diff options
Diffstat (limited to 'pkg/sentry/fsimpl')
114 files changed, 292 insertions, 21043 deletions
diff --git a/pkg/sentry/fsimpl/devtmpfs/BUILD b/pkg/sentry/fsimpl/devtmpfs/BUILD deleted file mode 100644 index aa0c2ad8c..000000000 --- a/pkg/sentry/fsimpl/devtmpfs/BUILD +++ /dev/null @@ -1,33 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -licenses(["notice"]) - -go_library( - name = "devtmpfs", - srcs = ["devtmpfs.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fspath", - "//pkg/sentry/fsimpl/tmpfs", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - "//pkg/sync", - ], -) - -go_test( - name = "devtmpfs_test", - size = "small", - srcs = ["devtmpfs_test.go"], - library = ":devtmpfs", - deps = [ - "//pkg/abi/linux", - "//pkg/fspath", - "//pkg/sentry/contexttest", - "//pkg/sentry/fsimpl/tmpfs", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - ], -) diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go deleted file mode 100644 index 64f1b142c..000000000 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go +++ /dev/null @@ -1,196 +0,0 @@ -// Copyright 2020 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 devtmpfs provides an implementation of /dev based on tmpfs, -// analogous to Linux's devtmpfs. -package devtmpfs - -import ( - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" -) - -// Name is the default filesystem name. -const Name = "devtmpfs" - -// FilesystemType implements vfs.FilesystemType. -type FilesystemType struct { - initOnce sync.Once - initErr error - - // fs is the tmpfs filesystem that backs all mounts of this FilesystemType. - // root is fs' root. fs and root are immutable. - fs *vfs.Filesystem - root *vfs.Dentry -} - -// Name implements vfs.FilesystemType.Name. -func (*FilesystemType) Name() string { - return Name -} - -// GetFilesystem implements vfs.FilesystemType.GetFilesystem. -func (fst *FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - fst.initOnce.Do(func() { - fs, root, err := tmpfs.FilesystemType{}.GetFilesystem(ctx, vfsObj, creds, "" /* source */, vfs.GetFilesystemOptions{ - Data: "mode=0755", // opts from drivers/base/devtmpfs.c:devtmpfs_init() - }) - if err != nil { - fst.initErr = err - return - } - fst.fs = fs - fst.root = root - }) - if fst.initErr != nil { - return nil, nil, fst.initErr - } - fst.fs.IncRef() - fst.root.IncRef() - return fst.fs, fst.root, nil -} - -// Accessor allows devices to create device special files in devtmpfs. -type Accessor struct { - vfsObj *vfs.VirtualFilesystem - mntns *vfs.MountNamespace - root vfs.VirtualDentry - creds *auth.Credentials -} - -// NewAccessor returns an Accessor that supports creation of device special -// files in the devtmpfs instance registered with name fsTypeName in vfsObj. -func NewAccessor(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, fsTypeName string) (*Accessor, error) { - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "devtmpfs" /* source */, fsTypeName, &vfs.GetFilesystemOptions{}) - if err != nil { - return nil, err - } - return &Accessor{ - vfsObj: vfsObj, - mntns: mntns, - root: mntns.Root(), - creds: creds, - }, nil -} - -// Release must be called when a is no longer in use. -func (a *Accessor) Release() { - a.root.DecRef() - a.mntns.DecRef() -} - -// accessorContext implements context.Context by extending an existing -// context.Context with an Accessor's values for VFS-relevant state. -type accessorContext struct { - context.Context - a *Accessor -} - -func (a *Accessor) wrapContext(ctx context.Context) *accessorContext { - return &accessorContext{ - Context: ctx, - a: a, - } -} - -// Value implements context.Context.Value. -func (ac *accessorContext) Value(key interface{}) interface{} { - switch key { - case vfs.CtxMountNamespace: - ac.a.mntns.IncRef() - return ac.a.mntns - case vfs.CtxRoot: - ac.a.root.IncRef() - return ac.a.root - default: - return ac.Context.Value(key) - } -} - -func (a *Accessor) pathOperationAt(pathname string) *vfs.PathOperation { - return &vfs.PathOperation{ - Root: a.root, - Start: a.root, - Path: fspath.Parse(pathname), - } -} - -// CreateDeviceFile creates a device special file at the given pathname in the -// devtmpfs instance accessed by the Accessor. -func (a *Accessor) CreateDeviceFile(ctx context.Context, pathname string, kind vfs.DeviceKind, major, minor uint32, perms uint16) error { - mode := (linux.FileMode)(perms) - switch kind { - case vfs.BlockDevice: - mode |= linux.S_IFBLK - case vfs.CharDevice: - mode |= linux.S_IFCHR - default: - panic(fmt.Sprintf("invalid vfs.DeviceKind: %v", kind)) - } - // NOTE: Linux's devtmpfs refuses to automatically delete files it didn't - // create, which it recognizes by storing a pointer to the kdevtmpfs struct - // thread in struct inode::i_private. Accessor doesn't yet support deletion - // of files at all, and probably won't as long as we don't need to support - // kernel modules, so this is moot for now. - return a.vfsObj.MknodAt(a.wrapContext(ctx), a.creds, a.pathOperationAt(pathname), &vfs.MknodOptions{ - Mode: mode, - DevMajor: major, - DevMinor: minor, - }) -} - -// UserspaceInit creates symbolic links and mount points in the devtmpfs -// instance accessed by the Accessor that are created by userspace in Linux. It -// does not create mounts. -func (a *Accessor) UserspaceInit(ctx context.Context) error { - actx := a.wrapContext(ctx) - - // systemd: src/shared/dev-setup.c:dev_setup() - for _, symlink := range []struct { - source string - target string - }{ - // /proc/kcore is not implemented. - {source: "fd", target: "/proc/self/fd"}, - {source: "stdin", target: "/proc/self/fd/0"}, - {source: "stdout", target: "/proc/self/fd/1"}, - {source: "stderr", target: "/proc/self/fd/2"}, - } { - if err := a.vfsObj.SymlinkAt(actx, a.creds, a.pathOperationAt(symlink.source), symlink.target); err != nil { - return fmt.Errorf("failed to create symlink %q => %q: %v", symlink.source, symlink.target, err) - } - } - - // systemd: src/core/mount-setup.c:mount_table - for _, dir := range []string{ - "shm", - "pts", - } { - if err := a.vfsObj.MkdirAt(actx, a.creds, a.pathOperationAt(dir), &vfs.MkdirOptions{ - // systemd: src/core/mount-setup.c:mount_one() - Mode: 0755, - }); err != nil { - return fmt.Errorf("failed to create directory %q: %v", dir, err) - } - } - - return nil -} diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go deleted file mode 100644 index b6d52c015..000000000 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go +++ /dev/null @@ -1,122 +0,0 @@ -// Copyright 2020 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 devtmpfs - -import ( - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -func TestDevtmpfs(t *testing.T) { - ctx := contexttest.Context(t) - creds := auth.CredentialsFromContext(ctx) - - vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - t.Fatalf("VFS init: %v", err) - } - // Register tmpfs just so that we can have a root filesystem that isn't - // devtmpfs. - vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - vfsObj.MustRegisterFilesystemType("devtmpfs", &FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - - // Create a test mount namespace with devtmpfs mounted at "/dev". - const devPath = "/dev" - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "tmpfs" /* source */, "tmpfs" /* fsTypeName */, &vfs.GetFilesystemOptions{}) - if err != nil { - t.Fatalf("failed to create tmpfs root mount: %v", err) - } - defer mntns.DecRef() - root := mntns.Root() - defer root.DecRef() - devpop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(devPath), - } - if err := vfsObj.MkdirAt(ctx, creds, &devpop, &vfs.MkdirOptions{ - Mode: 0755, - }); err != nil { - t.Fatalf("failed to create mount point: %v", err) - } - if err := vfsObj.MountAt(ctx, creds, "devtmpfs" /* source */, &devpop, "devtmpfs" /* fsTypeName */, &vfs.MountOptions{}); err != nil { - t.Fatalf("failed to mount devtmpfs: %v", err) - } - - a, err := NewAccessor(ctx, vfsObj, creds, "devtmpfs") - if err != nil { - t.Fatalf("failed to create devtmpfs.Accessor: %v", err) - } - defer a.Release() - - // Create "userspace-initialized" files using a devtmpfs.Accessor. - if err := a.UserspaceInit(ctx); err != nil { - t.Fatalf("failed to userspace-initialize devtmpfs: %v", err) - } - // Created files should be visible in the test mount namespace. - abspath := devPath + "/fd" - target, err := vfsObj.ReadlinkAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(abspath), - }) - if want := "/proc/self/fd"; err != nil || target != want { - t.Fatalf("readlink(%q): got (%q, %v), wanted (%q, nil)", abspath, target, err, want) - } - - // Create a dummy device special file using a devtmpfs.Accessor. - const ( - pathInDev = "dummy" - kind = vfs.CharDevice - major = 12 - minor = 34 - perms = 0600 - wantMode = linux.S_IFCHR | perms - ) - if err := a.CreateDeviceFile(ctx, pathInDev, kind, major, minor, perms); err != nil { - t.Fatalf("failed to create device file: %v", err) - } - // The device special file should be visible in the test mount namespace. - abspath = devPath + "/" + pathInDev - stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(abspath), - }, &vfs.StatOptions{ - Mask: linux.STATX_TYPE | linux.STATX_MODE, - }) - if err != nil { - t.Fatalf("failed to stat device file at %q: %v", abspath, err) - } - if stat.Mode != wantMode { - t.Errorf("device file mode: got %v, wanted %v", stat.Mode, wantMode) - } - if stat.RdevMajor != major { - t.Errorf("major device number: got %v, wanted %v", stat.RdevMajor, major) - } - if stat.RdevMinor != minor { - t.Errorf("minor device number: got %v, wanted %v", stat.RdevMinor, minor) - } -} diff --git a/pkg/sentry/fsimpl/ext/BUILD b/pkg/sentry/fsimpl/ext/BUILD deleted file mode 100644 index d83d75b3d..000000000 --- a/pkg/sentry/fsimpl/ext/BUILD +++ /dev/null @@ -1,89 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -package(licenses = ["notice"]) - -go_template_instance( - name = "dirent_list", - out = "dirent_list.go", - package = "ext", - prefix = "dirent", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*dirent", - "Linker": "*dirent", - }, -) - -go_library( - name = "ext", - srcs = [ - "block_map_file.go", - "dentry.go", - "directory.go", - "dirent_list.go", - "ext.go", - "extent_file.go", - "file_description.go", - "filesystem.go", - "inode.go", - "regular_file.go", - "symlink.go", - "utils.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/context", - "//pkg/fd", - "//pkg/fspath", - "//pkg/log", - "//pkg/safemem", - "//pkg/sentry/arch", - "//pkg/sentry/fs", - "//pkg/sentry/fsimpl/ext/disklayout", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/memmap", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/syscalls/linux", - "//pkg/sentry/vfs", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - "//pkg/waiter", - ], -) - -go_test( - name = "ext_test", - size = "small", - srcs = [ - "block_map_test.go", - "ext_test.go", - "extent_test.go", - ], - data = [ - "//pkg/sentry/fsimpl/ext:assets/bigfile.txt", - "//pkg/sentry/fsimpl/ext:assets/file.txt", - "//pkg/sentry/fsimpl/ext:assets/tiny.ext2", - "//pkg/sentry/fsimpl/ext:assets/tiny.ext3", - "//pkg/sentry/fsimpl/ext:assets/tiny.ext4", - ], - library = ":ext", - deps = [ - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/context", - "//pkg/fspath", - "//pkg/sentry/contexttest", - "//pkg/sentry/fsimpl/ext/disklayout", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - "//pkg/syserror", - "//pkg/usermem", - "//runsc/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/fsimpl/ext/README.md b/pkg/sentry/fsimpl/ext/README.md deleted file mode 100644 index af00cfda8..000000000 --- a/pkg/sentry/fsimpl/ext/README.md +++ /dev/null @@ -1,117 +0,0 @@ -## EXT(2/3/4) File System - -This is a filesystem driver which supports ext2, ext3 and ext4 filesystems. -Linux has specialized drivers for each variant but none which supports all. This -library takes advantage of ext's backward compatibility and understands the -internal organization of on-disk structures to support all variants. - -This driver implementation diverges from the Linux implementations in being more -forgiving about versioning. For instance, if a filesystem contains both extent -based inodes and classical block map based inodes, this driver will not complain -and interpret them both correctly. While in Linux this would be an issue. This -blurs the line between the three ext fs variants. - -Ext2 is considered deprecated as of Red Hat Enterprise Linux 7, and ext3 has -been superseded by ext4 by large performance gains. Thus it is recommended to -upgrade older filesystem images to ext4 using e2fsprogs for better performance. - -### Read Only - -This driver currently only allows read only operations. A lot of the design -decisions are based on this feature. There are plans to implement write (the -process for which is documented in the future work section). - -### Performance - -One of the biggest wins about this driver is that it directly talks to the -underlying block device (or whatever persistent storage is being used), instead -of making expensive RPCs to a gofer. - -Another advantage is that ext fs supports fast concurrent reads. Currently the -device is represented using a `io.ReaderAt` which allows for concurrent reads. -All reads are directly passed to the device driver which intelligently serves -the read requests in the optimal order. There is no congestion due to locking -while reading in the filesystem level. - -Reads are optimized further in the way file data is transferred over to user -memory. Ext fs directly copies over file data from disk into user memory with no -additional allocations on the way. We can only get faster by preloading file -data into memory (see future work section). - -The internal structures used to represent files, inodes and file descriptors use -a lot of inheritance. With the level of indirection that an interface adds with -an internal pointer, it can quickly fragment a structure across memory. As this -runs along side a full blown kernel (which is memory intensive), having a -fragmented struct might hurt performance. Hence these internal structures, -though interfaced, are tightly packed in memory using the same inheritance -pattern that pkg/sentry/vfs uses. The pkg/sentry/fsimpl/ext/disklayout package -makes an execption to this pattern for reasons documented in the package. - -### Security - -This driver also intends to help sandbox the container better by reducing the -surface of the host kernel that the application touches. It prevents the -application from exploiting vulnerabilities in the host filesystem driver. All -`io.ReaderAt.ReadAt()` calls are translated to `pread(2)` which are directly -passed to the device driver in the kernel. Hence this reduces the surface for -attack. - -The application can not affect any host filesystems other than the one passed -via block device by the user. - -### Future Work - -#### Write - -To support write operations we would need to modify the block device underneath. -Currently, the driver does not modify the device at all, not even for updating -the access times for reads. Modifying the filesystem incorrectly can corrupt it -and render it unreadable for other correct ext(x) drivers. Hence caution must be -maintained while modifying metadata structures. - -Ext4 specifically is built for performance and has added a lot of complexity as -to how metadata structures are modified. For instance, files that are organized -via an extent tree which must be balanced and file data blocks must be placed in -the same extent as much as possible to increase locality. Such properties must -be maintained while modifying the tree. - -Ext filesystems boast a lot about locality, which plays a big role in them being -performant. The block allocation algorithm in Linux does a good job in keeping -related data together. This behavior must be maintained as much as possible, -else we might end up degrading the filesystem performance over time. - -Ext4 also supports a wide variety of features which are specialized for varying -use cases. Implementing all of them can get difficult very quickly. - -Ext(x) checksums all its metadata structures to check for corruption, so -modification of any metadata struct must correspond with re-checksumming the -struct. Linux filesystem drivers also order on-disk updates intelligently to not -corrupt the filesystem and also remain performant. The in-memory metadata -structures must be kept in sync with what is on disk. - -There is also replication of some important structures across the filesystem. -All replicas must be updated when their original copy is updated. There is also -provisioning for snapshotting which must be kept in mind, although it should not -affect this implementation unless we allow users to create filesystem snapshots. - -Ext4 also introduced journaling (jbd2). The journal must be updated -appropriately. - -#### Performance - -To improve performance we should implement a buffer cache, and optionally, read -ahead for small files. While doing so we must also keep in mind the memory usage -and have a reasonable cap on how much file data we want to hold in memory. - -#### Features - -Our current implementation will work with most ext4 filesystems for readonly -purposed. However, the following features are not supported yet: - -- Journal -- Snapshotting -- Extended Attributes -- Hash Tree Directories -- Meta Block Groups -- Multiple Mount Protection -- Bigalloc diff --git a/pkg/sentry/fsimpl/ext/assets/README.md b/pkg/sentry/fsimpl/ext/assets/README.md deleted file mode 100644 index 6f1e81b3a..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/assets/bigfile.txt b/pkg/sentry/fsimpl/ext/assets/bigfile.txt deleted file mode 100644 index 3857cf516..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/assets/file.txt b/pkg/sentry/fsimpl/ext/assets/file.txt deleted file mode 100644 index 980a0d5f1..000000000 --- a/pkg/sentry/fsimpl/ext/assets/file.txt +++ /dev/null @@ -1 +0,0 @@ -Hello World! diff --git a/pkg/sentry/fsimpl/ext/assets/symlink.txt b/pkg/sentry/fsimpl/ext/assets/symlink.txt deleted file mode 120000 index 4c330738c..000000000 --- a/pkg/sentry/fsimpl/ext/assets/symlink.txt +++ /dev/null @@ -1 +0,0 @@ -file.txt
\ No newline at end of file diff --git a/pkg/sentry/fsimpl/ext/assets/tiny.ext2 b/pkg/sentry/fsimpl/ext/assets/tiny.ext2 Binary files differdeleted file mode 100644 index 381ade9bf..000000000 --- a/pkg/sentry/fsimpl/ext/assets/tiny.ext2 +++ /dev/null diff --git a/pkg/sentry/fsimpl/ext/assets/tiny.ext3 b/pkg/sentry/fsimpl/ext/assets/tiny.ext3 Binary files differdeleted file mode 100644 index 0e97a324c..000000000 --- a/pkg/sentry/fsimpl/ext/assets/tiny.ext3 +++ /dev/null diff --git a/pkg/sentry/fsimpl/ext/assets/tiny.ext4 b/pkg/sentry/fsimpl/ext/assets/tiny.ext4 Binary files differdeleted file mode 100644 index a6859736d..000000000 --- a/pkg/sentry/fsimpl/ext/assets/tiny.ext4 +++ /dev/null diff --git a/pkg/sentry/fsimpl/ext/benchmark/BUILD b/pkg/sentry/fsimpl/ext/benchmark/BUILD deleted file mode 100644 index 6c5a559fd..000000000 --- a/pkg/sentry/fsimpl/ext/benchmark/BUILD +++ /dev/null @@ -1,17 +0,0 @@ -load("//tools:defs.bzl", "go_test") - -package(licenses = ["notice"]) - -go_test( - name = "benchmark_test", - size = "small", - srcs = ["benchmark_test.go"], - deps = [ - "//pkg/context", - "//pkg/fspath", - "//pkg/sentry/contexttest", - "//pkg/sentry/fsimpl/ext", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - ], -) diff --git a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go deleted file mode 100644 index 89caee3df..000000000 --- a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go +++ /dev/null @@ -1,206 +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. - -// These benchmarks emulate memfs benchmarks. Ext4 images must be created -// before this benchmark is run using the `make_deep_ext4.sh` script at -// /tmp/image-{depth}.ext4 for all the depths tested below. -// -// The benchmark itself cannot run the script because the script requires -// sudo privileges to create the file system images. -package benchmark_test - -import ( - "fmt" - "os" - "runtime" - "strings" - "testing" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -var depths = []int{1, 2, 3, 8, 64, 100} - -const filename = "file.txt" - -// setUp opens imagePath as an ext Filesystem and returns all necessary -// elements required to run tests. If error is nil, it also returns a tear -// down function which must be called after the test is run for clean up. -func setUp(b *testing.B, imagePath string) (context.Context, *vfs.VirtualFilesystem, *vfs.VirtualDentry, func(), error) { - f, err := os.Open(imagePath) - if err != nil { - return nil, nil, nil, nil, err - } - - ctx := contexttest.Context(b) - creds := auth.CredentialsFromContext(ctx) - - // Create VFS. - vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - return nil, nil, nil, nil, err - } - vfsObj.MustRegisterFilesystemType("extfs", ext.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, imagePath, "extfs", &vfs.GetFilesystemOptions{InternalData: int(f.Fd())}) - if err != nil { - f.Close() - return nil, nil, nil, nil, err - } - - root := mntns.Root() - - tearDown := func() { - root.DecRef() - - if err := f.Close(); err != nil { - b.Fatalf("tearDown failed: %v", err) - } - } - return ctx, vfsObj, &root, tearDown, nil -} - -// mount mounts extfs at the path operation passed. Returns a tear down -// function which must be called after the test is run for clean up. -func mount(b *testing.B, imagePath string, vfsfs *vfs.VirtualFilesystem, pop *vfs.PathOperation) func() { - b.Helper() - - f, err := os.Open(imagePath) - if err != nil { - b.Fatalf("could not open image at %s: %v", imagePath, err) - } - - ctx := contexttest.Context(b) - creds := auth.CredentialsFromContext(ctx) - - if err := vfsfs.MountAt(ctx, creds, imagePath, pop, "extfs", &vfs.MountOptions{ - GetFilesystemOptions: vfs.GetFilesystemOptions{ - InternalData: int(f.Fd()), - }, - }); err != nil { - b.Fatalf("failed to mount tmpfs submount: %v", err) - } - return func() { - if err := f.Close(); err != nil { - b.Fatalf("tearDown failed: %v", err) - } - } -} - -// BenchmarkVFS2Ext4fsStat emulates BenchmarkVFS2MemfsStat. -func BenchmarkVFS2Ext4fsStat(b *testing.B) { - for _, depth := range depths { - b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { - ctx, vfsfs, root, tearDown, err := setUp(b, fmt.Sprintf("/tmp/image-%d.ext4", depth)) - if err != nil { - b.