - // Dirent should have 2 refs left. - if got, want := dirent.ReadRefs(), 2; int(got) != want { - t.Errorf("dirent.ReadRefs() got %d want %d", got, want) - } - - // Flush the dirent cache. - mns.FlushMountSourceRefs() - - // Dirent should have 1 ref left from the dirent cache. - if got, want := dirent.ReadRefs(), 1; int(got) != want { - t.Errorf("dirent.ReadRefs() got %d want %d", got, want) - } - - // Drop our ref. - dirent.DecRef() - - // We should be back to zero refs. - if got, want := dirent.ReadRefs(), 0; int(got) != want { - t.Errorf("dirent.ReadRefs() got %d want %d", got, want) - } - } - -} - -type dir struct { - fs.InodeOperations - - // List of negative child names. - negative []string - - // ReaddirCalled records whether Readdir was called on a file - // corresponding to this inode. - ReaddirCalled bool -} - -// Getxattr implements InodeOperations.Getxattr. -func (d *dir) Getxattr(inode *fs.Inode, name string) (string, error) { - for _, n := range d.negative { - if name == fs.XattrOverlayWhiteout(n) { - return "y", nil - } - } - return "", syserror.ENOATTR -} - -// GetFile implements InodeOperations.GetFile. -func (d *dir) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { - file, err := d.InodeOperations.GetFile(ctx, dirent, flags) - if err != nil { - return nil, err - } - defer file.DecRef() - // Wrap the file's FileOperations in a dirFile. - fops := &dirFile{ - FileOperations: file.FileOperations, - inode: d, - } - return fs.NewFile(ctx, dirent, flags, fops), nil -} - -type dirContent struct { - name string - dir bool -} - -type dirFile struct { - fs.FileOperations - inode *dir -} - -type inode struct { - fsutil.InodeGenericChecker `state:"nosave"` - fsutil.InodeNoExtendedAttributes `state:"nosave"` - fsutil.InodeNoopRelease `state:"nosave"` - fsutil.InodeNoopWriteOut `state:"nosave"` - fsutil.InodeNotAllocatable `state:"nosave"` - fsutil.InodeNotDirectory `state:"nosave"` - fsutil.InodeNotMappable `state:"nosave"` - fsutil.InodeNotSocket `state:"nosave"` - fsutil.InodeNotSymlink `state:"nosave"` - fsutil.InodeNotTruncatable `state:"nosave"` - fsutil.InodeNotVirtual `state:"nosave"` - - fsutil.InodeSimpleAttributes - fsutil.InodeStaticFileGetter -} - -// Readdir implements fs.FileOperations.Readdir. It sets the ReaddirCalled -// field on the inode. -func (f *dirFile) Readdir(ctx context.Context, file *fs.File, ser fs.DentrySerializer) (int64, error) { - f.inode.ReaddirCalled = true - return f.FileOperations.Readdir(ctx, file, ser) -} - -func newTestRamfsInode(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - inode := fs.NewInode(ctx, &inode{ - InodeStaticFileGetter: fsutil.InodeStaticFileGetter{ - Contents: []byte("foobar"), - }, - }, msrc, fs.StableAttr{Type: fs.RegularFile}) - return inode -} - -func newTestRamfsDir(ctx context.Context, contains []dirContent, negative []string) *fs.Inode { - msrc := fs.NewPseudoMountSource(ctx) - contents := make(map[string]*fs.Inode) - for _, c := range contains { - if c.dir { - contents[c.name] = newTestRamfsDir(ctx, nil, nil) - } else { - contents[c.name] = newTestRamfsInode(ctx, msrc) - } - } - dops := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermissions{ - User: fs.PermMask{Read: true, Execute: true}, - }) - return fs.NewInode(ctx, &dir{ - InodeOperations: dops, - negative: negative, - }, msrc, fs.StableAttr{Type: fs.Directory}) -} diff --git a/pkg/sentry/fs/lock/BUILD b/pkg/sentry/fs/lock/BUILD deleted file mode 100644 index 08d7c0c57..000000000 --- a/pkg/sentry/fs/lock/BUILD +++ /dev/null @@ -1,58 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library", "go_test") - -go_template_instance( - name = "lock_range", - out = "lock_range.go", - package = "lock", - prefix = "Lock", - template = "//pkg/segment:generic_range", - types = { - "T": "uint64", - }, -) - -go_template_instance( - name = "lock_set", - out = "lock_set.go", - consts = { - "minDegree": "3", - }, - package = "lock", - prefix = "Lock", - template = "//pkg/segment:generic_set", - types = { - "Key": "uint64", - "Range": "LockRange", - "Value": "Lock", - "Functions": "lockSetFunctions", - }, -) - -go_library( - name = "lock", - srcs = [ - "lock.go", - "lock_range.go", - "lock_set.go", - "lock_set_functions.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/lock", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/log", - "//pkg/waiter", - ], -) - -go_test( - name = "lock_test", - size = "small", - srcs = [ - "lock_range_test.go", - "lock_test.go", - ], - embed = [":lock"], -) diff --git a/pkg/sentry/fs/lock/lock_range.go b/pkg/sentry/fs/lock/lock_range.go new file mode 100755 index 000000000..7a6f77640 --- /dev/null +++ b/pkg/sentry/fs/lock/lock_range.go @@ -0,0 +1,62 @@ +package lock + +// A Range represents a contiguous range of T. +// +// +stateify savable +type LockRange struct { + // Start is the inclusive start of the range. + Start uint64 + + // End is the exclusive end of the range. + End uint64 +} + +// WellFormed returns true if r.Start <= r.End. All other methods on a Range +// require that the Range is well-formed. +func (r LockRange) WellFormed() bool { + return r.Start <= r.End +} + +// Length returns the length of the range. +func (r LockRange) Length() uint64 { + return r.End - r.Start +} + +// Contains returns true if r contains x. +func (r LockRange) Contains(x uint64) bool { + return r.Start <= x && x < r.End +} + +// Overlaps returns true if r and r2 overlap. +func (r LockRange) Overlaps(r2 LockRange) bool { + return r.Start < r2.End && r2.Start < r.End +} + +// IsSupersetOf returns true if r is a superset of r2; that is, the range r2 is +// contained within r. +func (r LockRange) IsSupersetOf(r2 LockRange) bool { + return r.Start <= r2.Start && r.End >= r2.End +} + +// Intersect returns a range consisting of the intersection between r and r2. +// If r and r2 do not overlap, Intersect returns a range with unspecified +// bounds, but for which Length() == 0. +func (r LockRange) Intersect(r2 LockRange) LockRange { + if r.Start < r2.Start { + r.Start = r2.Start + } + if r.End > r2.End { + r.End = r2.End + } + if r.End < r.Start { + r.End = r.Start + } + return r +} + +// CanSplitAt returns true if it is legal to split a segment spanning the range +// r at x; that is, splitting at x would produce two ranges, both of which have +// non-zero length. +func (r LockRange) CanSplitAt(x uint64) bool { + return r.Contains(x) && r.Start < x +} diff --git a/pkg/sentry/fs/lock/lock_range_test.go b/pkg/sentry/fs/lock/lock_range_test.go deleted file mode 100644 index 6221199d1..000000000 --- a/pkg/sentry/fs/lock/lock_range_test.go +++ /dev/null @@ -1,136 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package lock - -import ( - "syscall" - "testing" -) - -func TestComputeRange(t *testing.T) { - tests := []struct { - // Description of test. - name string - - // Requested start of the lock range. - start int64 - - // Requested length of the lock range, - // can be negative :( - length int64 - - // Pre-computed file offset based on whence. - // Will be added to start. - offset int64 - - // Expected error. - err error - - // If error is nil, the expected LockRange. - LockRange - }{ - { - name: "offset, start, and length all zero", - LockRange: LockRange{Start: 0, End: LockEOF}, - }, - { - name: "zero offset, zero start, positive length", - start: 0, - length: 4096, - offset: 0, - LockRange: LockRange{Start: 0, End: 4096}, - }, - { - name: "zero offset, negative start", - start: -4096, - offset: 0, - err: syscall.EINVAL, - }, - { - name: "large offset, negative start, positive length", - start: -2048, - length: 2048, - offset: 4096, - LockRange: LockRange{Start: 2048, End: 4096}, - }, - { - name: "large offset, negative start, zero length", - start: -2048, - length: 0, - offset: 4096, - LockRange: LockRange{Start: 2048, End: LockEOF}, - }, - { - name: "zero offset, zero start, negative length", - start: 0, - length: -4096, - offset: 0, - err: syscall.EINVAL, - }, - { - name: "large offset, zero start, negative length", - start: 0, - length: -4096, - offset: 4096, - LockRange: LockRange{Start: 0, End: 4096}, - }, - { - name: "offset, start, and length equal, length is negative", - start: 1024, - length: -1024, - offset: 1024, - LockRange: LockRange{Start: 1024, End: 2048}, - }, - { - name: "offset, start, and length equal, start is negative", - start: -1024, - length: 1024, - offset: 1024, - LockRange: LockRange{Start: 0, End: 1024}, - }, - { - name: "offset, start, and length equal, offset is negative", - start: 1024, - length: 1024, - offset: -1024, - LockRange: LockRange{Start: 0, End: 1024}, - }, - { - name: "offset, start, and length equal, all negative", - start: -1024, - length: -1024, - offset: -1024, - err: syscall.EINVAL, - }, - { - name: "offset, start, and length equal, all positive", - start: 1024, - length: 1024, - offset: 1024, - LockRange: LockRange{Start: 2048, End: 3072}, - }, - } - - for _, test := range tests { - rng, err := ComputeRange(test.start, test.length, test.offset) - if err != test.err { - t.Errorf("%s: lockRange(%d, %d, %d) got error %v, want %v", test.name, test.start, test.length, test.offset, err, test.err) - continue - } - if err == nil && rng != test.LockRange { - t.Errorf("%s: lockRange(%d, %d, %d) got LockRange %v, want %v", test.name, test.start, test.length, test.offset, rng, test.LockRange) - } - } -} diff --git a/pkg/sentry/fs/lock/lock_set.go b/pkg/sentry/fs/lock/lock_set.go new file mode 100755 index 000000000..2343ca0b4 --- /dev/null +++ b/pkg/sentry/fs/lock/lock_set.go @@ -0,0 +1,1270 @@ +package lock + +import ( + "bytes" + "fmt" +) + +const ( + // minDegree is the minimum degree of an internal node in a Set B-tree. + // + // - Any non-root node has at least minDegree-1 segments. + // + // - Any non-root internal (non-leaf) node has at least minDegree children. + // + // - The root node may have fewer than minDegree-1 segments, but it may + // only have 0 segments if the tree is empty. + // + // Our implementation requires minDegree >= 3. Higher values of minDegree + // usually improve performance, but increase memory usage for small sets. + LockminDegree = 3 + + LockmaxDegree = 2 * LockminDegree +) + +// A Set is a mapping of segments with non-overlapping Range keys. The zero +// value for a Set is an empty set. Set values are not safely movable nor +// copyable. Set is thread-compatible. +// +// +stateify savable +type LockSet struct { + root Locknode `state:".(*LockSegmentDataSlices)"` +} + +// IsEmpty returns true if the set contains no segments. +func (s *LockSet) IsEmpty() bool { + return s.root.nrSegments == 0 +} + +// IsEmptyRange returns true iff no segments in the set overlap the given +// range. This is semantically equivalent to s.SpanRange(r) == 0, but may be +// more efficient. +func (s *LockSet) IsEmptyRange(r LockRange) bool { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return true + } + _, gap := s.Find(r.Start) + if !gap.Ok() { + return false + } + return r.End <= gap.End() +} + +// Span returns the total size of all segments in the set. +func (s *LockSet) Span() uint64 { + var sz uint64 + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sz += seg.Range().Length() + } + return sz +} + +// SpanRange returns the total size of the intersection of segments in the set +// with the given range. +func (s *LockSet) SpanRange(r LockRange) uint64 { + switch { + case r.Length() < 0: + panic(fmt.Sprintf("invalid range %v", r)) + case r.Length() == 0: + return 0 + } + var sz uint64 + for seg := s.LowerBoundSegment(r.Start); seg.Ok() && seg.Start() < r.End; seg = seg.NextSegment() { + sz += seg.Range().Intersect(r).Length() + } + return sz +} + +// FirstSegment returns the first segment in the set. If the set is empty, +// FirstSegment returns a terminal iterator. +func (s *LockSet) FirstSegment() LockIterator { + if s.root.nrSegments == 0 { + return LockIterator{} + } + return s.root.firstSegment() +} + +// LastSegment returns the last segment in the set. If the set is empty, +// LastSegment returns a terminal iterator. +func (s *LockSet) LastSegment() LockIterator { + if s.root.nrSegments == 0 { + return LockIterator{} + } + return s.root.lastSegment() +} + +// FirstGap returns the first gap in the set. +func (s *LockSet) FirstGap() LockGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[0] + } + return LockGapIterator{n, 0} +} + +// LastGap returns the last gap in the set. +func (s *LockSet) LastGap() LockGapIterator { + n := &s.root + for n.hasChildren { + n = n.children[n.nrSegments] + } + return LockGapIterator{n, n.nrSegments} +} + +// Find returns the segment or gap whose range contains the given key. If a +// segment is found, the returned Iterator is non-terminal and the +// returned GapIterator is terminal. Otherwise, the returned Iterator is +// terminal and the returned GapIterator is non-terminal. +func (s *LockSet) Find(key uint64) (LockIterator, LockGapIterator) { + n := &s.root + for { + + lower := 0 + upper := n.nrSegments + for lower < upper { + i := lower + (upper-lower)/2 + if r := n.keys[i]; key < r.End { + if key >= r.Start { + return LockIterator{n, i}, LockGapIterator{} + } + upper = i + } else { + lower = i + 1 + } + } + i := lower + if !n.hasChildren { + return LockIterator{}, LockGapIterator{n, i} + } + n = n.children[i] + } +} + +// FindSegment returns the segment whose range contains the given key. If no +// such segment exists, FindSegment returns a terminal iterator. +func (s *LockSet) FindSegment(key uint64) LockIterator { + seg, _ := s.Find(key) + return seg +} + +// LowerBoundSegment returns the segment with the lowest range that contains a +// key greater than or equal to min. If no such segment exists, +// LowerBoundSegment returns a terminal iterator. +func (s *LockSet) LowerBoundSegment(min uint64) LockIterator { + seg, gap := s.Find(min) + if seg.Ok() { + return seg + } + return gap.NextSegment() +} + +// UpperBoundSegment returns the segment with the highest range that contains a +// key less than or equal to max. If no such segment exists, UpperBoundSegment +// returns a terminal iterator. +func (s *LockSet) UpperBoundSegment(max uint64) LockIterator { + seg, gap := s.Find(max) + if seg.Ok() { + return seg + } + return gap.PrevSegment() +} + +// FindGap returns the gap containing the given key. If no such gap exists +// (i.e. the set contains a segment containing that key), FindGap returns a +// terminal iterator. +func (s *LockSet) FindGap(key uint64) LockGapIterator { + _, gap := s.Find(key) + return gap +} + +// LowerBoundGap returns the gap with the lowest range that is greater than or +// equal to min. +func (s *LockSet) LowerBoundGap(min uint64) LockGapIterator { + seg, gap := s.Find(min) + if gap.Ok() { + return gap + } + return seg.NextGap() +} + +// UpperBoundGap returns the gap with the highest range that is less than or +// equal to max. +func (s *LockSet) UpperBoundGap(max uint64) LockGapIterator { + seg, gap := s.Find(max) + if gap.Ok() { + return gap + } + return seg.PrevGap() +} + +// Add inserts the given segment into the set and returns true. If the new +// segment can be merged with adjacent segments, Add will do so. If the new +// segment would overlap an existing segment, Add returns false. If Add +// succeeds, all existing iterators are invalidated. +func (s *LockSet) Add(r LockRange, val Lock) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.Insert(gap, r, val) + return true +} + +// AddWithoutMerging inserts the given segment into the set and returns true. +// If it would overlap an existing segment, AddWithoutMerging does nothing and +// returns false. If AddWithoutMerging succeeds, all existing iterators are +// invalidated. +func (s *LockSet) AddWithoutMerging(r LockRange, val Lock) bool { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + gap := s.FindGap(r.Start) + if !gap.Ok() { + return false + } + if r.End > gap.End() { + return false + } + s.InsertWithoutMergingUnchecked(gap, r, val) + return true +} + +// Insert inserts the given segment into the given gap. If the new segment can +// be merged with adjacent segments, Insert will do so. Insert returns an +// iterator to the segment containing the inserted value (which may have been +// merged with other values). All existing iterators (including gap, but not +// including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, Insert panics. +// +// Insert is semantically equivalent to a InsertWithoutMerging followed by a +// Merge, but may be more efficient. Note that there is no unchecked variant of +// Insert since Insert must retrieve and inspect gap's predecessor and +// successor segments regardless. +func (s *LockSet) Insert(gap LockGapIterator, r LockRange, val Lock) LockIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + prev, next := gap.PrevSegment(), gap.NextSegment() + if prev.Ok() && prev.End() > r.Start { + panic(fmt.Sprintf("new segment %v overlaps predecessor %v", r, prev.Range())) + } + if next.Ok() && next.Start() < r.End { + panic(fmt.Sprintf("new segment %v overlaps successor %v", r, next.Range())) + } + if prev.Ok() && prev.End() == r.Start { + if mval, ok := (lockSetFunctions{}).Merge(prev.Range(), prev.Value(), r, val); ok { + prev.SetEndUnchecked(r.End) + prev.SetValue(mval) + if next.Ok() && next.Start() == r.End { + val = mval + if mval, ok := (lockSetFunctions{}).Merge(prev.Range(), val, next.Range(), next.Value()); ok { + prev.SetEndUnchecked(next.End()) + prev.SetValue(mval) + return s.Remove(next).PrevSegment() + } + } + return prev + } + } + if next.Ok() && next.Start() == r.End { + if mval, ok := (lockSetFunctions{}).Merge(r, val, next.Range(), next.Value()); ok { + next.SetStartUnchecked(r.Start) + next.SetValue(mval) + return next + } + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMerging inserts the given segment into the given gap and +// returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// If the gap cannot accommodate the segment, or if r is invalid, +// InsertWithoutMerging panics. +func (s *LockSet) InsertWithoutMerging(gap LockGapIterator, r LockRange, val Lock) LockIterator { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if gr := gap.Range(); !gr.IsSupersetOf(r) { + panic(fmt.Sprintf("cannot insert segment range %v into gap range %v", r, gr)) + } + return s.InsertWithoutMergingUnchecked(gap, r, val) +} + +// InsertWithoutMergingUnchecked inserts the given segment into the given gap +// and returns an iterator to the inserted segment. All existing iterators +// (including gap, but not including the returned iterator) are invalidated. +// +// Preconditions: r.Start >= gap.Start(); r.End <= gap.End(). +func (s *LockSet) InsertWithoutMergingUnchecked(gap LockGapIterator, r LockRange, val Lock) LockIterator { + gap = gap.node.rebalanceBeforeInsert(gap) + copy(gap.node.keys[gap.index+1:], gap.node.keys[gap.index:gap.node.nrSegments]) + copy(gap.node.values[gap.index+1:], gap.node.values[gap.index:gap.node.nrSegments]) + gap.node.keys[gap.index] = r + gap.node.values[gap.index] = val + gap.node.nrSegments++ + return LockIterator{gap.node, gap.index} +} + +// Remove removes the given segment and returns an iterator to the vacated gap. +// All existing iterators (including seg, but not including the returned +// iterator) are invalidated. +func (s *LockSet) Remove(seg LockIterator) LockGapIterator { + + if seg.node.hasChildren { + + victim := seg.PrevSegment() + + seg.SetRangeUnchecked(victim.Range()) + seg.SetValue(victim.Value()) + return s.Remove(victim).NextGap() + } + copy(seg.node.keys[seg.index:], seg.node.keys[seg.index+1:seg.node.nrSegments]) + copy(seg.node.values[seg.index:], seg.node.values[seg.index+1:seg.node.nrSegments]) + lockSetFunctions{}.ClearValue(&seg.node.values[seg.node.nrSegments-1]) + seg.node.nrSegments-- + return seg.node.rebalanceAfterRemove(LockGapIterator{seg.node, seg.index}) +} + +// RemoveAll removes all segments from the set. All existing iterators are +// invalidated. +func (s *LockSet) RemoveAll() { + s.root = Locknode{} +} + +// RemoveRange removes all segments in the given range. An iterator to the +// newly formed gap is returned, and all existing iterators are invalidated. +func (s *LockSet) RemoveRange(r LockRange) LockGapIterator { + seg, gap := s.Find(r.Start) + if seg.Ok() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + for seg = gap.NextSegment(); seg.Ok() && seg.Start() < r.End; seg = gap.NextSegment() { + seg = s.Isolate(seg, r) + gap = s.Remove(seg) + } + return gap +} + +// Merge attempts to merge two neighboring segments. If successful, Merge +// returns an iterator to the merged segment, and all existing iterators are +// invalidated. Otherwise, Merge returns a terminal iterator. +// +// If first is not the predecessor of second, Merge panics. +func (s *LockSet) Merge(first, second LockIterator) LockIterator { + if first.NextSegment() != second { + panic(fmt.Sprintf("attempt to merge non-neighboring segments %v, %v", first.Range(), second.Range())) + } + return s.MergeUnchecked(first, second) +} + +// MergeUnchecked attempts to merge two neighboring segments. If successful, +// MergeUnchecked returns an iterator to the merged segment, and all existing +// iterators are invalidated. Otherwise, MergeUnchecked returns a terminal +// iterator. +// +// Precondition: first is the predecessor of second: first.NextSegment() == +// second, first == second.PrevSegment(). +func (s *LockSet) MergeUnchecked(first, second LockIterator) LockIterator { + if first.End() == second.Start() { + if mval, ok := (lockSetFunctions{}).Merge(first.Range(), first.Value(), second.Range(), second.Value()); ok { + + first.SetEndUnchecked(second.End()) + first.SetValue(mval) + return s.Remove(second).PrevSegment() + } + } + return LockIterator{} +} + +// MergeAll attempts to merge all adjacent segments in the set. All existing +// iterators are invalidated. +func (s *LockSet) MergeAll() { + seg := s.FirstSegment() + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeRange attempts to merge all adjacent segments that contain a key in the +// specific range. All existing iterators are invalidated. +func (s *LockSet) MergeRange(r LockRange) { + seg := s.LowerBoundSegment(r.Start) + if !seg.Ok() { + return + } + next := seg.NextSegment() + for next.Ok() && next.Range().Start < r.End { + if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { + seg, next = mseg, mseg.NextSegment() + } else { + seg, next = next, next.NextSegment() + } + } +} + +// MergeAdjacent attempts to merge the segment containing r.Start with its +// predecessor, and the segment containing r.End-1 with its successor. +func (s *LockSet) MergeAdjacent(r LockRange) { + first := s.FindSegment(r.Start) + if first.Ok() { + if prev := first.PrevSegment(); prev.Ok() { + s.Merge(prev, first) + } + } + last := s.FindSegment(r.End - 1) + if last.Ok() { + if next := last.NextSegment(); next.Ok() { + s.Merge(last, next) + } + } +} + +// Split splits the given segment at the given key and returns iterators to the +// two resulting segments. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +// +// If the segment cannot be split at split (because split is at the start or +// end of the segment's range, so splitting would produce a segment with zero +// length, or because split falls outside the segment's range altogether), +// Split panics. +func (s *LockSet) Split(seg LockIterator, split uint64) (LockIterator, LockIterator) { + if !seg.Range().CanSplitAt(split) { + panic(fmt.Sprintf("can't split %v at %v", seg.Range(), split)) + } + return s.SplitUnchecked(seg, split) +} + +// SplitUnchecked splits the given segment at the given key and returns +// iterators to the two resulting segments. All existing iterators (including +// seg, but not including the returned iterators) are invalidated. +// +// Preconditions: seg.Start() < key < seg.End(). +func (s *LockSet) SplitUnchecked(seg LockIterator, split uint64) (LockIterator, LockIterator) { + val1, val2 := (lockSetFunctions{}).Split(seg.Range(), seg.Value(), split) + end2 := seg.End() + seg.SetEndUnchecked(split) + seg.SetValue(val1) + seg2 := s.InsertWithoutMergingUnchecked(seg.NextGap(), LockRange{split, end2}, val2) + + return seg2.PrevSegment(), seg2 +} + +// SplitAt splits the segment straddling split, if one exists. SplitAt returns +// true if a segment was split and false otherwise. If SplitAt splits a +// segment, all existing iterators are invalidated. +func (s *LockSet) SplitAt(split uint64) bool { + if seg := s.FindSegment(split); seg.Ok() && seg.Range().CanSplitAt(split) { + s.SplitUnchecked(seg, split) + return true + } + return false +} + +// Isolate ensures that the given segment's range does not escape r by +// splitting at r.Start and r.End if necessary, and returns an updated iterator +// to the bounded segment. All existing iterators (including seg, but not +// including the returned iterators) are invalidated. +func (s *LockSet) Isolate(seg LockIterator, r LockRange) LockIterator { + if seg.Range().CanSplitAt(r.Start) { + _, seg = s.SplitUnchecked(seg, r.Start) + } + if seg.Range().CanSplitAt(r.End) { + seg, _ = s.SplitUnchecked(seg, r.End) + } + return seg +} + +// ApplyContiguous applies a function to a contiguous range of segments, +// splitting if necessary. The function is applied until the first gap is +// encountered, at which point the gap is returned. If the function is applied +// across the entire range, a terminal gap is returned. All existing iterators +// are invalidated. +// +// N.B. The Iterator must not be invalidated by the function. +func (s *LockSet) ApplyContiguous(r LockRange, fn func(seg LockIterator)) LockGapIterator { + seg, gap := s.Find(r.Start) + if !seg.Ok() { + return gap + } + for { + seg = s.Isolate(seg, r) + fn(seg) + if seg.End() >= r.End { + return LockGapIterator{} + } + gap = seg.NextGap() + if !gap.IsEmpty() { + return gap + } + seg = gap.NextSegment() + if !seg.Ok() { + + return LockGapIterator{} + } + } +} + +// +stateify savable +type Locknode struct { + // An internal binary tree node looks like: + // + // K + // / \ + // Cl Cr + // + // where all keys in the subtree rooted by Cl (the left subtree) are less + // than K (the key of the parent node), and all keys in the subtree rooted + // by Cr (the right subtree) are greater than K. + // + // An internal B-tree node's indexes work out to look like: + // + // K0 K1 K2 ... Kn-1 + // / \/ \/ \ ... / \ + // C0 C1 C2 C3 ... Cn-1 Cn + // + // where n is nrSegments. + nrSegments int + + // parent is a pointer to this node's parent. If this node is root, parent + // is nil. + parent *Locknode + + // parentIndex is the index of this node in parent.children. + parentIndex int + + // Flag for internal nodes that is technically redundant with "children[0] + // != nil", but is stored in the first cache line. "hasChildren" rather + // than "isLeaf" because false must be the correct value for an empty root. + hasChildren bool + + // Nodes store keys and values in separate arrays to maximize locality in + // the common case (scanning keys for lookup). + keys [LockmaxDegree - 1]LockRange + values [LockmaxDegree - 1]Lock + children [LockmaxDegree]*Locknode +} + +// firstSegment returns the first segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *Locknode) firstSegment() LockIterator { + for n.hasChildren { + n = n.children[0] + } + return LockIterator{n, 0} +} + +// lastSegment returns the last segment in the subtree rooted by n. +// +// Preconditions: n.nrSegments != 0. +func (n *Locknode) lastSegment() LockIterator { + for n.hasChildren { + n = n.children[n.nrSegments] + } + return LockIterator{n, n.nrSegments - 1} +} + +func (n *Locknode) prevSibling() *Locknode { + if n.parent == nil || n.parentIndex == 0 { + return nil + } + return n.parent.children[n.parentIndex-1] +} + +func (n *Locknode) nextSibling() *Locknode { + if n.parent == nil || n.parentIndex == n.parent.nrSegments { + return nil + } + return n.parent.children[n.parentIndex+1] +} + +// rebalanceBeforeInsert splits n and its ancestors if they are full, as +// required for insertion, and returns an updated iterator to the position +// represented by gap. +func (n *Locknode) rebalanceBeforeInsert(gap LockGapIterator) LockGapIterator { + if n.parent != nil { + gap = n.parent.rebalanceBeforeInsert(gap) + } + if n.nrSegments < LockmaxDegree-1 { + return gap + } + if n.parent == nil { + + left := &Locknode{ + nrSegments: LockminDegree - 1, + parent: n, + parentIndex: 0, + hasChildren: n.hasChildren, + } + right := &Locknode{ + nrSegments: LockminDegree - 1, + parent: n, + parentIndex: 1, + hasChildren: n.hasChildren, + } + copy(left.keys[:LockminDegree-1], n.keys[:LockminDegree-1]) + copy(left.values[:LockminDegree-1], n.values[:LockminDegree-1]) + copy(right.keys[:LockminDegree-1], n.keys[LockminDegree:]) + copy(right.values[:LockminDegree-1], n.values[LockminDegree:]) + n.keys[0], n.values[0] = n.keys[LockminDegree-1], n.values[LockminDegree-1] + LockzeroValueSlice(n.values[1:]) + if n.hasChildren { + copy(left.children[:LockminDegree], n.children[:LockminDegree]) + copy(right.children[:LockminDegree], n.children[LockminDegree:]) + LockzeroNodeSlice(n.children[2:]) + for i := 0; i < LockminDegree; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + right.children[i].parent = right + right.children[i].parentIndex = i + } + } + n.nrSegments = 1 + n.hasChildren = true + n.children[0] = left + n.children[1] = right + if gap.node != n { + return gap + } + if gap.index < LockminDegree { + return LockGapIterator{left, gap.index} + } + return LockGapIterator{right, gap.index - LockminDegree} + } + + copy(n.parent.keys[n.parentIndex+1:], n.parent.keys[n.parentIndex:n.parent.nrSegments]) + copy(n.parent.values[n.parentIndex+1:], n.parent.values[n.parentIndex:n.parent.nrSegments]) + n.parent.keys[n.parentIndex], n.parent.values[n.parentIndex] = n.keys[LockminDegree-1], n.values[LockminDegree-1] + copy(n.parent.children[n.parentIndex+2:], n.parent.children[n.parentIndex+1:n.parent.nrSegments+1]) + for i := n.parentIndex + 2; i < n.parent.nrSegments+2; i++ { + n.parent.children[i].parentIndex = i + } + sibling := &Locknode{ + nrSegments: LockminDegree - 1, + parent: n.parent, + parentIndex: n.parentIndex + 1, + hasChildren: n.hasChildren, + } + n.parent.children[n.parentIndex+1] = sibling + n.parent.nrSegments++ + copy(sibling.keys[:LockminDegree-1], n.keys[LockminDegree:]) + copy(sibling.values[:LockminDegree-1], n.values[LockminDegree:]) + LockzeroValueSlice(n.values[LockminDegree-1:]) + if n.hasChildren { + copy(sibling.children[:LockminDegree], n.children[LockminDegree:]) + LockzeroNodeSlice(n.children[LockminDegree:]) + for i := 0; i < LockminDegree; i++ { + sibling.children[i].parent = sibling + sibling.children[i].parentIndex = i + } + } + n.nrSegments = LockminDegree - 1 + + if gap.node != n { + return gap + } + if gap.index < LockminDegree { + return gap + } + return LockGapIterator{sibling, gap.index - LockminDegree} +} + +// rebalanceAfterRemove "unsplits" n and its ancestors if they are deficient +// (contain fewer segments than required by B-tree invariants), as required for +// removal, and returns an updated iterator to the position represented by gap. +// +// Precondition: n is the only node in the tree that may currently violate a +// B-tree invariant. +func (n *Locknode) rebalanceAfterRemove(gap LockGapIterator) LockGapIterator { + for { + if n.nrSegments >= LockminDegree-1 { + return gap + } + if n.parent == nil { + + return gap + } + + if sibling := n.prevSibling(); sibling != nil && sibling.nrSegments >= LockminDegree { + copy(n.keys[1:], n.keys[:n.nrSegments]) + copy(n.values[1:], n.values[:n.nrSegments]) + n.keys[0] = n.parent.keys[n.parentIndex-1] + n.values[0] = n.parent.values[n.parentIndex-1] + n.parent.keys[n.parentIndex-1] = sibling.keys[sibling.nrSegments-1] + n.parent.values[n.parentIndex-1] = sibling.values[sibling.nrSegments-1] + lockSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + copy(n.children[1:], n.children[:n.nrSegments+1]) + n.children[0] = sibling.children[sibling.nrSegments] + sibling.children[sibling.nrSegments] = nil + n.children[0].parent = n + n.children[0].parentIndex = 0 + for i := 1; i < n.nrSegments+2; i++ { + n.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling && gap.index == sibling.nrSegments { + return LockGapIterator{n, 0} + } + if gap.node == n { + return LockGapIterator{n, gap.index + 1} + } + return gap + } + if sibling := n.nextSibling(); sibling != nil && sibling.nrSegments >= LockminDegree { + n.keys[n.nrSegments] = n.parent.keys[n.parentIndex] + n.values[n.nrSegments] = n.parent.values[n.parentIndex] + n.parent.keys[n.parentIndex] = sibling.keys[0] + n.parent.values[n.parentIndex] = sibling.values[0] + copy(sibling.keys[:sibling.nrSegments-1], sibling.keys[1:]) + copy(sibling.values[:sibling.nrSegments-1], sibling.values[1:]) + lockSetFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) + if n.hasChildren { + n.children[n.nrSegments+1] = sibling.children[0] + copy(sibling.children[:sibling.nrSegments], sibling.children[1:]) + sibling.children[sibling.nrSegments] = nil + n.children[n.nrSegments+1].parent = n + n.children[n.nrSegments+1].parentIndex = n.nrSegments + 1 + for i := 0; i < sibling.nrSegments; i++ { + sibling.children[i].parentIndex = i + } + } + n.nrSegments++ + sibling.nrSegments-- + if gap.node == sibling { + if gap.index == 0 { + return LockGapIterator{n, n.nrSegments} + } + return LockGapIterator{sibling, gap.index - 1} + } + return gap + } + + p := n.parent + if p.nrSegments == 1 { + + left, right := p.children[0], p.children[1] + p.nrSegments = left.nrSegments + right.nrSegments + 1 + p.hasChildren = left.hasChildren + p.keys[left.nrSegments] = p.keys[0] + p.values[left.nrSegments] = p.values[0] + copy(p.keys[:left.nrSegments], left.keys[:left.nrSegments]) + copy(p.values[:left.nrSegments], left.values[:left.nrSegments]) + copy(p.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(p.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(p.children[:left.nrSegments+1], left.children[:left.nrSegments+1]) + copy(p.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := 0; i < p.nrSegments+1; i++ { + p.children[i].parent = p + p.children[i].parentIndex = i + } + } else { + p.children[0] = nil + p.children[1] = nil + } + if gap.node == left { + return LockGapIterator{p, gap.index} + } + if gap.node == right { + return LockGapIterator{p, gap.index + left.nrSegments + 1} + } + return gap + } + // Merge n and either sibling, along with the segment separating the + // two, into whichever of the two nodes comes first. This is the + // reverse of the non-root splitting case in + // node.rebalanceBeforeInsert. + var left, right *Locknode + if n.parentIndex > 0 { + left = n.prevSibling() + right = n + } else { + left = n + right = n.nextSibling() + } + + if gap.node == right { + gap = LockGapIterator{left, gap.index + left.nrSegments + 1} + } + left.keys[left.nrSegments] = p.keys[left.parentIndex] + left.values[left.nrSegments] = p.values[left.parentIndex] + copy(left.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) + copy(left.values[left.nrSegments+1:], right.values[:right.nrSegments]) + if left.hasChildren { + copy(left.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) + for i := left.nrSegments + 1; i < left.nrSegments+right.nrSegments+2; i++ { + left.children[i].parent = left + left.children[i].parentIndex = i + } + } + left.nrSegments += right.nrSegments + 1 + copy(p.keys[left.parentIndex:], p.keys[left.parentIndex+1:p.nrSegments]) + copy(p.values[left.parentIndex:], p.values[left.parentIndex+1:p.nrSegments]) + lockSetFunctions{}.ClearValue(&p.values[p.nrSegments-1]) + copy(p.children[left.parentIndex+1:], p.children[left.parentIndex+2:p.nrSegments+1]) + for i := 0; i < p.nrSegments; i++ { + p.children[i].parentIndex = i + } + p.children[p.nrSegments] = nil + p.nrSegments-- + + n = p + } +} + +// A Iterator is conceptually one of: +// +// - A pointer to a segment in a set; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Iterators are copyable values and are meaningfully equality-comparable. The +// zero value of Iterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type LockIterator struct { + // node is the node containing the iterated segment. If the iterator is + // terminal, node is nil. + node *Locknode + + // index is the index of the segment in node.keys/values. + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (seg LockIterator) Ok() bool { + return seg.node != nil +} + +// Range returns the iterated segment's range key. +func (seg LockIterator) Range() LockRange { + return seg.node.keys[seg.index] +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (seg LockIterator) Start() uint64 { + return seg.node.keys[seg.index].Start +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (seg LockIterator) End() uint64 { + return seg.node.keys[seg.index].End +} + +// SetRangeUnchecked mutates the iterated segment's range key. This operation +// does not invalidate any iterators. +// +// Preconditions: +// +// - r.Length() > 0. +// +// - The new range must not overlap an existing one: If seg.NextSegment().Ok(), +// then r.end <= seg.NextSegment().Start(); if seg.PrevSegment().Ok(), then +// r.start >= seg.PrevSegment().End(). +func (seg LockIterator) SetRangeUnchecked(r LockRange) { + seg.node.keys[seg.index] = r +} + +// SetRange mutates the iterated segment's range key. If the new range would +// cause the iterated segment to overlap another segment, or if the new range +// is invalid, SetRange panics. This operation does not invalidate any +// iterators. +func (seg LockIterator) SetRange(r LockRange) { + if r.Length() <= 0 { + panic(fmt.Sprintf("invalid segment range %v", r)) + } + if prev := seg.PrevSegment(); prev.Ok() && r.Start < prev.End() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, prev.Range())) + } + if next := seg.NextSegment(); next.Ok() && r.End > next.Start() { + panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, next.Range())) + } + seg.SetRangeUnchecked(r) +} + +// SetStartUnchecked mutates the iterated segment's start. This operation does +// not invalidate any iterators. +// +// Preconditions: The new start must be valid: start < seg.End(); if +// seg.PrevSegment().Ok(), then start >= seg.PrevSegment().End(). +func (seg LockIterator) SetStartUnchecked(start uint64) { + seg.node.keys[seg.index].Start = start +} + +// SetStart mutates the iterated segment's start. If the new start value would +// cause the iterated segment to overlap another segment, or would result in an +// invalid range, SetStart panics. This operation does not invalidate any +// iterators. +func (seg LockIterator) SetStart(start uint64) { + if start >= seg.End() { + panic(fmt.Sprintf("new start %v would invalidate segment range %v", start, seg.Range())) + } + if prev := seg.PrevSegment(); prev.Ok() && start < prev.End() { + panic(fmt.Sprintf("new start %v would cause segment range %v to overlap segment range %v", start, seg.Range(), prev.Range())) + } + seg.SetStartUnchecked(start) +} + +// SetEndUnchecked mutates the iterated segment's end. This operation does not +// invalidate any iterators. +// +// Preconditions: The new end must be valid: end > seg.Start(); if +// seg.NextSegment().Ok(), then end <= seg.NextSegment().Start(). +func (seg LockIterator) SetEndUnchecked(end uint64) { + seg.node.keys[seg.index].End = end +} + +// SetEnd mutates the iterated segment's end. If the new end value would cause +// the iterated segment to overlap another segment, or would result in an +// invalid range, SetEnd panics. This operation does not invalidate any +// iterators. +func (seg LockIterator) SetEnd(end uint64) { + if end <= seg.Start() { + panic(fmt.Sprintf("new end %v would invalidate segment range %v", end, seg.Range())) + } + if next := seg.NextSegment(); next.Ok() && end > next.Start() { + panic(fmt.Sprintf("new end %v would cause segment range %v to overlap segment range %v", end, seg.Range(), next.Range())) + } + seg.SetEndUnchecked(end) +} + +// Value returns a copy of the iterated segment's value. +func (seg LockIterator) Value() Lock { + return seg.node.values[seg.index] +} + +// ValuePtr returns a pointer to the iterated segment's value. The pointer is +// invalidated if the iterator is invalidated. This operation does not +// invalidate any iterators. +func (seg LockIterator) ValuePtr() *Lock { + return &seg.node.values[seg.index] +} + +// SetValue mutates the iterated segment's value. This operation does not +// invalidate any iterators. +func (seg LockIterator) SetValue(val Lock) { + seg.node.values[seg.index] = val +} + +// PrevSegment returns the iterated segment's predecessor. If there is no +// preceding segment, PrevSegment returns a terminal iterator. +func (seg LockIterator) PrevSegment() LockIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index].lastSegment() + } + if seg.index > 0 { + return LockIterator{seg.node, seg.index - 1} + } + if seg.node.parent == nil { + return LockIterator{} + } + return LocksegmentBeforePosition(seg.node.parent, seg.node.parentIndex) +} + +// NextSegment returns the iterated segment's successor. If there is no +// succeeding segment, NextSegment returns a terminal iterator. +func (seg LockIterator) NextSegment() LockIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment() + } + if seg.index < seg.node.nrSegments-1 { + return LockIterator{seg.node, seg.index + 1} + } + if seg.node.parent == nil { + return LockIterator{} + } + return LocksegmentAfterPosition(seg.node.parent, seg.node.parentIndex) +} + +// PrevGap returns the gap immediately before the iterated segment. +func (seg LockIterator) PrevGap() LockGapIterator { + if seg.node.hasChildren { + + return seg.node.children[seg.index].lastSegment().NextGap() + } + return LockGapIterator{seg.node, seg.index} +} + +// NextGap returns the gap immediately after the iterated segment. +func (seg LockIterator) NextGap() LockGapIterator { + if seg.node.hasChildren { + return seg.node.children[seg.index+1].firstSegment().PrevGap() + } + return LockGapIterator{seg.node, seg.index + 1} +} + +// PrevNonEmpty returns the iterated segment's predecessor if it is adjacent, +// or the gap before the iterated segment otherwise. If seg.Start() == +// Functions.MinKey(), PrevNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by PrevNonEmpty will be +// non-terminal. +func (seg LockIterator) PrevNonEmpty() (LockIterator, LockGapIterator) { + gap := seg.PrevGap() + if gap.Range().Length() != 0 { + return LockIterator{}, gap + } + return gap.PrevSegment(), LockGapIterator{} +} + +// NextNonEmpty returns the iterated segment's successor if it is adjacent, or +// the gap after the iterated segment otherwise. If seg.End() == +// Functions.MaxKey(), NextNonEmpty will return two terminal iterators. +// Otherwise, exactly one of the iterators returned by NextNonEmpty will be +// non-terminal. +func (seg LockIterator) NextNonEmpty() (LockIterator, LockGapIterator) { + gap := seg.NextGap() + if gap.Range().Length() != 0 { + return LockIterator{}, gap + } + return gap.NextSegment(), LockGapIterator{} +} + +// A GapIterator is conceptually one of: +// +// - A pointer to a position between two segments, before the first segment, or +// after the last segment in a set, called a *gap*; or +// +// - A terminal iterator, which is a sentinel indicating that the end of +// iteration has been reached. +// +// Note that the gap between two adjacent segments exists (iterators to it are +// non-terminal), but has a length of zero. GapIterator.IsEmpty returns true +// for such gaps. An empty set contains a single gap, spanning the entire range +// of the set's keys. +// +// GapIterators are copyable values and are meaningfully equality-comparable. +// The zero value of GapIterator is a terminal iterator. +// +// Unless otherwise specified, any mutation of a set invalidates all existing +// iterators into the set. +type LockGapIterator struct { + // The representation of a GapIterator is identical to that of an Iterator, + // except that index corresponds to positions between segments in the same + // way as for node.children (see comment for node.nrSegments). + node *Locknode + index int +} + +// Ok returns true if the iterator is not terminal. All other methods are only +// valid for non-terminal iterators. +func (gap LockGapIterator) Ok() bool { + return gap.node != nil +} + +// Range returns the range spanned by the iterated gap. +func (gap LockGapIterator) Range() LockRange { + return LockRange{gap.Start(), gap.End()} +} + +// Start is equivalent to Range().Start, but should be preferred if only the +// start of the range is needed. +func (gap LockGapIterator) Start() uint64 { + if ps := gap.PrevSegment(); ps.Ok() { + return ps.End() + } + return lockSetFunctions{}.MinKey() +} + +// End is equivalent to Range().End, but should be preferred if only the end of +// the range is needed. +func (gap LockGapIterator) End() uint64 { + if ns := gap.NextSegment(); ns.Ok() { + return ns.Start() + } + return lockSetFunctions{}.MaxKey() +} + +// IsEmpty returns true if the iterated gap is empty (that is, the "gap" is +// between two adjacent segments.) +func (gap LockGapIterator) IsEmpty() bool { + return gap.Range().Length() == 0 +} + +// PrevSegment returns the segment immediately before the iterated gap. If no +// such segment exists, PrevSegment returns a terminal iterator. +func (gap LockGapIterator) PrevSegment() LockIterator { + return LocksegmentBeforePosition(gap.node, gap.index) +} + +// NextSegment returns the segment immediately after the iterated gap. If no +// such segment exists, NextSegment returns a terminal iterator. +func (gap LockGapIterator) NextSegment() LockIterator { + return LocksegmentAfterPosition(gap.node, gap.index) +} + +// PrevGap returns the iterated gap's predecessor. If no such gap exists, +// PrevGap returns a terminal iterator. +func (gap LockGapIterator) PrevGap() LockGapIterator { + seg := gap.PrevSegment() + if !seg.Ok() { + return LockGapIterator{} + } + return seg.PrevGap() +} + +// NextGap returns the iterated gap's successor. If no such gap exists, NextGap +// returns a terminal iterator. +func (gap LockGapIterator) NextGap() LockGapIterator { + seg := gap.NextSegment() + if !seg.Ok() { + return LockGapIterator{} + } + return seg.NextGap() +} + +// segmentBeforePosition returns the predecessor segment of the position given +// by n.children[i], which may or may not contain a child. If no such segment +// exists, segmentBeforePosition returns a terminal iterator. +func LocksegmentBeforePosition(n *Locknode, i int) LockIterator { + for i == 0 { + if n.parent == nil { + return LockIterator{} + } + n, i = n.parent, n.parentIndex + } + return LockIterator{n, i - 1} +} + +// segmentAfterPosition returns the successor segment of the position given by +// n.children[i], which may or may not contain a child. If no such segment +// exists, segmentAfterPosition returns a terminal iterator. +func LocksegmentAfterPosition(n *Locknode, i int) LockIterator { + for i == n.nrSegments { + if n.parent == nil { + return LockIterator{} + } + n, i = n.parent, n.parentIndex + } + return LockIterator{n, i} +} + +func LockzeroValueSlice(slice []Lock) { + + for i := range slice { + lockSetFunctions{}.ClearValue(&slice[i]) + } +} + +func LockzeroNodeSlice(slice []*Locknode) { + for i := range slice { + slice[i] = nil + } +} + +// String stringifies a Set for debugging. +func (s *LockSet) String() string { + return s.root.String() +} + +// String stringifies a node (and all of its children) for debugging. +func (n *Locknode) String() string { + var buf bytes.Buffer + n.writeDebugString(&buf, "") + return buf.String() +} + +func (n *Locknode) writeDebugString(buf *bytes.Buffer, prefix string) { + if n.hasChildren != (n.nrSegments > 0 && n.children[0] != nil) { + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent value of hasChildren: got %v, want %v\n", n.hasChildren, !n.hasChildren)) + } + for i := 0; i < n.nrSegments; i++ { + if child := n.children[i]; child != nil { + cprefix := fmt.Sprintf("%s- % 3d ", prefix, i) + if child.parent != n || child.parentIndex != i { + buf.WriteString(cprefix) + buf.WriteString(fmt.Sprintf("WARNING: inconsistent linkage to parent: got (%p, %d), want (%p, %d)\n", child.parent, child.parentIndex, n, i)) + } + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, i)) + } + buf.WriteString(prefix) + buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) + } + if child := n.children[n.nrSegments]; child != nil { + child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, n.nrSegments)) + } +} + +// SegmentDataSlices represents segments from a set as slices of start, end, and +// values. SegmentDataSlices is primarily used as an intermediate representation +// for save/restore and the layout here is optimized for that. +// +// +stateify savable +type LockSegmentDataSlices struct { + Start []uint64 + End []uint64 + Values []Lock +} + +// ExportSortedSlice returns a copy of all segments in the given set, in ascending +// key order. +func (s *LockSet) ExportSortedSlices() *LockSegmentDataSlices { + var sds LockSegmentDataSlices + for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { + sds.Start = append(sds.Start, seg.Start()) + sds.End = append(sds.End, seg.End()) + sds.Values = append(sds.Values, seg.Value()) + } + sds.Start = sds.Start[:len(sds.Start):len(sds.Start)] + sds.End = sds.End[:len(sds.End):len(sds.End)] + sds.Values = sds.Values[:len(sds.Values):len(sds.Values)] + return &sds +} + +// ImportSortedSlice initializes the given set from the given slice. +// +// Preconditions: s must be empty. sds must represent a valid set (the segments +// in sds must have valid lengths that do not overlap). The segments in sds +// must be sorted in ascending key order. +func (s *LockSet) ImportSortedSlices(sds *LockSegmentDataSlices) error { + if !s.IsEmpty() { + return fmt.Errorf("cannot import into non-empty set %v", s) + } + gap := s.FirstGap() + for i := range sds.Start { + r := LockRange{sds.Start[i], sds.End[i]} + if !gap.Range().IsSupersetOf(r) { + return fmt.Errorf("segment overlaps a preceding segment or is incorrectly sorted: [%d, %d) => %v", sds.Start[i], sds.End[i], sds.Values[i]) + } + gap = s.InsertWithoutMerging(gap, r, sds.Values[i]).NextGap() + } + return nil +} +func (s *LockSet) saveRoot() *LockSegmentDataSlices { + return s.ExportSortedSlices() +} + +func (s *LockSet) loadRoot(sds *LockSegmentDataSlices) { + if err := s.ImportSortedSlices(sds); err != nil { + panic(err) + } +} diff --git a/pkg/sentry/fs/lock/lock_state_autogen.go b/pkg/sentry/fs/lock/lock_state_autogen.go new file mode 100755 index 000000000..cb69e2cd0 --- /dev/null +++ b/pkg/sentry/fs/lock/lock_state_autogen.go @@ -0,0 +1,106 @@ +// automatically generated by stateify. + +package lock + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *Lock) beforeSave() {} +func (x *Lock) save(m state.Map) { + x.beforeSave() + m.Save("Readers", &x.Readers) + m.Save("HasWriter", &x.HasWriter) + m.Save("Writer", &x.Writer) +} + +func (x *Lock) afterLoad() {} +func (x *Lock) load(m state.Map) { + m.Load("Readers", &x.Readers) + m.Load("HasWriter", &x.HasWriter) + m.Load("Writer", &x.Writer) +} + +func (x *Locks) beforeSave() {} +func (x *Locks) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.blockedQueue) { m.Failf("blockedQueue is %v, expected zero", x.blockedQueue) } + m.Save("locks", &x.locks) +} + +func (x *Locks) afterLoad() {} +func (x *Locks) load(m state.Map) { + m.Load("locks", &x.locks) +} + +func (x *LockRange) beforeSave() {} +func (x *LockRange) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) +} + +func (x *LockRange) afterLoad() {} +func (x *LockRange) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) +} + +func (x *LockSet) beforeSave() {} +func (x *LockSet) save(m state.Map) { + x.beforeSave() + var root *LockSegmentDataSlices = x.saveRoot() + m.SaveValue("root", root) +} + +func (x *LockSet) afterLoad() {} +func (x *LockSet) load(m state.Map) { + m.LoadValue("root", new(*LockSegmentDataSlices), func(y interface{}) { x.loadRoot(y.(*LockSegmentDataSlices)) }) +} + +func (x *Locknode) beforeSave() {} +func (x *Locknode) save(m state.Map) { + x.beforeSave() + m.Save("nrSegments", &x.nrSegments) + m.Save("parent", &x.parent) + m.Save("parentIndex", &x.parentIndex) + m.Save("hasChildren", &x.hasChildren) + m.Save("keys", &x.keys) + m.Save("values", &x.values) + m.Save("children", &x.children) +} + +func (x *Locknode) afterLoad() {} +func (x *Locknode) load(m state.Map) { + m.Load("nrSegments", &x.nrSegments) + m.Load("parent", &x.parent) + m.Load("parentIndex", &x.parentIndex) + m.Load("hasChildren", &x.hasChildren) + m.Load("keys", &x.keys) + m.Load("values", &x.values) + m.Load("children", &x.children) +} + +func (x *LockSegmentDataSlices) beforeSave() {} +func (x *LockSegmentDataSlices) save(m state.Map) { + x.beforeSave() + m.Save("Start", &x.Start) + m.Save("End", &x.End) + m.Save("Values", &x.Values) +} + +func (x *LockSegmentDataSlices) afterLoad() {} +func (x *LockSegmentDataSlices) load(m state.Map) { + m.Load("Start", &x.Start) + m.Load("End", &x.End) + m.Load("Values", &x.Values) +} + +func init() { + state.Register("lock.Lock", (*Lock)(nil), state.Fns{Save: (*Lock).save, Load: (*Lock).load}) + state.Register("lock.Locks", (*Locks)(nil), state.Fns{Save: (*Locks).save, Load: (*Locks).load}) + state.Register("lock.LockRange", (*LockRange)(nil), state.Fns{Save: (*LockRange).save, Load: (*LockRange).load}) + state.Register("lock.LockSet", (*LockSet)(nil), state.Fns{Save: (*LockSet).save, Load: (*LockSet).load}) + state.Register("lock.Locknode", (*Locknode)(nil), state.Fns{Save: (*Locknode).save, Load: (*Locknode).load}) + state.Register("lock.LockSegmentDataSlices", (*LockSegmentDataSlices)(nil), state.Fns{Save: (*LockSegmentDataSlices).save, Load: (*LockSegmentDataSlices).load}) +} diff --git a/pkg/sentry/fs/lock/lock_test.go b/pkg/sentry/fs/lock/lock_test.go deleted file mode 100644 index ba002aeb7..000000000 --- a/pkg/sentry/fs/lock/lock_test.go +++ /dev/null @@ -1,1059 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package lock - -import ( - "reflect" - "testing" -) - -type entry struct { - Lock - LockRange -} - -func equals(e0, e1 []entry) bool { - if len(e0) != len(e1) { - return false - } - for i := range e0 { - for k := range e0[i].Lock.Readers { - if !e1[i].Lock.Readers[k] { - return false - } - } - for k := range e1[i].Lock.Readers { - if !e0[i].Lock.Readers[k] { - return false - } - } - if !reflect.DeepEqual(e0[i].LockRange, e1[i].LockRange) { - return false - } - if e0[i].Lock.HasWriter != e1[i].Lock.HasWriter { - return false - } - if e0[i].Lock.Writer != e1[i].Lock.Writer { - return false - } - } - return true -} - -// fill a LockSet with consecutive region locks. Will panic if -// LockRanges are not consecutive. -func fill(entries []entry) LockSet { - l := LockSet{} - for _, e := range entries { - gap := l.FindGap(e.LockRange.Start) - if !gap.Ok() { - panic("cannot insert into existing segment") - } - l.Insert(gap, e.LockRange, e.Lock) - } - return l -} - -func TestCanLockEmpty(t *testing.T) { - l := LockSet{} - - // Expect to be able to take any locks given that the set is empty. - eof := l.FirstGap().End() - r := LockRange{0, eof} - if !l.canLock(1, ReadLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", ReadLock, r, 1) - } - if !l.canLock(2, ReadLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", ReadLock, r, 2) - } - if !l.canLock(1, WriteLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", WriteLock, r, 1) - } - if !l.canLock(2, WriteLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", WriteLock, r, 2) - } -} - -func TestCanLock(t *testing.T) { - // + -------------- + ---------- + -------------- + --------- + - // | Readers 1 & 2 | Readers 1 | Readers 1 & 3 | Writer 1 | - // + ------------- + ---------- + -------------- + --------- + - // 0 1024 2048 3072 4096 - l := fill([]entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{1: true, 2: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{1: true}}, - LockRange: LockRange{1024, 2048}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{1: true, 3: true}}, - LockRange: LockRange{2048, 3072}, - }, - { - Lock: Lock{HasWriter: true, Writer: 1}, - LockRange: LockRange{3072, 4096}, - }, - }) - - // Now that we have a mildly interesting layout, try some checks on different - // ranges, uids, and lock types. - // - // Expect to be able to extend the read lock, despite the writer lock, because - // the writer has the same uid as the requested read lock. - r := LockRange{0, 8192} - if !l.canLock(1, ReadLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", ReadLock, r, 1) - } - // Expect to *not* be able to extend the read lock since there is an overlapping - // writer region locked by someone other than the uid. - if l.canLock(2, ReadLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got true, want false", ReadLock, r, 2) - } - // Expect to be able to extend the read lock if there are only other readers in - // the way. - r = LockRange{64, 3072} - if !l.canLock(2, ReadLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", ReadLock, r, 2) - } - // Expect to be able to set a read lock beyond the range of any existing locks. - r = LockRange{4096, 10240} - if !l.canLock(2, ReadLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", ReadLock, r, 2) - } - - // Expect to not be able to take a write lock with other readers in the way. - r = LockRange{0, 8192} - if l.canLock(1, WriteLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got true, want false", WriteLock, r, 1) - } - // Expect to be able to extend the write lock for the same uid. - r = LockRange{3072, 8192} - if !l.canLock(1, WriteLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", WriteLock, r, 1) - } - // Expect to not be able to overlap a write lock for two different uids. - if l.canLock(2, WriteLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got true, want false", WriteLock, r, 2) - } - // Expect to be able to set a write lock that is beyond the range of any - // existing locks. - r = LockRange{8192, 10240} - if !l.canLock(2, WriteLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", WriteLock, r, 2) - } - // Expect to be able to upgrade a read lock (any portion of it). - r = LockRange{1024, 2048} - if !l.canLock(1, WriteLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", WriteLock, r, 1) - } - r = LockRange{1080, 2000} - if !l.canLock(1, WriteLock, r) { - t.Fatalf("canLock type %d for range %v and uid %d got false, want true", WriteLock, r, 1) - } -} - -func TestSetLock(t *testing.T) { - tests := []struct { - // description of test. - name string - - // LockSet entries to pre-fill. - before []entry - - // Description of region to lock: - // - // start is the file offset of the lock. - start uint64 - // end is the end file offset of the lock. - end uint64 - // uid of lock attempter. - uid UniqueID - // lock type requested. - lockType LockType - - // success is true if taking the above - // lock should succeed. - success bool - - // Expected layout of the set after locking - // if success is true. - after []entry - }{ - { - name: "set zero length ReadLock on empty set", - start: 0, - end: 0, - uid: 0, - lockType: ReadLock, - success: true, - }, - { - name: "set zero length WriteLock on empty set", - start: 0, - end: 0, - uid: 0, - lockType: WriteLock, - success: true, - }, - { - name: "set ReadLock on empty set", - start: 0, - end: LockEOF, - uid: 0, - lockType: ReadLock, - success: true, - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - }, - { - name: "set WriteLock on empty set", - start: 0, - end: LockEOF, - uid: 0, - lockType: WriteLock, - success: true, - // + ----------------------------------------- + - // | Writer 0 | - // + ----------------------------------------- + - // 0 max uint64 - after: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, LockEOF}, - }, - }, - }, - { - name: "set ReadLock on WriteLock same uid", - // + ----------------------------------------- + - // | Writer 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: 4096, - uid: 0, - lockType: ReadLock, - success: true, - // + ----------- + --------------------------- + - // | Readers 0 | Writer 0 | - // + ----------- + --------------------------- + - // 0 4096 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, 4096}, - }, - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "set WriteLock on ReadLock same uid", - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: 4096, - uid: 0, - lockType: WriteLock, - success: true, - // + ----------- + --------------------------- + - // | Writer 0 | Readers 0 | - // + ----------- + --------------------------- + - // 0 4096 max uint64 - after: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, 4096}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "set ReadLock on WriteLock different uid", - // + ----------------------------------------- + - // | Writer 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: 4096, - uid: 1, - lockType: ReadLock, - success: false, - }, - { - name: "set WriteLock on ReadLock different uid", - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: 4096, - uid: 1, - lockType: WriteLock, - success: false, - }, - { - name: "split ReadLock for overlapping