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - creds := auth.CredentialsFromContext(ctx) - var filePathBuilder strings.Builder - filePathBuilder.WriteByte('/') - for i := 1; i <= depth; i++ { - filePathBuilder.WriteString(fmt.Sprintf("%d", i)) - filePathBuilder.WriteByte('/') - } - filePathBuilder.WriteString(filename) - filePath := filePathBuilder.String() - - runtime.GC() - b.ResetTimer() - for i := 0; i < b.N; i++ { - stat, err := vfsfs.StatAt(ctx, creds, &vfs.PathOperation{ - Root: *root, - Start: *root, - Path: fspath.Parse(filePath), - FollowFinalSymlink: true, - }, &vfs.StatOptions{}) - if err != nil { - b.Fatalf("stat(%q) failed: %v", filePath, err) - } - // Sanity check. - if stat.Size > 0 { - b.Fatalf("got wrong file size (%d)", stat.Size) - } - } - }) - } -} - -// BenchmarkVFS2ExtfsMountStat emulates BenchmarkVFS2MemfsMountStat. -func BenchmarkVFS2ExtfsMountStat(b *testing.B) { - for _, depth := range depths { - b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { - // Create root extfs with depth 1 so we can mount extfs again at /1/. - ctx, vfsfs, root, tearDown, err := setUp(b, fmt.Sprintf("/tmp/image-%d.ext4", 1)) - if err != nil { - b.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - creds := auth.CredentialsFromContext(ctx) - mountPointName := "/1/" - pop := vfs.PathOperation{ - Root: *root, - Start: *root, - Path: fspath.Parse(mountPointName), - } - - // Save the mount point for later use. - mountPoint, err := vfsfs.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) - if err != nil { - b.Fatalf("failed to walk to mount point: %v", err) - } - defer mountPoint.DecRef() - - // Create extfs submount. - mountTearDown := mount(b, fmt.Sprintf("/tmp/image-%d.ext4", depth), vfsfs, &pop) - defer mountTearDown() - - var filePathBuilder strings.Builder - filePathBuilder.WriteString(mountPointName) - for i := 1; i <= depth; i++ { - filePathBuilder.WriteString(fmt.Sprintf("%d", i)) - filePathBuilder.WriteByte('/') - } - filePathBuilder.WriteString(filename) - filePath := filePathBuilder.String() - - runtime.GC() - b.ResetTimer() - for i := 0; i < b.N; i++ { - stat, err := vfsfs.StatAt(ctx, creds, &vfs.PathOperation{ - Root: *root, - Start: *root, - Path: fspath.Parse(filePath), - FollowFinalSymlink: true, - }, &vfs.StatOptions{}) - if err != nil { - b.Fatalf("stat(%q) failed: %v", filePath, err) - } - // Sanity check. touch(1) always creates files of size 0 (empty). - if stat.Size > 0 { - b.Fatalf("got wrong file size (%d)", stat.Size) - } - } - }) - } -} diff --git a/pkg/sentry/fsimpl/ext/benchmark/make_deep_ext4.sh b/pkg/sentry/fsimpl/ext/benchmark/make_deep_ext4.sh deleted file mode 100755 index d0910da1f..000000000 --- a/pkg/sentry/fsimpl/ext/benchmark/make_deep_ext4.sh +++ /dev/null @@ -1,72 +0,0 @@ -#!/bin/bash - -# 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. - -# This script creates an ext4 image with $1 depth of directories and a file in -# the inner most directory. The created file is at path /1/2/.../depth/file.txt. -# The ext4 image is written to $2. The image is temporarily mounted at -# /tmp/mountpoint. This script must be run with sudo privileges. - -# Usage: -# sudo bash make_deep_ext4.sh {depth} {output path} - -# Check positional arguments. -if [ "$#" -ne 2 ]; then - echo "Usage: sudo bash make_deep_ext4.sh {depth} {output path}" - exit 1 -fi - -# Make sure depth is a non-negative number. -if ! [[ "$1" =~ ^[0-9]+$ ]]; then - echo "Depth must be a non-negative number." - exit 1 -fi - -# Create a 1 MB filesystem image at the requested output path. -rm -f $2 -fallocate -l 1M $2 -if [ $? -ne 0 ]; then - echo "fallocate failed" - exit $? -fi - -# Convert that blank into an ext4 image. -mkfs.ext4 -j $2 -if [ $? -ne 0 ]; then - echo "mkfs.ext4 failed" - exit $? -fi - -# Mount the image. -MOUNTPOINT=/tmp/mountpoint -mkdir -p $MOUNTPOINT -mount -o loop $2 $MOUNTPOINT -if [ $? -ne 0 ]; then - echo "mount failed" - exit $? -fi - -# Create nested directories and the file. -if [ "$1" -eq 0 ]; then - FILEPATH=$MOUNTPOINT/file.txt -else - FILEPATH=$MOUNTPOINT/$(seq -s '/' 1 $1)/file.txt -fi -mkdir -p $(dirname $FILEPATH) || exit -touch $FILEPATH - -# Clean up. -umount $MOUNTPOINT -rm -rf $MOUNTPOINT diff --git a/pkg/sentry/fsimpl/ext/block_map_file.go b/pkg/sentry/fsimpl/ext/block_map_file.go deleted file mode 100644 index a2d8c3ad6..000000000 --- a/pkg/sentry/fsimpl/ext/block_map_file.go +++ /dev/null @@ -1,200 +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" - "math" - - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/syserror" -) - -const ( - // numDirectBlks is the number of direct blocks in ext block map inodes. - numDirectBlks = 12 -) - -// 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 - - // directBlks are the direct blocks numbers. The physical blocks pointed by - // these holds file data. Contains file blocks 0 to 11. - directBlks [numDirectBlks]uint32 - - // indirectBlk is the physical block which contains (blkSize/4) direct block - // numbers (as uint32 integers). - indirectBlk uint32 - - // doubleIndirectBlk is the physical block which contains (blkSize/4) indirect - // block numbers (as uint32 integers). - doubleIndirectBlk uint32 - - // tripleIndirectBlk is the physical block which contains (blkSize/4) doubly - // indirect block numbers (as uint32 integers). - tripleIndirectBlk uint32 - - // coverage at (i)th index indicates the amount of file data a node at - // height (i) covers. Height 0 is the direct block. - coverage [4]uint64 -} - -// Compiles only if blockMapFile implements io.ReaderAt. -var _ io.ReaderAt = (*blockMapFile)(nil) - -// newBlockMapFile is the blockMapFile constructor. It initializes the file to -// physical blocks map with (at most) the first 12 (direct) blocks. -func newBlockMapFile(regFile regularFile) (*blockMapFile, error) { - file := &blockMapFile{regFile: regFile} - file.regFile.impl = file - - for i := uint(0); i < 4; i++ { - file.coverage[i] = getCoverage(regFile.inode.blkSize, i) - } - - blkMap := regFile.inode.diskInode.Data() - binary.Unmarshal(blkMap[:numDirectBlks*4], binary.LittleEndian, &file.directBlks) - binary.Unmarshal(blkMap[numDirectBlks*4:(numDirectBlks+1)*4], binary.LittleEndian, &file.indirectBlk) - binary.Unmarshal(blkMap[(numDirectBlks+1)*4:(numDirectBlks+2)*4], binary.LittleEndian, &file.doubleIndirectBlk) - binary.Unmarshal(blkMap[(numDirectBlks+2)*4:(numDirectBlks+3)*4], binary.LittleEndian, &file.tripleIndirectBlk) - return file, nil -} - -// ReadAt implements io.ReaderAt.ReadAt. -func (f *blockMapFile) ReadAt(dst []byte, off int64) (int, error) { - if len(dst) == 0 { - return 0, nil - } - - if off < 0 { - return 0, syserror.EINVAL - } - - offset := uint64(off) - size := f.regFile.inode.diskInode.Size() - if offset >= size { - return 0, io.EOF - } - - // dirBlksEnd is the file offset until which direct blocks cover file data. - // Direct blocks cover 0 <= file offset < dirBlksEnd. - dirBlksEnd := numDirectBlks * f.coverage[0] - - // indirBlkEnd is the file offset until which the indirect block covers file - // data. The indirect block covers dirBlksEnd <= file offset < indirBlkEnd. - indirBlkEnd := dirBlksEnd + f.coverage[1] - - // doubIndirBlkEnd is the file offset until which the double indirect block - // covers file data. The double indirect block covers the range - // indirBlkEnd <= file offset < doubIndirBlkEnd. - doubIndirBlkEnd := indirBlkEnd + f.coverage[2] - - read := 0 - toRead := len(dst) - if uint64(toRead)+offset > size { - toRead = int(size - offset) - } - for read < toRead { - var err error - var curR int - - // Figure out which block to delegate the read to. - switch { - case offset < dirBlksEnd: - // Direct block. - curR, err = f.read(f.directBlks[offset/f.regFile.inode.blkSize], offset%f.regFile.inode.blkSize, 0, dst[read:]) - case offset < indirBlkEnd: - // Indirect block. - curR, err = f.read(f.indirectBlk, offset-dirBlksEnd, 1, dst[read:]) - case offset < doubIndirBlkEnd: - // Doubly indirect block. - curR, err = f.read(f.doubleIndirectBlk, offset-indirBlkEnd, 2, dst[read:]) - default: - // Triply indirect block. - curR, err = f.read(f.tripleIndirectBlk, offset-doubIndirBlkEnd, 3, dst[read:]) - } - - read += curR - offset += uint64(curR) - if err != nil { - return read, err - } - } - - if read < len(dst) { - return read, io.EOF - } - return read, nil -} - -// read is the recursive step of the ReadAt function. It relies on knowing the -// current node's location on disk (curPhyBlk) and its height in the block map -// tree. A height of 0 shows that the current node is actually holding file -// data. relFileOff tells the offset from which we need to start to reading -// under the current node. It is completely relative to the current node. -func (f *blockMapFile) read(curPhyBlk uint32, relFileOff uint64, height uint, dst []byte) (int, error) { - curPhyBlkOff := int64(curPhyBlk) * int64(f.regFile.inode.blkSize) - if height == 0 { - toRead := int(f.regFile.inode.blkSize - relFileOff) - if len(dst) < toRead { - toRead = len(dst) - } - - n, _ := f.regFile.inode.fs.dev.ReadAt(dst[:toRead], curPhyBlkOff+int64(relFileOff)) - if n < toRead { - return n, syserror.EIO - } - return n, nil - } - - childCov := f.coverage[height-1] - startIdx := relFileOff / childCov - endIdx := f.regFile.inode.blkSize / 4 // This is exclusive. - wantEndIdx := (relFileOff + uint64(len(dst))) / childCov - wantEndIdx++ // Make this exclusive. - if wantEndIdx < endIdx { - endIdx = wantEndIdx - } - - read := 0 - curChildOff := relFileOff % childCov - for i := startIdx; i < endIdx; i++ { - var childPhyBlk uint32 - err := readFromDisk(f.regFile.inode.fs.dev, curPhyBlkOff+int64(i*4), &childPhyBlk) - if err != nil { - return read, err - } - - n, err := f.read(childPhyBlk, curChildOff, height-1, dst[read:]) - read += n - if err != nil { - return read, err - } - - curChildOff = 0 - } - - return read, nil -} - -// getCoverage returns the number of bytes a node at the given height covers. -// Height 0 is the file data block itself. Height 1 is the indirect block. -// -// Formula: blkSize * ((blkSize / 4)^height) -func getCoverage(blkSize uint64, height uint) uint64 { - return blkSize * uint64(math.Pow(float64(blkSize/4), float64(height))) -} diff --git a/pkg/sentry/fsimpl/ext/block_map_test.go b/pkg/sentry/fsimpl/ext/block_map_test.go deleted file mode 100644 index 181727ef7..000000000 --- a/pkg/sentry/fsimpl/ext/block_map_test.go +++ /dev/null @@ -1,159 +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" - "math/rand" - "testing" - - "github.com/google/go-cmp/cmp" - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" -) - -// These consts are for mocking the block map tree. -const ( - mockBMBlkSize = uint32(16) - mockBMDiskSize = 2500 -) - -// TestBlockMapReader stress tests block map reader functionality. It performs -// random length reads from all possible positions in the block map structure. -func TestBlockMapReader(t *testing.T) { - mockBMFile, want := blockMapSetUp(t) - n := len(want) - - for from := 0; from < n; from++ { - got := make([]byte, n-from) - - if read, err := mockBMFile.ReadAt(got, int64(from)); err != nil { - t.Fatalf("file read operation from offset %d to %d only read %d bytes: %v", from, n, read, err) - } - - if diff := cmp.Diff(got, want[from:]); diff != "" { - t.Fatalf("file data from offset %d to %d mismatched (-want +got):\n%s", from, n, diff) - } - } -} - -// blkNumGen is a number generator which gives block numbers for building the -// block map file on disk. It gives unique numbers in a random order which -// facilitates in creating an extremely fragmented filesystem. -type blkNumGen struct { - nums []uint32 -} - -// newBlkNumGen is the blkNumGen constructor. -func newBlkNumGen() *blkNumGen { - blkNums := &blkNumGen{} - lim := mockBMDiskSize / mockBMBlkSize - blkNums.nums = make([]uint32, lim) - for i := uint32(0); i < lim; i++ { - blkNums.nums[i] = i - } - - rand.Shuffle(int(lim), func(i, j int) { - blkNums.nums[i], blkNums.nums[j] = blkNums.nums[j], blkNums.nums[i] - }) - return blkNums -} - -// next returns the next random block number. -func (n *blkNumGen) next() uint32 { - ret := n.nums[0] - n.nums = n.nums[1:] - return ret -} - -// blockMapSetUp creates a mock disk and a block map file. It initializes the -// block map file with 12 direct block, 1 indirect block, 1 double indirect -// block and 1 triple indirect block (basically fill it till the rim). It -// initializes the disk to reflect the inode. Also returns the file data that -// the inode covers and that is written to disk. -func blockMapSetUp(t *testing.T) (*blockMapFile, []byte) { - mockDisk := make([]byte, mockBMDiskSize) - regFile := regularFile{ - inode: inode{ - fs: &filesystem{ - dev: bytes.NewReader(mockDisk), - }, - diskInode: &disklayout.InodeNew{ - InodeOld: disklayout.InodeOld{ - SizeLo: getMockBMFileFize(), - }, - }, - blkSize: uint64(mockBMBlkSize), - }, - } - - var fileData []byte - blkNums := newBlkNumGen() - var data []byte - - // Write the direct blocks. - for i := 0; i < numDirectBlks; i++ { - curBlkNum := blkNums.next() - data = binary.Marshal(data, binary.LittleEndian, curBlkNum) - fileData = append(fileData, writeFileDataToBlock(mockDisk, curBlkNum, 0, blkNums)...) - } - - // Write to indirect block. - indirectBlk := blkNums.next() - data = binary.Marshal(data, binary.LittleEndian, indirectBlk) - fileData = append(fileData, writeFileDataToBlock(mockDisk, indirectBlk, 1, blkNums)...) - - // Write to indirect block. - doublyIndirectBlk := blkNums.next() - data = binary.Marshal(data, binary.LittleEndian, doublyIndirectBlk) - fileData = append(fileData, writeFileDataToBlock(mockDisk, doublyIndirectBlk, 2, blkNums)...) - - // Write to indirect block. - triplyIndirectBlk := blkNums.next() - data = binary.Marshal(data, binary.LittleEndian, triplyIndirectBlk) - fileData = append(fileData, writeFileDataToBlock(mockDisk, triplyIndirectBlk, 3, blkNums)...) - - copy(regFile.inode.diskInode.Data(), data) - - mockFile, err := newBlockMapFile(regFile) - if err != nil { - t.Fatalf("newBlockMapFile failed: %v", err) - } - return mockFile, fileData -} - -// writeFileDataToBlock writes random bytes to the block on disk. -func writeFileDataToBlock(disk []byte, blkNum uint32, height uint, blkNums *blkNumGen) []byte { - if height == 0 { - start := blkNum * mockBMBlkSize - end := start + mockBMBlkSize - rand.Read(disk[start:end]) - return disk[start:end] - } - - var fileData []byte - for off := blkNum * mockBMBlkSize; off < (blkNum+1)*mockBMBlkSize; off += 4 { - curBlkNum := blkNums.next() - copy(disk[off:off+4], binary.Marshal(nil, binary.LittleEndian, curBlkNum)) - fileData = append(fileData, writeFileDataToBlock(disk, curBlkNum, height-1, blkNums)...) - } - return fileData -} - -// getMockBMFileFize gets the size of the mock block map file which is used for -// testing. -func getMockBMFileFize() uint32 { - return uint32(numDirectBlks*getCoverage(uint64(mockBMBlkSize), 0) + getCoverage(uint64(mockBMBlkSize), 1) + getCoverage(uint64(mockBMBlkSize), 2) + getCoverage(uint64(mockBMBlkSize), 3)) -} diff --git a/pkg/sentry/fsimpl/ext/dentry.go b/pkg/sentry/fsimpl/ext/dentry.go deleted file mode 100644 index a080cb189..000000000 --- a/pkg/sentry/fsimpl/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 -} - -// 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() { - d.inode.incRef() -} - -// TryIncRef implements vfs.DentryImpl.TryIncRef. -func (d *dentry) TryIncRef() bool { - return d.inode.tryIncRef() -} - -// DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef() { - // FIXME(b/134676337): filesystem.mu may not be locked as required by - // inode.decRef(). - d.inode.decRef() -} diff --git a/pkg/sentry/fsimpl/ext/directory.go b/pkg/sentry/fsimpl/ext/directory.go deleted file mode 100644 index bd6ede995..000000000 --- a/pkg/sentry/fsimpl/ext/directory.go +++ /dev/null @@ -1,307 +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/abi/linux" - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserror" -) - -// directory represents a directory inode. It holds the childList in memory. -type directory struct { - inode inode - - // mu serializes the changes to childList. - // Lock Order (outermost locks must be taken first): - // directory.mu - // filesystem.mu - mu sync.Mutex - - // childList is a list containing (1) child dirents and (2) fake dirents - // (with diskDirent == nil) that represent the iteration position of - // directoryFDs. childList is used to support directoryFD.IterDirents() - // efficiently. childList is protected by mu. - childList direntList - - // childMap maps the child's filename to the dirent structure stored in - // childList. This adds some data replication but helps in faster path - // traversal. For consistency, key == childMap[key].diskDirent.FileName(). - // Immutable. - childMap map[string]*dirent -} - -// newDirectroy is the directory constructor. -func newDirectroy(inode inode, newDirent bool) (*directory, error) { - file := &directory{inode: inode, childMap: make(map[string]*dirent)} - file.inode.impl = file - - // Initialize childList by reading dirents from the underlying file. - if inode.diskInode.Flags().Index { - // TODO(b/134676337): Support hash tree directories. Currently only the '.' - // and '..' entries are read in. - - // Users cannot navigate this hash tree directory yet. - log.Warningf("hash tree directory being used which is unsupported") - return file, nil - } - - // The dirents are organized in a linear array in the file data. - // Extract the file data and decode the dirents. - regFile, err := newRegularFile(inode) - if err != nil { - return nil, err - } - - // buf is used as scratch space for reading in dirents from disk and - // unmarshalling them into dirent structs. - buf := make([]byte, disklayout.DirentSize) - size := inode.diskInode.Size() - for off, inc := uint64(0), uint64(0); off < size; off += inc { - toRead := size - off - if toRead > disklayout.DirentSize { - toRead = disklayout.DirentSize - } - if n, err := regFile.impl.ReadAt(buf[:toRead], int64(off)); uint64(n) < toRead { - return nil, err - } - - var curDirent dirent - if newDirent { - curDirent.diskDirent = &disklayout.DirentNew{} - } else { - curDirent.diskDirent = &disklayout.DirentOld{} - } - binary.Unmarshal(buf, binary.LittleEndian, curDirent.diskDirent) - - if curDirent.diskDirent.Inode() != 0 && len(curDirent.diskDirent.FileName()) != 0 { - // Inode number and name length fields being set to 0 is used to indicate - // an unused dirent. - file.childList.PushBack(&curDirent) - file.childMap[curDirent.diskDirent.FileName()] = &curDirent - } - - // The next dirent is placed exactly after this dirent record on disk. - inc = uint64(curDirent.diskDirent.RecordSize()) - } - - return file, nil -} - -func (i *inode) isDir() bool { - _, ok := i.impl.(*directory) - return ok -} - -// dirent is the directory.childList node. -type dirent struct { - diskDirent disklayout.Dirent - - // direntEntry links dirents into their parent directory.childList. - direntEntry -} - -// directoryFD represents a directory file description. It implements -// vfs.FileDescriptionImpl. -type directoryFD struct { - fileDescription - vfs.DirectoryFileDescriptionDefaultImpl - - // Protected by directory.mu. - iter *dirent - off int64 -} - -// Compiles only if directoryFD implements vfs.FileDescriptionImpl. -var _ vfs.FileDescriptionImpl = (*directoryFD)(nil) - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *directoryFD) Release() { - if fd.iter == nil { - return - } - - dir := fd.inode().impl.(*directory) - dir.mu.Lock() - dir.childList.Remove(fd.iter) - dir.mu.Unlock() - fd.iter = nil -} - -// IterDirents implements vfs.FileDescriptionImpl.IterDirents. -func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { - extfs := fd.filesystem() - dir := fd.inode().impl.(*directory) - - dir.mu.Lock() - defer dir.mu.Unlock() - - // Ensure that fd.iter exists and is not linked into dir.childList. - var child *dirent - if fd.iter == nil { - // Start iteration at the beginning of dir. - child = dir.childList.Front() - fd.iter = &dirent{} - } else { - // Continue iteration from where we left off. - child = fd.iter.Next() - dir.childList.Remove(fd.iter) - } - for ; child != nil; child = child.Next() { - // Skip other directoryFD iterators. - if child.diskDirent != nil { - childType, ok := child.diskDirent.FileType() - if !ok { - // We will need to read the inode off disk. Do not increment - // ref count here because this inode is not being added to the - // dentry tree. - extfs.mu.Lock() - childInode, err := extfs.getOrCreateInodeLocked(child.diskDirent.Inode()) - extfs.mu.Unlock() - if err != nil { - // Usage of the file description after the error is - // undefined. This implementation would continue reading - // from the next dirent. - fd.off++ - dir.childList.InsertAfter(child, fd.iter) - return err - } - childType = fs.ToInodeType(childInode.diskInode.Mode().FileType()) - } - - if err := cb.Handle(vfs.Dirent{ - Name: child.diskDirent.FileName(), - Type: fs.ToDirentType(childType), - Ino: uint64(child.diskDirent.Inode()), - NextOff: fd.off + 1, - }); err != nil { - dir.childList.InsertBefore(child, fd.iter) - return err - } - fd.off++ - } - } - dir.childList.PushBack(fd.iter) - return nil -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - if whence != linux.SEEK_SET && whence != linux.SEEK_CUR { - return 0, syserror.EINVAL - } - - dir := fd.inode().impl.(*directory) - - dir.mu.Lock() - defer dir.mu.Unlock() - - // Find resulting offset. - if whence == linux.SEEK_CUR { - offset += fd.off - } - - if offset < 0 { - // lseek(2) specifies that EINVAL should be returned if the resulting offset - // is negative. - return 0, syserror.EINVAL - } - - n := int64(len(dir.childMap)) - realWantOff := offset - if realWantOff > n { - realWantOff = n - } - realCurOff := fd.off - if realCurOff > n { - realCurOff = n - } - - // Ensure that fd.iter exists and is linked into dir.childList so we can - // intelligently seek from the optimal position. - if fd.iter == nil { - fd.iter = &dirent{} - dir.childList.PushFront(fd.iter) - } - - // Guess that iterating from the current position is optimal. - child := fd.iter - diff := realWantOff - realCurOff // Shows direction and magnitude of travel. - - // See if starting from the beginning or end is better. - abDiff := diff - if diff < 0 { - abDiff = -diff - } - if abDiff > realWantOff { - // Starting from the beginning is best. - child = dir.childList.Front() - diff = realWantOff - } else if abDiff > (n - realWantOff) { - // Starting from the end is best. - child = dir.childList.Back() - // (n - 1) because the last non-nil dirent represents the (n-1)th offset. - diff = realWantOff - (n - 1) - } - - for child != nil { - // Skip other directoryFD iterators. - if child.diskDirent != nil { - if diff == 0 { - if child != fd.iter { - dir.childList.Remove(fd.iter) - dir.childList.InsertBefore(child, fd.iter) - } - - fd.off = offset - return offset, nil - } - - if diff < 0 { - diff++ - child = child.Prev() - } else { - diff-- - child = child.Next() - } - continue - } - - if diff < 0 { - child = child.Prev() - } else { - child = child.Next() - } - } - - // Reaching here indicates that the offset is beyond the end of the childList. - dir.childList.Remove(fd.iter) - dir.childList.PushBack(fd.iter) - fd.off = offset - return offset, nil -} - -// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. -func (fd *directoryFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { - // mmap(2) specifies that EACCESS should be returned for non-regular file fds. - return syserror.EACCES -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/BUILD b/pkg/sentry/fsimpl/ext/disklayout/BUILD deleted file mode 100644 index 9bd9c76c0..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/BUILD +++ /dev/null @@ -1,47 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -package(licenses = ["notice"]) - -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", - ], - 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", - ], - library = ":disklayout", - deps = ["//pkg/sentry/kernel/time"], -) diff --git a/pkg/sentry/fsimpl/ext/disklayout/block_group.go b/pkg/sentry/fsimpl/ext/disklayout/block_group.go deleted file mode 100644 index ad6f4fef8..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/disklayout/block_group_32.go b/pkg/sentry/fsimpl/ext/disklayout/block_group_32.go deleted file mode 100644 index 3e16c76db..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/disklayout/block_group_64.go b/pkg/sentry/fsimpl/ext/disklayout/block_group_64.go deleted file mode 100644 index 9a809197a..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/disklayout/block_group_test.go b/pkg/sentry/fsimpl/ext/disklayout/block_group_test.go deleted file mode 100644 index 0ef4294c0..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/disklayout/dirent.go b/pkg/sentry/fsimpl/ext/disklayout/dirent.go deleted file mode 100644 index 417b6cf65..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/dirent.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 - -import ( - "gvisor.dev/gvisor/pkg/sentry/fs" -) - -const ( - // MaxFileName is the maximum length of an ext fs file's name. - MaxFileName = 255 - - // DirentSize is the size of ext dirent structures. - DirentSize = 263 -) - -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/fsimpl/ext/disklayout/dirent_new.go b/pkg/sentry/fsimpl/ext/disklayout/dirent_new.go deleted file mode 100644 index 29ae4a5c2..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/disklayout/dirent_old.go b/pkg/sentry/fsimpl/ext/disklayout/dirent_old.go deleted file mode 100644 index 6fff12a6e..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/disklayout/dirent_test.go b/pkg/sentry/fsimpl/ext/disklayout/dirent_test.go deleted file mode 100644 index 934919f8a..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/dirent_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" -) - -// TestDirentSize tests that the dirent structs are of the correct -// size. -func TestDirentSize(t *testing.T) { - assertSize(t, DirentOld{}, uintptr(DirentSize)) - assertSize(t, DirentNew{}, uintptr(DirentSize)) -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/disklayout.go b/pkg/sentry/fsimpl/ext/disklayout/disklayout.go deleted file mode 100644 index bdf4e2132..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/disklayout/extent.go b/pkg/sentry/fsimpl/ext/disklayout/extent.go deleted file mode 100644 index 4110649ab..000000000 --- a/pkg/sentry/fsimpl/ext/disklayout/extent.go +++ /dev/null @@ -1,143 +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 ( - // ExtentHeaderSize is the size of the header of an extent tree node. - ExtentHeaderSize = 12 - - // ExtentEntrySize is the size of an entry in an extent tree node. - // This size is the same for both leaf and internal nodes. - ExtentEntrySize = 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 represents 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/fsimpl/ext/disklayout/extent_test.go b/pkg/sentry/fsimpl/ext/disklayout/extent_test.go deleted file mode 100644 index 8762b90db..000000000 --- a/pkg/sentry/fsimpl/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{}, ExtentHeaderSize) - assertSize(t, ExtentIdx{}, ExtentEntrySize) - assertSize(t, Extent{}, ExtentEntrySize) -} diff --git a/pkg/sentry/fsimpl/ext/disklayout/inode.go b/pkg/sentry/fsimpl/ext/disklayout/inode.go deleted file mode 100644 index 88ae913f5..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/disklayout/inode_new.go b/pkg/sentry/fsimpl/ext/disklayout/inode_new.go deleted file mode 100644 index 8f9f574ce..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/disklayout/inode_old.go b/pkg/sentry/fsimpl/ext/disklayout/inode_old.go deleted file mode 100644 index db25b11b6..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/disklayout/inode_test.go b/pkg/sentry/fsimpl/ext/disklayout/inode_test.go deleted file mode 100644 index dd03ee50e..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/disklayout/superblock.go b/pkg/sentry/fsimpl/ext/disklayout/superblock.go deleted file mode 100644 index 8bb327006..000000000 --- a/pkg/sentry/fsimpl/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 DirentNew 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/fsimpl/ext/disklayout/superblock_32.go b/pkg/sentry/fsimpl/ext/disklayout/superblock_32.go deleted file mode 100644 index 53e515fd3..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/disklayout/superblock_64.go b/pkg/sentry/fsimpl/ext/disklayout/superblock_64.go deleted file mode 100644 index 7c1053fb4..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/disklayout/superblock_old.go b/pkg/sentry/fsimpl/ext/disklayout/superblock_old.go deleted file mode 100644 index 9221e0251..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/disklayout/superblock_test.go b/pkg/sentry/fsimpl/ext/disklayout/superblock_test.go deleted file mode 100644 index 463b5ba21..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/disklayout/test_utils.go b/pkg/sentry/fsimpl/ext/disklayout/test_utils.go deleted file mode 100644 index 9c63f04c0..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/ext/ext.go b/pkg/sentry/fsimpl/ext/ext.go deleted file mode 100644 index 7176af6d1..000000000 --- a/pkg/sentry/fsimpl/ext/ext.go +++ /dev/null @@ -1,143 +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/context" - "gvisor.dev/gvisor/pkg/fd" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// Name is the name of this filesystem. -const Name = "ext" - -// 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.GetFilesystemOptions) (io.ReaderAt, error) { - if opts.InternalData == nil { - // User mount call. - // TODO(b/134676337): Open the device specified by `source` and return that. - panic("unimplemented") - } - - // GetFilesystem 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 -} - -// isCompatible checks if the superblock has feature sets which are compatible. -// We only need to check the superblock incompatible feature set since we are -// mounting readonly. We will also need to check readonly compatible feature -// set when mounting for read/write. -func isCompatible(sb disklayout.SuperBlock) bool { - // Please note that what is being checked is limited based on the fact that we - // are mounting readonly and that we are not journaling. When mounting - // read/write or with a journal, this must be reevaluated. - incompatFeatures := sb.IncompatibleFeatures() - if incompatFeatures.MetaBG { - log.Warningf("ext fs: meta block groups are not supported") - return false - } - if incompatFeatures.MMP { - log.Warningf("ext fs: multiple mount protection is not supported") - return false - } - if incompatFeatures.Encrypted { - log.Warningf("ext fs: encrypted inodes not supported") - return false - } - if incompatFeatures.InlineData { - log.Warningf("ext fs: inline files not supported") - return false - } - return true -} - -// Name implements vfs.FilesystemType.Name. -func (FilesystemType) Name() string { - return Name -} - -// GetFilesystem implements vfs.FilesystemType.GetFilesystem. -func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - // TODO(b/134676337): Ensure that the user is mounting readonly. If not, - // EACCESS should be returned according to mount(2). Filesystem independent - // flags (like readonly) are currently not available in pkg/sentry/vfs. - - dev, err := getDeviceFd(source, opts) - if err != nil { - return nil, nil, err - } - - fs := filesystem{dev: dev, inodeCache: make(map[uint32]*inode)} - fs.vfsfs.Init(vfsObj, &fsType, &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 - } - - // Refuse to mount if the filesystem is incompatible. - if !isCompatible(fs.sb) { - return nil, nil, syserror.EINVAL - } - - fs.bgs, err = readBlockGroups(dev, fs.sb) - if err != nil { - return nil, nil, err - } - - rootInode, err := fs.getOrCreateInodeLocked(disklayout.RootDirInode) - if err != nil { - return nil, nil, err - } - rootInode.incRef() - - return &fs.vfsfs, &newDentry(rootInode).vfsd, nil -} diff --git a/pkg/sentry/fsimpl/ext/ext_test.go b/pkg/sentry/fsimpl/ext/ext_test.go deleted file mode 100644 index 29bb73765..000000000 --- a/pkg/sentry/fsimpl/ext/ext_test.go +++ /dev/null @@ -1,922 +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" - "io" - "os" - "path" - "sort" - "testing" - - "github.com/google/go-cmp/cmp" - "github.com/google/go-cmp/cmp/cmpopts" - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" - - "gvisor.dev/gvisor/runsc/testutil" -) - -const ( - assetsDir = "pkg/sentry/fsimpl/ext/assets" -) - -var ( - ext2ImagePath = path.Join(assetsDir, "tiny.ext2") - ext3ImagePath = path.Join(assetsDir, "tiny.ext3") - ext4ImagePath = path.Join(assetsDir, "tiny.ext4") -) - -// 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.VirtualFilesystem, *vfs.VirtualDentry, 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 - } - - ctx := contexttest.Context(t) - creds := auth.CredentialsFromContext(ctx) - - // Create VFS. - vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - t.Fatalf("VFS init: %v", err) - } - vfsObj.MustRegisterFilesystemType("extfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, localImagePath, "extfs", &vfs.GetFilesystemOptions{InternalData: int(f.Fd())}) - if err != nil { - f.Close() - return nil, nil, nil, nil, err - } - - root := mntns.Root() - - tearDown := func() { - root.DecRef() - - if err := f.Close(); err != nil { - t.Fatalf("tearDown failed: %v", err) - } - } - return ctx, vfsObj, &root, tearDown, nil -} - -// TODO(b/134676337): Test vfs.FilesystemImpl.ReadlinkAt and -// vfs.FilesystemImpl.StatFSAt which are not implemented in -// vfs.VirtualFilesystem yet. - -// TestSeek tests vfs.FileDescriptionImpl.Seek functionality. -func TestSeek(t *testing.T) { - type seekTest struct { - name string - image string - path string - } - - tests := []seekTest{ - { - name: "ext4 root dir seek", - image: ext4ImagePath, - path: "/", - }, - { - name: "ext3 root dir seek", - image: ext3ImagePath, - path: "/", - }, - { - name: "ext2 root dir seek", - image: ext2ImagePath, - path: "/", - }, - { - name: "ext4 reg file seek", - image: ext4ImagePath, - path: "/file.txt", - }, - { - name: "ext3 reg file seek", - image: ext3ImagePath, - path: "/file.txt", - }, - { - name: "ext2 reg file seek", - image: ext2ImagePath, - path: "/file.txt", - }, - } - - for _, test := range tests { - t.Run(test.name, func(t *testing.T) { - ctx, vfsfs, root, tearDown, err := setUp(t, test.image) - if err != nil { - t.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - fd, err := vfsfs.OpenAt( - ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: *root, Start: *root, Path: fspath.Parse(test.path)}, - &vfs.OpenOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.OpenAt failed: %v", err) - } - - if n, err := fd.Seek(ctx, 0, linux.SEEK_SET); n != 0 || err != nil { - t.Errorf("expected seek position 0, got %d and error %v", n, err) - } - - stat, err := fd.Stat(ctx, vfs.StatOptions{}) - if err != nil { - t.Errorf("fd.stat failed for file %s in image %s: %v", test.path, test.image, err) - } - - // We should be able to seek beyond the end of file. - size := int64(stat.Size) - if n, err := fd.Seek(ctx, size, linux.SEEK_SET); n != size || err != nil { - t.Errorf("expected seek position %d, got %d and error %v", size, n, err) - } - - // EINVAL should be returned if the resulting offset is negative. - if _, err := fd.Seek(ctx, -1, linux.SEEK_SET); err != syserror.EINVAL { - t.Errorf("expected error EINVAL but got %v", err) - } - - if n, err := fd.Seek(ctx, 3, linux.SEEK_CUR); n != size+3 || err != nil { - t.Errorf("expected seek position %d, got %d and error %v", size+3, n, err) - } - - // Make sure negative offsets work with SEEK_CUR. - if n, err := fd.Seek(ctx, -2, linux.SEEK_CUR); n != size+1 || err != nil { - t.Errorf("expected seek position %d, got %d and error %v", size+1, n, err) - } - - // EINVAL should be returned if the resulting offset is negative. - if _, err := fd.Seek(ctx, -(size + 2), linux.SEEK_CUR); err != syserror.EINVAL { - t.Errorf("expected error EINVAL but got %v", err) - } - - // Make sure SEEK_END works with regular files. - if _, ok := fd.Impl().(*regularFileFD); ok { - // Seek back to 0. - if n, err := fd.Seek(ctx, -size, linux.SEEK_END); n != 0 || err != nil { - t.Errorf("expected seek position %d, got %d and error %v", 0, n, err) - } - - // Seek forward beyond EOF. - if n, err := fd.Seek(ctx, 1, linux.SEEK_END); n != size+1 || err != nil { - t.Errorf("expected seek position %d, got %d and error %v", size+1, n, err) - } - - // EINVAL should be returned if the resulting offset is negative. - if _, err := fd.Seek(ctx, -(size + 1), linux.SEEK_END); err != syserror.EINVAL { - t.Errorf("expected error EINVAL but got %v", err) - } - } - }) - } -} - -// TestStatAt tests filesystem.StatAt functionality. -func TestStatAt(t *testing.T) { - type statAtTest struct { - name string - image string - path string - want linux.Statx - } - - tests := []statAtTest{ - { - name: "ext4 statx small file", - image: ext4ImagePath, - path: "/file.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0644 | linux.ModeRegular, - Size: 13, - }, - }, - { - name: "ext3 statx small file", - image: ext3ImagePath, - path: "/file.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0644 | linux.ModeRegular, - Size: 13, - }, - }, - { - name: "ext2 statx small file", - image: ext2ImagePath, - path: "/file.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0644 | linux.ModeRegular, - Size: 13, - }, - }, - { - name: "ext4 statx big file", - image: ext4ImagePath, - path: "/bigfile.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0644 | linux.ModeRegular, - Size: 13042, - }, - }, - { - name: "ext3 statx big file", - image: ext3ImagePath, - path: "/bigfile.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0644 | linux.ModeRegular, - Size: 13042, - }, - }, - { - name: "ext2 statx big file", - image: ext2ImagePath, - path: "/bigfile.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0644 | linux.ModeRegular, - Size: 13042, - }, - }, - { - name: "ext4 statx symlink file", - image: ext4ImagePath, - path: "/symlink.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0777 | linux.ModeSymlink, - Size: 8, - }, - }, - { - name: "ext3 statx symlink file", - image: ext3ImagePath, - path: "/symlink.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0777 | linux.ModeSymlink, - Size: 8, - }, - }, - { - name: "ext2 statx symlink file", - image: ext2ImagePath, - path: "/symlink.txt", - want: linux.Statx{ - Blksize: 0x400, - Nlink: 1, - UID: 0, - GID: 0, - Mode: 0777 | linux.ModeSymlink, - Size: 8, - }, - }, - } - - // Ignore the fields that are not supported by filesystem.StatAt yet and - // those which are likely to change as the image does. - ignoredFields := map[string]bool{ - "Attributes": true, - "AttributesMask": true, - "Atime": true, - "Blocks": true, - "Btime": true, - "Ctime": true, - "DevMajor": true, - "DevMinor": true, - "Ino": true, - "Mask": true, - "Mtime": true, - "RdevMajor": true, - "RdevMinor": true, - } - - for _, test := range tests { - t.Run(test.name, func(t *testing.T) { - ctx, vfsfs, root, tearDown, err := setUp(t, test.image) - if err != nil { - t.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - got, err := vfsfs.StatAt(ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: *root, Start: *root, Path: fspath.Parse(test.path)}, - &vfs.StatOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.StatAt failed for file %s in image %s: %v", test.path, test.image, err) - } - - cmpIgnoreFields := cmp.FilterPath(func(p cmp.Path) bool { - _, ok := ignoredFields[p.String()] - return ok - }, cmp.Ignore()) - if diff := cmp.Diff(got, test.want, cmpIgnoreFields, cmpopts.IgnoreUnexported(linux.Statx{})); diff != "" { - t.Errorf("stat mismatch (-want +got):\n%s", diff) - } - }) - } -} - -// TestRead tests the read functionality for vfs file descriptions. -func TestRead(t *testing.T) { - type readTest struct { - name string - image string - absPath string - } - - tests := []readTest{ - { - name: "ext4 read small file", - image: ext4ImagePath, - absPath: "/file.txt", - }, - { - name: "ext3 read small file", - image: ext3ImagePath, - absPath: "/file.txt", - }, - { - name: "ext2 read small file", - image: ext2ImagePath, - absPath: "/file.txt", - }, - { - name: "ext4 read big file", - image: ext4ImagePath, - absPath: "/bigfile.txt", - }, - { - name: "ext3 read big file", - image: ext3ImagePath, - absPath: "/bigfile.txt", - }, - { - name: "ext2 read big file", - image: ext2ImagePath, - absPath: "/bigfile.txt", - }, - } - - for _, test := range tests { - t.Run(test.name, func(t *testing.T) { - ctx, vfsfs, root, tearDown, err := setUp(t, test.image) - if err != nil { - t.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - fd, err := vfsfs.OpenAt( - ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: *root, Start: *root, Path: fspath.Parse(test.absPath)}, - &vfs.OpenOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.OpenAt failed: %v", err) - } - - // Get a local file descriptor and compare its functionality with a vfs file - // description for the same file. - localFile, err := testutil.FindFile(path.Join(assetsDir, test.absPath)) - if err != nil { - t.Fatalf("testutil.FindFile failed for %s: %v", test.absPath, err) - } - - f, err := os.Open(localFile) - if err != nil { - t.Fatalf("os.Open failed for %s: %v", localFile, err) - } - defer f.Close() - - // Read the entire file by reading one byte repeatedly. Doing this stress - // tests the underlying file reader implementation. - got := make([]byte, 1) - want := make([]byte, 1) - for { - n, err := f.Read(want) - fd.Read(ctx, usermem.BytesIOSequence(got), vfs.ReadOptions{}) - - if diff := cmp.Diff(got, want); diff != "" { - t.Errorf("file data mismatch (-want +got):\n%s", diff) - } - - // Make sure there is no more file data left after getting EOF. - if n == 0 || err == io.EOF { - if n, _ := fd.Read(ctx, usermem.BytesIOSequence(got), vfs.ReadOptions{}); n != 0 { - t.Errorf("extra unexpected file data in file %s in image %s", test.absPath, test.image) - } - - break - } - - if err != nil { - t.Fatalf("read failed: %v", err) - } - } - }) - } -} - -// iterDirentsCb is a simple callback which just keeps adding the dirents to an -// internal list. Implements vfs.IterDirentsCallback. -type iterDirentsCb struct { - dirents []vfs.Dirent -} - -// Compiles only if iterDirentCb implements vfs.IterDirentsCallback. -var _ vfs.IterDirentsCallback = (*iterDirentsCb)(nil) - -// newIterDirentsCb is the iterDirent -func newIterDirentCb() *iterDirentsCb { - return &iterDirentsCb{dirents: make([]vfs.Dirent, 0)} -} - -// Handle implements vfs.IterDirentsCallback.Handle. -func (cb *iterDirentsCb) Handle(dirent vfs.Dirent) error { - cb.dirents = append(cb.dirents, dirent) - return nil -} - -// TestIterDirents tests the FileDescriptionImpl.IterDirents functionality. -func TestIterDirents(t *testing.T) { - type iterDirentTest struct { - name string - image string - path string - want []vfs.Dirent - } - - wantDirents := []vfs.Dirent{ - { - Name: ".", - Type: linux.DT_DIR, - }, - { - Name: "..", - Type: linux.DT_DIR, - }, - { - Name: "lost+found", - Type: linux.DT_DIR, - }, - { - Name: "file.txt", - Type: linux.DT_REG, - }, - { - Name: "bigfile.txt", - Type: linux.DT_REG, - }, - { - Name: "symlink.txt", - Type: linux.DT_LNK, - }, - } - tests := []iterDirentTest{ - { - name: "ext4 root dir iteration", - image: ext4ImagePath, - path: "/", - want: wantDirents, - }, - { - name: "ext3 root dir iteration", - image: ext3ImagePath, - path: "/", - want: wantDirents, - }, - { - name: "ext2 root dir iteration", - image: ext2ImagePath, - path: "/", - want: wantDirents, - }, - } - - for _, test := range tests { - t.Run(test.name, func(t *testing.T) { - ctx, vfsfs, root, tearDown, err := setUp(t, test.image) - if err != nil { - t.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - fd, err := vfsfs.OpenAt( - ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: *root, Start: *root, Path: fspath.Parse(test.path)}, - &vfs.OpenOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.OpenAt failed: %v", err) - } - - cb := &iterDirentsCb{} - if err = fd.IterDirents(ctx, cb); err != nil { - t.Fatalf("dir fd.IterDirents() failed: %v", err) - } - - sort.Slice(cb.dirents, func(i int, j int) bool { return cb.dirents[i].Name < cb.dirents[j].Name }) - sort.Slice(test.want, func(i int, j int) bool { return test.want[i].Name < test.want[j].Name }) - - // Ignore the inode number and offset of dirents because those are likely to - // change as the underlying image changes. - cmpIgnoreFields := cmp.FilterPath(func(p cmp.Path) bool { - return p.String() == "Ino" || p.String() == "NextOff" - }, cmp.Ignore()) - if diff := cmp.Diff(cb.dirents, test.want, cmpIgnoreFields); diff != "" { - t.Errorf("dirents mismatch (-want +got):\n%s", diff) - } - }) - } -} - -// 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) { - _, _, vd, tearDown, err := setUp(t, test.image) - if err != nil { - t.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - d, ok := vd.Dentry().Impl().(*dentry) - if !ok { - t.Fatalf("ext dentry of incorrect type: %T", vd.Dentry().