lock at start 0", - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: 4096, - uid: 1, - lockType: ReadLock, - success: true, - // + -------------- + --------------------------- + - // | Readers 0 & 1 | Readers 0 | - // + -------------- + --------------------------- + - // 0 4096 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{0, 4096}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "split ReadLock for overlapping lock at non-zero start", - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 4096, - end: 8192, - uid: 1, - lockType: ReadLock, - success: true, - // + ---------- + -------------- + ----------- + - // | Readers 0 | Readers 0 & 1 | Readers 0 | - // + ---------- + -------------- + ----------- + - // 0 4096 8192 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, 4096}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{4096, 8192}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{8192, LockEOF}, - }, - }, - }, - { - name: "fill front gap with ReadLock", - // + --------- + ---------------------------- + - // | gap | Readers 0 | - // + --------- + ---------------------------- + - // 0 1024 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{1024, LockEOF}, - }, - }, - start: 0, - end: 8192, - uid: 0, - lockType: ReadLock, - success: true, - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - }, - { - name: "fill end gap with ReadLock", - // + ---------------------------- + - // | Readers 0 | - // + ---------------------------- + - // 0 4096 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, 4096}, - }, - }, - start: 1024, - end: LockEOF, - uid: 0, - lockType: ReadLock, - success: true, - // Note that this is not merged after lock does a Split. This is - // fine because the two locks will still *behave* as one. In other - // words we can fragment any lock all we want and semantically it - // makes no difference. - // - // + ----------- + --------------------------- + - // | Readers 0 | Readers 0 | - // + ----------- + --------------------------- + - // 0 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{1024, LockEOF}, - }, - }, - }, - { - name: "fill gap with ReadLock and split", - // + --------- + ---------------------------- + - // | gap | Readers 0 | - // + --------- + ---------------------------- + - // 0 1024 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{1024, LockEOF}, - }, - }, - start: 0, - end: 4096, - uid: 1, - lockType: ReadLock, - success: true, - // + --------- + ------------- + ------------- + - // | Reader 1 | Readers 0 & 1 | Reader 0 | - // + ----------+ ------------- + ------------- + - // 0 1024 4096 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{1: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{1024, 4096}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "upgrade ReadLock to WriteLock for single uid fill gap", - // + ------------- + --------- + --- + ------------- + - // | Readers 0 & 1 | Readers 0 | gap | Readers 0 & 2 | - // + ------------- + --------- + --- + ------------- + - // 0 1024 2048 4096 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{1024, 2048}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 2: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - start: 1024, - end: 4096, - uid: 0, - lockType: WriteLock, - success: true, - // + ------------- + -------- + ------------- + - // | Readers 0 & 1 | Writer 0 | Readers 0 & 2 | - // + ------------- + -------- + ------------- + - // 0 1024 4096 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{1024, 4096}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 2: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "upgrade ReadLock to WriteLock for single uid keep gap", - // + ------------- + --------- + --- + ------------- + - // | Readers 0 & 1 | Readers 0 | gap | Readers 0 & 2 | - // + ------------- + --------- + --- + ------------- + - // 0 1024 2048 4096 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{1024, 2048}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 2: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - start: 1024, - end: 3072, - uid: 0, - lockType: WriteLock, - success: true, - // + ------------- + -------- + --- + ------------- + - // | Readers 0 & 1 | Writer 0 | gap | Readers 0 & 2 | - // + ------------- + -------- + --- + ------------- + - // 0 1024 3072 4096 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{1024, 3072}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 2: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "fail to upgrade ReadLock to WriteLock with conflicting Reader", - // + ------------- + --------- + - // | Readers 0 & 1 | Readers 0 | - // + ------------- + --------- + - // 0 1024 2048 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{1024, 2048}, - }, - }, - start: 0, - end: 2048, - uid: 0, - lockType: WriteLock, - success: false, - }, - { - name: "take WriteLock on whole file if all uids are the same", - // + ------------- + --------- + --------- + ---------- + - // | Writer 0 | Readers 0 | Readers 0 | Readers 0 | - // + ------------- + --------- + --------- + ---------- + - // 0 1024 2048 4096 max uint64 - before: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{1024, 2048}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{2048, 4096}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - start: 0, - end: LockEOF, - uid: 0, - lockType: WriteLock, - success: true, - // We do not manually merge locks. Semantically a fragmented lock - // held by the same uid will behave as one lock so it makes no difference. - // - // + ------------- + ---------------------------- + - // | Writer 0 | Writer 0 | - // + ------------- + ---------------------------- + - // 0 1024 max uint64 - after: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{1024, LockEOF}, - }, - }, - }, - } - - for _, test := range tests { - l := fill(test.before) - - r := LockRange{Start: test.start, End: test.end} - success := l.lock(test.uid, test.lockType, r) - var got []entry - for seg := l.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { - got = append(got, entry{ - Lock: seg.Value(), - LockRange: seg.Range(), - }) - } - - if success != test.success { - t.Errorf("%s: setlock(%v, %+v, %d, %d) got success %v, want %v", test.name, test.before, r, test.uid, test.lockType, success, test.success) - continue - } - - if success { - if !equals(got, test.after) { - t.Errorf("%s: got set %+v, want %+v", test.name, got, test.after) - } - } - } -} - -func TestUnlock(t *testing.T) { - tests := []struct { - // description of test. - name string - - // LockSet entries to pre-fill. - before []entry - - // Description of region to unlock: - // - // start is the file start of the lock. - start uint64 - // end is the end file start of the lock. - end uint64 - // uid of lock holder. - uid UniqueID - - // Expected layout of the set after unlocking. - after []entry - }{ - { - name: "unlock zero length on empty set", - start: 0, - end: 0, - uid: 0, - }, - { - name: "unlock on empty set (no-op)", - start: 0, - end: LockEOF, - uid: 0, - }, - { - name: "unlock uid not locked (no-op)", - // + --------------------------- + - // | Readers 1 & 2 | - // + --------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{1: true, 2: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 1024, - end: 4096, - uid: 0, - // + --------------------------- + - // | Readers 1 & 2 | - // + --------------------------- + - // 0 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{1: true, 2: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - }, - { - name: "unlock ReadLock over entire file", - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: LockEOF, - uid: 0, - }, - { - name: "unlock WriteLock over entire file", - // + ----------------------------------------- + - // | Writer 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: LockEOF, - uid: 0, - }, - { - name: "unlock partial ReadLock (start)", - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: 4096, - uid: 0, - // + ------ + --------------------------- + - // | gap | Readers 0 | - // +------- + --------------------------- + - // 0 4096 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "unlock partial WriteLock (start)", - // + ----------------------------------------- + - // | Writer 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 0, - end: 4096, - uid: 0, - // + ------ + --------------------------- + - // | gap | Writer 0 | - // +------- + --------------------------- + - // 0 4096 max uint64 - after: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "unlock partial ReadLock (end)", - // + ----------------------------------------- + - // | Readers 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 4096, - end: LockEOF, - uid: 0, - // + --------------------------- + - // | Readers 0 | - // +---------------------------- + - // 0 4096 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true}}, - LockRange: LockRange{0, 4096}, - }, - }, - }, - { - name: "unlock partial WriteLock (end)", - // + ----------------------------------------- + - // | Writer 0 | - // + ----------------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 4096, - end: LockEOF, - uid: 0, - // + --------------------------- + - // | Writer 0 | - // +---------------------------- + - // 0 4096 - after: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, 4096}, - }, - }, - }, - { - name: "unlock for single uid", - // + ------------- + --------- + ------------------- + - // | Readers 0 & 1 | Writer 0 | Readers 0 & 1 & 2 | - // + ------------- + --------- + ------------------- + - // 0 1024 4096 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{1024, 4096}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true, 2: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - start: 0, - end: LockEOF, - uid: 0, - // + --------- + --- + --------------- + - // | Readers 1 | gap | Readers 1 & 2 | - // + --------- + --- + --------------- + - // 0 1024 4096 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{1: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{1: true, 2: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "unlock subsection locked", - // + ------------------------------- + - // | Readers 0 & 1 & 2 | - // + ------------------------------- + - // 0 max uint64 - before: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true, 2: true}}, - LockRange: LockRange{0, LockEOF}, - }, - }, - start: 1024, - end: 4096, - uid: 0, - // + ----------------- + ------------- + ----------------- + - // | Readers 0 & 1 & 2 | Readers 1 & 2 | Readers 0 & 1 & 2 | - // + ----------------- + ------------- + ----------------- + - // 0 1024 4096 max uint64 - after: []entry{ - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true, 2: true}}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{1: true, 2: true}}, - LockRange: LockRange{1024, 4096}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true, 2: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "unlock mid-gap to increase gap", - // + --------- + ----- + ------------------- + - // | Writer 0 | gap | Readers 0 & 1 | - // + --------- + ----- + ------------------- + - // 0 1024 4096 max uint64 - before: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - start: 8, - end: 2048, - uid: 0, - // + --------- + ----- + ------------------- + - // | Writer 0 | gap | Readers 0 & 1 | - // + --------- + ----- + ------------------- + - // 0 8 4096 max uint64 - after: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, 8}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - }, - { - name: "unlock split region on uid mid-gap", - // + --------- + ----- + ------------------- + - // | Writer 0 | gap | Readers 0 & 1 | - // + --------- + ----- + ------------------- + - // 0 1024 4096 max uint64 - before: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{4096, LockEOF}, - }, - }, - start: 2048, - end: 8192, - uid: 0, - // + --------- + ----- + --------- + ------------- + - // | Writer 0 | gap | Readers 1 | Readers 0 & 1 | - // + --------- + ----- + --------- + ------------- + - // 0 1024 4096 8192 max uint64 - after: []entry{ - { - Lock: Lock{HasWriter: true, Writer: 0}, - LockRange: LockRange{0, 1024}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{1: true}}, - LockRange: LockRange{4096, 8192}, - }, - { - Lock: Lock{Readers: map[UniqueID]bool{0: true, 1: true}}, - LockRange: LockRange{8192, LockEOF}, - }, - }, - }, - } - - for _, test := range tests { - l := fill(test.before) - - r := LockRange{Start: test.start, End: test.end} - l.unlock(test.uid, r) - var got []entry - for seg := l.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { - got = append(got, entry{ - Lock: seg.Value(), - LockRange: seg.Range(), - }) - } - if !equals(got, test.after) { - t.Errorf("%s: got set %+v, want %+v", test.name, got, test.after) - } - } -} diff --git a/pkg/sentry/fs/mount_test.go b/pkg/sentry/fs/mount_test.go deleted file mode 100644 index 0b84732aa..000000000 --- a/pkg/sentry/fs/mount_test.go +++ /dev/null @@ -1,271 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fs - -import ( - "fmt" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" -) - -// cacheReallyContains iterates through the dirent cache to determine whether -// it contains the given dirent. -func cacheReallyContains(cache *DirentCache, d *Dirent) bool { - for i := cache.list.Front(); i != nil; i = i.Next() { - if i == d { - return true - } - } - return false -} - -func mountPathsAre(root *Dirent, got []*Mount, want ...string) error { - gotPaths := make(map[string]struct{}, len(got)) - gotStr := make([]string, len(got)) - for i, g := range got { - groot := g.Root() - name, _ := groot.FullName(root) - groot.DecRef() - gotStr[i] = name - gotPaths[name] = struct{}{} - } - if len(got) != len(want) { - return fmt.Errorf("mount paths are different, got: %q, want: %q", gotStr, want) - } - for _, w := range want { - if _, ok := gotPaths[w]; !ok { - return fmt.Errorf("no mount with path %q found", w) - } - } - return nil -} - -// TestMountSourceOnlyCachedOnce tests that a Dirent that is mounted over only ends -// up in a single Dirent Cache. NOTE(b/63848693): Having a dirent in multiple -// caches causes major consistency issues. -func TestMountSourceOnlyCachedOnce(t *testing.T) { - ctx := contexttest.Context(t) - - rootCache := NewDirentCache(100) - rootInode := NewMockInode(ctx, NewMockMountSource(rootCache), StableAttr{ - Type: Directory, - }) - mm, err := NewMountNamespace(ctx, rootInode) - if err != nil { - t.Fatalf("NewMountNamespace failed: %v", err) - } - rootDirent := mm.Root() - defer rootDirent.DecRef() - - // Get a child of the root which we will mount over. Note that the - // MockInodeOperations causes Walk to always succeed. - child, err := rootDirent.Walk(ctx, rootDirent, "child") - if err != nil { - t.Fatalf("failed to walk to child dirent: %v", err) - } - child.maybeExtendReference() // Cache. - - // Ensure that the root cache contains the child. - if !cacheReallyContains(rootCache, child) { - t.Errorf("wanted rootCache to contain child dirent, but it did not") - } - - // Create a new cache and inode, and mount it over child. - submountCache := NewDirentCache(100) - submountInode := NewMockInode(ctx, NewMockMountSource(submountCache), StableAttr{ - Type: Directory, - }) - if err := mm.Mount(ctx, child, submountInode); err != nil { - t.Fatalf("failed to mount over child: %v", err) - } - - // Walk to the child again. - child2, err := rootDirent.Walk(ctx, rootDirent, "child") - if err != nil { - t.Fatalf("failed to walk to child dirent: %v", err) - } - - // Should have a different Dirent than before. - if child == child2 { - t.Fatalf("expected %v not equal to %v, but they are the same", child, child2) - } - - // Neither of the caches should no contain the child. - if cacheReallyContains(rootCache, child) { - t.Errorf("wanted rootCache not to contain child dirent, but it did") - } - if cacheReallyContains(submountCache, child) { - t.Errorf("wanted submountCache not to contain child dirent, but it did") - } -} - -func TestAllMountsUnder(t *testing.T) { - ctx := contexttest.Context(t) - - rootCache := NewDirentCache(100) - rootInode := NewMockInode(ctx, NewMockMountSource(rootCache), StableAttr{ - Type: Directory, - }) - mm, err := NewMountNamespace(ctx, rootInode) - if err != nil { - t.Fatalf("NewMountNamespace failed: %v", err) - } - rootDirent := mm.Root() - defer rootDirent.DecRef() - - // Add mounts at the following paths: - paths := []string{ - "/foo", - "/foo/bar", - "/foo/bar/baz", - "/foo/qux", - "/waldo", - } - - var maxTraversals uint - for _, p := range paths { - maxTraversals = 0 - d, err := mm.FindLink(ctx, rootDirent, nil, p, &maxTraversals) - if err != nil { - t.Fatalf("could not find path %q in mount manager: %v", p, err) - } - - submountInode := NewMockInode(ctx, NewMockMountSource(nil), StableAttr{ - Type: Directory, - }) - if err := mm.Mount(ctx, d, submountInode); err != nil { - t.Fatalf("could not mount at %q: %v", p, err) - } - d.DecRef() - } - - // mm root should contain all submounts (and does not include the root mount). - rootMnt := mm.FindMount(rootDirent) - submounts := mm.AllMountsUnder(rootMnt) - allPaths := append(paths, "/") - if err := mountPathsAre(rootDirent, submounts, allPaths...); err != nil { - t.Error(err) - } - - // Each mount should have a unique ID. - foundIDs := make(map[uint64]struct{}) - for _, m := range submounts { - if _, ok := foundIDs[m.ID]; ok { - t.Errorf("got multiple mounts with id %d", m.ID) - } - foundIDs[m.ID] = struct{}{} - } - - // Root mount should have no parent. - if p := rootMnt.ParentID; p != invalidMountID { - t.Errorf("root.Parent got %v wanted nil", p) - } - - // Check that "foo" mount has 3 children. - maxTraversals = 0 - d, err := mm.FindLink(ctx, rootDirent, nil, "/foo", &maxTraversals) - if err != nil { - t.Fatalf("could not find path %q in mount manager: %v", "/foo", err) - } - defer d.DecRef() - submounts = mm.AllMountsUnder(mm.FindMount(d)) - if err := mountPathsAre(rootDirent, submounts, "/foo", "/foo/bar", "/foo/qux", "/foo/bar/baz"); err != nil { - t.Error(err) - } - - // "waldo" mount should have no children. - maxTraversals = 0 - waldo, err := mm.FindLink(ctx, rootDirent, nil, "/waldo", &maxTraversals) - if err != nil { - t.Fatalf("could not find path %q in mount manager: %v", "/waldo", err) - } - defer waldo.DecRef() - submounts = mm.AllMountsUnder(mm.FindMount(waldo)) - if err := mountPathsAre(rootDirent, submounts, "/waldo"); err != nil { - t.Error(err) - } -} - -func TestUnmount(t *testing.T) { - ctx := contexttest.Context(t) - - rootCache := NewDirentCache(100) - rootInode := NewMockInode(ctx, NewMockMountSource(rootCache), StableAttr{ - Type: Directory, - }) - mm, err := NewMountNamespace(ctx, rootInode) - if err != nil { - t.Fatalf("NewMountNamespace failed: %v", err) - } - rootDirent := mm.Root() - defer rootDirent.DecRef() - - // Add mounts at the following paths: - paths := []string{ - "/foo", - "/foo/bar", - "/foo/bar/goo", - "/foo/bar/goo/abc", - "/foo/abc", - "/foo/def", - "/waldo", - "/wally", - } - - var maxTraversals uint - for _, p := range paths { - maxTraversals = 0 - d, err := mm.FindLink(ctx, rootDirent, nil, p, &maxTraversals) - if err != nil { - t.Fatalf("could not find path %q in mount manager: %v", p, err) - } - - submountInode := NewMockInode(ctx, NewMockMountSource(nil), StableAttr{ - Type: Directory, - }) - if err := mm.Mount(ctx, d, submountInode); err != nil { - t.Fatalf("could not mount at %q: %v", p, err) - } - d.DecRef() - } - - allPaths := make([]string, len(paths)+1) - allPaths[0] = "/" - copy(allPaths[1:], paths) - - rootMnt := mm.FindMount(rootDirent) - for i := len(paths) - 1; i >= 0; i-- { - maxTraversals = 0 - p := paths[i] - d, err := mm.FindLink(ctx, rootDirent, nil, p, &maxTraversals) - if err != nil { - t.Fatalf("could not find path %q in mount manager: %v", p, err) - } - - if err := mm.Unmount(ctx, d, false); err != nil { - t.Fatalf("could not unmount at %q: %v", p, err) - } - d.DecRef() - - // Remove the path that has been unmounted and the check that the remaining - // mounts are still there. - allPaths = allPaths[:len(allPaths)-1] - submounts := mm.AllMountsUnder(rootMnt) - if err := mountPathsAre(rootDirent, submounts, allPaths...); err != nil { - t.Error(err) - } - } -} diff --git a/pkg/sentry/fs/mounts_test.go b/pkg/sentry/fs/mounts_test.go deleted file mode 100644 index c4c771f2c..000000000 --- a/pkg/sentry/fs/mounts_test.go +++ /dev/null @@ -1,105 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fs_test - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" - "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" -) - -// Creates a new MountNamespace with filesystem: -// / (root dir) -// |-foo (dir) -// |-bar (file) -func createMountNamespace(ctx context.Context) (*fs.MountNamespace, error) { - perms := fs.FilePermsFromMode(0777) - m := fs.NewPseudoMountSource(ctx) - - barFile := fsutil.NewSimpleFileInode(ctx, fs.RootOwner, perms, 0) - fooDir := ramfs.NewDir(ctx, map[string]*fs.Inode{ - "bar": fs.NewInode(ctx, barFile, m, fs.StableAttr{Type: fs.RegularFile}), - }, fs.RootOwner, perms) - rootDir := ramfs.NewDir(ctx, map[string]*fs.Inode{ - "foo": fs.NewInode(ctx, fooDir, m, fs.StableAttr{Type: fs.Directory}), - }, fs.RootOwner, perms) - - return fs.NewMountNamespace(ctx, fs.NewInode(ctx, rootDir, m, fs.StableAttr{Type: fs.Directory})) -} - -func TestFindLink(t *testing.T) { - ctx := contexttest.Context(t) - mm, err := createMountNamespace(ctx) - if err != nil { - t.Fatalf("createMountNamespace failed: %v", err) - } - - root := mm.Root() - defer root.DecRef() - foo, err := root.Walk(ctx, root, "foo") - if err != nil { - t.Fatalf("Error walking to foo: %v", err) - } - - // Positive cases. - for _, tc := range []struct { - findPath string - wd *fs.Dirent - wantPath string - }{ - {".", root, "/"}, - {".", foo, "/foo"}, - {"..", foo, "/"}, - {"../../..", foo, "/"}, - {"///foo", foo, "/foo"}, - {"/foo", foo, "/foo"}, - {"/foo/bar", foo, "/foo/bar"}, - {"/foo/.