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) { - _, _, vd, tearDown, err := setUp(t, test.image) - if err != nil { - t.Fatalf("setUp failed: %v", err) - } - defer tearDown() - - fs, ok := vd.Mount().Filesystem().Impl().(*filesystem) - if !ok { - t.Fatalf("ext filesystem of incorrect type: %T", vd.Mount().Filesystem().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/fsimpl/ext/extent_file.go b/pkg/sentry/fsimpl/ext/extent_file.go deleted file mode 100644 index 11dcc0346..000000000 --- a/pkg/sentry/fsimpl/ext/extent_file.go +++ /dev/null @@ -1,237 +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/fsimpl/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 io.ReaderAt. -var _ io.ReaderAt = (*extentFile)(nil) - -// 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(regFile regularFile) (*extentFile, error) { - file := &extentFile{regFile: regFile} - file.regFile.impl = file - err := file.buildExtTree() - 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() error { - rootNodeData := f.regFile.inode.diskInode.Data() - - binary.Unmarshal(rootNodeData[:disklayout.ExtentHeaderSize], 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.ExtentEntrySize; i < f.root.Header.NumEntries; i, off = i+1, off+disklayout.ExtentEntrySize { - 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.ExtentEntrySize], 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 = f.buildExtTreeFromDisk(f.root.Entries[i].Entry); 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 (f *extentFile) buildExtTreeFromDisk(entry disklayout.ExtentEntry) (*disklayout.ExtentNode, error) { - var header disklayout.ExtentHeader - off := entry.PhysicalBlock() * f.regFile.inode.blkSize - err := readFromDisk(f.regFile.inode.fs.dev, int64(off), &header) - if err != nil { - return nil, err - } - - entries := make([]disklayout.ExtentEntryPair, header.NumEntries) - for i, off := uint16(0), off+disklayout.ExtentEntrySize; i < header.NumEntries; i, off = i+1, off+disklayout.ExtentEntrySize { - var curEntry disklayout.ExtentEntry - if header.Height == 0 { - // Leaf node. - curEntry = &disklayout.Extent{} - } else { - // Internal node. - curEntry = &disklayout.ExtentIdx{} - } - - err := readFromDisk(f.regFile.inode.fs.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 = f.buildExtTreeFromDisk(entries[i].Entry) - if err != nil { - return nil, err - } - } - } - - return &disklayout.ExtentNode{header, entries}, nil -} - -// ReadAt implements io.ReaderAt.ReadAt. -func (f *extentFile) ReadAt(dst []byte, off int64) (int, error) { - if len(dst) == 0 { - return 0, nil - } - - if off < 0 { - return 0, syserror.EINVAL - } - - if uint64(off) >= f.regFile.inode.diskInode.Size() { - return 0, io.EOF - } - - n, err := f.read(&f.root, uint64(off), dst) - if n < len(dst) && err == nil { - err = io.EOF - } - return n, err -} - -// read is the recursive step of extentFile.ReadAt which traverses the extent -// tree from the node passed and reads file data. -func (f *extentFile) read(node *disklayout.ExtentNode, off uint64, 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 := uint32(off / f.regFile.inode.blkSize) - 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 = f.readFromExtent(node.Entries[i].Entry.(*disklayout.Extent), off, dst[read:]) - } else { - curR, err = f.read(node.Entries[i].Node, off, dst[read:]) - } - - read += curR - off += uint64(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. -// -// 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 (f *extentFile) readFromExtent(ex *disklayout.Extent, off uint64, dst []byte) (int, error) { - curFileBlk := uint32(off / f.regFile.inode.blkSize) - 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*f.regFile.inode.blkSize + (off % f.regFile.inode.blkSize) - - endPhyBlk := ex.PhysicalBlock() + uint64(ex.Length) - extentEnd := endPhyBlk * f.regFile.inode.blkSize // This is exclusive. - - toRead := int(extentEnd - readStart) - if len(dst) < toRead { - toRead = len(dst) - } - - n, _ := f.regFile.inode.fs.dev.ReadAt(dst[:toRead], int64(readStart)) - if n < toRead { - return n, syserror.EIO - } - return n, nil -} diff --git a/pkg/sentry/fsimpl/ext/extent_test.go b/pkg/sentry/fsimpl/ext/extent_test.go deleted file mode 100644 index a2382daa3..000000000 --- a/pkg/sentry/fsimpl/ext/extent_test.go +++ /dev/null @@ -1,267 +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" - "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/fsimpl/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 stress tests extentReader functionality. It performs random -// length reads from all possible positions in the extent tree. -func TestExtentReader(t *testing.T) { - mockExtentFile, want := extentTreeSetUp(t, node0) - n := len(want) - - for from := 0; from < n; from++ { - got := make([]byte, n-from) - - if read, err := mockExtentFile.ReadAt(got, int64(from)); err != nil { - t.Fatalf("file read operation from offset %d to %d only read %d bytes: %v", from, n, read, err) - } - - if diff := cmp.Diff(got, want[from:]); diff != "" { - t.Fatalf("file data from offset %d to %d mismatched (-want +got):\n%s", from, n, 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) (*extentFile, []byte) { - t.Helper() - - mockDisk := make([]byte, mockExtentBlkSize*10) - mockExtentFile := &extentFile{ - regFile: regularFile{ - inode: inode{ - fs: &filesystem{ - dev: bytes.NewReader(mockDisk), - }, - diskInode: &disklayout.InodeNew{ - InodeOld: disklayout.InodeOld{ - SizeLo: uint32(mockExtentBlkSize) * getNumPhyBlks(root), - }, - }, - blkSize: mockExtentBlkSize, - }, - }, - } - - fileData := writeTree(&mockExtentFile.regFile.inode, mockDisk, node0, mockExtentBlkSize) - - if err := mockExtentFile.buildExtTree(); err != nil { - t.Fatalf("inode.buildExtTree failed: %v", err) - } - return 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, writeFileDataToExtent(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, writeFileDataToExtent(disk, ep.Entry.(*disklayout.Extent))...) - } else { - fileData = append(fileData, writeTreeToDisk(disk, ep)...) - } - } - return fileData -} - -// writeFileDataToExtent writes random bytes to the blocks on disk that the -// passed extent points to. -func writeFileDataToExtent(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/fsimpl/ext/file_description.go b/pkg/sentry/fsimpl/ext/file_description.go deleted file mode 100644 index 92f7da40d..000000000 --- a/pkg/sentry/fsimpl/ext/file_description.go +++ /dev/null @@ -1,64 +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/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// fileDescription is embedded by ext implementations of -// vfs.FileDescriptionImpl. -type fileDescription struct { - vfsfd vfs.FileDescription - vfs.FileDescriptionDefaultImpl -} - -func (fd *fileDescription) filesystem() *filesystem { - return fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) -} - -func (fd *fileDescription) inode() *inode { - return fd.vfsfd.Dentry().Impl().(*dentry).inode -} - -// Stat implements vfs.FileDescriptionImpl.Stat. -func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { - var stat linux.Statx - fd.inode().statTo(&stat) - return stat, nil -} - -// SetStat implements vfs.FileDescriptionImpl.SetStat. -func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { - if opts.Stat.Mask == 0 { - return nil - } - return syserror.EPERM -} - -// SetStat implements vfs.FileDescriptionImpl.StatFS. -func (fd *fileDescription) StatFS(ctx context.Context) (linux.Statfs, error) { - var stat linux.Statfs - fd.filesystem().statTo(&stat) - return stat, nil -} - -// Sync implements vfs.FileDescriptionImpl.Sync. -func (fd *fileDescription) Sync(ctx context.Context) error { - return nil -} diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go deleted file mode 100644 index 48eaccdbc..000000000 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ /dev/null @@ -1,519 +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 ( - "errors" - "io" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserror" -) - -var ( - // errResolveDirent indicates that the vfs.ResolvingPath.Component() does - // not exist on the dentry tree but does exist on disk. So it has to be read in - // using the in-memory dirent and added to the dentry tree. Usually indicates - // the need to lock filesystem.mu for writing. - errResolveDirent = errors.New("resolve path component using dirent") -) - -// filesystem implements vfs.FilesystemImpl. -type filesystem struct { - vfsfs vfs.Filesystem - - // mu serializes changes to the Dentry tree. - mu sync.RWMutex - - // dev represents 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 most additions (see IterDirents) and all removals - // 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) - -// stepLocked resolves rp.Component() in parent directory vfsd. The write -// parameter passed tells if the caller has acquired filesystem.mu for writing -// or not. If set to true, an existing inode on disk can be added to the dentry -// tree if not present already. -// -// stepLocked is loosely analogous to fs/namei.c:walk_component(). -// -// Preconditions: -// - filesystem.mu must be locked (for writing if write param is true). -// - !rp.Done(). -// - inode == vfsd.Impl().(*Dentry).inode. -func stepLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, inode *inode, write bool) (*vfs.Dentry, *inode, error) { - if !inode.isDir() { - return nil, nil, syserror.ENOTDIR - } - if err := inode.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { - return nil, nil, err - } - - for { - nextVFSD, err := rp.ResolveComponent(vfsd) - if err != nil { - return nil, nil, err - } - if nextVFSD == nil { - // Since the Dentry tree is not the sole source of truth for extfs, if it's - // not in the Dentry tree, it might need to be pulled from disk. - childDirent, ok := inode.impl.(*directory).childMap[rp.Component()] - if !ok { - // The underlying inode does not exist on disk. - return nil, nil, syserror.ENOENT - } - - if !write { - // filesystem.mu must be held for writing to add to the dentry tree. - return nil, nil, errResolveDirent - } - - // Create and add the component's dirent to the dentry tree. - fs := rp.Mount().Filesystem().Impl().(*filesystem) - childInode, err := fs.getOrCreateInodeLocked(childDirent.diskDirent.Inode()) - if err != nil { - return nil, nil, err - } - // incRef because this is being added to the dentry tree. - childInode.incRef() - child := newDentry(childInode) - vfsd.InsertChild(&child.vfsd, rp.Component()) - - // Continue as usual now that nextVFSD is not nil. - nextVFSD = &child.vfsd - } - nextInode := nextVFSD.Impl().(*dentry).inode - if nextInode.isSymlink() && rp.ShouldFollowSymlink() { - if err := rp.HandleSymlink(inode.impl.(*symlink).target); err != nil { - return nil, nil, err - } - continue - } - rp.Advance() - return nextVFSD, nextInode, nil - } -} - -// walkLocked resolves rp to an existing file. The write parameter -// passed tells if the caller has acquired filesystem.mu for writing or not. -// If set to true, additions can be made to the dentry tree while walking. -// If errResolveDirent is returned, the walk needs to be continued with an -// upgraded filesystem.mu. -// -// walkLocked is loosely analogous to Linux's fs/namei.c:path_lookupat(). -// -// Preconditions: -// - filesystem.mu must be locked (for writing if write param is true). -func walkLocked(rp *vfs.ResolvingPath, write bool) (*vfs.Dentry, *inode, error) { - vfsd := rp.Start() - inode := vfsd.Impl().(*dentry).inode - for !rp.Done() { - var err error - vfsd, inode, err = stepLocked(rp, vfsd, inode, write) - if err != nil { - return nil, nil, err - } - } - if rp.MustBeDir() && !inode.isDir() { - return nil, nil, syserror.ENOTDIR - } - return vfsd, inode, nil -} - -// walkParentLocked resolves all but the last path component of rp to an -// existing directory. It does not check that the returned directory is -// searchable by the provider of rp. The write parameter passed tells if the -// caller has acquired filesystem.mu for writing or not. If set to true, -// additions can be made to the dentry tree while walking. -// If errResolveDirent is returned, the walk needs to be continued with an -// upgraded filesystem.mu. -// -// walkParentLocked is loosely analogous to Linux's fs/namei.c:path_parentat(). -// -// Preconditions: -// - filesystem.mu must be locked (for writing if write param is true). -// - !rp.Done(). -func walkParentLocked(rp *vfs.ResolvingPath, write bool) (*vfs.Dentry, *inode, error) { - vfsd := rp.Start() - inode := vfsd.Impl().(*dentry).inode - for !rp.Final() { - var err error - vfsd, inode, err = stepLocked(rp, vfsd, inode, write) - if err != nil { - return nil, nil, err - } - } - if !inode.isDir() { - return nil, nil, syserror.ENOTDIR - } - return vfsd, inode, nil -} - -// walk resolves rp to an existing file. If parent is set to true, it resolves -// the rp till the parent of the last component which should be an existing -// directory. If parent is false then resolves rp entirely. Attemps to resolve -// the path as far as it can with a read lock and upgrades the lock if needed. -func (fs *filesystem) walk(rp *vfs.ResolvingPath, parent bool) (*vfs.Dentry, *inode, error) { - var ( - vfsd *vfs.Dentry - inode *inode - err error - ) - - // Try walking with the hopes that all dentries have already been pulled out - // of disk. This reduces congestion (allows concurrent walks). - fs.mu.RLock() - if parent { - vfsd, inode, err = walkParentLocked(rp, false) - } else { - vfsd, inode, err = walkLocked(rp, false) - } - fs.mu.RUnlock() - - if err == errResolveDirent { - // Upgrade lock and continue walking. Lock upgrading in the middle of the - // walk is fine as this is a read only filesystem. - fs.mu.Lock() - if parent { - vfsd, inode, err = walkParentLocked(rp, true) - } else { - vfsd, inode, err = walkLocked(rp, true) - } - fs.mu.Unlock() - } - - return vfsd, inode, err -} - -// getOrCreateInodeLocked 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. -// The caller must increment the ref count if adding this to the dentry tree. -// -// Precondition: must be holding fs.mu for writing. -func (fs *filesystem) getOrCreateInodeLocked(inodeNum uint32) (*inode, error) { - if in, ok := fs.inodeCache[inodeNum]; ok { - return in, nil - } - - in, err := newInode(fs, inodeNum) - if err != nil { - return nil, err - } - - fs.inodeCache[inodeNum] = in - return in, nil -} - -// statTo writes the statfs fields to the output parameter. -func (fs *filesystem) statTo(stat *linux.Statfs) { - stat.Type = uint64(fs.sb.Magic()) - stat.BlockSize = int64(fs.sb.BlockSize()) - stat.Blocks = fs.sb.BlocksCount() - stat.BlocksFree = fs.sb.FreeBlocksCount() - stat.BlocksAvailable = fs.sb.FreeBlocksCount() - stat.Files = uint64(fs.sb.InodesCount()) - stat.FilesFree = uint64(fs.sb.FreeInodesCount()) - stat.NameLength = disklayout.MaxFileName - stat.FragmentSize = int64(fs.sb.BlockSize()) - // TODO(b/134676337): Set Statfs.Flags and Statfs.FSID. -} - -// AccessAt implements vfs.Filesystem.Impl.AccessAt. -func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { - _, inode, err := fs.walk(rp, false) - if err != nil { - return err - } - return inode.checkPermissions(rp.Credentials(), ats) -} - -// GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. -func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { - vfsd, inode, err := fs.walk(rp, false) - if err != nil { - return nil, err - } - - if opts.CheckSearchable { - if !inode.isDir() { - return nil, syserror.ENOTDIR - } - if err := inode.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { - return nil, err - } - } - - inode.incRef() - return vfsd, nil -} - -// GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. -func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { - vfsd, inode, err := fs.walk(rp, true) - if err != nil { - return nil, err - } - inode.incRef() - return vfsd, nil -} - -// OpenAt implements vfs.FilesystemImpl.OpenAt. -func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - vfsd, inode, err := fs.walk(rp, false) - if err != nil { - return nil, err - } - - // EROFS is returned if write access is needed. - if vfs.MayWriteFileWithOpenFlags(opts.Flags) || opts.Flags&(linux.O_CREAT|linux.O_EXCL|linux.O_TMPFILE) != 0 { - return nil, syserror.EROFS - } - return inode.open(rp, vfsd, &opts) -} - -// ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. -func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { - _, inode, err := fs.walk(rp, false) - if err != nil { - return "", err - } - symlink, ok := inode.impl.(*symlink) - if !ok { - return "", syserror.EINVAL - } - return symlink.target, nil -} - -// StatAt implements vfs.FilesystemImpl.StatAt. -func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { - _, inode, err := fs.walk(rp, false) - if err != nil { - return linux.Statx{}, err - } - var stat linux.Statx - inode.statTo(&stat) - return stat, nil -} - -// StatFSAt implements vfs.FilesystemImpl.StatFSAt. -func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { - if _, _, err := fs.walk(rp, false); err != nil { - return linux.Statfs{}, err - } - - var stat linux.Statfs - fs.statTo(&stat) - return stat, 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 -// this read-only filesystem. - -// LinkAt implements vfs.FilesystemImpl.LinkAt. -func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { - if rp.Done() { - return syserror.EEXIST - } - - if _, _, err := fs.walk(rp, true); err != nil { - return err - } - - return syserror.EROFS -} - -// MkdirAt implements vfs.FilesystemImpl.MkdirAt. -func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { - if rp.Done() { - return syserror.EEXIST - } - - if _, _, err := fs.walk(rp, true); err != nil { - return err - } - - return syserror.EROFS -} - -// MknodAt implements vfs.FilesystemImpl.MknodAt. -func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { - if rp.Done() { - return syserror.EEXIST - } - - _, _, err := fs.walk(rp, true) - if err != nil { - return err - } - - return syserror.EROFS -} - -// RenameAt implements vfs.FilesystemImpl.RenameAt. -func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldParentVD vfs.VirtualDentry, oldName string, opts vfs.RenameOptions) error { - if rp.Done() { - return syserror.ENOENT - } - - _, _, err := fs.walk(rp, false) - if err != nil { - return err - } - - return syserror.EROFS -} - -// RmdirAt implements vfs.FilesystemImpl.RmdirAt. -func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { - _, inode, err := fs.walk(rp, false) - if err != nil { - return err - } - - if !inode.isDir() { - return syserror.ENOTDIR - } - - return syserror.EROFS -} - -// SetStatAt implements vfs.FilesystemImpl.SetStatAt. -func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { - _, _, err := fs.walk(rp, false) - if err != nil { - return err - } - - return syserror.EROFS -} - -// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. -func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { - if rp.Done() { - return syserror.EEXIST - } - - _, _, err := fs.walk(rp, true) - if err != nil { - return err - } - - return syserror.EROFS -} - -// UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. -func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { - _, inode, err := fs.walk(rp, false) - if err != nil { - return err - } - - if inode.isDir() { - return syserror.EISDIR - } - - return syserror.EROFS -} - -// BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. -func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) { - _, _, err := fs.walk(rp, false) - if err != nil { - return nil, err - } - - // TODO(b/134676337): Support sockets. - return nil, syserror.ECONNREFUSED -} - -// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. -func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { - _, _, err := fs.walk(rp, false) - if err != nil { - return nil, err - } - return nil, syserror.ENOTSUP -} - -// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. -func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { - _, _, err := fs.walk(rp, false) - if err != nil { - return "", err - } - return "", syserror.ENOTSUP -} - -// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. -func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { - _, _, err := fs.walk(rp, false) - if err != nil { - return err - } - return syserror.ENOTSUP -} - -// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. -func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { - _, _, err := fs.walk(rp, false) - if err != nil { - return err - } - return syserror.ENOTSUP -} - -// PrependPath implements vfs.FilesystemImpl.PrependPath. -func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - return vfs.GenericPrependPath(vfsroot, vd, b) -} diff --git a/pkg/sentry/fsimpl/ext/inode.go b/pkg/sentry/fsimpl/ext/inode.go deleted file mode 100644 index a39a37318..000000000 --- a/pkg/sentry/fsimpl/ext/inode.go +++ /dev/null @@ -1,220 +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" - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/ext/disklayout" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "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 -- -// |-- dir -// |-- symlink -// |-- regular-- -// |-- extent file -// |-- block map file -type inode struct { - // refs is a reference count. refs is accessed using atomic memory operations. - refs int64 - - // fs is the containing filesystem. - fs *filesystem - - // inodeNum is the inode number of this inode on disk. This is used to - // identify inodes within the ext filesystem. - inodeNum uint32 - - // blkSize is the fs data block size. Same as filesystem.sb.BlockSize(). - blkSize uint64 - - // 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 filesystem.mu. -func (in *inode) decRef() { - if refs := atomic.AddInt64(&in.refs, -1); refs == 0 { - delete(in.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(fs *filesystem, inodeNum uint32) (*inode, error) { - if inodeNum == 0 { - panic("inode number 0 on ext filesystems is not possible") - } - - inodeRecordSize := fs.