///./bar", foo, "/foo/bar"}, - {"/foo///bar", foo, "/foo/bar"}, - {"/foo/../foo/bar", foo, "/foo/bar"}, - {"foo/bar", root, "/foo/bar"}, - {"foo////bar", root, "/foo/bar"}, - {"bar", foo, "/foo/bar"}, - } { - wdPath, _ := tc.wd.FullName(root) - maxTraversals := uint(0) - if d, err := mm.FindLink(ctx, root, tc.wd, tc.findPath, &maxTraversals); err != nil { - t.Errorf("FindLink(%q, wd=%q) failed: %v", tc.findPath, wdPath, err) - } else if got, _ := d.FullName(root); got != tc.wantPath { - t.Errorf("FindLink(%q, wd=%q) got dirent %q, want %q", tc.findPath, wdPath, got, tc.wantPath) - } - } - - // Negative cases. - for _, tc := range []struct { - findPath string - wd *fs.Dirent - }{ - {"bar", root}, - {"/bar", root}, - {"/foo/../../bar", root}, - {"foo", foo}, - } { - wdPath, _ := tc.wd.FullName(root) - maxTraversals := uint(0) - if _, err := mm.FindLink(ctx, root, tc.wd, tc.findPath, &maxTraversals); err == nil { - t.Errorf("FindLink(%q, wd=%q) did not return error", tc.findPath, wdPath) - } - } -} diff --git a/pkg/sentry/fs/path_test.go b/pkg/sentry/fs/path_test.go deleted file mode 100644 index e6f57ebba..000000000 --- a/pkg/sentry/fs/path_test.go +++ /dev/null @@ -1,289 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fs - -import ( - "testing" -) - -// TestSplitLast tests variants of path splitting. -func TestSplitLast(t *testing.T) { - cases := []struct { - path string - dir string - file string - }{ - {path: "/", dir: "/", file: "."}, - {path: "/.", dir: "/", file: "."}, - {path: "/./", dir: "/", file: "."}, - {path: "/./.", dir: "/.", file: "."}, - {path: "/././", dir: "/.", file: "."}, - {path: "/./..", dir: "/.", file: ".."}, - {path: "/./../", dir: "/.", file: ".."}, - {path: "/..", dir: "/", file: ".."}, - {path: "/../", dir: "/", file: ".."}, - {path: "/../.", dir: "/..", file: "."}, - {path: "/.././", dir: "/..", file: "."}, - {path: "/../..", dir: "/..", file: ".."}, - {path: "/../../", dir: "/..", file: ".."}, - - {path: "", dir: ".", file: "."}, - {path: ".", dir: ".", file: "."}, - {path: "./", dir: ".", file: "."}, - {path: "./.", dir: ".", file: "."}, - {path: "././", dir: ".", file: "."}, - {path: "./..", dir: ".", file: ".."}, - {path: "./../", dir: ".", file: ".."}, - {path: "..", dir: ".", file: ".."}, - {path: "../", dir: ".", file: ".."}, - {path: "../.", dir: "..", file: "."}, - {path: ".././", dir: "..", file: "."}, - {path: "../..", dir: "..", file: ".."}, - {path: "../../", dir: "..", file: ".."}, - - {path: "/foo", dir: "/", file: "foo"}, - {path: "/foo/", dir: "/", file: "foo"}, - {path: "/foo/.", dir: "/foo", file: "."}, - {path: "/foo/./", dir: "/foo", file: "."}, - {path: "/foo/./.", dir: "/foo/.", file: "."}, - {path: "/foo/./..", dir: "/foo/.", file: ".."}, - {path: "/foo/..", dir: "/foo", file: ".."}, - {path: "/foo/../", dir: "/foo", file: ".."}, - {path: "/foo/../.", dir: "/foo/..", file: "."}, - {path: "/foo/../..", dir: "/foo/..", file: ".."}, - - {path: "/foo/bar", dir: "/foo", file: "bar"}, - {path: "/foo/bar/", dir: "/foo", file: "bar"}, - {path: "/foo/bar/.", dir: "/foo/bar", file: "."}, - {path: "/foo/bar/./", dir: "/foo/bar", file: "."}, - {path: "/foo/bar/./.", dir: "/foo/bar/.", file: "."}, - {path: "/foo/bar/./..", dir: "/foo/bar/.", file: ".."}, - {path: "/foo/bar/..", dir: "/foo/bar", file: ".."}, - {path: "/foo/bar/../", dir: "/foo/bar", file: ".."}, - {path: "/foo/bar/../.", dir: "/foo/bar/..", file: "."}, - {path: "/foo/bar/../..", dir: "/foo/bar/..", file: ".."}, - - {path: "foo", dir: ".", file: "foo"}, - {path: "foo", dir: ".", file: "foo"}, - {path: "foo/", dir: ".", file: "foo"}, - {path: "foo/.", dir: "foo", file: "."}, - {path: "foo/./", dir: "foo", file: "."}, - {path: "foo/./.", dir: "foo/.", file: "."}, - {path: "foo/./..", dir: "foo/.", file: ".."}, - {path: "foo/..", dir: "foo", file: ".."}, - {path: "foo/../", dir: "foo", file: ".."}, - {path: "foo/../.", dir: "foo/..", file: "."}, - {path: "foo/../..", dir: "foo/..", file: ".."}, - {path: "foo/", dir: ".", file: "foo"}, - {path: "foo/.", dir: "foo", file: "."}, - - {path: "foo/bar", dir: "foo", file: "bar"}, - {path: "foo/bar/", dir: "foo", file: "bar"}, - {path: "foo/bar/.", dir: "foo/bar", file: "."}, - {path: "foo/bar/./", dir: "foo/bar", file: "."}, - {path: "foo/bar/./.", dir: "foo/bar/.", file: "."}, - {path: "foo/bar/./..", dir: "foo/bar/.", file: ".."}, - {path: "foo/bar/..", dir: "foo/bar", file: ".."}, - {path: "foo/bar/../", dir: "foo/bar", file: ".."}, - {path: "foo/bar/../.", dir: "foo/bar/..", file: "."}, - {path: "foo/bar/../..", dir: "foo/bar/..", file: ".."}, - {path: "foo/bar/", dir: "foo", file: "bar"}, - {path: "foo/bar/.", dir: "foo/bar", file: "."}, - } - - for _, c := range cases { - dir, file := SplitLast(c.path) - if dir != c.dir || file != c.file { - t.Errorf("SplitLast(%q) got (%q, %q), expected (%q, %q)", c.path, dir, file, c.dir, c.file) - } - } -} - -// TestSplitFirst tests variants of path splitting. -func TestSplitFirst(t *testing.T) { - cases := []struct { - path string - first string - remainder string - }{ - {path: "/", first: "/", remainder: ""}, - {path: "/.", first: "/", remainder: "."}, - {path: "///.", first: "/", remainder: "//."}, - {path: "/.///", first: "/", remainder: "."}, - {path: "/./.", first: "/", remainder: "./."}, - {path: "/././", first: "/", remainder: "./."}, - {path: "/./..", first: "/", remainder: "./.."}, - {path: "/./../", first: "/", remainder: "./.."}, - {path: "/..", first: "/", remainder: ".."}, - {path: "/../", first: "/", remainder: ".."}, - {path: "/../.", first: "/", remainder: "../."}, - {path: "/.././", first: "/", remainder: "../."}, - {path: "/../..", first: "/", remainder: "../.."}, - {path: "/../../", first: "/", remainder: "../.."}, - - {path: "", first: ".", remainder: ""}, - {path: ".", first: ".", remainder: ""}, - {path: "./", first: ".", remainder: ""}, - {path: ".///", first: ".", remainder: ""}, - {path: "./.", first: ".", remainder: "."}, - {path: "././", first: ".", remainder: "."}, - {path: "./..", first: ".", remainder: ".."}, - {path: "./../", first: ".", remainder: ".."}, - {path: "..", first: "..", remainder: ""}, - {path: "../", first: "..", remainder: ""}, - {path: "../.", first: "..", remainder: "."}, - {path: ".././", first: "..", remainder: "."}, - {path: "../..", first: "..", remainder: ".."}, - {path: "../../", first: "..", remainder: ".."}, - - {path: "/foo", first: "/", remainder: "foo"}, - {path: "/foo/", first: "/", remainder: "foo"}, - {path: "/foo///", first: "/", remainder: "foo"}, - {path: "/foo/.", first: "/", remainder: "foo/."}, - {path: "/foo/./", first: "/", remainder: "foo/."}, - {path: "/foo/./.", first: "/", remainder: "foo/./."}, - {path: "/foo/./..", first: "/", remainder: "foo/./.."}, - {path: "/foo/..", first: "/", remainder: "foo/.."}, - {path: "/foo/../", first: "/", remainder: "foo/.."}, - {path: "/foo/../.", first: "/", remainder: "foo/../."}, - {path: "/foo/../..", first: "/", remainder: "foo/../.."}, - - {path: "/foo/bar", first: "/", remainder: "foo/bar"}, - {path: "///foo/bar", first: "/", remainder: "//foo/bar"}, - {path: "/foo///bar", first: "/", remainder: "foo///bar"}, - {path: "/foo/bar/.", first: "/", remainder: "foo/bar/."}, - {path: "/foo/bar/./", first: "/", remainder: "foo/bar/."}, - {path: "/foo/bar/./.", first: "/", remainder: "foo/bar/./."}, - {path: "/foo/bar/./..", first: "/", remainder: "foo/bar/./.."}, - {path: "/foo/bar/..", first: "/", remainder: "foo/bar/.."}, - {path: "/foo/bar/../", first: "/", remainder: "foo/bar/.."}, - {path: "/foo/bar/../.", first: "/", remainder: "foo/bar/../."}, - {path: "/foo/bar/../..", first: "/", remainder: "foo/bar/../.."}, - - {path: "foo", first: "foo", remainder: ""}, - {path: "foo", first: "foo", remainder: ""}, - {path: "foo/", first: "foo", remainder: ""}, - {path: "foo///", first: "foo", remainder: ""}, - {path: "foo/.", first: "foo", remainder: "."}, - {path: "foo/./", first: "foo", remainder: "."}, - {path: "foo/./.", first: "foo", remainder: "./."}, - {path: "foo/./..", first: "foo", remainder: "./.."}, - {path: "foo/..", first: "foo", remainder: ".."}, - {path: "foo/../", first: "foo", remainder: ".."}, - {path: "foo/../.", first: "foo", remainder: "../."}, - {path: "foo/../..", first: "foo", remainder: "../.."}, - {path: "foo/", first: "foo", remainder: ""}, - {path: "foo/.", first: "foo", remainder: "."}, - - {path: "foo/bar", first: "foo", remainder: "bar"}, - {path: "foo///bar", first: "foo", remainder: "bar"}, - {path: "foo/bar/", first: "foo", remainder: "bar"}, - {path: "foo/bar/.", first: "foo", remainder: "bar/."}, - {path: "foo/bar/./", first: "foo", remainder: "bar/."}, - {path: "foo/bar/./.", first: "foo", remainder: "bar/./."}, - {path: "foo/bar/./..", first: "foo", remainder: "bar/./.."}, - {path: "foo/bar/..", first: "foo", remainder: "bar/.."}, - {path: "foo/bar/../", first: "foo", remainder: "bar/.."}, - {path: "foo/bar/../.", first: "foo", remainder: "bar/../."}, - {path: "foo/bar/../..", first: "foo", remainder: "bar/../.."}, - {path: "foo/bar/", first: "foo", remainder: "bar"}, - {path: "foo/bar/.", first: "foo", remainder: "bar/."}, - } - - for _, c := range cases { - first, remainder := SplitFirst(c.path) - if first != c.first || remainder != c.remainder { - t.Errorf("SplitFirst(%q) got (%q, %q), expected (%q, %q)", c.path, first, remainder, c.first, c.remainder) - } - } -} - -// TestIsSubpath tests the IsSubpath method. -func TestIsSubpath(t *testing.T) { - tcs := []struct { - // Two absolute paths. - pathA string - pathB string - - // Whether pathA is a subpath of pathB. - wantIsSubpath bool - - // Relative path from pathA to pathB. Only checked if - // wantIsSubpath is true. - wantRelpath string - }{ - { - pathA: "/foo/bar/baz", - pathB: "/foo", - wantIsSubpath: true, - wantRelpath: "bar/baz", - }, - { - pathA: "/foo", - pathB: "/foo/bar/baz", - wantIsSubpath: false, - }, - { - pathA: "/foo", - pathB: "/foo", - wantIsSubpath: false, - }, - { - pathA: "/foobar", - pathB: "/foo", - wantIsSubpath: false, - }, - { - pathA: "/foo", - pathB: "/foobar", - wantIsSubpath: false, - }, - { - pathA: "/foo", - pathB: "/foobar", - wantIsSubpath: false, - }, - { - pathA: "/", - pathB: "/foo", - wantIsSubpath: false, - }, - { - pathA: "/foo", - pathB: "/", - wantIsSubpath: true, - wantRelpath: "foo", - }, - { - pathA: "/foo/bar/../bar", - pathB: "/foo", - wantIsSubpath: true, - wantRelpath: "bar", - }, - { - pathA: "/foo/bar", - pathB: "/foo/../foo", - wantIsSubpath: true, - wantRelpath: "bar", - }, - } - - for _, tc := range tcs { - gotRelpath, gotIsSubpath := IsSubpath(tc.pathA, tc.pathB) - if gotRelpath != tc.wantRelpath || gotIsSubpath != tc.wantIsSubpath { - t.Errorf("IsSubpath(%q, %q) got %q %t, want %q %t", tc.pathA, tc.pathB, gotRelpath, gotIsSubpath, tc.wantRelpath, tc.wantIsSubpath) - } - } -} diff --git a/pkg/sentry/fs/proc/BUILD b/pkg/sentry/fs/proc/BUILD deleted file mode 100644 index 70ed854a8..000000000 --- a/pkg/sentry/fs/proc/BUILD +++ /dev/null @@ -1,73 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_stateify:defs.bzl", "go_library", "go_test") - -go_library( - name = "proc", - srcs = [ - "cgroup.go", - "cpuinfo.go", - "exec_args.go", - "fds.go", - "filesystems.go", - "fs.go", - "inode.go", - "loadavg.go", - "meminfo.go", - "mounts.go", - "net.go", - "proc.go", - "rpcinet_proc.go", - "stat.go", - "sys.go", - "sys_net.go", - "sys_net_state.go", - "task.go", - "uid_gid_map.go", - "uptime.go", - "version.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/proc", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/log", - "//pkg/sentry/context", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fs/proc/device", - "//pkg/sentry/fs/proc/seqfile", - "//pkg/sentry/fs/ramfs", - "//pkg/sentry/inet", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sentry/limits", - "//pkg/sentry/mm", - "//pkg/sentry/socket", - "//pkg/sentry/socket/rpcinet", - "//pkg/sentry/socket/unix", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usage", - "//pkg/sentry/usermem", - "//pkg/syserror", - "//pkg/waiter", - ], -) - -go_test( - name = "proc_test", - size = "small", - srcs = [ - "net_test.go", - "sys_net_test.go", - ], - embed = [":proc"], - deps = [ - "//pkg/abi/linux", - "//pkg/sentry/context", - "//pkg/sentry/inet", - "//pkg/sentry/usermem", - ], -) diff --git a/pkg/sentry/fs/proc/README.md b/pkg/sentry/fs/proc/README.md deleted file mode 100644 index 5d4ec6c7b..000000000 --- a/pkg/sentry/fs/proc/README.md +++ /dev/null @@ -1,332 +0,0 @@ -This document tracks what is implemented in procfs. Refer to -Documentation/filesystems/proc.txt in the Linux project for information about -procfs generally. - -**NOTE**: This document is not guaranteed to be up to date. If you find an -inconsistency, please file a bug. - -[TOC] - -## Kernel data - -The following files are implemented: - -| File /proc/ | Content | -| :------------------------ | :---------------------------------------------------- | -| [cpuinfo](#cpuinfo) | Info about the CPU | -| [filesystems](#filesystems) | Supported filesystems | -| [loadavg](#loadavg) | Load average of last 1, 5 & 15 minutes | -| [meminfo](#meminfo) | Overall memory info | -| [stat](#stat) | Overall kernel statistics | -| [sys](#sys) | Change parameters within the kernel | -| [uptime](#uptime) | Wall clock since boot, combined idle time of all cpus | -| [version](#version) | Kernel version | - -### cpuinfo - -```bash -$ cat /proc/cpuinfo -processor : 0 -vendor_id : GenuineIntel -cpu family : 6 -model : 45 -model name : unknown -stepping : unknown -cpu MHz : 1234.588 -fpu : yes -fpu_exception : yes -cpuid level : 13 -wp : yes -flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic popcnt tsc_deadline_timer aes xsave avx xsaveopt -bogomips : 1234.59 -clflush size : 64 -cache_alignment : 64 -address sizes : 46 bits physical, 48 bits virtual -power management: - -... -``` - -Notable divergences: - -Field name | Notes -:--------------- | :--------------------------------------- -model name | Always unknown -stepping | Always unknown -fpu | Always yes -fpu_exception | Always yes -wp | Always yes -bogomips | Bogus value (matches cpu MHz) -clflush size | Always 64 -cache_alignment | Always 64 -address sizes | Always 46 bits physical, 48 bits virtual -power management | Always blank - -Otherwise fields are derived from the sentry configuration. - -### filesystems - -```bash -$ cat /proc/filesystems -nodev 9p -nodev devpts -nodev devtmpfs -nodev proc -nodev sysfs -nodev tmpfs -``` - -### loadavg - -```bash -$ cat /proc/loadavg -0.00 0.00 0.00 0/0 0 -``` - -Column | Notes -:------------------------------------ | :---------- -CPU.IO utilization in last 1 minute | Always zero -CPU.IO utilization in last 5 minutes | Always zero -CPU.IO utilization in last 10 minutes | Always zero -Num currently running processes | Always zero -Total num processes | Always zero - -TODO(b/62345059): Populate the columns with accurate statistics. - -### meminfo - -```bash -$ cat /proc/meminfo -MemTotal: 2097152 kB -MemFree: 2083540 kB -MemAvailable: 2083540 kB -Buffers: 0 kB -Cached: 4428 kB -SwapCache: 0 kB -Active: 10812 kB -Inactive: 2216 kB -Active(anon): 8600 kB -Inactive(anon): 0 kB -Active(file): 2212 kB -Inactive(file): 2216 kB -Unevictable: 0 kB -Mlocked: 0 kB -SwapTotal: 0 kB -SwapFree: 0 kB -Dirty: 0 kB -Writeback: 0 kB -AnonPages: 8600 kB -Mapped: 4428 kB -Shmem: 0 kB - -``` - -Notable divergences: - -Field name | Notes -:---------------- | :----------------------------------------------------- -Buffers | Always zero, no block devices -SwapCache | Always zero, no swap -Inactive(anon) | Always zero, see SwapCache -Unevictable | Always zero TODO(b/31823263) -Mlocked | Always zero TODO(b/31823263) -SwapTotal | Always zero, no swap -SwapFree | Always zero, no swap -Dirty | Always zero TODO(b/31823263) -Writeback | Always zero TODO(b/31823263) -MemAvailable | Uses the same value as MemFree since there is no swap. -Slab | Missing -SReclaimable | Missing -SUnreclaim | Missing -KernelStack | Missing -PageTables | Missing -NFS_Unstable | Missing -Bounce | Missing -WritebackTmp | Missing -CommitLimit | Missing -Committed_AS | Missing -VmallocTotal | Missing -VmallocUsed | Missing -VmallocChunk | Missing -HardwareCorrupted | Missing -AnonHugePages | Missing -ShmemHugePages | Missing -ShmemPmdMapped | Missing -HugePages_Total | Missing -HugePages_Free | Missing -HugePages_Rsvd | Missing -HugePages_Surp | Missing -Hugepagesize | Missing -DirectMap4k | Missing -DirectMap2M | Missing -DirectMap1G | Missing - -### stat - -```bash -$ cat /proc/stat -cpu 0 0 0 0 0 0 0 0 0 0 -cpu0 0 0 0 0 0 0 0 0 0 0 -cpu1 0 0 0 0 0 0 0 0 0 0 -cpu2 0 0 0 0 0 0 0 0 0 0 -cpu3 0 0 0 0 0 0 0 0 0 0 -cpu4 0 0 0 0 0 0 0 0 0 0 -cpu5 0 0 0 0 0 0 0 0 0 0 -cpu6 0 0 0 0 0 0 0 0 0 0 -cpu7 0 0 0 0 0 0 0 0 0 0 -intr 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -ctxt 0 -btime 1504040968 -processes 0 -procs_running 0 -procs_blokkcked 0 -softirq 0 0 0 0 0 0 0 0 0 0 0 -``` - -All fields except for `btime` are always zero. - -TODO(b/37226836): Populate with accurate fields. - -### sys - -```bash -$ ls /proc/sys -kernel vm -``` - -Directory | Notes -:-------- | :---------------------------- -abi | Missing -debug | Missing -dev | Missing -fs | Missing -kernel | Contains hostname (only) -net | Missing -user | Missing -vm | Contains mmap_min_addr (only) - -### uptime - -```bash -$ cat /proc/uptime -3204.62 0.00 -``` - -Column | Notes -:------------------------------- | :---------------------------- -Total num seconds system running | Time since procfs was mounted -Number of seconds idle | Always zero - -### version - -```bash -$ cat /proc/version -Linux version 4.4 #1 SMP Sun Jan 10 15:06:54 PST 2016 -``` - -## Process-specific data - -The following files are implemented: - -File /proc/PID | Content -:---------------------- | :--------------------------------------------------- -[auxv](#auxv) | Copy of auxiliary vector for the process -[cmdline](#cmdline) | Command line arguments -[comm](#comm) | Command name associated with the process -[environ](#environ) | Process environment -[exe](#exe) | Symlink to the process's executable -[fd](#fd) | Directory containing links to open file descriptors -[fdinfo](#fdinfo) | Information associated with open file descriptors -[gid_map](#gid_map) | Mappings for group IDs inside the user namespace -[io](#io) | IO statistics -[maps](#maps) | Memory mappings (anon, executables, library files) -[mounts](#mounts) | Mounted filesystems -[mountinfo](#mountinfo) | Information about mounts -[ns](#ns) | Directory containing info about supported namespaces -[stat](#stat) | Process statistics -[statm](#statm) | Process memory statistics -[status](#status) | Process status in human readable format -[task](#task) | Directory containing info about running threads -[uid_map](#uid_map) | Mappings for user IDs inside the user namespace - -### auxv - -TODO - -### cmdline - -TODO - -### comm - -TODO - -### environment - -TODO - -### exe - -TODO - -### fd - -TODO - -### fdinfo - -TODO - -### gid_map - -TODO - -### io - -Only has data for rchar, wchar, syscr, and syscw. - -TODO: add more detail. - -### maps - -TODO - -### mounts - -TODO - -### mountinfo - -TODO - -### ns - -TODO - -### stat - -Only has data for pid, comm, state, ppid, utime, stime, cutime, cstime, -num_threads, and exit_signal. - -TODO: add more detail. - -### statm - -Only has data for vss and rss. - -TODO: add more detail. - -### status - -Contains data for Name, State, Tgid, Pid, Ppid, TracerPid, FDSize, VmSize, -VmRSS, Threads, CapInh, CapPrm, CapEff, CapBnd, Seccomp. - -TODO: add more detail. - -### task - -TODO - -### uid_map - -TODO diff --git a/pkg/sentry/fs/proc/device/BUILD b/pkg/sentry/fs/proc/device/BUILD deleted file mode 100644 index 0394451d4..000000000 --- a/pkg/sentry/fs/proc/device/BUILD +++ /dev/null @@ -1,11 +0,0 @@ -load("//tools/go_stateify:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "device", - srcs = ["device.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/proc/device", - visibility = ["//pkg/sentry:internal"], - deps = ["//pkg/sentry/device"], -) diff --git a/pkg/sentry/fs/proc/device/device_state_autogen.go b/pkg/sentry/fs/proc/device/device_state_autogen.go new file mode 100755 index 000000000..be407ac45 --- /dev/null +++ b/pkg/sentry/fs/proc/device/device_state_autogen.go @@ -0,0 +1,4 @@ +// automatically generated by stateify. + +package device + diff --git a/pkg/sentry/fs/proc/net_test.go b/pkg/sentry/fs/proc/net_test.go deleted file mode 100644 index f18681405..000000000 --- a/pkg/sentry/fs/proc/net_test.go +++ /dev/null @@ -1,74 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "reflect" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/inet" -) - -func newIPv6TestStack() *inet.TestStack { - s := inet.NewTestStack() - s.SupportsIPv6Flag = true - return s -} - -func TestIfinet6NoAddresses(t *testing.T) { - n := &ifinet6{s: newIPv6TestStack()} - if got := n.contents(); got != nil { - t.Errorf("Got n.contents() = %v, want = %v", got, nil) - } -} - -func TestIfinet6(t *testing.T) { - s := newIPv6TestStack() - s.InterfacesMap[1] = inet.Interface{Name: "eth0"} - s.InterfaceAddrsMap[1] = []inet.InterfaceAddr{ - { - Family: linux.AF_INET6, - PrefixLen: 128, - Addr: []byte("\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f"), - }, - } - s.InterfacesMap[2] = inet.Interface{Name: "eth1"} - s.InterfaceAddrsMap[2] = []inet.InterfaceAddr{ - { - Family: linux.AF_INET6, - PrefixLen: 128, - Addr: []byte("\x10\x11\x12\x13\x14\x15\x16\x17\x18\x19\x1a\x1b\x1c\x1d\x1e\x1f"), - }, - } - want := map[string]struct{}{ - "000102030405060708090a0b0c0d0e0f 01 80 00 00 eth0\n": {}, - "101112131415161718191a1b1c1d1e1f 02 80 00 00 eth1\n": {}, - } - - n := &ifinet6{s: s} - contents := n.contents() - if len(contents) != len(want) { - t.Errorf("Got len(n.contents()) = %d, want = %d", len(contents), len(want)) - } - got := map[string]struct{}{} - for _, l := range contents { - got[l] = struct{}{} - } - - if !reflect.DeepEqual(got, want) { - t.Errorf("Got n.contents() = %v, want = %v", got, want) - } -} diff --git a/pkg/sentry/fs/proc/proc_state_autogen.go b/pkg/sentry/fs/proc/proc_state_autogen.go new file mode 100755 index 000000000..2fe57901a --- /dev/null +++ b/pkg/sentry/fs/proc/proc_state_autogen.go @@ -0,0 +1,669 @@ +// automatically generated by stateify. + +package proc + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *execArgInode) beforeSave() {} +func (x *execArgInode) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) + m.Save("arg", &x.arg) + m.Save("t", &x.t) +} + +func (x *execArgInode) afterLoad() {} +func (x *execArgInode) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) + m.Load("arg", &x.arg) + m.Load("t", &x.t) +} + +func (x *execArgFile) beforeSave() {} +func (x *execArgFile) save(m state.Map) { + x.beforeSave() + m.Save("arg", &x.arg) + m.Save("t", &x.t) +} + +func (x *execArgFile) afterLoad() {} +func (x *execArgFile) load(m state.Map) { + m.Load("arg", &x.arg) + m.Load("t", &x.t) +} + +func (x *fdDir) beforeSave() {} +func (x *fdDir) save(m state.Map) { + x.beforeSave() + m.Save("Dir", &x.Dir) + m.Save("t", &x.t) +} + +func (x *fdDir) afterLoad() {} +func (x *fdDir) load(m state.Map) { + m.Load("Dir", &x.Dir) + m.Load("t", &x.t) +} + +func (x *fdDirFile) beforeSave() {} +func (x *fdDirFile) save(m state.Map) { + x.beforeSave() + m.Save("isInfoFile", &x.isInfoFile) + m.Save("t", &x.t) +} + +func (x *fdDirFile) afterLoad() {} +func (x *fdDirFile) load(m state.Map) { + m.Load("isInfoFile", &x.isInfoFile) + m.Load("t", &x.t) +} + +func (x *fdInfoDir) beforeSave() {} +func (x *fdInfoDir) save(m state.Map) { + x.beforeSave() + m.Save("Dir", &x.Dir) + m.Save("t", &x.t) +} + +func (x *fdInfoDir) afterLoad() {} +func (x *fdInfoDir) load(m state.Map) { + m.Load("Dir", &x.Dir) + m.Load("t", &x.t) +} + +func (x *filesystemsData) beforeSave() {} +func (x *filesystemsData) save(m state.Map) { + x.beforeSave() +} + +func (x *filesystemsData) afterLoad() {} +func (x *filesystemsData) load(m state.Map) { +} + +func (x *filesystem) beforeSave() {} +func (x *filesystem) save(m state.Map) { + x.beforeSave() +} + +func (x *filesystem) afterLoad() {} +func (x *filesystem) load(m state.Map) { +} + +func (x *taskOwnedInodeOps) beforeSave() {} +func (x *taskOwnedInodeOps) save(m state.Map) { + x.beforeSave() + m.Save("InodeOperations", &x.InodeOperations) + m.Save("t", &x.t) +} + +func (x *taskOwnedInodeOps) afterLoad() {} +func (x *taskOwnedInodeOps) load(m state.Map) { + m.Load("InodeOperations", &x.InodeOperations) + m.Load("t", &x.t) +} + +func (x *staticFileInodeOps) beforeSave() {} +func (x *staticFileInodeOps) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Save("InodeStaticFileGetter", &x.InodeStaticFileGetter) +} + +func (x *staticFileInodeOps) afterLoad() {} +func (x *staticFileInodeOps) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Load("InodeStaticFileGetter", &x.InodeStaticFileGetter) +} + +func (x *loadavgData) beforeSave() {} +func (x *loadavgData) save(m state.Map) { + x.beforeSave() +} + +func (x *loadavgData) afterLoad() {} +func (x *loadavgData) load(m state.Map) { +} + +func (x *meminfoData) beforeSave() {} +func (x *meminfoData) save(m state.Map) { + x.beforeSave() + m.Save("k", &x.k) +} + +func (x *meminfoData) afterLoad() {} +func (x *meminfoData) load(m state.Map) { + m.Load("k", &x.k) +} + +func (x *mountInfoFile) beforeSave() {} +func (x *mountInfoFile) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) +} + +func (x *mountInfoFile) afterLoad() {} +func (x *mountInfoFile) load(m state.Map) { + m.Load("t", &x.t) +} + +func (x *mountsFile) beforeSave() {} +func (x *mountsFile) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) +} + +func (x *mountsFile) afterLoad() {} +func (x *mountsFile) load(m state.Map) { + m.Load("t", &x.t) +} + +func (x *ifinet6) beforeSave() {} +func (x *ifinet6) save(m state.Map) { + x.beforeSave() + m.Save("s", &x.s) +} + +func (x *ifinet6) afterLoad() {} +func (x *ifinet6) load(m state.Map) { + m.Load("s", &x.s) +} + +func (x *netDev) beforeSave() {} +func (x *netDev) save(m state.Map) { + x.beforeSave() + m.Save("s", &x.s) +} + +func (x *netDev) afterLoad() {} +func (x *netDev) load(m state.Map) { + m.Load("s", &x.s) +} + +func (x *netUnix) beforeSave() {} +func (x *netUnix) save(m state.Map) { + x.beforeSave() + m.Save("k", &x.k) +} + +func (x *netUnix) afterLoad() {} +func (x *netUnix) load(m state.Map) { + m.Load("k", &x.k) +} + +func (x *netTCP) beforeSave() {} +func (x *netTCP) save(m state.Map) { + x.beforeSave() + m.Save("k", &x.k) +} + +func (x *netTCP) afterLoad() {} +func (x *netTCP) load(m state.Map) { + m.Load("k", &x.k) +} + +func (x *proc) beforeSave() {} +func (x *proc) save(m state.Map) { + x.beforeSave() + m.Save("Dir", &x.Dir) + m.Save("k", &x.k) + m.Save("pidns", &x.pidns) + m.Save("cgroupControllers", &x.cgroupControllers) +} + +func (x *proc) afterLoad() {} +func (x *proc) load(m state.Map) { + m.Load("Dir", &x.Dir) + m.Load("k", &x.k) + m.Load("pidns", &x.pidns) + m.Load("cgroupControllers", &x.cgroupControllers) +} + +func (x *self) beforeSave() {} +func (x *self) save(m state.Map) { + x.beforeSave() + m.Save("Symlink", &x.Symlink) + m.Save("pidns", &x.pidns) +} + +func (x *self) afterLoad() {} +func (x *self) load(m state.Map) { + m.Load("Symlink", &x.Symlink) + m.Load("pidns", &x.pidns) +} + +func (x *threadSelf) beforeSave() {} +func (x *threadSelf) save(m state.Map) { + x.beforeSave() + m.Save("Symlink", &x.Symlink) + m.Save("pidns", &x.pidns) +} + +func (x *threadSelf) afterLoad() {} +func (x *threadSelf) load(m state.Map) { + m.Load("Symlink", &x.Symlink) + m.Load("pidns", &x.pidns) +} + +func (x *rootProcFile) beforeSave() {} +func (x *rootProcFile) save(m state.Map) { + x.beforeSave() + m.Save("iops", &x.iops) +} + +func (x *rootProcFile) afterLoad() {} +func (x *rootProcFile) load(m state.Map) { + m.Load("iops", &x.iops) +} + +func (x *statData) beforeSave() {} +func (x *statData) save(m state.Map) { + x.beforeSave() + m.Save("k", &x.k) +} + +func (x *statData) afterLoad() {} +func (x *statData) load(m state.Map) { + m.Load("k", &x.k) +} + +func (x *mmapMinAddrData) beforeSave() {} +func (x *mmapMinAddrData) save(m state.Map) { + x.beforeSave() + m.Save("k", &x.k) +} + +func (x *mmapMinAddrData) afterLoad() {} +func (x *mmapMinAddrData) load(m state.Map) { + m.Load("k", &x.k) +} + +func (x *overcommitMemory) beforeSave() {} +func (x *overcommitMemory) save(m state.Map) { + x.beforeSave() +} + +func (x *overcommitMemory) afterLoad() {} +func (x *overcommitMemory) load(m state.Map) { +} + +func (x *hostname) beforeSave() {} +func (x *hostname) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) +} + +func (x *hostname) afterLoad() {} +func (x *hostname) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) +} + +func (x *hostnameFile) beforeSave() {} +func (x *hostnameFile) save(m state.Map) { + x.beforeSave() +} + +func (x *hostnameFile) afterLoad() {} +func (x *hostnameFile) load(m state.Map) { +} + +func (x *tcpMemInode) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) + m.Save("dir", &x.dir) + m.Save("s", &x.s) + m.Save("size", &x.size) +} + +func (x *tcpMemInode) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) + m.Load("dir", &x.dir) + m.LoadWait("s", &x.s) + m.Load("size", &x.size) + m.AfterLoad(x.afterLoad) +} + +func (x *tcpMemFile) beforeSave() {} +func (x *tcpMemFile) save(m state.Map) { + x.beforeSave() + m.Save("tcpMemInode", &x.tcpMemInode) +} + +func (x *tcpMemFile) afterLoad() {} +func (x *tcpMemFile) load(m state.Map) { + m.Load("tcpMemInode", &x.tcpMemInode) +} + +func (x *tcpSack) beforeSave() {} +func (x *tcpSack) save(m state.Map) { + x.beforeSave() + m.Save("stack", &x.stack) + m.Save("enabled", &x.enabled) + m.Save("SimpleFileInode", &x.SimpleFileInode) +} + +func (x *tcpSack) load(m state.Map) { + m.LoadWait("stack", &x.stack) + m.Load("enabled", &x.enabled) + m.Load("SimpleFileInode", &x.SimpleFileInode) + m.AfterLoad(x.afterLoad) +} + +func (x *tcpSackFile) beforeSave() {} +func (x *tcpSackFile) save(m state.Map) { + x.beforeSave() + m.Save("tcpSack", &x.tcpSack) + m.Save("stack", &x.stack) +} + +func (x *tcpSackFile) afterLoad() {} +func (x *tcpSackFile) load(m state.Map) { + m.Load("tcpSack", &x.tcpSack) + m.LoadWait("stack", &x.stack) +} + +func (x *taskDir) beforeSave() {} +func (x *taskDir) save(m state.Map) { + x.beforeSave() + m.Save("Dir", &x.Dir) + m.Save("t", &x.t) + m.Save("pidns", &x.pidns) +} + +func (x *taskDir) afterLoad() {} +func (x *taskDir) load(m state.Map) { + m.Load("Dir", &x.Dir) + m.Load("t", &x.t) + m.Load("pidns", &x.pidns) +} + +func (x *subtasks) beforeSave() {} +func (x *subtasks) save(m state.Map) { + x.beforeSave() + m.Save("Dir", &x.Dir) + m.Save("t", &x.t) + m.Save("p", &x.p) +} + +func (x *subtasks) afterLoad() {} +func (x *subtasks) load(m state.Map) { + m.Load("Dir", &x.Dir) + m.Load("t", &x.t) + m.Load("p", &x.p) +} + +func (x *subtasksFile) beforeSave() {} +func (x *subtasksFile) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) + m.Save("pidns", &x.pidns) +} + +func (x *subtasksFile) afterLoad() {} +func (x *subtasksFile) load(m state.Map) { + m.Load("t", &x.t) + m.Load("pidns", &x.pidns) +} + +func (x *exe) beforeSave() {} +func (x *exe) save(m state.Map) { + x.beforeSave() + m.Save("Symlink", &x.Symlink) + m.Save("t", &x.t) +} + +func (x *exe) afterLoad() {} +func (x *exe) load(m state.Map) { + m.Load("Symlink", &x.Symlink) + m.Load("t", &x.t) +} + +func (x *namespaceSymlink) beforeSave() {} +func (x *namespaceSymlink) save(m state.Map) { + x.beforeSave() + m.Save("Symlink", &x.Symlink) + m.Save("t", &x.t) +} + +func (x *namespaceSymlink) afterLoad() {} +func (x *namespaceSymlink) load(m state.Map) { + m.Load("Symlink", &x.Symlink) + m.Load("t", &x.t) +} + +func (x *mapsData) beforeSave() {} +func (x *mapsData) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) +} + +func (x *mapsData) afterLoad() {} +func (x *mapsData) load(m state.Map) { + m.Load("t", &x.t) +} + +func (x *smapsData) beforeSave() {} +func (x *smapsData) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) +} + +func (x *smapsData) afterLoad() {} +func (x *smapsData) load(m state.Map) { + m.Load("t", &x.t) +} + +func (x *taskStatData) beforeSave() {} +func (x *taskStatData) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) + m.Save("tgstats", &x.tgstats) + m.Save("pidns", &x.pidns) +} + +func (x *taskStatData) afterLoad() {} +func (x *taskStatData) load(m state.Map) { + m.Load("t", &x.t) + m.Load("tgstats", &x.tgstats) + m.Load("pidns", &x.pidns) +} + +func (x *statmData) beforeSave() {} +func (x *statmData) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) +} + +func (x *statmData) afterLoad() {} +func (x *statmData) load(m state.Map) { + m.Load("t", &x.t) +} + +func (x *statusData) beforeSave() {} +func (x *statusData) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) + m.Save("pidns", &x.pidns) +} + +func (x *statusData) afterLoad() {} +func (x *statusData) load(m state.Map) { + m.Load("t", &x.t) + m.Load("pidns", &x.pidns) +} + +func (x *ioData) beforeSave() {} +func (x *ioData) save(m state.Map) { + x.beforeSave() + m.Save("ioUsage", &x.ioUsage) +} + +func (x *ioData) afterLoad() {} +func (x *ioData) load(m state.Map) { + m.Load("ioUsage", &x.ioUsage) +} + +func (x *comm) beforeSave() {} +func (x *comm) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) + m.Save("t", &x.t) +} + +func (x *comm) afterLoad() {} +func (x *comm) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) + m.Load("t", &x.t) +} + +func (x *commFile) beforeSave() {} +func (x *commFile) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) +} + +func (x *commFile) afterLoad() {} +func (x *commFile) load(m state.Map) { + m.Load("t", &x.t) +} + +func (x *auxvec) beforeSave() {} +func (x *auxvec) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) + m.Save("t", &x.t) +} + +func (x *auxvec) afterLoad() {} +func (x *auxvec) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) + m.Load("t", &x.t) +} + +func (x *auxvecFile) beforeSave() {} +func (x *auxvecFile) save(m state.Map) { + x.beforeSave() + m.Save("t", &x.t) +} + +func (x *auxvecFile) afterLoad() {} +func (x *auxvecFile) load(m state.Map) { + m.Load("t", &x.t) +} + +func (x *idMapInodeOperations) beforeSave() {} +func (x *idMapInodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Save("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Save("t", &x.t) + m.Save("gids", &x.gids) +} + +func (x *idMapInodeOperations) afterLoad() {} +func (x *idMapInodeOperations) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Load("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Load("t", &x.t) + m.Load("gids", &x.gids) +} + +func (x *idMapFileOperations) beforeSave() {} +func (x *idMapFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("iops", &x.iops) +} + +func (x *idMapFileOperations) afterLoad() {} +func (x *idMapFileOperations) load(m state.Map) { + m.Load("iops", &x.iops) +} + +func (x *uptime) beforeSave() {} +func (x *uptime) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) + m.Save("startTime", &x.startTime) +} + +func (x *uptime) afterLoad() {} +func (x *uptime) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) + m.Load("startTime", &x.startTime) +} + +func (x *uptimeFile) beforeSave() {} +func (x *uptimeFile) save(m state.Map) { + x.beforeSave() + m.Save("startTime", &x.startTime) +} + +func (x *uptimeFile) afterLoad() {} +func (x *uptimeFile) load(m state.Map) { + m.Load("startTime", &x.startTime) +} + +func (x *versionData) beforeSave() {} +func (x *versionData) save(m state.Map) { + x.beforeSave() + m.Save("k", &x.k) +} + +func (x *versionData) afterLoad() {} +func (x *versionData) load(m state.Map) { + m.Load("k", &x.k) +} + +func init() { + state.Register("proc.execArgInode", (*execArgInode)(nil), state.Fns{Save: (*execArgInode).save, Load: (*execArgInode).load}) + state.Register("proc.execArgFile", (*execArgFile)(nil), state.Fns{Save: (*execArgFile).save, Load: (*execArgFile).load}) + state.Register("proc.fdDir", (*fdDir)(nil), state.Fns{Save: (*fdDir).save, Load: (*fdDir).load}) + state.Register("proc.fdDirFile", (*fdDirFile)(nil), state.Fns{Save: (*fdDirFile).save, Load: (*fdDirFile).load}) + state.Register("proc.fdInfoDir", (*fdInfoDir)(nil), state.Fns{Save: (*fdInfoDir).save, Load: (*fdInfoDir).load}) + state.Register("proc.filesystemsData", (*filesystemsData)(nil), state.Fns{Save: (*filesystemsData).save, Load: (*filesystemsData).load}) + state.Register("proc.filesystem", (*filesystem)(nil), state.Fns{Save: (*filesystem).save, Load: (*filesystem).load}) + state.Register("proc.taskOwnedInodeOps", (*taskOwnedInodeOps)(nil), state.Fns{Save: (*taskOwnedInodeOps).save, Load: (*taskOwnedInodeOps).load}) + state.Register("proc.staticFileInodeOps", (*staticFileInodeOps)(nil), state.Fns{Save: (*staticFileInodeOps).save, Load: (*staticFileInodeOps).load}) + state.Register("proc.loadavgData", (*loadavgData)(nil), state.Fns{Save: (*loadavgData).save, Load: (*loadavgData).load}) + state.Register("proc.meminfoData", (*meminfoData)(nil), state.Fns{Save: (*meminfoData).save, Load: (*meminfoData).load}) + state.Register("proc.mountInfoFile", (*mountInfoFile)(nil), state.Fns{Save: (*mountInfoFile).save, Load: (*mountInfoFile).load}) + state.Register("proc.mountsFile", (*mountsFile)(nil), state.Fns{Save: (*mountsFile).save, Load: (*mountsFile).load}) + state.Register("proc.ifinet6", (*ifinet6)(nil), state.Fns{Save: (*ifinet6).save, Load: (*ifinet6).load}) + state.Register("proc.netDev", (*netDev)(nil), state.Fns{Save: (*netDev).save, Load: (*netDev).load}) + state.Register("proc.netUnix", (*netUnix)(nil), state.Fns{Save: (*netUnix).save, Load: (*netUnix).load}) + state.Register("proc.netTCP", (*netTCP)(nil), state.Fns{Save: (*netTCP).save, Load: (*netTCP).load}) + state.Register("proc.proc", (*proc)(nil), state.Fns{Save: (*proc).save, Load: (*proc).load}) + state.Register("proc.self", (*self)(nil), state.Fns{Save: (*self).save, Load: (*self).load}) + state.Register("proc.threadSelf", (*threadSelf)(nil), state.Fns{Save: (*threadSelf).save, Load: (*threadSelf).load}) + state.Register("proc.rootProcFile", (*rootProcFile)(nil), state.Fns{Save: (*rootProcFile).save, Load: (*rootProcFile).load}) + state.Register("proc.statData", (*statData)(nil), state.Fns{Save: (*statData).save, Load: (*statData).load}) + state.Register("proc.mmapMinAddrData", (*mmapMinAddrData)(nil), state.Fns{Save: (*mmapMinAddrData).save, Load: (*mmapMinAddrData).load}) + state.Register("proc.overcommitMemory", (*overcommitMemory)(nil), state.Fns{Save: (*overcommitMemory).save, Load: (*overcommitMemory).load}) + state.Register("proc.hostname", (*hostname)(nil), state.Fns{Save: (*hostname).save, Load: (*hostname).load}) + state.Register("proc.hostnameFile", (*hostnameFile)(nil), state.Fns{Save: (*hostnameFile).save, Load: (*hostnameFile).load}) + state.Register("proc.tcpMemInode", (*tcpMemInode)(nil), state.Fns{Save: (*tcpMemInode).save, Load: (*tcpMemInode).load}) + state.Register("proc.tcpMemFile", (*tcpMemFile)(nil), state.Fns{Save: (*tcpMemFile).save, Load: (*tcpMemFile).load}) + state.Register("proc.tcpSack", (*tcpSack)(nil), state.Fns{Save: (*tcpSack).save, Load: (*tcpSack).load}) + state.Register("proc.tcpSackFile", (*tcpSackFile)(nil), state.Fns{Save: (*tcpSackFile).save, Load: (*tcpSackFile).load}) + state.Register("proc.taskDir", (*taskDir)(nil), state.Fns{Save: (*taskDir).save, Load: (*taskDir).load}) + state.Register("proc.subtasks", (*subtasks)(nil), state.Fns{Save: (*subtasks).save, Load: (*subtasks).load}) + state.Register("proc.subtasksFile", (*subtasksFile)(nil), state.Fns{Save: (*subtasksFile).save, Load: (*subtasksFile).load}) + state.Register("proc.exe", (*exe)(nil), state.Fns{Save: (*exe).save, Load: (*exe).load}) + state.Register("proc.namespaceSymlink", (*namespaceSymlink)(nil), state.Fns{Save: (*namespaceSymlink).save, Load: (*namespaceSymlink).load}) + state.Register("proc.mapsData", (*mapsData)(nil), state.Fns{Save: (*mapsData).save, Load: (*mapsData).load}) + state.Register("proc.smapsData", (*smapsData)(nil), state.Fns{Save: (*smapsData).save, Load: (*smapsData).load}) + state.Register("proc.taskStatData", (*taskStatData)(nil), state.Fns{Save: (*taskStatData).save, Load: (*taskStatData).load}) + state.Register("proc.statmData", (*statmData)(nil), state.Fns{Save: (*statmData).save, Load: (*statmData).load}) + state.Register("proc.statusData", (*statusData)(nil), state.Fns{Save: (*statusData).save, Load: (*statusData).load}) + state.Register("proc.ioData", (*ioData)(nil), state.Fns{Save: (*ioData).save, Load: (*ioData).load}) + state.Register("proc.comm", (*comm)(nil), state.Fns{Save: (*comm).save, Load: (*comm).load}) + state.Register("proc.commFile", (*commFile)(nil), state.Fns{Save: (*commFile).save, Load: (*commFile).load}) + state.Register("proc.auxvec", (*auxvec)(nil), state.Fns{Save: (*auxvec).save, Load: (*auxvec).load}) + state.Register("proc.auxvecFile", (*auxvecFile)(nil), state.Fns{Save: (*auxvecFile).save, Load: (*auxvecFile).load}) + state.Register("proc.idMapInodeOperations", (*idMapInodeOperations)(nil), state.Fns{Save: (*idMapInodeOperations).save, Load: (*idMapInodeOperations).load}) + state.Register("proc.idMapFileOperations", (*idMapFileOperations)(nil), state.Fns{Save: (*idMapFileOperations).save, Load: (*idMapFileOperations).load}) + state.Register("proc.uptime", (*uptime)(nil), state.Fns{Save: (*uptime).save, Load: (*uptime).load}) + state.Register("proc.uptimeFile", (*uptimeFile)(nil), state.Fns{Save: (*uptimeFile).save, Load: (*uptimeFile).load}) + state.Register("proc.versionData", (*versionData)(nil), state.Fns{Save: (*versionData).save, Load: (*versionData).load}) +} diff --git a/pkg/sentry/fs/proc/seqfile/BUILD b/pkg/sentry/fs/proc/seqfile/BUILD deleted file mode 100644 index 20c3eefc8..000000000 --- a/pkg/sentry/fs/proc/seqfile/BUILD +++ /dev/null @@ -1,35 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_stateify:defs.bzl", "go_library", "go_test") - -go_library( - name = "seqfile", - srcs = ["seqfile.