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 := fs.sb.InodesPerGroup() - blkSize := fs.sb.BlockSize() - inodeTableOff := fs.bgs[getBGNum(inodeNum, inodesPerGrp)].InodeTable() * blkSize - inodeOff := inodeTableOff + uint64(uint32(inodeRecordSize)*getBGOff(inodeNum, inodesPerGrp)) - - if err := readFromDisk(fs.dev, int64(inodeOff), diskInode); err != nil { - return nil, err - } - - // Build the inode based on its type. - inode := inode{ - fs: fs, - inodeNum: inodeNum, - blkSize: blkSize, - diskInode: diskInode, - } - - switch diskInode.Mode().FileType() { - case linux.ModeSymlink: - f, err := newSymlink(inode) - if err != nil { - return nil, err - } - return &f.inode, nil - case linux.ModeRegular: - f, err := newRegularFile(inode) - if err != nil { - return nil, err - } - return &f.inode, nil - case linux.ModeDirectory: - f, err := newDirectroy(inode, fs.sb.IncompatibleFeatures().DirentFileType) - if err != nil { - return nil, err - } - return &f.inode, nil - default: - // TODO(b/134676337): Return appropriate errors for sockets, pipes and devices. - return nil, syserror.EINVAL - } -} - -// open creates and returns a file description for the dentry passed in. -func (in *inode) open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { - ats := vfs.AccessTypesForOpenFlags(opts) - if err := in.checkPermissions(rp.Credentials(), ats); err != nil { - return nil, err - } - mnt := rp.Mount() - switch in.impl.(type) { - case *regularFile: - var fd regularFileFD - if err := fd.vfsfd.Init(&fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{}); err != nil { - return nil, err - } - return &fd.vfsfd, nil - case *directory: - // Can't open directories writably. This check is not necessary for a read - // only filesystem but will be required when write is implemented. - if ats&vfs.MayWrite != 0 { - return nil, syserror.EISDIR - } - var fd directoryFD - if err := fd.vfsfd.Init(&fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{}); err != nil { - return nil, err - } - return &fd.vfsfd, nil - case *symlink: - if opts.Flags&linux.O_PATH == 0 { - // Can't open symlinks without O_PATH. - return nil, syserror.ELOOP - } - var fd symlinkFD - fd.vfsfd.Init(&fd, opts.Flags, mnt, vfsd, &vfs.FileDescriptionOptions{}) - return &fd.vfsfd, nil - default: - panic(fmt.Sprintf("unknown inode type: %T", in.impl)) - } -} - -func (in *inode) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { - return vfs.GenericCheckPermissions(creds, ats, in.diskInode.Mode(), in.diskInode.UID(), in.diskInode.GID()) -} - -// statTo writes the statx fields to the output parameter. -func (in *inode) statTo(stat *linux.Statx) { - stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | - linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_SIZE | - linux.STATX_ATIME | linux.STATX_CTIME | linux.STATX_MTIME - stat.Blksize = uint32(in.blkSize) - stat.Mode = uint16(in.diskInode.Mode()) - stat.Nlink = uint32(in.diskInode.LinksCount()) - stat.UID = uint32(in.diskInode.UID()) - stat.GID = uint32(in.diskInode.GID()) - stat.Ino = uint64(in.inodeNum) - stat.Size = in.diskInode.Size() - stat.Atime = in.diskInode.AccessTime().StatxTimestamp() - stat.Ctime = in.diskInode.ChangeTime().StatxTimestamp() - stat.Mtime = in.diskInode.ModificationTime().StatxTimestamp() - // TODO(b/134676337): Set stat.Blocks which is the number of 512 byte blocks - // (including metadata blocks) required to represent this file. -} - -// 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/fsimpl/ext/regular_file.go b/pkg/sentry/fsimpl/ext/regular_file.go deleted file mode 100644 index 30135ddb0..000000000 --- a/pkg/sentry/fsimpl/ext/regular_file.go +++ /dev/null @@ -1,159 +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/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// 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. - // io.ReaderAt is more strict than io.Reader in the sense that a partial read - // is always accompanied by an error. If a read spans past the end of file, a - // partial read (within file range) is done and io.EOF is returned. - impl io.ReaderAt -} - -// newRegularFile is the regularFile constructor. It figures out what kind of -// file this is and initializes the fileReader. -func newRegularFile(inode inode) (*regularFile, error) { - regFile := regularFile{ - inode: inode, - } - - inodeFlags := inode.diskInode.Flags() - - if inodeFlags.Extents { - file, err := newExtentFile(regFile) - if err != nil { - return nil, err - } - - file.regFile.inode.impl = &file.regFile - return &file.regFile, nil - } - - file, err := newBlockMapFile(regFile) - if err != nil { - return nil, err - } - file.regFile.inode.impl = &file.regFile - return &file.regFile, nil -} - -func (in *inode) isRegular() bool { - _, ok := in.impl.(*regularFile) - return ok -} - -// directoryFD represents a directory file description. It implements -// vfs.FileDescriptionImpl. -type regularFileFD struct { - fileDescription - - // off is the file offset. off is accessed using atomic memory operations. - off int64 - - // offMu serializes operations that may mutate off. - offMu sync.Mutex -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *regularFileFD) Release() {} - -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - safeReader := safemem.FromIOReaderAt{ - ReaderAt: fd.inode().impl.(*regularFile).impl, - Offset: offset, - } - - // Copies data from disk directly into usermem without any intermediate - // allocations (if dst is converted into BlockSeq such that it does not need - // safe copying). - return dst.CopyOutFrom(ctx, safeReader) -} - -// Read implements vfs.FileDescriptionImpl.Read. -func (fd *regularFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - n, err := fd.PRead(ctx, dst, fd.off, opts) - fd.offMu.Lock() - fd.off += n - fd.offMu.Unlock() - return n, err -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - // write(2) specifies that EBADF must be returned if the fd is not open for - // writing. - return 0, syserror.EBADF -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *regularFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - n, err := fd.PWrite(ctx, src, fd.off, opts) - fd.offMu.Lock() - fd.off += n - fd.offMu.Unlock() - return n, err -} - -// IterDirents implements vfs.FileDescriptionImpl.IterDirents. -func (fd *regularFileFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { - return syserror.ENOTDIR -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *regularFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - fd.offMu.Lock() - defer fd.offMu.Unlock() - switch whence { - case linux.SEEK_SET: - // Use offset as specified. - case linux.SEEK_CUR: - offset += fd.off - case linux.SEEK_END: - offset += int64(fd.inode().diskInode.Size()) - default: - return 0, syserror.EINVAL - } - if offset < 0 { - return 0, syserror.EINVAL - } - fd.off = offset - return offset, nil -} - -// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. -func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { - // TODO(b/134676337): Implement mmap(2). - return syserror.ENODEV -} diff --git a/pkg/sentry/fsimpl/ext/symlink.go b/pkg/sentry/fsimpl/ext/symlink.go deleted file mode 100644 index 1447a4dc1..000000000 --- a/pkg/sentry/fsimpl/ext/symlink.go +++ /dev/null @@ -1,111 +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/context" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// 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(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(inode) - if err != nil { - return nil, err - } - - link = make([]byte, size) - if n, err := regFile.impl.ReadAt(link, 0); uint64(n) < size { - return nil, err - } - } - - file = &symlink{inode: inode, target: string(link)} - file.inode.impl = file - return file, nil -} - -func (in *inode) isSymlink() bool { - _, ok := in.impl.(*symlink) - return ok -} - -// symlinkFD represents a symlink file description and implements implements -// vfs.FileDescriptionImpl. which may only be used if open options contains -// O_PATH. For this reason most of the functions return EBADF. -type symlinkFD struct { - fileDescription -} - -// Compiles only if symlinkFD implements vfs.FileDescriptionImpl. -var _ vfs.FileDescriptionImpl = (*symlinkFD)(nil) - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *symlinkFD) Release() {} - -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *symlinkFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - return 0, syserror.EBADF -} - -// Read implements vfs.FileDescriptionImpl.Read. -func (fd *symlinkFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - return 0, syserror.EBADF -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *symlinkFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - return 0, syserror.EBADF -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *symlinkFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - return 0, syserror.EBADF -} - -// IterDirents implements vfs.FileDescriptionImpl.IterDirents. -func (fd *symlinkFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { - return syserror.ENOTDIR -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *symlinkFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - return 0, syserror.EBADF -} - -// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. -func (fd *symlinkFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { - return syserror.EBADF -} diff --git a/pkg/sentry/fsimpl/ext/utils.go b/pkg/sentry/fsimpl/ext/utils.go deleted file mode 100644 index d8b728f8c..000000000 --- a/pkg/sentry/fsimpl/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/fsimpl/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/fsimpl/gofer/BUILD b/pkg/sentry/fsimpl/gofer/BUILD deleted file mode 100644 index 99d1e3f8f..000000000 --- a/pkg/sentry/fsimpl/gofer/BUILD +++ /dev/null @@ -1,66 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -licenses(["notice"]) - -go_template_instance( - name = "dentry_list", - out = "dentry_list.go", - package = "gofer", - prefix = "dentry", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*dentry", - "Linker": "*dentry", - }, -) - -go_library( - name = "gofer", - srcs = [ - "dentry_list.go", - "directory.go", - "filesystem.go", - "gofer.go", - "handle.go", - "handle_unsafe.go", - "p9file.go", - "pagemath.go", - "regular_file.go", - "special_file.go", - "symlink.go", - "time.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fd", - "//pkg/fspath", - "//pkg/log", - "//pkg/p9", - "//pkg/safemem", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", - "//pkg/sentry/memmap", - "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usage", - "//pkg/sentry/vfs", - "//pkg/syserror", - "//pkg/unet", - "//pkg/usermem", - ], -) - -go_test( - name = "gofer_test", - srcs = ["gofer_test.go"], - library = ":gofer", - deps = [ - "//pkg/p9", - "//pkg/sentry/contexttest", - ], -) diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go deleted file mode 100644 index 49d9f859b..000000000 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ /dev/null @@ -1,199 +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 gofer - -import ( - "sync" - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -func (d *dentry) isDir() bool { - return d.fileType() == linux.S_IFDIR -} - -// Preconditions: d.dirMu must be locked. d.isDir(). fs.opts.interop != -// InteropModeShared. -func (d *dentry) cacheNegativeChildLocked(name string) { - if d.negativeChildren == nil { - d.negativeChildren = make(map[string]struct{}) - } - d.negativeChildren[name] = struct{}{} -} - -type directoryFD struct { - fileDescription - vfs.DirectoryFileDescriptionDefaultImpl - - mu sync.Mutex - off int64 - dirents []vfs.Dirent -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *directoryFD) Release() { -} - -// IterDirents implements vfs.FileDescriptionImpl.IterDirents. -func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { - fd.mu.Lock() - defer fd.mu.Unlock() - - d := fd.dentry() - if fd.dirents == nil { - ds, err := d.getDirents(ctx) - if err != nil { - return err - } - fd.dirents = ds - } - - if d.fs.opts.interop != InteropModeShared { - d.touchAtime(fd.vfsfd.Mount()) - } - - for fd.off < int64(len(fd.dirents)) { - if err := cb.Handle(fd.dirents[fd.off]); err != nil { - return err - } - fd.off++ - } - return nil -} - -// Preconditions: d.isDir(). There exists at least one directoryFD representing d. -func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { - // 9P2000.L's readdir does not specify behavior in the presence of - // concurrent mutation of an iterated directory, so implementations may - // duplicate or omit entries in this case, which violates POSIX semantics. - // Thus we read all directory entries while holding d.dirMu to exclude - // directory mutations. (Note that it is impossible for the client to - // exclude concurrent mutation from other remote filesystem users. Since - // there is no way to detect if the server has incorrectly omitted - // directory entries, we simply assume that the server is well-behaved - // under InteropModeShared.) This is inconsistent with Linux (which appears - // to assume that directory fids have the correct semantics, and translates - // struct file_operations::readdir calls directly to readdir RPCs), but is - // consistent with VFS1. - // - // NOTE(b/135560623): In particular, some gofer implementations may not - // retain state between calls to Readdir, so may not provide a coherent - // directory stream across in the presence of mutation. - - d.fs.renameMu.RLock() - defer d.fs.renameMu.RUnlock() - d.dirMu.Lock() - defer d.dirMu.Unlock() - if d.dirents != nil { - return d.dirents, nil - } - - // It's not clear if 9P2000.L's readdir is expected to return "." and "..", - // so we generate them here. - parent := d.vfsd.ParentOrSelf().Impl().(*dentry) - dirents := []vfs.Dirent{ - { - Name: ".", - Type: linux.DT_DIR, - Ino: d.ino, - NextOff: 1, - }, - { - Name: "..", - Type: uint8(atomic.LoadUint32(&parent.mode) >> 12), - Ino: parent.ino, - NextOff: 2, - }, - } - off := uint64(0) - const count = 64 * 1024 // for consistency with the vfs1 client - d.handleMu.RLock() - defer d.handleMu.RUnlock() - if !d.handleReadable { - // This should not be possible because a readable handle should have - // been opened when the calling directoryFD was opened. - panic("gofer.dentry.getDirents called without a readable handle") - } - for { - p9ds, err := d.handle.file.readdir(ctx, off, count) - if err != nil { - return nil, err - } - if len(p9ds) == 0 { - // Cache dirents for future directoryFDs if permitted. - if d.fs.opts.interop != InteropModeShared { - d.dirents = dirents - } - return dirents, nil - } - for _, p9d := range p9ds { - if p9d.Name == "." || p9d.Name == ".." { - continue - } - dirent := vfs.Dirent{ - Name: p9d.Name, - Ino: p9d.QID.Path, - NextOff: int64(len(dirents) + 1), - } - // p9 does not expose 9P2000.U's DMDEVICE, DMNAMEDPIPE, or - // DMSOCKET. - switch p9d.Type { - case p9.TypeSymlink: - dirent.Type = linux.DT_LNK - case p9.TypeDir: - dirent.Type = linux.DT_DIR - default: - dirent.Type = linux.DT_REG - } - dirents = append(dirents, dirent) - } - off = p9ds[len(p9ds)-1].Offset - } -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - fd.mu.Lock() - defer fd.mu.Unlock() - - switch whence { - case linux.SEEK_SET: - if offset < 0 { - return 0, syserror.EINVAL - } - if offset == 0 { - // Ensure that the next call to fd.IterDirents() calls - // fd.dentry().getDirents(). - fd.dirents = nil - } - fd.off = offset - return fd.off, nil - case linux.SEEK_CUR: - offset += fd.off - if offset < 0 { - return 0, syserror.EINVAL - } - // Don't clear fd.dirents in this case, even if offset == 0. - fd.off = offset - return fd.off, nil - default: - return 0, syserror.EINVAL - } -} diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go deleted file mode 100644 index 137260898..000000000 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ /dev/null @@ -1,1135 +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 gofer - -import ( - "sync" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// Sync implements vfs.FilesystemImpl.Sync. -func (fs *filesystem) Sync(ctx context.Context) error { - // Snapshot current dentries and special files. - fs.syncMu.Lock() - ds := make([]*dentry, 0, len(fs.dentries)) - for d := range fs.dentries { - ds = append(ds, d) - } - sffds := make([]*specialFileFD, 0, len(fs.specialFileFDs)) - for sffd := range fs.specialFileFDs { - sffds = append(sffds, sffd) - } - fs.syncMu.Unlock() - - // Return the first error we encounter, but sync everything we can - // regardless. - var retErr error - - // Sync regular files. - for _, d := range ds { - if !d.TryIncRef() { - continue - } - err := d.syncSharedHandle(ctx) - d.DecRef() - if err != nil && retErr == nil { - retErr = err - } - } - - // Sync special files, which may be writable but do not use dentry shared - // handles (so they won't be synced by the above). - for _, sffd := range sffds { - if !sffd.vfsfd.TryIncRef() { - continue - } - err := sffd.Sync(ctx) - sffd.vfsfd.DecRef() - if err != nil && retErr == nil { - retErr = err - } - } - - return retErr -} - -// maxFilenameLen is the maximum length of a filename. This is dictated by 9P's -// encoding of strings, which uses 2 bytes for the length prefix. -const maxFilenameLen = (1 << 16) - 1 - -// dentrySlicePool is a pool of *[]*dentry used to store dentries for which -// dentry.checkCachingLocked() must be called. The pool holds pointers to -// slices because Go lacks generics, so sync.Pool operates on interface{}, so -// every call to (what should be) sync.Pool<[]*dentry>.Put() allocates a copy -// of the slice header on the heap. -var dentrySlicePool = sync.Pool{ - New: func() interface{} { - ds := make([]*dentry, 0, 4) // arbitrary non-zero initial capacity - return &ds - }, -} - -func appendDentry(ds *[]*dentry, d *dentry) *[]*dentry { - if ds == nil { - ds = dentrySlicePool.Get().(*[]*dentry) - } - *ds = append(*ds, d) - return ds -} - -// Preconditions: ds != nil. -func putDentrySlice(ds *[]*dentry) { - // Allow dentries to be GC'd. - for i := range *ds { - (*ds)[i] = nil - } - *ds = (*ds)[:0] - dentrySlicePool.Put(ds) -} - -// stepLocked resolves rp.Component() to an existing file, starting from the -// given directory. -// -// Dentries which may become cached as a result of the traversal are appended -// to *ds. -// -// Preconditions: fs.renameMu must be locked. d.dirMu must be locked. -// !rp.Done(). If fs.opts.interop == InteropModeShared, then d's cached -// metadata must be up to date. -func (fs *filesystem) stepLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { - if !d.isDir() { - return nil, syserror.ENOTDIR - } - if err := d.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { - return nil, err - } -afterSymlink: - name := rp.Component() - if name == "." { - rp.Advance() - return d, nil - } - if name == ".." { - parentVFSD, err := rp.ResolveParent(&d.vfsd) - if err != nil { - return nil, err - } - parent := parentVFSD.Impl().(*dentry) - if fs.opts.interop == InteropModeShared { - // We must assume that parentVFSD is correct, because if d has been - // moved elsewhere in the remote filesystem so that its parent has - // changed, we have no way of determining its new parent's location - // in the filesystem. Get updated metadata for parentVFSD. - _, attrMask, attr, err := parent.file.getAttr(ctx, dentryAttrMask()) - if err != nil { - return nil, err - } - parent.updateFromP9Attrs(attrMask, &attr) - } - rp.Advance() - return parent, nil - } - childVFSD, err := rp.ResolveChild(&d.vfsd, name) - if err != nil { - return nil, err - } - // FIXME(jamieliu): Linux performs revalidation before mount lookup - // (fs/namei.c:lookup_fast() => __d_lookup_rcu(), d_revalidate(), - // __follow_mount_rcu()). - child, err := fs.revalidateChildLocked(ctx, rp.VirtualFilesystem(), d, name, childVFSD, ds) - if err != nil { - return nil, err - } - if child == nil { - return nil, syserror.ENOENT - } - if child.isSymlink() && rp.ShouldFollowSymlink() { - target, err := child.readlink(ctx, rp.Mount()) - if err != nil { - return nil, err - } - if err := rp.HandleSymlink(target); err != nil { - return nil, err - } - goto afterSymlink // don't check the current directory again - } - rp.Advance() - return child, nil -} - -// revalidateChildLocked must be called after a call to parent.vfsd.Child(name) -// or vfs.ResolvingPath.ResolveChild(name) returns childVFSD (which may be -// nil) to verify that the returned child (or lack thereof) is correct. If no file -// exists at name, revalidateChildLocked returns (nil, nil). -// -// Preconditions: fs.renameMu must be locked. parent.dirMu must be locked. -// parent.isDir(). name is not "." or "..". -// -// Postconditions: If revalidateChildLocked returns a non-nil dentry, its -// cached metadata is up to date. -func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.VirtualFilesystem, parent *dentry, name string, childVFSD *vfs.Dentry, ds **[]*dentry) (*dentry, error) { - if childVFSD != nil && fs.opts.interop != InteropModeShared { - // We have a cached dentry that is assumed to be correct. - return childVFSD.Impl().(*dentry), nil - } - // We either don't have a cached dentry or need to verify that it's still - // correct, either of which requires a remote lookup. Check if this name is - // valid before performing the lookup. - if len(name) > maxFilenameLen { - return nil, syserror.ENAMETOOLONG - } - // Check if we've already cached this lookup with a negative result. - if _, ok := parent.negativeChildren[name]; ok { - return nil, nil - } - // Perform the remote lookup. - qid, file, attrMask, attr, err := parent.file.walkGetAttrOne(ctx, name) - if err != nil && err != syserror.ENOENT { - return nil, err - } - if childVFSD != nil { - child := childVFSD.Impl().(*dentry) - if !file.isNil() && qid.Path == child.ino { - // The file at this path hasn't changed. Just update cached - // metadata. - file.close(ctx) - child.updateFromP9Attrs(attrMask, &attr) - return child, nil - } - // The file at this path has changed or no longer exists. Remove - // the stale dentry from the tree, and re-evaluate its caching - // status (i.e. if it has 0 references, drop it). - vfsObj.ForceDeleteDentry(childVFSD) - *ds = appendDentry(*ds, child) - childVFSD = nil - } - if file.isNil() { - // No file exists at this path now. Cache the negative lookup if - // allowed. - if fs.opts.interop != InteropModeShared { - parent.cacheNegativeChildLocked(name) - } - return nil, nil - } - // Create a new dentry representing the file. - child, err := fs.newDentry(ctx, file, qid, attrMask, &attr) - if err != nil { - file.close(ctx) - return nil, err - } - parent.IncRef() // reference held by child on its parent - parent.vfsd.InsertChild(&child.vfsd, name) - // For now, child has 0 references, so our caller should call - // child.checkCachingLocked(). - *ds = appendDentry(*ds, child) - return child, nil -} - -// walkParentDirLocked resolves all but the last path component of rp to an -// existing directory, starting from the given directory (which is usually -// rp.Start().Impl().