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/sentry/context", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fs/proc/device", - "//pkg/sentry/kernel/time", - "//pkg/sentry/usermem", - "//pkg/syserror", - "//pkg/waiter", - ], -) - -go_test( - name = "seqfile_test", - size = "small", - srcs = ["seqfile_test.go"], - embed = [":seqfile"], - deps = [ - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", - "//pkg/sentry/fs", - "//pkg/sentry/fs/ramfs", - "//pkg/sentry/usermem", - ], -) diff --git a/pkg/sentry/fs/proc/seqfile/seqfile_state_autogen.go b/pkg/sentry/fs/proc/seqfile/seqfile_state_autogen.go new file mode 100755 index 000000000..db9f7ceb9 --- /dev/null +++ b/pkg/sentry/fs/proc/seqfile/seqfile_state_autogen.go @@ -0,0 +1,58 @@ +// automatically generated by stateify. + +package seqfile + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *SeqData) beforeSave() {} +func (x *SeqData) save(m state.Map) { + x.beforeSave() + m.Save("Buf", &x.Buf) + m.Save("Handle", &x.Handle) +} + +func (x *SeqData) afterLoad() {} +func (x *SeqData) load(m state.Map) { + m.Load("Buf", &x.Buf) + m.Load("Handle", &x.Handle) +} + +func (x *SeqFile) beforeSave() {} +func (x *SeqFile) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Save("SeqSource", &x.SeqSource) + m.Save("source", &x.source) + m.Save("generation", &x.generation) + m.Save("lastRead", &x.lastRead) +} + +func (x *SeqFile) afterLoad() {} +func (x *SeqFile) load(m state.Map) { + m.Load("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Load("SeqSource", &x.SeqSource) + m.Load("source", &x.source) + m.Load("generation", &x.generation) + m.Load("lastRead", &x.lastRead) +} + +func (x *seqFileOperations) beforeSave() {} +func (x *seqFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("seqFile", &x.seqFile) +} + +func (x *seqFileOperations) afterLoad() {} +func (x *seqFileOperations) load(m state.Map) { + m.Load("seqFile", &x.seqFile) +} + +func init() { + state.Register("seqfile.SeqData", (*SeqData)(nil), state.Fns{Save: (*SeqData).save, Load: (*SeqData).load}) + state.Register("seqfile.SeqFile", (*SeqFile)(nil), state.Fns{Save: (*SeqFile).save, Load: (*SeqFile).load}) + state.Register("seqfile.seqFileOperations", (*seqFileOperations)(nil), state.Fns{Save: (*seqFileOperations).save, Load: (*seqFileOperations).load}) +} diff --git a/pkg/sentry/fs/proc/seqfile/seqfile_test.go b/pkg/sentry/fs/proc/seqfile/seqfile_test.go deleted file mode 100644 index ebfeee835..000000000 --- a/pkg/sentry/fs/proc/seqfile/seqfile_test.go +++ /dev/null @@ -1,279 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package seqfile - -import ( - "bytes" - "fmt" - "io" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" - "gvisor.dev/gvisor/pkg/sentry/usermem" -) - -type seqTest struct { - actual []SeqData - update bool -} - -func (s *seqTest) Init() { - var sq []SeqData - // Create some SeqData. - for i := 0; i < 10; i++ { - var b []byte - for j := 0; j < 10; j++ { - b = append(b, byte(i)) - } - sq = append(sq, SeqData{ - Buf: b, - Handle: &testHandle{i: i}, - }) - } - s.actual = sq -} - -// NeedsUpdate reports whether we need to update the data we've previously read. -func (s *seqTest) NeedsUpdate(int64) bool { - return s.update -} - -// ReadSeqFiledata returns a slice of SeqData which contains elements -// greater than the handle. -func (s *seqTest) ReadSeqFileData(ctx context.Context, handle SeqHandle) ([]SeqData, int64) { - if handle == nil { - return s.actual, 0 - } - h := *handle.(*testHandle) - var ret []SeqData - for _, b := range s.actual { - // We want the next one. - h2 := *b.Handle.(*testHandle) - if h2.i > h.i { - ret = append(ret, b) - } - } - return ret, 0 -} - -// Flatten a slice of slices into one slice. -func flatten(buf ...[]byte) []byte { - var flat []byte - for _, b := range buf { - flat = append(flat, b...) - } - return flat -} - -type testHandle struct { - i int -} - -type testTable struct { - offset int64 - readBufferSize int - expectedData []byte - expectedError error -} - -func runTableTests(ctx context.Context, table []testTable, dirent *fs.Dirent) error { - for _, tt := range table { - file, err := dirent.Inode.InodeOperations.GetFile(ctx, dirent, fs.FileFlags{Read: true}) - if err != nil { - return fmt.Errorf("GetFile returned error: %v", err) - } - - data := make([]byte, tt.readBufferSize) - resultLen, err := file.Preadv(ctx, usermem.BytesIOSequence(data), tt.offset) - if err != tt.expectedError { - return fmt.Errorf("t.Preadv(len: %v, offset: %v) (error) => %v expected %v", tt.readBufferSize, tt.offset, err, tt.expectedError) - } - expectedLen := int64(len(tt.expectedData)) - if resultLen != expectedLen { - // We make this just an error so we wall through and print the data below. - return fmt.Errorf("t.Preadv(len: %v, offset: %v) (size) => %v expected %v", tt.readBufferSize, tt.offset, resultLen, expectedLen) - } - if !bytes.Equal(data[:expectedLen], tt.expectedData) { - return fmt.Errorf("t.Preadv(len: %v, offset: %v) (data) => %v expected %v", tt.readBufferSize, tt.offset, data[:expectedLen], tt.expectedData) - } - } - return nil -} - -func TestSeqFile(t *testing.T) { - testSource := &seqTest{} - testSource.Init() - - // Create a file that can be R/W. - ctx := contexttest.Context(t) - m := fs.NewPseudoMountSource(ctx) - contents := map[string]*fs.Inode{ - "foo": NewSeqFileInode(ctx, testSource, m), - } - root := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0777)) - - // How about opening it? - inode := fs.NewInode(ctx, root, m, fs.StableAttr{Type: fs.Directory}) - dirent2, err := root.Lookup(ctx, inode, "foo") - if err != nil { - t.Fatalf("failed to walk to foo for n2: %v", err) - } - n2 := dirent2.Inode.InodeOperations - file2, err := n2.GetFile(ctx, dirent2, fs.FileFlags{Read: true, Write: true}) - if err != nil { - t.Fatalf("GetFile returned error: %v", err) - } - - // Writing? - if _, err := file2.Writev(ctx, usermem.BytesIOSequence([]byte("test"))); err == nil { - t.Fatalf("managed to write to n2: %v", err) - } - - // How about reading? - dirent3, err := root.Lookup(ctx, inode, "foo") - if err != nil { - t.Fatalf("failed to walk to foo: %v", err) - } - n3 := dirent3.Inode.InodeOperations - if n2 != n3 { - t.Error("got n2 != n3, want same") - } - - testSource.update = true - - table := []testTable{ - // Read past the end. - {100, 4, []byte{}, io.EOF}, - {110, 4, []byte{}, io.EOF}, - {200, 4, []byte{}, io.EOF}, - // Read a truncated first line. - {0, 4, testSource.actual[0].Buf[:4], nil}, - // Read the whole first line. - {0, 10, testSource.actual[0].Buf, nil}, - // Read the whole first line + 5 bytes of second line. - {0, 15, flatten(testSource.actual[0].Buf, testSource.actual[1].Buf[:5]), nil}, - // First 4 bytes of the second line. - {10, 4, testSource.actual[1].Buf[:4], nil}, - // Read the two first lines. - {0, 20, flatten(testSource.actual[0].Buf, testSource.actual[1].Buf), nil}, - // Read three lines. - {0, 30, flatten(testSource.actual[0].Buf, testSource.actual[1].Buf, testSource.actual[2].Buf), nil}, - // Read everything, but use a bigger buffer than necessary. - {0, 150, flatten(testSource.actual[0].Buf, testSource.actual[1].Buf, testSource.actual[2].Buf, testSource.actual[3].Buf, testSource.actual[4].Buf, testSource.actual[5].Buf, testSource.actual[6].Buf, testSource.actual[7].Buf, testSource.actual[8].Buf, testSource.actual[9].Buf), nil}, - // Read the last 3 bytes. - {97, 10, testSource.actual[9].Buf[7:], nil}, - } - if err := runTableTests(ctx, table, dirent2); err != nil { - t.Errorf("runTableTest failed with testSource.update = %v : %v", testSource.update, err) - } - - // Disable updates and do it again. - testSource.update = false - if err := runTableTests(ctx, table, dirent2); err != nil { - t.Errorf("runTableTest failed with testSource.update = %v: %v", testSource.update, err) - } -} - -// Test that we behave correctly when the file is updated. -func TestSeqFileFileUpdated(t *testing.T) { - testSource := &seqTest{} - testSource.Init() - testSource.update = true - - // Create a file that can be R/W. - ctx := contexttest.Context(t) - m := fs.NewPseudoMountSource(ctx) - contents := map[string]*fs.Inode{ - "foo": NewSeqFileInode(ctx, testSource, m), - } - root := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0777)) - - // How about opening it? - inode := fs.NewInode(ctx, root, m, fs.StableAttr{Type: fs.Directory}) - dirent2, err := root.Lookup(ctx, inode, "foo") - if err != nil { - t.Fatalf("failed to walk to foo for dirent2: %v", err) - } - - table := []testTable{ - {0, 16, flatten(testSource.actual[0].Buf, testSource.actual[1].Buf[:6]), nil}, - } - if err := runTableTests(ctx, table, dirent2); err != nil { - t.Errorf("runTableTest failed: %v", err) - } - // Delete the first entry. - cut := testSource.actual[0].Buf - testSource.actual = testSource.actual[1:] - - table = []testTable{ - // Try reading buffer 0 with an offset. This will not delete the old data. - {1, 5, cut[1:6], nil}, - // Reset our file by reading at offset 0. - {0, 10, testSource.actual[0].Buf, nil}, - {16, 14, flatten(testSource.actual[1].Buf[6:], testSource.actual[2].Buf), nil}, - // Read the same data a second time. - {16, 14, flatten(testSource.actual[1].Buf[6:], testSource.actual[2].Buf), nil}, - // Read the following two lines. - {30, 20, flatten(testSource.actual[3].Buf, testSource.actual[4].Buf), nil}, - } - if err := runTableTests(ctx, table, dirent2); err != nil { - t.Errorf("runTableTest failed after removing first entry: %v", err) - } - - // Add a new duplicate line in the middle (6666...) - after := testSource.actual[5:] - testSource.actual = testSource.actual[:4] - // Note the list must be sorted. - testSource.actual = append(testSource.actual, after[0]) - testSource.actual = append(testSource.actual, after...) - - table = []testTable{ - {50, 20, flatten(testSource.actual[4].Buf, testSource.actual[5].Buf), nil}, - } - if err := runTableTests(ctx, table, dirent2); err != nil { - t.Errorf("runTableTest failed after adding middle entry: %v", err) - } - // This will be used in a later test. - oldTestData := testSource.actual - - // Delete everything. - testSource.actual = testSource.actual[:0] - table = []testTable{ - {20, 20, []byte{}, io.EOF}, - } - if err := runTableTests(ctx, table, dirent2); err != nil { - t.Errorf("runTableTest failed after removing all entries: %v", err) - } - // Restore some of the data. - testSource.actual = oldTestData[:1] - table = []testTable{ - {6, 20, testSource.actual[0].Buf[6:], nil}, - } - if err := runTableTests(ctx, table, dirent2); err != nil { - t.Errorf("runTableTest failed after adding first entry back: %v", err) - } - - // Re-extend the data - testSource.actual = oldTestData - table = []testTable{ - {30, 20, flatten(testSource.actual[3].Buf, testSource.actual[4].Buf), nil}, - } - if err := runTableTests(ctx, table, dirent2); err != nil { - t.Errorf("runTableTest failed after extending testSource: %v", err) - } -} diff --git a/pkg/sentry/fs/proc/sys_net_test.go b/pkg/sentry/fs/proc/sys_net_test.go deleted file mode 100644 index 6abae7a60..000000000 --- a/pkg/sentry/fs/proc/sys_net_test.go +++ /dev/null @@ -1,125 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/inet" - "gvisor.dev/gvisor/pkg/sentry/usermem" -) - -func TestQuerySendBufferSize(t *testing.T) { - ctx := context.Background() - s := inet.NewTestStack() - s.TCPSendBufSize = inet.TCPBufferSize{100, 200, 300} - tmi := &tcpMemInode{s: s, dir: tcpWMem} - tmf := &tcpMemFile{tcpMemInode: tmi} - - buf := make([]byte, 100) - dst := usermem.BytesIOSequence(buf) - n, err := tmf.Read(ctx, nil, dst, 0) - if err != nil { - t.Fatalf("Read failed: %v", err) - } - - if got, want := string(buf[:n]), "100\t200\t300\n"; got != want { - t.Fatalf("Bad string: got %v, want %v", got, want) - } -} - -func TestQueryRecvBufferSize(t *testing.T) { - ctx := context.Background() - s := inet.NewTestStack() - s.TCPRecvBufSize = inet.TCPBufferSize{100, 200, 300} - tmi := &tcpMemInode{s: s, dir: tcpRMem} - tmf := &tcpMemFile{tcpMemInode: tmi} - - buf := make([]byte, 100) - dst := usermem.BytesIOSequence(buf) - n, err := tmf.Read(ctx, nil, dst, 0) - if err != nil { - t.Fatalf("Read failed: %v", err) - } - - if got, want := string(buf[:n]), "100\t200\t300\n"; got != want { - t.Fatalf("Bad string: got %v, want %v", got, want) - } -} - -var cases = []struct { - str string - initial inet.TCPBufferSize - final inet.TCPBufferSize -}{ - { - str: "", - initial: inet.TCPBufferSize{1, 2, 3}, - final: inet.TCPBufferSize{1, 2, 3}, - }, - { - str: "100\n", - initial: inet.TCPBufferSize{1, 100, 200}, - final: inet.TCPBufferSize{100, 100, 200}, - }, - { - str: "100 200 300\n", - initial: inet.TCPBufferSize{1, 2, 3}, - final: inet.TCPBufferSize{100, 200, 300}, - }, -} - -func TestConfigureSendBufferSize(t *testing.T) { - ctx := context.Background() - s := inet.NewTestStack() - for _, c := range cases { - s.TCPSendBufSize = c.initial - tmi := &tcpMemInode{s: s, dir: tcpWMem} - tmf := &tcpMemFile{tcpMemInode: tmi} - - // Write the values. - src := usermem.BytesIOSequence([]byte(c.str)) - if n, err := tmf.Write(ctx, nil, src, 0); n != int64(len(c.str)) || err != nil { - t.Errorf("Write, case = %q: got (%d, %v), wanted (%d, nil)", c.str, n, err, len(c.str)) - } - - // Read the values from the stack and check them. - if s.TCPSendBufSize != c.final { - t.Errorf("TCPSendBufferSize, case = %q: got %v, wanted %v", c.str, s.TCPSendBufSize, c.final) - } - } -} - -func TestConfigureRecvBufferSize(t *testing.T) { - ctx := context.Background() - s := inet.NewTestStack() - for _, c := range cases { - s.TCPRecvBufSize = c.initial - tmi := &tcpMemInode{s: s, dir: tcpRMem} - tmf := &tcpMemFile{tcpMemInode: tmi} - - // Write the values. - src := usermem.BytesIOSequence([]byte(c.str)) - if n, err := tmf.Write(ctx, nil, src, 0); n != int64(len(c.str)) || err != nil { - t.Errorf("Write, case = %q: got (%d, %v), wanted (%d, nil)", c.str, n, err, len(c.str)) - } - - // Read the values from the stack and check them. - if s.TCPRecvBufSize != c.final { - t.Errorf("TCPRecvBufferSize, case = %q: got %v, wanted %v", c.str, s.TCPRecvBufSize, c.final) - } - } -} diff --git a/pkg/sentry/fs/ramfs/BUILD b/pkg/sentry/fs/ramfs/BUILD deleted file mode 100644 index 516efcc4c..000000000 --- a/pkg/sentry/fs/ramfs/BUILD +++ /dev/null @@ -1,37 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_stateify:defs.bzl", "go_library", "go_test") - -go_library( - name = "ramfs", - srcs = [ - "dir.go", - "socket.go", - "symlink.go", - "tree.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/ramfs", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/sentry/context", - "//pkg/sentry/fs", - "//pkg/sentry/fs/anon", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usermem", - "//pkg/syserror", - "//pkg/waiter", - ], -) - -go_test( - name = "ramfs_test", - size = "small", - srcs = ["tree_test.go"], - embed = [":ramfs"], - deps = [ - "//pkg/sentry/context/contexttest", - "//pkg/sentry/fs", - ], -) diff --git a/pkg/sentry/fs/ramfs/ramfs_state_autogen.go b/pkg/sentry/fs/ramfs/ramfs_state_autogen.go new file mode 100755 index 000000000..bc86a01a1 --- /dev/null +++ b/pkg/sentry/fs/ramfs/ramfs_state_autogen.go @@ -0,0 +1,94 @@ +// automatically generated by stateify. + +package ramfs + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *Dir) beforeSave() {} +func (x *Dir) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Save("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Save("children", &x.children) + m.Save("dentryMap", &x.dentryMap) +} + +func (x *Dir) afterLoad() {} +func (x *Dir) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Load("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Load("children", &x.children) + m.Load("dentryMap", &x.dentryMap) +} + +func (x *dirFileOperations) beforeSave() {} +func (x *dirFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("dirCursor", &x.dirCursor) + m.Save("dir", &x.dir) +} + +func (x *dirFileOperations) afterLoad() {} +func (x *dirFileOperations) load(m state.Map) { + m.Load("dirCursor", &x.dirCursor) + m.Load("dir", &x.dir) +} + +func (x *Socket) beforeSave() {} +func (x *Socket) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Save("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Save("ep", &x.ep) +} + +func (x *Socket) afterLoad() {} +func (x *Socket) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Load("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Load("ep", &x.ep) +} + +func (x *socketFileOperations) beforeSave() {} +func (x *socketFileOperations) save(m state.Map) { + x.beforeSave() +} + +func (x *socketFileOperations) afterLoad() {} +func (x *socketFileOperations) load(m state.Map) { +} + +func (x *Symlink) beforeSave() {} +func (x *Symlink) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Save("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Save("Target", &x.Target) +} + +func (x *Symlink) afterLoad() {} +func (x *Symlink) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Load("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Load("Target", &x.Target) +} + +func (x *symlinkFileOperations) beforeSave() {} +func (x *symlinkFileOperations) save(m state.Map) { + x.beforeSave() +} + +func (x *symlinkFileOperations) afterLoad() {} +func (x *symlinkFileOperations) load(m state.Map) { +} + +func init() { + state.Register("ramfs.Dir", (*Dir)(nil), state.Fns{Save: (*Dir).save, Load: (*Dir).load}) + state.Register("ramfs.dirFileOperations", (*dirFileOperations)(nil), state.Fns{Save: (*dirFileOperations).save, Load: (*dirFileOperations).load}) + state.Register("ramfs.Socket", (*Socket)(nil), state.Fns{Save: (*Socket).save, Load: (*Socket).load}) + state.Register("ramfs.socketFileOperations", (*socketFileOperations)(nil), state.Fns{Save: (*socketFileOperations).save, Load: (*socketFileOperations).load}) + state.Register("ramfs.Symlink", (*Symlink)(nil), state.Fns{Save: (*Symlink).save, Load: (*Symlink).load}) + state.Register("ramfs.symlinkFileOperations", (*symlinkFileOperations)(nil), state.Fns{Save: (*symlinkFileOperations).save, Load: (*symlinkFileOperations).load}) +} diff --git a/pkg/sentry/fs/ramfs/tree_test.go b/pkg/sentry/fs/ramfs/tree_test.go deleted file mode 100644 index 61a7e2900..000000000 --- a/pkg/sentry/fs/ramfs/tree_test.go +++ /dev/null @@ -1,80 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ramfs - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" -) - -func TestMakeDirectoryTree(t *testing.T) { - - for _, test := range []struct { - name string - subdirs []string - }{ - { - name: "abs paths", - subdirs: []string{ - "/tmp", - "/tmp/a/b", - "/tmp/a/c/d", - "/tmp/c", - "/proc", - "/dev/a/b", - "/tmp", - }, - }, - { - name: "rel paths", - subdirs: []string{ - "tmp", - "tmp/a/b", - "tmp/a/c/d", - "tmp/c", - "proc", - "dev/a/b", - "tmp", - }, - }, - } { - ctx := contexttest.Context(t) - mount := fs.NewPseudoMountSource(ctx) - tree, err := MakeDirectoryTree(ctx, mount, test.subdirs) - if err != nil { - t.Errorf("%s: failed to make ramfs tree, got error %v, want nil", test.name, err) - continue - } - - // Expect to be able to find each of the paths. - mm, err := fs.NewMountNamespace(ctx, tree) - if err != nil { - t.Errorf("%s: failed to create mount manager: %v", test.name, err) - continue - } - root := mm.Root() - defer mm.DecRef() - - for _, p := range test.subdirs { - maxTraversals := uint(0) - if _, err := mm.FindInode(ctx, root, nil, p, &maxTraversals); err != nil { - t.Errorf("%s: failed to find node %s: %v", test.name, p, err) - break - } - } - } -} diff --git a/pkg/sentry/fs/sys/BUILD b/pkg/sentry/fs/sys/BUILD deleted file mode 100644 index 70fa3af89..000000000 --- a/pkg/sentry/fs/sys/BUILD +++ /dev/null @@ -1,25 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_stateify:defs.bzl", "go_library") - -go_library( - name = "sys", - srcs = [ - "device.go", - "devices.go", - "fs.go", - "sys.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/sys", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/sentry/context", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fs/ramfs", - "//pkg/sentry/kernel", - "//pkg/sentry/usermem", - ], -) diff --git a/pkg/sentry/fs/sys/sys_state_autogen.go b/pkg/sentry/fs/sys/sys_state_autogen.go new file mode 100755 index 000000000..603057309 --- /dev/null +++ b/pkg/sentry/fs/sys/sys_state_autogen.go @@ -0,0 +1,34 @@ +// automatically generated by stateify. + +package sys + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *cpunum) beforeSave() {} +func (x *cpunum) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Save("InodeStaticFileGetter", &x.InodeStaticFileGetter) +} + +func (x *cpunum) afterLoad() {} +func (x *cpunum) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Load("InodeStaticFileGetter", &x.InodeStaticFileGetter) +} + +func (x *filesystem) beforeSave() {} +func (x *filesystem) save(m state.Map) { + x.beforeSave() +} + +func (x *filesystem) afterLoad() {} +func (x *filesystem) load(m state.Map) { +} + +func init() { + state.Register("sys.cpunum", (*cpunum)(nil), state.Fns{Save: (*cpunum).save, Load: (*cpunum).load}) + state.Register("sys.filesystem", (*filesystem)(nil), state.Fns{Save: (*filesystem).save, Load: (*filesystem).load}) +} diff --git a/pkg/sentry/fs/timerfd/BUILD b/pkg/sentry/fs/timerfd/BUILD deleted file mode 100644 index 1d80daeaf..000000000 --- a/pkg/sentry/fs/timerfd/BUILD +++ /dev/null @@ -1,20 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_stateify:defs.bzl", "go_library") - -go_library( - name = "timerfd", - srcs = ["timerfd.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/timerfd", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/sentry/context", - "//pkg/sentry/fs", - "//pkg/sentry/fs/anon", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/kernel/time", - "//pkg/sentry/usermem", - "//pkg/syserror", - "//pkg/waiter", - ], -) diff --git a/pkg/sentry/fs/timerfd/timerfd_state_autogen.go b/pkg/sentry/fs/timerfd/timerfd_state_autogen.go new file mode 100755 index 000000000..e8d98af97 --- /dev/null +++ b/pkg/sentry/fs/timerfd/timerfd_state_autogen.go @@ -0,0 +1,25 @@ +// automatically generated by stateify. + +package timerfd + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *TimerOperations) beforeSave() {} +func (x *TimerOperations) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.events) { m.Failf("events is %v, expected zero", x.events) } + m.Save("timer", &x.timer) + m.Save("val", &x.val) +} + +func (x *TimerOperations) afterLoad() {} +func (x *TimerOperations) load(m state.Map) { + m.Load("timer", &x.timer) + m.Load("val", &x.val) +} + +func init() { + state.Register("timerfd.TimerOperations", (*TimerOperations)(nil), state.Fns{Save: (*TimerOperations).save, Load: (*TimerOperations).load}) +} diff --git a/pkg/sentry/fs/tmpfs/BUILD b/pkg/sentry/fs/tmpfs/BUILD deleted file mode 100644 index 8f7eb5757..000000000 --- a/pkg/sentry/fs/tmpfs/BUILD +++ /dev/null @@ -1,50 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_stateify:defs.bzl", "go_library", "go_test") - -go_library( - name = "tmpfs", - srcs = [ - "device.go", - "file_regular.go", - "fs.go", - "inode_file.go", - "tmpfs.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/metric", - "//pkg/sentry/context", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fs/ramfs", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/pipe", - "//pkg/sentry/kernel/time", - "//pkg/sentry/memmap", - "//pkg/sentry/safemem", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usage", - "//pkg/sentry/usermem", - "//pkg/syserror", - "//pkg/waiter", - ], -) - -go_test( - name = "tmpfs_test", - size = "small", - srcs = ["file_test.go"], - embed = [":tmpfs"], - deps = [ - "//pkg/sentry/context", - "//pkg/sentry/fs", - "//pkg/sentry/kernel/contexttest", - "//pkg/sentry/usage", - "//pkg/sentry/usermem", - ], -) diff --git a/pkg/sentry/fs/tmpfs/file_test.go b/pkg/sentry/fs/tmpfs/file_test.go deleted file mode 100644 index 0075ef023..000000000 --- a/pkg/sentry/fs/tmpfs/file_test.go +++ /dev/null @@ -1,72 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package tmpfs - -import ( - "bytes" - "testing" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" - "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" -) - -func newFileInode(ctx context.Context) *fs.Inode { - m := fs.NewCachingMountSource(ctx, &Filesystem{}, fs.MountSourceFlags{}) - iops := NewInMemoryFile(ctx, usage.Tmpfs, fs.WithCurrentTime(ctx, fs.UnstableAttr{})) - return fs.NewInode(ctx, iops, m, fs.StableAttr{ - DeviceID: tmpfsDevice.DeviceID(), - InodeID: tmpfsDevice.NextIno(), - BlockSize: usermem.PageSize, - Type: fs.RegularFile, - }) -} - -func newFile(ctx context.Context) *fs.File { - inode := newFileInode(ctx) - f, _ := inode.GetFile(ctx, fs.NewDirent(ctx, inode, "stub"), fs.FileFlags{Read: true, Write: true}) - return f -} - -// Allocate once, write twice. -func TestGrow(t *testing.T) { - ctx := contexttest.Context(t) - f := newFile(ctx) - defer f.DecRef() - - abuf := bytes.Repeat([]byte{'a'}, 68) - n, err := f.Pwritev(ctx, usermem.BytesIOSequence(abuf), 0) - if n != int64(len(abuf)) || err != nil { - t.Fatalf("Pwritev got (%d, %v) want (%d, nil)", n, err, len(abuf)) - } - - bbuf := bytes.Repeat([]byte{'b'}, 856) - n, err = f.Pwritev(ctx, usermem.BytesIOSequence(bbuf), 68) - if n != int64(len(bbuf)) || err != nil { - t.Fatalf("Pwritev got (%d, %v) want (%d, nil)", n, err, len(bbuf)) - } - - rbuf := make([]byte, len(abuf)+len(bbuf)) - n, err = f.Preadv(ctx, usermem.BytesIOSequence(rbuf), 0) - if n != int64(len(rbuf)) || err != nil { - t.Fatalf("Preadv got (%d, %v) want (%d, nil)", n, err, len(rbuf)) - } - - if want := append(abuf, bbuf...); !bytes.Equal(rbuf, want) { - t.Fatalf("Read %v, want %v", rbuf, want) - } -} diff --git a/pkg/sentry/fs/tmpfs/tmpfs_state_autogen.go b/pkg/sentry/fs/tmpfs/tmpfs_state_autogen.go new file mode 100755 index 000000000..7d73d1c77 --- /dev/null +++ b/pkg/sentry/fs/tmpfs/tmpfs_state_autogen.go @@ -0,0 +1,108 @@ +// automatically generated by stateify. + +package tmpfs + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *regularFileOperations) beforeSave() {} +func (x *regularFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("iops", &x.iops) +} + +func (x *regularFileOperations) afterLoad() {} +func (x *regularFileOperations) load(m state.Map) { + m.Load("iops", &x.iops) +} + +func (x *Filesystem) beforeSave() {} +func (x *Filesystem) save(m state.Map) { + x.beforeSave() +} + +func (x *Filesystem) afterLoad() {} +func (x *Filesystem) load(m state.Map) { +} + +func (x *fileInodeOperations) beforeSave() {} +func (x *fileInodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Save("kernel", &x.kernel) + m.Save("memUsage", &x.memUsage) + m.Save("attr", &x.attr) + m.Save("mappings", &x.mappings) + m.Save("writableMappingPages", &x.writableMappingPages) + m.Save("data", &x.data) + m.Save("seals", &x.seals) +} + +func (x *fileInodeOperations) afterLoad() {} +func (x *fileInodeOperations) load(m state.Map) { + m.Load("InodeSimpleExtendedAttributes", &x.InodeSimpleExtendedAttributes) + m.Load("kernel", &x.kernel) + m.Load("memUsage", &x.memUsage) + m.Load("attr", &x.attr) + m.Load("mappings", &x.mappings) + m.Load("writableMappingPages", &x.writableMappingPages) + m.Load("data", &x.data) + m.Load("seals", &x.seals) +} + +func (x *Dir) beforeSave() {} +func (x *Dir) save(m state.Map) { + x.beforeSave() + m.Save("ramfsDir", &x.ramfsDir) + m.Save("kernel", &x.kernel) +} + +func (x *Dir) load(m state.Map) { + m.Load("ramfsDir", &x.ramfsDir) + m.Load("kernel", &x.kernel) + m.AfterLoad(x.afterLoad) +} + +func (x *Symlink) beforeSave() {} +func (x *Symlink) save(m state.Map) { + x.beforeSave() + m.Save("Symlink", &x.Symlink) +} + +func (x *Symlink) afterLoad() {} +func (x *Symlink) load(m state.Map) { + m.Load("Symlink", &x.Symlink) +} + +func (x *Socket) beforeSave() {} +func (x *Socket) save(m state.Map) { + x.beforeSave() + m.Save("Socket", &x.Socket) +} + +func (x *Socket) afterLoad() {} +func (x *Socket) load(m state.Map) { + m.Load("Socket", &x.Socket) +} + +func (x *Fifo) beforeSave() {} +func (x *Fifo) save(m state.Map) { + x.beforeSave() + m.Save("InodeOperations", &x.InodeOperations) +} + +func (x *Fifo) afterLoad() {} +func (x *Fifo) load(m state.Map) { + m.Load("InodeOperations", &x.InodeOperations) +} + +func init() { + state.Register("tmpfs.regularFileOperations", (*regularFileOperations)(nil), state.Fns{Save: (*regularFileOperations).save, Load: (*regularFileOperations).load}) + state.Register("tmpfs.Filesystem", (*Filesystem)(nil), state.Fns{Save: (*Filesystem).save, Load: (*Filesystem).load}) + state.Register("tmpfs.fileInodeOperations", (*fileInodeOperations)(nil), state.Fns{Save: (*fileInodeOperations).save, Load: (*fileInodeOperations).load}) + state.Register("tmpfs.Dir", (*Dir)(nil), state.Fns{Save: (*Dir).save, Load: (*Dir).load}) + state.Register("tmpfs.Symlink", (*Symlink)(nil), state.Fns{Save: (*Symlink).save, Load: (*Symlink).load}) + state.Register("tmpfs.Socket", (*Socket)(nil), state.Fns{Save: (*Socket).save, Load: (*Socket).load}) + state.Register("tmpfs.Fifo", (*Fifo)(nil), state.Fns{Save: (*Fifo).save, Load: (*Fifo).load}) +} diff --git a/pkg/sentry/fs/tty/BUILD b/pkg/sentry/fs/tty/BUILD deleted file mode 100644 index 5e9327aec..000000000 --- a/pkg/sentry/fs/tty/BUILD +++ /dev/null @@ -1,46 +0,0 @@ -package(licenses = ["notice"]) - -load("//tools/go_stateify:defs.bzl", "go_library", "go_test") - -go_library( - name = "tty", - srcs = [ - "dir.go", - "fs.go", - "line_discipline.go", - "master.go", - "queue.go", - "slave.go", - "terminal.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/fs/tty", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/refs", - "//pkg/sentry/arch", - "//pkg/sentry/context", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/safemem", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/unimpl", - "//pkg/sentry/usermem", - "//pkg/syserror", - "//pkg/waiter", - ], -) - -go_test( - name = "tty_test", - size = "small", - srcs = ["tty_test.go"], - embed = [":tty"], - deps = [ - "//pkg/abi/linux", - "//pkg/sentry/context/contexttest", - "//pkg/sentry/usermem", - ], -) diff --git a/pkg/sentry/fs/tty/tty_state_autogen.go b/pkg/sentry/fs/tty/tty_state_autogen.go new file mode 100755 index 000000000..6c9845627 --- /dev/null +++ b/pkg/sentry/fs/tty/tty_state_autogen.go @@ -0,0 +1,202 @@ +// automatically generated by stateify. + +package tty + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *dirInodeOperations) beforeSave() {} +func (x *dirInodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Save("msrc", &x.msrc) + m.Save("master", &x.master) + m.Save("slaves", &x.slaves) + m.Save("dentryMap", &x.dentryMap) + m.Save("next", &x.next) +} + +func (x *dirInodeOperations) afterLoad() {} +func (x *dirInodeOperations) load(m state.Map) { + m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) + m.Load("msrc", &x.msrc) + m.Load("master", &x.master) + m.Load("slaves", &x.slaves) + m.Load("dentryMap", &x.dentryMap) + m.Load("next", &x.next) +} + +func (x *dirFileOperations) beforeSave() {} +func (x *dirFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("di", &x.di) + m.Save("dirCursor", &x.dirCursor) +} + +func (x *dirFileOperations) afterLoad() {} +func (x *dirFileOperations) load(m state.Map) { + m.Load("di", &x.di) + m.Load("dirCursor", &x.dirCursor) +} + +func (x *filesystem) beforeSave() {} +func (x *filesystem) save(m state.Map) { + x.beforeSave() +} + +func (x *filesystem) afterLoad() {} +func (x *filesystem) load(m state.Map) { +} + +func (x *superOperations) beforeSave() {} +func (x *superOperations) save(m state.Map) { + x.beforeSave() +} + +func (x *superOperations) afterLoad() {} +func (x *superOperations) load(m state.Map) { +} + +func (x *lineDiscipline) beforeSave() {} +func (x *lineDiscipline) save(m state.Map) { + x.beforeSave() + if !state.IsZeroValue(x.masterWaiter) { m.Failf("masterWaiter is %v, expected zero", x.masterWaiter) } + if !state.IsZeroValue(x.slaveWaiter) { m.Failf("slaveWaiter is %v, expected zero", x.slaveWaiter) } + m.Save("size", &x.size) + m.Save("inQueue", &x.inQueue) + m.Save("outQueue", &x.outQueue) + m.Save("termios", &x.termios) + m.Save("column", &x.column) +} + +func (x *lineDiscipline) afterLoad() {} +func (x *lineDiscipline) load(m state.Map) { + m.Load("size", &x.size) + m.Load("inQueue", &x.inQueue) + m.Load("outQueue", &x.outQueue) + m.Load("termios", &x.termios) + m.Load("column", &x.column) +} + +func (x *outputQueueTransformer) beforeSave() {} +func (x *outputQueueTransformer) save(m state.Map) { + x.beforeSave() +} + +func (x *outputQueueTransformer) afterLoad() {} +func (x *outputQueueTransformer) load(m state.Map) { +} + +func (x *inputQueueTransformer) beforeSave() {} +func (x *inputQueueTransformer) save(m state.Map) { + x.beforeSave() +} + +func (x *inputQueueTransformer) afterLoad() {} +func (x *inputQueueTransformer) load(m state.Map) { +} + +func (x *masterInodeOperations) beforeSave() {} +func (x *masterInodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) + m.Save("d", &x.d) +} + +func (x *masterInodeOperations) afterLoad() {} +func (x *masterInodeOperations) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) + m.Load("d", &x.d) +} + +func (x *masterFileOperations) beforeSave() {} +func (x *masterFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("d", &x.d) + m.Save("t", &x.t) +} + +func (x *masterFileOperations) afterLoad() {} +func (x *masterFileOperations) load(m state.Map) { + m.Load("d", &x.d) + m.Load("t", &x.t) +} + +func (x *queue) beforeSave() {} +func (x *queue) save(m state.Map) { + x.beforeSave() + m.Save("readBuf", &x.readBuf) + m.Save("waitBuf", &x.waitBuf) + m.Save("waitBufLen", &x.waitBufLen) + m.Save("readable", &x.readable) + m.Save("transformer", &x.transformer) +} + +func (x *queue) afterLoad() {} +func (x *queue) load(m state.Map) { + m.Load("readBuf", &x.readBuf) + m.Load("waitBuf", &x.waitBuf) + m.Load("waitBufLen", &x.waitBufLen) + m.Load("readable", &x.readable) + m.Load("transformer", &x.transformer) +} + +func (x *slaveInodeOperations) beforeSave() {} +func (x *slaveInodeOperations) save(m state.Map) { + x.beforeSave() + m.Save("SimpleFileInode", &x.SimpleFileInode) + m.Save("d", &x.d) + m.Save("t", &x.t) +} + +func (x *slaveInodeOperations) afterLoad() {} +func (x *slaveInodeOperations) load(m state.Map) { + m.Load("SimpleFileInode", &x.SimpleFileInode) + m.Load("d", &x.d) + m.Load("t", &x.t) +} + +func (x *slaveFileOperations) beforeSave() {} +func (x *slaveFileOperations) save(m state.Map) { + x.beforeSave() + m.Save("si", &x.si) +} + +func (x *slaveFileOperations) afterLoad() {} +func (x *slaveFileOperations) load(m state.Map) { + m.Load("si", &x.si) +} + +func (x *Terminal) beforeSave() {} +func (x *Terminal) save(m state.Map) { + x.beforeSave() + m.Save("AtomicRefCount", &x.AtomicRefCount) + m.Save("n", &x.n) + m.Save("d", &x.d) + m.Save("ld", &x.ld) +} + +func (x *Terminal) afterLoad() {} +func (x *Terminal) load(m state.Map) { + m.Load("AtomicRefCount", &x.AtomicRefCount) + m.Load("n", &x.n) + m.Load("d", &x.d) + m.Load("ld", &x.ld) +} + +func init() { + state.Register("tty.dirInodeOperations", (*dirInodeOperations)(nil), state.Fns{Save: (*dirInodeOperations).save, Load: (*dirInodeOperations).load}) + state.Register("tty.dirFileOperations", (*dirFileOperations)(nil), state.Fns{Save: (*dirFileOperations).save, Load: (*dirFileOperations).load}) + state.Register("tty.filesystem", (*filesystem)(nil), state.Fns{Save: (*filesystem).save, Load: (*filesystem).load}) + state.Register("tty.superOperations", (*superOperations)(nil), state.Fns{Save: (*superOperations).save, Load: (*superOperations).load}) + state.Register("tty.lineDiscipline", (*lineDiscipline)(nil), state.Fns{Save: (*lineDiscipline).save, Load: (*lineDiscipline).load}) + state.Register("tty.outputQueueTransformer", (*outputQueueTransformer)(nil), state.Fns{Save: (*outputQueueTransformer).save, Load: (*outputQueueTransformer).load}) + state.Register("tty.inputQueueTransformer", (*inputQueueTransformer)(nil), state.Fns{Save: (*inputQueueTransformer).save, Load: (*inputQueueTransformer).load}) + state.Register("tty.masterInodeOperations", (*masterInodeOperations)(nil), state.Fns{Save: (*masterInodeOperations).save, Load: (*masterInodeOperations).load}) + state.Register("tty.masterFileOperations", (*masterFileOperations)(nil), state.Fns{Save: (*masterFileOperations).save, Load: (*masterFileOperations).load}) + state.Register("tty.queue", (*queue)(nil), state.Fns{Save: (*queue).save, Load: (*queue).load}) + state.Register("tty.slaveInodeOperations", (*slaveInodeOperations)(nil), state.Fns{Save: (*slaveInodeOperations).save, Load: (*slaveInodeOperations).load}) + state.Register("tty.slaveFileOperations", (*slaveFileOperations)(nil), state.Fns{Save: (*slaveFileOperations).save, Load: (*slaveFileOperations).load}) + state.Register("tty.Terminal", (*Terminal)(nil), state.Fns{Save: (*Terminal).save, Load: (*Terminal).load}) +} diff --git a/pkg/sentry/fs/tty/tty_test.go b/pkg/sentry/fs/tty/tty_test.go deleted file mode 100644 index 59f07ff8e..000000000 --- a/pkg/sentry/fs/tty/tty_test.go +++ /dev/null @@ -1,56 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package tty - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" - "gvisor.dev/gvisor/pkg/sentry/usermem" -) - -func TestSimpleMasterToSlave(t *testing.T) { - ld := newLineDiscipline(linux.DefaultSlaveTermios) - ctx := contexttest.Context(t) - inBytes := []byte("hello, tty\n") - src := usermem.BytesIOSequence(inBytes) - outBytes := make([]byte, 32) - dst := usermem.BytesIOSequence(outBytes) - - // Write to the input queue. - nw, err := ld.inputQueueWrite(ctx, src) - if err != nil { - t.Fatalf("error writing to input queue: %v", err) - } - if nw != int64(len(inBytes)) { - t.Fatalf("wrote wrong length: got %d, want %d", nw, len(inBytes)) - } - - // Read from the input queue. - nr, err := ld.inputQueueRead(ctx, dst) - if err != nil { - t.Fatalf("error reading from input queue: %v", err) - } - if nr != int64(len(inBytes)) { - t.Fatalf("read wrong length: got %d, want %d", nr, len(inBytes)) - } - - outStr := string(outBytes[:nr]) - inStr := string(inBytes) - if outStr != inStr { - t.Fatalf("written and read strings do not match: got %q, want %q", outStr, inStr) - } -} |