(*dentry)). It does not check that the returned directory -// is searchable by the provider of rp. -// -// Preconditions: fs.renameMu must be locked. !rp.Done(). If fs.opts.interop == -// InteropModeShared, then d's cached metadata must be up to date. -func (fs *filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry, ds **[]*dentry) (*dentry, error) { - for !rp.Final() { - d.dirMu.Lock() - next, err := fs.stepLocked(ctx, rp, d, ds) - d.dirMu.Unlock() - if err != nil { - return nil, err - } - d = next - } - if !d.isDir() { - return nil, syserror.ENOTDIR - } - return d, nil -} - -// resolveLocked resolves rp to an existing file. -// -// Preconditions: fs.renameMu must be locked. -func (fs *filesystem) resolveLocked(ctx context.Context, rp *vfs.ResolvingPath, ds **[]*dentry) (*dentry, error) { - d := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { - // Get updated metadata for rp.Start() as required by fs.stepLocked(). - if err := d.updateFromGetattr(ctx); err != nil { - return nil, err - } - } - for !rp.Done() { - d.dirMu.Lock() - next, err := fs.stepLocked(ctx, rp, d, ds) - d.dirMu.Unlock() - if err != nil { - return nil, err - } - d = next - } - if rp.MustBeDir() && !d.isDir() { - return nil, syserror.ENOTDIR - } - return d, nil -} - -// doCreateAt checks that creating a file at rp is permitted, then invokes -// create to do so. -// -// Preconditions: !rp.Done(). For the final path component in rp, -// !rp.ShouldFollowSymlink(). -func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, create func(parent *dentry, name string) error) error { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - start := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { - // Get updated metadata for start as required by - // fs.walkParentDirLocked(). - if err := start.updateFromGetattr(ctx); err != nil { - return err - } - } - parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) - if err != nil { - return err - } - if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - if parent.isDeleted() { - return syserror.ENOENT - } - name := rp.Component() - if name == "." || name == ".." { - return syserror.EEXIST - } - if len(name) > maxFilenameLen { - return syserror.ENAMETOOLONG - } - if !dir && rp.MustBeDir() { - return syserror.ENOENT - } - mnt := rp.Mount() - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - parent.dirMu.Lock() - defer parent.dirMu.Unlock() - if fs.opts.interop == InteropModeShared { - // The existence of a dentry at name would be inconclusive because the - // file it represents may have been deleted from the remote filesystem, - // so we would need to make an RPC to revalidate the dentry. Just - // attempt the file creation RPC instead. If a file does exist, the RPC - // will fail with EEXIST like we would have. If the RPC succeeds, and a - // stale dentry exists, the dentry will fail revalidation next time - // it's used. - return create(parent, name) - } - if parent.vfsd.Child(name) != nil { - return syserror.EEXIST - } - // No cached dentry exists; however, there might still be an existing file - // at name. As above, we attempt the file creation RPC anyway. - if err := create(parent, name); err != nil { - return err - } - if fs.opts.interop != InteropModeShared { - parent.touchCMtime() - } - delete(parent.negativeChildren, name) - parent.dirents = nil - return nil -} - -// Preconditions: !rp.Done(). -func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool) error { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - start := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { - // Get updated metadata for start as required by - // fs.walkParentDirLocked(). - if err := start.updateFromGetattr(ctx); err != nil { - return err - } - } - parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) - if err != nil { - return err - } - if err := parent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - if err := rp.Mount().CheckBeginWrite(); err != nil { - return err - } - defer rp.Mount().EndWrite() - - name := rp.Component() - if dir { - if name == "." { - return syserror.EINVAL - } - if name == ".." { - return syserror.ENOTEMPTY - } - } else { - if name == "." || name == ".." { - return syserror.EISDIR - } - } - vfsObj := rp.VirtualFilesystem() - mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() - parent.dirMu.Lock() - defer parent.dirMu.Unlock() - childVFSD := parent.vfsd.Child(name) - var child *dentry - // We only need a dentry representing the file at name if it can be a mount - // point. If childVFSD is nil, then it can't be a mount point. If childVFSD - // is non-nil but stale, the actual file can't be a mount point either; we - // detect this case by just speculatively calling PrepareDeleteDentry and - // only revalidating the dentry if that fails (indicating that the existing - // dentry is a mount point). - if childVFSD != nil { - child = childVFSD.Impl().(*dentry) - if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { - child, err = fs.revalidateChildLocked(ctx, vfsObj, parent, name, childVFSD, &ds) - if err != nil { - return err - } - if child != nil { - childVFSD = &child.vfsd - if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { - return err - } - } else { - childVFSD = nil - } - } - } else if _, ok := parent.negativeChildren[name]; ok { - return syserror.ENOENT - } - flags := uint32(0) - if dir { - if child != nil && !child.isDir() { - vfsObj.AbortDeleteDentry(childVFSD) - return syserror.ENOTDIR - } - flags = linux.AT_REMOVEDIR - } else { - if child != nil && child.isDir() { - vfsObj.AbortDeleteDentry(childVFSD) - return syserror.EISDIR - } - if rp.MustBeDir() { - if childVFSD != nil { - vfsObj.AbortDeleteDentry(childVFSD) - } - return syserror.ENOTDIR - } - } - err = parent.file.unlinkAt(ctx, name, flags) - if err != nil { - if childVFSD != nil { - vfsObj.AbortDeleteDentry(childVFSD) - } - return err - } - if fs.opts.interop != InteropModeShared { - parent.touchCMtime() - if dir { - parent.decLinks() - } - parent.cacheNegativeChildLocked(name) - parent.dirents = nil - } - if child != nil { - child.setDeleted() - vfsObj.CommitDeleteDentry(childVFSD) - ds = appendDentry(ds, child) - } - return nil -} - -// renameMuRUnlockAndCheckCaching calls fs.renameMu.RUnlock(), then calls -// dentry.checkCachingLocked on all dentries in *ds with fs.renameMu locked for -// writing. -// -// ds is a pointer-to-pointer since defer evaluates its arguments immediately, -// but dentry slices are allocated lazily, and it's much easier to say "defer -// fs.renameMuRUnlockAndCheckCaching(&ds)" than "defer func() { -// fs.renameMuRUnlockAndCheckCaching(ds) }()" to work around this. -func (fs *filesystem) renameMuRUnlockAndCheckCaching(ds **[]*dentry) { - fs.renameMu.RUnlock() - if *ds == nil { - return - } - if len(**ds) != 0 { - fs.renameMu.Lock() - for _, d := range **ds { - d.checkCachingLocked() - } - fs.renameMu.Unlock() - } - putDentrySlice(*ds) -} - -func (fs *filesystem) renameMuUnlockAndCheckCaching(ds **[]*dentry) { - if *ds == nil { - fs.renameMu.Unlock() - return - } - for _, d := range **ds { - d.checkCachingLocked() - } - fs.renameMu.Unlock() - putDentrySlice(*ds) -} - -// AccessAt implements vfs.Filesystem.Impl.AccessAt. -func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return err - } - return d.checkPermissions(creds, ats) -} - -// GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. -func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return nil, err - } - if opts.CheckSearchable { - if !d.isDir() { - return nil, syserror.ENOTDIR - } - if err := d.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { - return nil, err - } - } - d.IncRef() - return &d.vfsd, nil -} - -// GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. -func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - start := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { - // Get updated metadata for start as required by - // fs.walkParentDirLocked(). - if err := start.updateFromGetattr(ctx); err != nil { - return nil, err - } - } - d, err := fs.walkParentDirLocked(ctx, rp, start, &ds) - if err != nil { - return nil, err - } - d.IncRef() - return &d.vfsd, nil -} - -// LinkAt implements vfs.FilesystemImpl.LinkAt. -func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { - return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, childName string) error { - if rp.Mount() != vd.Mount() { - return syserror.EXDEV - } - // 9P2000.L supports hard links, but we don't. - return syserror.EPERM - }) -} - -// MkdirAt implements vfs.FilesystemImpl.MkdirAt. -func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { - return fs.doCreateAt(ctx, rp, true /* dir */, func(parent *dentry, name string) error { - creds := rp.Credentials() - if _, err := parent.file.mkdir(ctx, name, (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)); err != nil { - return err - } - if fs.opts.interop != InteropModeShared { - parent.incLinks() - } - return nil - }) -} - -// MknodAt implements vfs.FilesystemImpl.MknodAt. -func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { - return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, name string) error { - creds := rp.Credentials() - _, err := parent.file.mknod(ctx, name, (p9.FileMode)(opts.Mode), opts.DevMajor, opts.DevMinor, (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) - return err - }) -} - -// OpenAt implements vfs.FilesystemImpl.OpenAt. -func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - // Reject O_TMPFILE, which is not supported; supporting it correctly in the - // presence of other remote filesystem users requires remote filesystem - // support, and it isn't clear that there's any way to implement this in - // 9P. - if opts.Flags&linux.O_TMPFILE != 0 { - return nil, syserror.EOPNOTSUPP - } - mayCreate := opts.Flags&linux.O_CREAT != 0 - mustCreate := opts.Flags&(linux.O_CREAT|linux.O_EXCL) == (linux.O_CREAT | linux.O_EXCL) - - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - - start := rp.Start().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { - // Get updated metadata for start as required by fs.stepLocked(). - if err := start.updateFromGetattr(ctx); err != nil { - return nil, err - } - } - if rp.Done() { - return start.openLocked(ctx, rp, &opts) - } - -afterTrailingSymlink: - parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) - if err != nil { - return nil, err - } - // Check for search permission in the parent directory. - if err := parent.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { - return nil, err - } - // Determine whether or not we need to create a file. - parent.dirMu.Lock() - child, err := fs.stepLocked(ctx, rp, parent, &ds) - if err == syserror.ENOENT && mayCreate { - fd, err := parent.createAndOpenChildLocked(ctx, rp, &opts) - parent.dirMu.Unlock() - return fd, err - } - if err != nil { - parent.dirMu.Unlock() - return nil, err - } - // Open existing child or follow symlink. - parent.dirMu.Unlock() - if mustCreate { - return nil, syserror.EEXIST - } - if child.isSymlink() && rp.ShouldFollowSymlink() { - target, err := child.readlink(ctx, rp.Mount()) - if err != nil { - return nil, err - } - if err := rp.HandleSymlink(target); err != nil { - return nil, err - } - start = parent - goto afterTrailingSymlink - } - return child.openLocked(ctx, rp, &opts) -} - -// Preconditions: fs.renameMu must be locked. -func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { - ats := vfs.AccessTypesForOpenFlags(opts) - if err := d.checkPermissions(rp.Credentials(), ats); err != nil { - return nil, err - } - mnt := rp.Mount() - filetype := d.fileType() - switch { - case filetype == linux.S_IFREG && !d.fs.opts.regularFilesUseSpecialFileFD: - if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, opts.Flags&linux.O_TRUNC != 0); err != nil { - return nil, err - } - fd := ®ularFileFD{} - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{ - AllowDirectIO: true, - }); err != nil { - return nil, err - } - return &fd.vfsfd, nil - case filetype == linux.S_IFDIR: - // Can't open directories with O_CREAT. - if opts.Flags&linux.O_CREAT != 0 { - return nil, syserror.EISDIR - } - // Can't open directories writably. - if ats&vfs.MayWrite != 0 { - return nil, syserror.EISDIR - } - if opts.Flags&linux.O_DIRECT != 0 { - return nil, syserror.EINVAL - } - if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, false /* write */, false /* trunc */); err != nil { - return nil, err - } - fd := &directoryFD{} - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { - return nil, err - } - return &fd.vfsfd, nil - case filetype == linux.S_IFLNK: - // Can't open symlinks without O_PATH (which is unimplemented). - return nil, syserror.ELOOP - default: - if opts.Flags&linux.O_DIRECT != 0 { - return nil, syserror.EINVAL - } - h, err := openHandle(ctx, d.file, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, opts.Flags&linux.O_TRUNC != 0) - if err != nil { - return nil, err - } - fd := &specialFileFD{ - handle: h, - } - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { - h.close(ctx) - return nil, err - } - return &fd.vfsfd, nil - } -} - -// Preconditions: d.fs.renameMu must be locked. d.dirMu must be locked. -func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions) (*vfs.FileDescription, error) { - if err := d.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { - return nil, err - } - if d.isDeleted() { - return nil, syserror.ENOENT - } - mnt := rp.Mount() - if err := mnt.CheckBeginWrite(); err != nil { - return nil, err - } - defer mnt.EndWrite() - - // 9P2000.L's lcreate takes a fid representing the parent directory, and - // converts it into an open fid representing the created file, so we need - // to duplicate the directory fid first. - _, dirfile, err := d.file.walk(ctx, nil) - if err != nil { - return nil, err - } - creds := rp.Credentials() - name := rp.Component() - fdobj, openFile, createQID, _, err := dirfile.create(ctx, name, (p9.OpenFlags)(opts.Flags), (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) - if err != nil { - dirfile.close(ctx) - return nil, err - } - // Then we need to walk to the file we just created to get a non-open fid - // representing it, and to get its metadata. This must use d.file since, as - // explained above, dirfile was invalidated by dirfile.Create(). - walkQID, nonOpenFile, attrMask, attr, err := d.file.walkGetAttrOne(ctx, name) - if err != nil { - openFile.close(ctx) - if fdobj != nil { - fdobj.Close() - } - return nil, err - } - // Sanity-check that we walked to the file we created. - if createQID.Path != walkQID.Path { - // Probably due to concurrent remote filesystem mutation? - ctx.Warningf("gofer.dentry.createAndOpenChildLocked: created file has QID %v before walk, QID %v after (interop=%v)", createQID, walkQID, d.fs.opts.interop) - nonOpenFile.close(ctx) - openFile.close(ctx) - if fdobj != nil { - fdobj.Close() - } - return nil, syserror.EAGAIN - } - - // Construct the new dentry. - child, err := d.fs.newDentry(ctx, nonOpenFile, createQID, attrMask, &attr) - if err != nil { - nonOpenFile.close(ctx) - openFile.close(ctx) - if fdobj != nil { - fdobj.Close() - } - return nil, err - } - // Incorporate the fid that was opened by lcreate. - useRegularFileFD := child.fileType() == linux.S_IFREG && !d.fs.opts.regularFilesUseSpecialFileFD - if useRegularFileFD { - child.handleMu.Lock() - child.handle.file = openFile - if fdobj != nil { - child.handle.fd = int32(fdobj.Release()) - } - child.handleReadable = vfs.MayReadFileWithOpenFlags(opts.Flags) - child.handleWritable = vfs.MayWriteFileWithOpenFlags(opts.Flags) - child.handleMu.Unlock() - } - // Take a reference on the new dentry to be held by the new file - // description. (This reference also means that the new dentry is not - // eligible for caching yet, so we don't need to append to a dentry slice.) - child.refs = 1 - // Insert the dentry into the tree. - d.IncRef() // reference held by child on its parent d - d.vfsd.InsertChild(&child.vfsd, name) - if d.fs.opts.interop != InteropModeShared { - delete(d.negativeChildren, name) - d.dirents = nil - } - - // Finally, construct a file description representing the created file. - var childVFSFD *vfs.FileDescription - mnt.IncRef() - if useRegularFileFD { - fd := ®ularFileFD{} - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &child.vfsd, &vfs.FileDescriptionOptions{ - AllowDirectIO: true, - }); err != nil { - return nil, err - } - childVFSFD = &fd.vfsfd - } else { - fd := &specialFileFD{ - handle: handle{ - file: openFile, - fd: -1, - }, - } - if fdobj != nil { - fd.handle.fd = int32(fdobj.Release()) - } - if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &child.vfsd, &vfs.FileDescriptionOptions{}); err != nil { - fd.handle.close(ctx) - return nil, err - } - childVFSFD = &fd.vfsfd - } - if d.fs.opts.interop != InteropModeShared { - d.touchCMtime() - } - return childVFSFD, nil -} - -// ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. -func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return "", err - } - if !d.isSymlink() { - return "", syserror.EINVAL - } - return d.readlink(ctx, rp.Mount()) -} - -// RenameAt implements vfs.FilesystemImpl.RenameAt. -func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldParentVD vfs.VirtualDentry, oldName string, opts vfs.RenameOptions) error { - if opts.Flags != 0 { - // Requires 9P support. - return syserror.EINVAL - } - - var ds *[]*dentry - fs.renameMu.Lock() - defer fs.renameMuUnlockAndCheckCaching(&ds) - newParent, err := fs.walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry), &ds) - if err != nil { - return err - } - newName := rp.Component() - if newName == "." || newName == ".." { - return syserror.EBUSY - } - mnt := rp.Mount() - if mnt != oldParentVD.Mount() { - return syserror.EXDEV - } - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - - oldParent := oldParentVD.Dentry().Impl().(*dentry) - if fs.opts.interop == InteropModeShared { - if err := oldParent.updateFromGetattr(ctx); err != nil { - return err - } - } - if err := oldParent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - vfsObj := rp.VirtualFilesystem() - // We need a dentry representing the renamed file since, if it's a - // directory, we need to check for write permission on it. - oldParent.dirMu.Lock() - defer oldParent.dirMu.Unlock() - renamed, err := fs.revalidateChildLocked(ctx, vfsObj, oldParent, oldName, oldParent.vfsd.Child(oldName), &ds) - if err != nil { - return err - } - if renamed == nil { - return syserror.ENOENT - } - if renamed.isDir() { - if renamed == newParent || renamed.vfsd.IsAncestorOf(&newParent.vfsd) { - return syserror.EINVAL - } - if oldParent != newParent { - if err := renamed.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { - return err - } - } - } else { - if opts.MustBeDir || rp.MustBeDir() { - return syserror.ENOTDIR - } - } - - if oldParent != newParent { - if err := newParent.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - newParent.dirMu.Lock() - defer newParent.dirMu.Unlock() - } - if newParent.isDeleted() { - return syserror.ENOENT - } - replacedVFSD := newParent.vfsd.Child(newName) - var replaced *dentry - // This is similar to unlinkAt, except: - // - // - We revalidate the replaced dentry unconditionally for simplicity. - // - // - If rp.MustBeDir(), then we need a dentry representing the replaced - // file regardless to confirm that it's a directory. - if replacedVFSD != nil || rp.MustBeDir() { - replaced, err = fs.revalidateChildLocked(ctx, vfsObj, newParent, newName, replacedVFSD, &ds) - if err != nil { - return err - } - if replaced != nil { - if replaced.isDir() { - if !renamed.isDir() { - return syserror.EISDIR - } - } else { - if rp.MustBeDir() || renamed.isDir() { - return syserror.ENOTDIR - } - } - replacedVFSD = &replaced.vfsd - } else { - replacedVFSD = nil - } - } - - if oldParent == newParent && oldName == newName { - return nil - } - mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() - if err := vfsObj.PrepareRenameDentry(mntns, &renamed.vfsd, replacedVFSD); err != nil { - return err - } - if err := renamed.file.rename(ctx, newParent.file, newName); err != nil { - vfsObj.AbortRenameDentry(&renamed.vfsd, replacedVFSD) - return err - } - if fs.opts.interop != InteropModeShared { - oldParent.cacheNegativeChildLocked(oldName) - oldParent.dirents = nil - delete(newParent.negativeChildren, newName) - newParent.dirents = nil - if renamed.isDir() { - oldParent.decLinks() - newParent.incLinks() - } - oldParent.touchCMtime() - newParent.touchCMtime() - renamed.touchCtime() - } - vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, &newParent.vfsd, newName, replacedVFSD) - return nil -} - -// RmdirAt implements vfs.FilesystemImpl.RmdirAt. -func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { - return fs.unlinkAt(ctx, rp, true /* dir */) -} - -// SetStatAt implements vfs.FilesystemImpl.SetStatAt. -func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return err - } - return d.setStat(ctx, rp.Credentials(), &opts.Stat, rp.Mount()) -} - -// StatAt implements vfs.FilesystemImpl.StatAt. -func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return linux.Statx{}, err - } - // Since walking updates metadata for all traversed dentries under - // InteropModeShared, including the returned one, we can return cached - // metadata here regardless of fs.opts.interop. - var stat linux.Statx - d.statTo(&stat) - return stat, nil -} - -// StatFSAt implements vfs.FilesystemImpl.StatFSAt. -func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return linux.Statfs{}, err - } - fsstat, err := d.file.statFS(ctx) - if err != nil { - return linux.Statfs{}, err - } - nameLen := uint64(fsstat.NameLength) - if nameLen > maxFilenameLen { - nameLen = maxFilenameLen - } - return linux.Statfs{ - // This is primarily for distinguishing a gofer file system in - // tests. Testing is important, so instead of defining - // something completely random, use a standard value. - Type: linux.V9FS_MAGIC, - BlockSize: int64(fsstat.BlockSize), - Blocks: fsstat.Blocks, - BlocksFree: fsstat.BlocksFree, - BlocksAvailable: fsstat.BlocksAvailable, - Files: fsstat.Files, - FilesFree: fsstat.FilesFree, - NameLength: nameLen, - }, nil -} - -// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. -func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { - return fs.doCreateAt(ctx, rp, false /* dir */, func(parent *dentry, name string) error { - creds := rp.Credentials() - _, err := parent.file.symlink(ctx, target, name, (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID)) - return err - }) -} - -// UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. -func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { - return fs.unlinkAt(ctx, rp, false /* dir */) -} - -// BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. -// -// TODO(gvisor.dev/issue/1476): Implement BoundEndpointAt. -func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) { - return nil, syserror.ECONNREFUSED -} - -// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. -func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return nil, err - } - return d.listxattr(ctx) -} - -// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. -func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return "", err - } - return d.getxattr(ctx, name) -} - -// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. -func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return err - } - return d.setxattr(ctx, &opts) -} - -// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. -func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { - var ds *[]*dentry - fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) - d, err := fs.resolveLocked(ctx, rp, &ds) - if err != nil { - return err - } - return d.removexattr(ctx, name) -} - -// PrependPath implements vfs.FilesystemImpl.PrependPath. -func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { - fs.renameMu.RLock() - defer fs.renameMu.RUnlock() - return vfs.GenericPrependPath(vfsroot, vd, b) -} diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go deleted file mode 100644 index 20edaf643..000000000 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ /dev/null @@ -1,1209 +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 gofer provides a filesystem implementation that is backed by a 9p -// server, interchangably referred to as "gofers" throughout this package. -// -// Lock order: -// regularFileFD/directoryFD.mu -// filesystem.renameMu -// dentry.dirMu -// filesystem.syncMu -// dentry.metadataMu -// *** "memmap.Mappable locks" below this point -// dentry.mapsMu -// *** "memmap.Mappable locks taken by Translate" below this point -// dentry.handleMu -// dentry.dataMu -// -// Locking dentry.dirMu in multiple dentries requires holding -// filesystem.renameMu for writing. -package gofer - -import ( - "fmt" - "strconv" - "sync" - "sync/atomic" - "syscall" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/unet" - "gvisor.dev/gvisor/pkg/usermem" -) - -// Name is the default filesystem name. -const Name = "9p" - -// FilesystemType implements vfs.FilesystemType. -type FilesystemType struct{} - -// filesystem implements vfs.FilesystemImpl. -type filesystem struct { - vfsfs vfs.Filesystem - - // mfp is used to allocate memory that caches regular file contents. mfp is - // immutable. - mfp pgalloc.MemoryFileProvider - - // Immutable options. - opts filesystemOptions - - // client is the client used by this filesystem. client is immutable. - client *p9.Client - - // clock is a realtime clock used to set timestamps in file operations. - clock ktime.Clock - - // uid and gid are the effective KUID and KGID of the filesystem's creator, - // and are used as the owner and group for files that don't specify one. - // uid and gid are immutable. - uid auth.KUID - gid auth.KGID - - // renameMu serves two purposes: - // - // - It synchronizes path resolution with renaming initiated by this - // client. - // - // - It is held by path resolution to ensure that reachable dentries remain - // valid. A dentry is reachable by path resolution if it has a non-zero - // reference count (such that it is usable as vfs.ResolvingPath.Start() or - // is reachable from its children), or if it is a child dentry (such that - // it is reachable from its parent). - renameMu sync.RWMutex - - // cachedDentries contains all dentries with 0 references. (Due to race - // conditions, it may also contain dentries with non-zero references.) - // cachedDentriesLen is the number of dentries in cachedDentries. These - // fields are protected by renameMu. - cachedDentries dentryList - cachedDentriesLen uint64 - - // dentries contains all dentries in this filesystem. specialFileFDs - // contains all open specialFileFDs. These fields are protected by syncMu. - syncMu sync.Mutex - dentries map[*dentry]struct{} - specialFileFDs map[*specialFileFD]struct{} -} - -type filesystemOptions struct { - // "Standard" 9P options. - fd int - aname string - interop InteropMode // derived from the "cache" mount option - msize uint32 - version string - - // maxCachedDentries is the maximum number of dentries with 0 references - // retained by the client. - maxCachedDentries uint64 - - // If forcePageCache is true, host FDs may not be used for application - // memory mappings even if available; instead, the client must perform its - // own caching of regular file pages. This is primarily useful for testing. - forcePageCache bool - - // If limitHostFDTranslation is true, apply maxFillRange() constraints to - // host FD mappings returned by dentry.(memmap.Mappable).Translate(). This - // makes memory accounting behavior more consistent between cases where - // host FDs are / are not available, but may increase the frequency of - // sentry-handled page faults on files for which a host FD is available. - limitHostFDTranslation bool - - // If overlayfsStaleRead is true, O_RDONLY host FDs provided by the remote - // filesystem may not be coherent with writable host FDs opened later, so - // mappings of the former must be replaced by mappings of the latter. This - // is usually only the case when the remote filesystem is an overlayfs - // mount on Linux < 4.19. - overlayfsStaleRead bool - - // If regularFilesUseSpecialFileFD is true, application FDs representing - // regular files will use distinct file handles for each FD, in the same - // way that application FDs representing "special files" such as sockets - // do. Note that this disables client caching and mmap for regular files. - regularFilesUseSpecialFileFD bool -} - -// InteropMode controls the client's interaction with other remote filesystem -// users. -type InteropMode uint32 - -const ( - // InteropModeExclusive is appropriate when the filesystem client is the - // only user of the remote filesystem. - // - // - The client may cache arbitrary filesystem state (file data, metadata, - // filesystem structure, etc.). - // - // - Client changes to filesystem state may be sent to the remote - // filesystem asynchronously, except when server permission checks are - // necessary. - // - // - File timestamps are based on client clocks. This ensures that users of - // the client observe timestamps that are coherent with their own clocks - // and consistent with Linux's semantics. However, since it is not always - // possible for clients to set arbitrary atimes and mtimes, and never - // possible for clients to set arbitrary ctimes, file timestamp changes are - // stored in the client only and never sent to the remote filesystem. - InteropModeExclusive InteropMode = iota - - // InteropModeWritethrough is appropriate when there are read-only users of - // the remote filesystem that expect to observe changes made by the - // filesystem client. - // - // - The client may cache arbitrary filesystem state. - // - // - Client changes to filesystem state must be sent to the remote - // filesystem synchronously. - // - // - File timestamps are based on client clocks. As a corollary, access - // timestamp changes from other remote filesystem users will not be visible - // to the client. - InteropModeWritethrough - - // InteropModeShared is appropriate when there are users of the remote - // filesystem that may mutate its state other than the client. - // - // - The client must verify cached filesystem state before using it. - // - // - Client changes to filesystem state must be sent to the remote - // filesystem synchronously. - // - // - File timestamps are based on server clocks. This is necessary to - // ensure that timestamp changes are synchronized between remote filesystem - // users. - // - // Note that the correctness of InteropModeShared depends on the server - // correctly implementing 9P fids (i.e. each fid immutably represents a - // single filesystem object), even in the presence of remote filesystem - // mutations from other users. If this is violated, the behavior of the - // client is undefined. - InteropModeShared -) - -// Name implements vfs.FilesystemType.Name. -func (FilesystemType) Name() string { - return Name -} - -// GetFilesystem implements vfs.FilesystemType.GetFilesystem. -func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - mfp := pgalloc.MemoryFileProviderFromContext(ctx) - if mfp == nil { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: context does not provide a pgalloc.MemoryFileProvider") - return nil, nil, syserror.EINVAL - } - - mopts := vfs.GenericParseMountOptions(opts.Data) - var fsopts filesystemOptions - - // Check that the transport is "fd". - trans, ok := mopts["trans"] - if !ok { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: transport must be specified as 'trans=fd'") - return nil, nil, syserror.EINVAL - } - delete(mopts, "trans") - if trans != "fd" { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: unsupported transport: trans=%s", trans) - return nil, nil, syserror.EINVAL - } - - // Check that read and write FDs are provided and identical. - rfdstr, ok := mopts["rfdno"] - if !ok { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: read FD must be specified as 'rfdno=<file descriptor>") - return nil, nil, syserror.EINVAL - } - delete(mopts, "rfdno") - rfd, err := strconv.Atoi(rfdstr) - if err != nil { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid read FD: rfdno=%s", rfdstr) - return nil, nil, syserror.EINVAL - } - wfdstr, ok := mopts["wfdno"] - if !ok { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: write FD must be specified as 'wfdno=<file descriptor>") - return nil, nil, syserror.EINVAL - } - delete(mopts, "wfdno") - wfd, err := strconv.Atoi(wfdstr) - if err != nil { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid write FD: wfdno=%s", wfdstr) - return nil, nil, syserror.EINVAL - } - if rfd != wfd { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: read FD (%d) and write FD (%d) must be equal", rfd, wfd) - return nil, nil, syserror.EINVAL - } - fsopts.fd = rfd - - // Get the attach name. - fsopts.aname = "/" - if aname, ok := mopts["aname"]; ok { - delete(mopts, "aname") - fsopts.aname = aname - } - - // Parse the cache policy. For historical reasons, this defaults to the - // least generally-applicable option, InteropModeExclusive. - fsopts.interop = InteropModeExclusive - if cache, ok := mopts["cache"]; ok { - delete(mopts, "cache") - switch cache { - case "fscache": - fsopts.interop = InteropModeExclusive - case "fscache_writethrough": - fsopts.interop = InteropModeWritethrough - case "none": - fsopts.regularFilesUseSpecialFileFD = true - fallthrough - case "remote_revalidating": - fsopts.interop = InteropModeShared - default: - ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid cache policy: cache=%s", cache) - return nil, nil, syserror.EINVAL - } - } - - // Parse the 9P message size. - fsopts.msize = 1024 * 1024 // 1M, tested to give good enough performance up to 64M - if msizestr, ok := mopts["msize"]; ok { - delete(mopts, "msize") - msize, err := strconv.ParseUint(msizestr, 10, 32) - if err != nil { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid message size: msize=%s", msizestr) - return nil, nil, syserror.EINVAL - } - fsopts.msize = uint32(msize) - } - - // Parse the 9P protocol version. - fsopts.version = p9.HighestVersionString() - if version, ok := mopts["version"]; ok { - delete(mopts, "version") - fsopts.version = version - } - - // Parse the dentry cache limit. - fsopts.maxCachedDentries = 1000 - if str, ok := mopts["dentry_cache_limit"]; ok { - delete(mopts, "dentry_cache_limit") - maxCachedDentries, err := strconv.ParseUint(str, 10, 64) - if err != nil { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: invalid dentry cache limit: dentry_cache_limit=%s", str) - return nil, nil, syserror.EINVAL - } - fsopts.maxCachedDentries = maxCachedDentries - } - - // Handle simple flags. - if _, ok := mopts["force_page_cache"]; ok { - delete(mopts, "force_page_cache") - fsopts.forcePageCache = true - } - if _, ok := mopts["limit_host_fd_translation"]; ok { - delete(mopts, "limit_host_fd_translation") - fsopts.limitHostFDTranslation = true - } - if _, ok := mopts["overlayfs_stale_read"]; ok { - delete(mopts, "overlayfs_stale_read") - fsopts.overlayfsStaleRead = true - } - // fsopts.regularFilesUseSpecialFileFD can only be enabled by specifying - // "cache=none". - - // Check for unparsed options. - if len(mopts) != 0 { - ctx.Warningf("gofer.FilesystemType.GetFilesystem: unknown options: %v", mopts) - return nil, nil, syserror.EINVAL - } - - // Establish a connection with the server. - conn, err := unet.NewSocket(fsopts.fd) - if err != nil { - return nil, nil, err - } - - // Perform version negotiation with the server. - ctx.UninterruptibleSleepStart(false) - client, err := p9.NewClient(conn, fsopts.msize, fsopts.version) - ctx.UninterruptibleSleepFinish(false) - if err != nil { - conn.Close() - return nil, nil, err - } - // Ownership of conn has been transferred to client. - - // Perform attach to obtain the filesystem root. - ctx.UninterruptibleSleepStart(false) - attached, err := client.Attach(fsopts.aname) - ctx.UninterruptibleSleepFinish(false) - if err != nil { - client.Close() - return nil, nil, err - } - attachFile := p9file{attached} - qid, attrMask, attr, err := attachFile.getAttr(ctx, dentryAttrMask()) - if err != nil { - attachFile.close(ctx) - client.Close() - return nil, nil, err - } - - // Construct the filesystem object. - fs := &filesystem{ - mfp: mfp, - opts: fsopts, - uid: creds.EffectiveKUID, - gid: creds.EffectiveKGID, - client: client, - clock: ktime.RealtimeClockFromContext(ctx), - dentries: make(map[*dentry]struct{}), - specialFileFDs: make(map[*specialFileFD]struct{}), - } - fs.vfsfs.Init(vfsObj, &fstype, fs) - - // Construct the root dentry. - root, err := fs.newDentry(ctx, attachFile, qid, attrMask, &attr) - if err != nil { - attachFile.close(ctx) - fs.vfsfs.DecRef() - return nil, nil, err - } - // Set the root's reference count to 2. One reference is returned to the - // caller, and the other is deliberately leaked to prevent the root from - // being "cached" and subsequently evicted. Its resources will still be - // cleaned up by fs.Release(). - root.refs = 2 - - return &fs.vfsfs, &root.vfsd, nil -} - -// Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { - ctx := context.Background() - mf := fs.mfp.MemoryFile() - - fs.syncMu.Lock() - for d := range fs.dentries { - d.handleMu.Lock() - d.dataMu.Lock() - if d.handleWritable { - // Write dirty cached data to the remote file. - if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, fs.mfp.MemoryFile(), d.handle.writeFromBlocksAt); err != nil { - log.Warningf("gofer.filesystem.Release: failed to flush dentry: %v", err) - } - // TODO(jamieliu): Do we need to flushf/fsync d? - } - // Discard cached pages. - d.cache.DropAll(mf) - d.dirty.RemoveAll() - d.dataMu.Unlock() - // Close the host fd if one exists. - if d.handle.fd >= 0 { - syscall.Close(int(d.handle.fd)) - d.handle.fd = -1 - } - d.handleMu.Unlock() - } - // There can't be any specialFileFDs still using fs, since each such - // FileDescription would hold a reference on a Mount holding a reference on - // fs. - fs.syncMu.Unlock() - - // Close the connection to the server. This implicitly clunks all fids. - fs.client.Close() -} - -// dentry implements vfs.DentryImpl. -type dentry struct { - vfsd vfs.Dentry - - // refs is the reference count. Each dentry holds a reference on its - // parent, even if disowned. refs is accessed using atomic memory - // operations. When refs reaches 0, the dentry may be added to the cache or - // destroyed. If refs==-1 the dentry has already been destroyed. - refs int64 - - // fs is the owning filesystem. fs is immutable. - fs *filesystem - - // We don't support hard links, so each dentry maps 1:1 to an inode. - - // file is the unopened p9.File that backs this dentry. file is immutable. - file p9file - - // If deleted is non-zero, the file represented by this dentry has been - // deleted. deleted is accessed using atomic memory operations. - deleted uint32 - - // If cached is true, dentryEntry links dentry into - // filesystem.cachedDentries. cached and dentryEntry are protected by - // filesystem.renameMu. - cached bool - dentryEntry - - dirMu sync.Mutex - - // If this dentry represents a directory, and InteropModeShared is not in - // effect, negativeChildren is a set of child names in this directory that - // are known not to exist. negativeChildren is protected by dirMu. - negativeChildren map[string]struct{} - - // If this dentry represents a directory, InteropModeShared is not in - // effect, and dirents is not nil, it is a cache of all entries in the - // directory, in the order they were returned by the server. dirents is - // protected by dirMu. - dirents []vfs.Dirent - - // Cached metadata; protected by metadataMu and accessed using atomic - // memory operations unless otherwise specified. - metadataMu sync.Mutex - ino uint64 // immutable - mode uint32 // type is immutable, perms are mutable - uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic - gid uint32 // auth.KGID, but ... - blockSize uint32 // 0 if unknown - // Timestamps, all nsecs from the Unix epoch. - atime int64 - mtime int64 - ctime int64 - btime int64 - // File size, protected by both metadataMu and dataMu (i.e. both must be - // locked to mutate it). - size uint64 - - // nlink counts the number of hard links to this dentry. It's updated and - // accessed using atomic operations. It's not protected by metadataMu like the - // other metadata fields. - nlink uint32 - - mapsMu sync.Mutex - - // If this dentry represents a regular file, mappings tracks mappings of - // the file into memmap.MappingSpaces. mappings is protected by mapsMu. - mappings memmap.MappingSet - - // If this dentry represents a regular file or directory: - // - // - handle is the I/O handle used by all regularFileFDs/directoryFDs - // representing this dentry. - // - // - handleReadable is true if handle is readable. - // - // - handleWritable is true if handle is writable. - // - // Invariants: - // - // - If handleReadable == handleWritable == false, then handle.file == nil - // (i.e. there is no open handle). Conversely, if handleReadable || - // handleWritable == true, then handle.file != nil (i.e. there is an open - // handle). - // - // - handleReadable and handleWritable cannot transition from true to false - // (i.e. handles may not be downgraded). - // - // These fields are protected by handleMu. - handleMu sync.RWMutex - handle handle - handleReadable bool - handleWritable bool - - dataMu sync.RWMutex - - // If this dentry represents a regular file that is client-cached, cache - // maps offsets into the cached file to offsets into - // filesystem.mfp.MemoryFile() that store the file's data. cache is - // protected by dataMu. - cache fsutil.FileRangeSet - - // If this dentry represents a regular file that is client-cached, dirty - // tracks dirty segments in cache. dirty is protected by dataMu. - dirty fsutil.DirtySet - - // pf implements platform.File for mappings of handle.fd. - pf dentryPlatformFile - - // If this dentry represents a symbolic link, InteropModeShared is not in - // effect, and haveTarget is true, target is the symlink target. haveTarget - // and target are protected by dataMu. - haveTarget bool - target string -} - -// dentryAttrMask returns a p9.AttrMask enabling all attributes used by the -// gofer client. -func dentryAttrMask() p9.AttrMask { - return p9.AttrMask{ - Mode: true, - UID: true, - GID: true, - ATime: true, - MTime: true, - CTime: true, - Size: true, - BTime: true, - } -} - -// newDentry creates a new dentry representing the given file. The dentry -// initially has no references, but is not cached; it is the caller's -// responsibility to set the dentry's reference count and/or call -// dentry.checkCachingLocked() as appropriate. -func (fs *filesystem) newDentry(ctx context.Context, file p9file, qid p9.QID, mask p9.AttrMask, attr *p9.Attr) (*dentry, error) { - if !mask.Mode { - ctx.Warningf("can't create gofer.dentry without file type") - return nil, syserror.EIO - } - if attr.Mode.FileType() == p9.ModeRegular && !mask.Size { - ctx.Warningf("can't create regular file gofer.dentry without file size") - return nil, syserror.EIO - } - - d := &dentry{ - fs: fs, - file: file, - ino: qid.Path, - mode: uint32(attr.Mode), - uid: uint32(fs.uid), - gid: uint32(fs.gid), - blockSize: usermem.PageSize, - handle: handle{ - fd: -1, - }, - } - d.pf.dentry = d - if mask.UID { - d.uid = uint32(attr.UID) - } - if mask.GID { - d.gid = uint32(attr.GID) - } - if mask.Size { - d.size = attr.Size - } - if attr.BlockSize != 0 { - d.blockSize = uint32(attr.BlockSize) - } - if mask.ATime { - d.atime = dentryTimestampFromP9(attr.ATimeSeconds, attr.ATimeNanoSeconds) - } - if mask.MTime { - d.mtime = dentryTimestampFromP9(attr.MTimeSeconds, attr.MTimeNanoSeconds) - } - if mask.CTime { - d.ctime = dentryTimestampFromP9(attr.CTimeSeconds, attr.CTimeNanoSeconds) - } - if mask.BTime { - d.btime = dentryTimestampFromP9(attr.BTimeSeconds, attr.BTimeNanoSeconds) - } - if mask.NLink { - d.nlink = uint32(attr.NLink) - } - d.vfsd.Init(d) - - fs.syncMu.Lock() - fs.dentries[d] = struct{}{} - fs.syncMu.Unlock() - return d, nil -} - -// updateFromP9Attrs is called to update d's metadata after an update from the -// remote filesystem. -func (d *dentry) updateFromP9Attrs(mask p9.AttrMask, attr *p9.Attr) { - d.metadataMu.Lock() - if mask.Mode { - if got, want := uint32(attr.Mode.FileType()), d.fileType(); got != want { - d.metadataMu.Unlock() - panic(fmt.Sprintf("gofer.dentry file type changed from %#o to %#o", want, got)) - } - atomic.StoreUint32(&d.mode, uint32(attr.Mode)) - } - if mask.UID { - atomic.StoreUint32(&d.uid, uint32(attr.UID)) - } - if mask.GID { - atomic.StoreUint32(&d.gid, uint32(attr.GID)) - } - // There is no P9_GETATTR_* bit for I/O block size. - if attr.BlockSize != 0 { - atomic.StoreUint32(&d.blockSize, uint32(attr.BlockSize)) - } - if mask.ATime { - atomic.StoreInt64(&d.atime, dentryTimestampFromP9(attr.ATimeSeconds, attr.ATimeNanoSeconds)) - } - if mask.MTime { - atomic.StoreInt64(&d.mtime, dentryTimestampFromP9(attr.MTimeSeconds, attr.MTimeNanoSeconds)) - } - if mask.CTime { - atomic.StoreInt64(&d.ctime, dentryTimestampFromP9(attr.CTimeSeconds, attr.CTimeNanoSeconds)) - } - if mask.BTime { - atomic.StoreInt64(&d.btime, dentryTimestampFromP9(attr.BTimeSeconds, attr.BTimeNanoSeconds)) - } - if mask.NLink { - atomic.StoreUint32(&d.nlink, uint32(attr.NLink)) - } - if mask.Size { - d.dataMu.Lock() - atomic.StoreUint64(&d.size, attr.Size) - d.dataMu.Unlock() - } - d.metadataMu.Unlock() -} - -func (d *dentry) updateFromGetattr(ctx context.Context) error { - // Use d.handle.file, which represents a 9P fid that has been opened, in - // preference to d.file, which represents a 9P fid that has not. This may - // be significantly more efficient in some implementations. - var ( - file p9file - handleMuRLocked bool - ) - d.handleMu.RLock() - if !d.handle.file.isNil() { - file = d.handle.file - handleMuRLocked = true - } else { - file = d.file - d.handleMu.RUnlock() - } - _, attrMask, attr, err := file.getAttr(ctx, dentryAttrMask()) - if handleMuRLocked { - d.handleMu.RUnlock() - } - if err != nil { - return err - } - d.updateFromP9Attrs(attrMask, &attr) - return nil -} - -func (d *dentry) fileType() uint32 { - return atomic.LoadUint32(&d.mode) & linux.S_IFMT -} - -func (d *dentry) statTo(stat *linux.Statx) { - stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_ATIME | linux.STATX_MTIME | linux.STATX_CTIME | linux.STATX_INO | linux.STATX_SIZE | linux.STATX_BLOCKS | linux.STATX_BTIME - stat.Blksize = atomic.LoadUint32(&d.blockSize) - stat.Nlink = atomic.LoadUint32(&d.nlink) - stat.UID = atomic.LoadUint32(&d.uid) - stat.GID = atomic.LoadUint32(&d.gid) - stat.Mode = uint16(atomic.LoadUint32(&d.mode)) - stat.Ino = d.ino - stat.Size = atomic.LoadUint64(&d.size) - // This is consistent with regularFileFD.Seek(), which treats regular files - // as having no holes. - stat.Blocks = (stat.Size + 511) / 512 - stat.Atime = statxTimestampFromDentry(atomic.LoadInt64(&d.atime)) - stat.Btime = statxTimestampFromDentry(atomic.LoadInt64(&d.btime)) - stat.Ctime = statxTimestampFromDentry(atomic.LoadInt64(&d.ctime)) - stat.Mtime = statxTimestampFromDentry(atomic.LoadInt64(&d.mtime)) - // TODO(gvisor.dev/issue/1198): device number -} - -func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx, mnt *vfs.Mount) error { - if stat.Mask == 0 { - return nil - } - if stat.Mask&^(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID|linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_SIZE) != 0 { - return syserror.EPERM - } - mode := linux.FileMode(atomic.LoadUint32(&d.mode)) - if err := vfs.CheckSetStat(ctx, creds, stat, mode, auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))); err != nil { - return err - } - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - setLocalAtime := false - setLocalMtime := false - if d.fs.opts.interop != InteropModeShared { - // Timestamp updates will be handled locally. - setLocalAtime = stat.Mask&linux.STATX_ATIME != 0 - setLocalMtime = stat.Mask&linux.STATX_MTIME != 0 - stat.Mask &^= linux.STATX_ATIME | linux.STATX_MTIME - if !setLocalMtime && (stat.Mask&linux.STATX_SIZE != 0) { - // Truncate updates mtime. - setLocalMtime = true - stat.Mtime.Nsec = linux.UTIME_NOW - } - } - d.metadataMu.Lock() - defer d.metadataMu.Unlock() - if stat.Mask != 0 { - if err := d.file.setAttr(ctx, p9.SetAttrMask{ - Permissions: stat.Mask&linux.STATX_MODE != 0, - UID: stat.Mask&linux.STATX_UID != 0, - GID: stat.Mask&linux.STATX_GID != 0, - Size: stat.Mask&linux.STATX_SIZE != 0, - ATime: stat.Mask&linux.STATX_ATIME != 0, - MTime: stat.Mask&linux.STATX_MTIME != 0, - ATimeNotSystemTime: stat.Atime.Nsec != linux.UTIME_NOW, - MTimeNotSystemTime: stat.Mtime.Nsec != linux.UTIME_NOW, - }, p9.SetAttr{ - Permissions: p9.FileMode(stat.Mode), - UID: p9.UID(stat.UID), - GID: p9.GID(stat.GID), - Size: stat.Size, - ATimeSeconds: uint64(stat.Atime.Sec), - ATimeNanoSeconds: uint64(stat.Atime.Nsec), - MTimeSeconds: uint64(stat.Mtime.Sec), - MTimeNanoSeconds: uint64(stat.Mtime.Nsec), - }); err != nil { - return err - } - } - if d.fs.opts.interop == InteropModeShared { - // There's no point to updating d's metadata in this case since it'll - // be overwritten by revalidation before the next time it's used - // anyway. (InteropModeShared inhibits client caching of regular file - // data, so there's no cache to truncate either.) - return nil - } - now := d.fs.clock.Now().Nanoseconds() - if stat.Mask&linux.STATX_MODE != 0 { - atomic.StoreUint32(&d.mode, d.fileType()|uint32(stat.Mode)) - } - if stat.Mask&linux.STATX_UID != 0 { - atomic.StoreUint32(&d.uid, stat.UID) - } - if stat.Mask&linux.STATX_GID != 0 { - atomic.StoreUint32(&d.gid, stat.GID) - } - if setLocalAtime { - if stat.Atime.Nsec == linux.UTIME_NOW { - atomic.StoreInt64(&d.atime, now) - } else { - atomic.StoreInt64(&d.atime, dentryTimestampFromStatx(stat.Atime)) - } - } - if setLocalMtime { - if stat.Mtime.Nsec == linux.UTIME_NOW { - atomic.StoreInt64(&d.mtime, now) - } else { - atomic.StoreInt64(&d.mtime, dentryTimestampFromStatx(stat.Mtime)) - } - } - atomic.StoreInt64(&d.ctime, now) - if stat.Mask&linux.STATX_SIZE != 0 { - d.dataMu.Lock() - oldSize := d.size - d.size = stat.Size - // d.dataMu must be unlocked to lock d.mapsMu and invalidate mappings - // below. This allows concurrent calls to Read/Translate/etc. These - // functions synchronize with truncation by refusing to use cache - // contents beyond the new d.size. (We are still holding d.metadataMu, - // so we can't race with Write or another truncate.) - d.dataMu.Unlock() - if d.size < oldSize { - oldpgend := pageRoundUp(oldSize) - newpgend := pageRoundUp(d.size) - if oldpgend != newpgend { - d.mapsMu.Lock() - d.mappings.Invalidate(memmap.MappableRange{newpgend, oldpgend}, memmap.InvalidateOpts{ - // Compare Linux's mm/truncate.c:truncate_setsize() => - // truncate_pagecache() => - // mm/memory.c:unmap_mapping_range(evencows=1). - InvalidatePrivate: true, - }) - d.mapsMu.Unlock() - } - // We are now guaranteed that there are no translations of - // truncated pages, and can remove them from the cache. Since - // truncated pages have been removed from the remote file, they - // should be dropped without being written back. - d.dataMu.Lock() - d.cache.Truncate(d.size, d.fs.mfp.MemoryFile()) - d.dirty.KeepClean(memmap.MappableRange{d.size, oldpgend}) - d.dataMu.Unlock() - } - } - return nil -} - -func (d *dentry) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { - return vfs.GenericCheckPermissions(creds, ats, linux.FileMode(atomic.LoadUint32(&d.mode)), auth.KUID(atomic.LoadUint32(&d.uid)), auth.KGID(atomic.LoadUint32(&d.gid))) -} - -// IncRef implements vfs.DentryImpl.IncRef. -func (d *dentry) IncRef() { - // d.refs may be 0 if d.fs.renameMu is locked, which serializes against - // d.checkCachingLocked(). - atomic.AddInt64(&d.refs, 1) -} - -// TryIncRef implements vfs.DentryImpl.TryIncRef. -func (d *dentry) TryIncRef() bool { - for { - refs := atomic.LoadInt64(&d.refs) - if refs <= 0 { - return false - } - if atomic.CompareAndSwapInt64(&d.refs, refs, refs+1) { - return true - } - } -} - -// DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef() { - if refs := atomic.AddInt64(&d.refs, -1); refs == 0 { - d.fs.renameMu.Lock() - d.checkCachingLocked() - d.fs.renameMu.Unlock() - } else if refs < 0 { - panic("gofer.dentry.DecRef() called without holding a reference") - } -} - -// checkCachingLocked should be called after d's reference count becomes 0 or it -// becomes disowned. -// -// It may be called on a destroyed dentry. For example, -// renameMu[R]UnlockAndCheckCaching may call checkCachingLocked multiple times -// for the same dentry when the dentry is visited more than once in the same -// operation. One of the calls may destroy the dentry, so subsequent calls will -// do nothing. -// -// Preconditions: d.fs.renameMu must be locked for writing. -func (d *dentry) checkCachingLocked() { - // Dentries with a non-zero reference count must be retained. (The only way - // to obtain a reference on a dentry with zero references is via path - // resolution, which requires renameMu, so if d.refs is zero then it will - // remain zero while we hold renameMu for writing.) - refs := atomic.LoadInt64(&d.refs) - if refs > 0 { - if d.cached { - d.fs.cachedDentries.Remove(d) - d.fs.cachedDentriesLen-- - d.cached = false - } - return - } - if refs == -1 { - // Dentry has already been destroyed. - return - } - // Non-child dentries with zero references are no longer reachable by path - // resolution and should be dropped immediately. - if d.vfsd.Parent() == nil || d.vfsd.IsDisowned() { - if d.cached { - d.fs.cachedDentries.Remove(d) - d.fs.cachedDentriesLen-- - d.cached = false - } - d.destroyLocked() - return - } - // If d is already cached, just move it to the front of the LRU. - if d.cached { - d.fs.cachedDentries.Remove(d) - d.fs.cachedDentries.PushFront(d) - return - } - // Cache the dentry, then evict the least recently used cached dentry if - // the cache becomes over-full. - d.fs.cachedDentries.PushFront(d) - d.fs.cachedDentriesLen++ - d.cached = true - if d.fs.cachedDentriesLen > d.fs.opts.maxCachedDentries { - victim := d.fs.cachedDentries.Back() - d.fs.cachedDentries.Remove(victim) - d.fs.cachedDentriesLen-- - victim.cached = false - // victim.refs may have become non-zero from an earlier path - // resolution since it was inserted into fs.cachedDentries; see - // dentry.incRefLocked(). Either way, we brought - // fs.cachedDentriesLen back down to fs.opts.maxCachedDentries, so - // we don't loop. - if atomic.LoadInt64(&victim.refs) == 0 { - if victimParentVFSD := victim.vfsd.Parent(); victimParentVFSD != nil { - victimParent := victimParentVFSD.Impl().(*dentry) - victimParent.dirMu.Lock() - if !victim.vfsd.IsDisowned() { - // victim can't be a mount point (in any mount - // namespace), since VFS holds references on mount - // points. - d.fs.vfsfs.VirtualFilesystem().ForceDeleteDentry(&victim.vfsd) - // We're only deleting the dentry, not the file it - // represents, so we don't need to update - // victimParent.dirents etc. - } - victimParent.dirMu.Unlock() - } - victim.destroyLocked() - } - } -} - -// destroyLocked destroys the dentry. It may flushes dirty pages from cache, -// close p9 file and remove reference on parent dentry. -// -// Preconditions: d.fs.renameMu must be locked for writing. d.refs == 0. d is -// not a child dentry. -func (d *dentry) destroyLocked() { - switch atomic.LoadInt64(&d.refs) { - case 0: - // Mark the dentry destroyed. - atomic.StoreInt64(&d.refs, -1) - case -1: - panic("dentry.destroyLocked() called on already destroyed dentry") - default: - panic("dentry.destroyLocked() called with references on the dentry") - } - - ctx := context.Background() - d.handleMu.Lock() - if !d.handle.file.isNil() { - mf := d.fs.mfp.MemoryFile() - d.dataMu.Lock() - // Write dirty pages back to the remote filesystem. - if d.handleWritable { - if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, mf, d.handle.writeFromBlocksAt); err != nil { - log.Warningf("gofer.dentry.DecRef: failed to write dirty data back: %v", err) - } - } - // Discard cached data. - d.cache.DropAll(mf) - d.dirty.RemoveAll() - d.dataMu.Unlock() - // Clunk open fids and close open host FDs. - d.handle.close(ctx) - } - d.handleMu.Unlock() - if !d.file.isNil() { - d.file.close(ctx) - d.file = p9file{} - } - // Remove d from the set of all dentries. - d.fs.syncMu.Lock() - delete(d.fs.dentries, d) - d.fs.syncMu.Unlock() - // Drop the reference held by d on its parent. - if parentVFSD := d.vfsd.Parent(); parentVFSD != nil { - parent := parentVFSD.Impl().(*dentry) - // This is parent.DecRef() without recursive locking of d.fs.renameMu. - if refs := atomic.AddInt64(&parent.refs, -1); refs == 0 { - parent.checkCachingLocked() - } else if refs < 0 { - panic("gofer.dentry.DecRef() called without holding a reference") - } - } -} - -func (d *dentry) isDeleted() bool { - return atomic.LoadUint32(&d.deleted) != 0 -} - -func (d *dentry) setDeleted() { - atomic.StoreUint32(&d.deleted, 1) -} - -func (d *dentry) listxattr(ctx context.Context) ([]string, error) { - return nil, syserror.ENOTSUP -} - -func (d *dentry) getxattr(ctx context.Context, name string) (string, error) { - // TODO(jamieliu): add vfs.GetxattrOptions.Size - return d.file.getXattr(ctx, name, linux.XATTR_SIZE_MAX) -} - -func (d *dentry) setxattr(ctx context.Context, opts *vfs.SetxattrOptions) error { - return d.file.setXattr(ctx, opts.Name, opts.Value, opts.Flags) -} - -func (d *dentry) removexattr(ctx context.Context, name string) error { - return syserror.ENOTSUP -} - -// Preconditions: d.isRegularFile() || d.isDirectory(). -func (d *dentry) ensureSharedHandle(ctx context.Context, read, write, trunc bool) error { - // O_TRUNC unconditionally requires us to obtain a new handle (opened with - // O_TRUNC). - if !trunc { - d.handleMu.RLock() - if (!read || d.handleReadable) && (!write || d.handleWritable) { - // The current handle is sufficient. - d.handleMu.RUnlock() - return nil - } - d.handleMu.RUnlock() - } - - haveOldFD := false - d.handleMu.Lock() - if (read && !d.handleReadable) || (write && !d.handleWritable) || trunc { - // Get a new handle. - wantReadable := d.handleReadable || read - wantWritable := d.handleWritable || write - h, err := openHandle(ctx, d.file, wantReadable, wantWritable, trunc) - if err != nil { - d.handleMu.Unlock() - return err - } - if !d.handle.file.isNil() { - // Check that old and new handles are compatible: If the old handle - // includes a host file descriptor but the new one does not, or - // vice versa, old and new memory mappings may be incoherent. - haveOldFD = d.handle.fd >= 0 - haveNewFD := h.fd >= 0 - if haveOldFD != haveNewFD { - d.handleMu.Unlock() - ctx.Warningf("gofer.dentry.ensureSharedHandle: can't change host FD availability from %v to %v across dentry handle upgrade", haveOldFD, haveNewFD) - h.close(ctx) - return syserror.EIO - } - if haveOldFD { - // We may have raced with callers of d.pf.FD() that are now - // using the old file descriptor, preventing us from safely - // closing it. We could handle this by invalidating existing - // memmap.Translations, but this is expensive. Instead, use - // dup3 to make the old file descriptor refer to the new file - // description, then close the new file descriptor (which is no - // longer needed). Racing callers may use the old or new file - // description, but this doesn't matter since they refer to the - // same file (unless d.fs.opts.overlayfsStaleRead is true, - // which we handle separately). - if err := syscall.Dup3(int(h.fd), int(d.handle.fd), 0); err != nil { - d.handleMu.Unlock() - ctx.Warningf("gofer.dentry.ensureSharedHandle: failed to dup fd %d to fd %d: %v", h.fd, d.handle.fd, err) - h.close(ctx) - return err - } - syscall.Close(int(h.fd)) - h.fd = d.handle.fd - if d.fs.opts.overlayfsStaleRead { - // Replace sentry mappings of the old FD with mappings of - // the new FD, since the two are not necessarily coherent. - if err := d.pf.hostFileMapper.RegenerateMappings(int(h.fd)); err != nil { - d.handleMu.Unlock() - ctx.Warningf("gofer.dentry.ensureSharedHandle: failed to replace sentry mappings of old FD with mappings of new FD: %v", err) - h.close(ctx) - return err - } - } - // Clunk the old fid before making the new handle visible (by - // unlocking d.handleMu). - d.handle.file.close(ctx) - } - } - // Switch to the new handle. - d.handle = h - d.handleReadable = wantReadable - d.handleWritable = wantWritable - } - d.handleMu.Unlock() - - if d.fs.opts.overlayfsStaleRead && haveOldFD { - // Invalidate application mappings that may be using the old FD; they - // will be replaced with mappings using the new FD after future calls - // to d.Translate(). This requires holding d.mapsMu, which precedes - // d.handleMu in the lock order. - d.mapsMu.Lock() - d.mappings.InvalidateAll(memmap.InvalidateOpts{}) - d.mapsMu.Unlock() - } - - return nil -} - -// incLinks increments link count. -// -// Preconditions: d.nlink != 0 && d.nlink < math.MaxUint32. -func (d *dentry) incLinks() { - v := atomic.AddUint32(&d.nlink, 1) - if v < 2 { - panic(fmt.Sprintf("dentry.nlink is invalid (was 0 or overflowed): %d", v)) - } -} - -// decLinks decrements link count. -// -// Preconditions: d.nlink > 1. -func (d *dentry) decLinks() { - v := atomic.AddUint32(&d.nlink, ^uint32(0)) - if v == 0 { - panic(fmt.Sprintf("dentry.nlink must be greater than 0: %d", v)) - } -} - -// fileDescription is embedded by gofer implementations of -// vfs.FileDescriptionImpl. -type fileDescription struct { - vfsfd vfs.FileDescription - vfs.FileDescriptionDefaultImpl -} - -func (fd *fileDescription) filesystem() *filesystem { - return fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) -} - -func (fd *fileDescription) dentry() *dentry { - return fd.vfsfd.Dentry().Impl().(*dentry) -} - -// Stat implements vfs.FileDescriptionImpl.Stat. -func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { - d := fd.dentry() - const validMask = uint32(linux.STATX_MODE | linux.STATX_UID | linux.STATX_GID | linux.STATX_ATIME | linux.STATX_MTIME | linux.STATX_CTIME | linux.STATX_SIZE | linux.STATX_BLOCKS | linux.STATX_BTIME) - if d.fs.opts.interop == InteropModeShared && opts.Mask&(validMask) != 0 && opts.Sync != linux.AT_STATX_DONT_SYNC { - // TODO(jamieliu): Use specialFileFD.handle.file for the getattr if - // available? - if err := d.updateFromGetattr(ctx); err != nil { - return linux.Statx{}, err - } - } - var stat linux.Statx - d.statTo(&stat) - return stat, nil -} - -// SetStat implements vfs.FileDescriptionImpl.SetStat. -func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { - return fd.dentry().setStat(ctx, auth.CredentialsFromContext(ctx), &opts.Stat, fd.vfsfd.Mount()) -} - -// Listxattr implements vfs.FileDescriptionImpl.Listxattr. -func (fd *fileDescription) Listxattr(ctx context.Context) ([]string, error) { - return fd.dentry().listxattr(ctx) -} - -// Getxattr implements vfs.FileDescriptionImpl.Getxattr. -func (fd *fileDescription) Getxattr(ctx context.Context, name string) (string, error) { - return fd.dentry().getxattr(ctx, name) -} - -// Setxattr implements vfs.FileDescriptionImpl.Setxattr. -func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOptions) error { - return fd.dentry().setxattr(ctx, &opts) -} - -// Removexattr implements vfs.FileDescriptionImpl.Removexattr. -func (fd *fileDescription) Removexattr(ctx context.Context, name string) error { - return fd.dentry().removexattr(ctx, name) -} diff --git a/pkg/sentry/fsimpl/gofer/gofer_test.go b/pkg/sentry/fsimpl/gofer/gofer_test.go deleted file mode 100644 index 82bc239db..000000000 --- a/pkg/sentry/fsimpl/gofer/gofer_test.go +++ /dev/null @@ -1,64 +0,0 @@ -// Copyright 2020 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 ( - "sync/atomic" - "testing" - - "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/sentry/contexttest" -) - -func TestDestroyIdempotent(t *testing.T) { - fs := filesystem{ - dentries: make(map[*dentry]struct{}), - opts: filesystemOptions{ - // Test relies on no dentry being held in the cache. - maxCachedDentries: 0, - }, - } - - ctx := contexttest.Context(t) - attr := &p9.Attr{ - Mode: p9.ModeRegular, - } - mask := p9.AttrMask{ - Mode: true, - Size: true, - } - parent, err := fs.newDentry(ctx, p9file{}, p9.QID{}, mask, attr) - if err != nil { - t.Fatalf("fs.newDentry(): %v", err) - } - - child, err := fs.newDentry(ctx, p9file{}, p9.QID{}, mask, attr) - if err != nil { - t.Fatalf("fs.newDentry(): %v", err) - } - parent.IncRef() // reference held by child on its parent. - parent.vfsd.InsertChild(&child.vfsd, "child") - - child.checkCachingLocked() - if got := atomic.LoadInt64(&child.refs); got != -1 { - t.Fatalf("child.refs=%d, want: -1", got) - } - // Parent will also be destroyed when child reference is removed. - if got := atomic.LoadInt64(&parent.refs); got != -1 { - t.Fatalf("parent.refs=%d, want: -1", got) - } - child.checkCachingLocked() - child.checkCachingLocked() -} diff --git a/pkg/sentry/fsimpl/gofer/handle.go b/pkg/sentry/fsimpl/gofer/handle.go deleted file mode 100644 index cfe66f797..000000000 --- a/pkg/sentry/fsimpl/gofer/handle.go +++ /dev/null @@ -1,135 +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 gofer - -import ( - "syscall" - - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/safemem" -) - -// handle represents a remote "open file descriptor", consisting of an opened -// fid (p9.File) and optionally a host file descriptor. -type handle struct { - file p9file - fd int32 // -1 if unavailable -} - -// Preconditions: read || write. -func openHandle(ctx context.Context, file p9file, read, write, trunc bool) (handle, error) { - _, newfile, err := file.walk(ctx, nil) - if err != nil { - return handle{fd: -1}, err - } - var flags p9.OpenFlags - switch { - case read && !write: - flags = p9.ReadOnly - case !read && write: - flags = p9.WriteOnly - case read && write: - flags = p9.ReadWrite - } - if trunc { - flags |= p9.OpenTruncate - } - fdobj, _, _, err := newfile.open(ctx, flags) - if err != nil { - newfile.close(ctx) - return handle{fd: -1}, err - } - fd := int32(-1) - if fdobj != nil { - fd = int32(fdobj.Release()) - } - return handle{ - file: newfile, - fd: fd, - }, nil -} - -func (h *handle) close(ctx context.Context) { - h.file.close(ctx) - h.file = p9file{} - if h.fd >= 0 { - syscall.Close(int(h.fd)) - h.fd = -1 - } -} - -func (h *handle) readToBlocksAt(ctx context.Context, dsts safemem.BlockSeq, offset uint64) (uint64, error) { - if dsts.IsEmpty() { - return 0, nil - } - if h.fd >= 0 { - ctx.UninterruptibleSleepStart(false) - n, err := hostPreadv(h.fd, dsts, int64(offset)) - ctx.UninterruptibleSleepFinish(false) - return n, err - } - if dsts.NumBlocks() == 1 && !dsts.Head().NeedSafecopy() { - n, err := h.file.readAt(ctx, dsts.Head().ToSlice(), offset) - return uint64(n), err - } - // Buffer the read since p9.File.ReadAt() takes []byte. - buf := make([]byte, dsts.NumBytes()) - n, err := h.file.readAt(ctx, buf, offset) - if n == 0 { - return 0, err - } - if cp, cperr := safemem.CopySeq(dsts, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf[:n]))); cperr != nil { - return cp, cperr - } - return uint64(n), err -} - -func (h *handle) writeFromBlocksAt(ctx context.Context, srcs safemem.BlockSeq, offset uint64) (uint64, error) { - if srcs.IsEmpty() { - return 0, nil - } - if h.fd >= 0 { - ctx.UninterruptibleSleepStart(false) - n, err := hostPwritev(h.fd, srcs, int64(offset)) - ctx.UninterruptibleSleepFinish(false) - return n, err - } - if srcs.NumBlocks() == 1 && !srcs.Head().NeedSafecopy() { - n, err := h.file.writeAt(ctx, srcs.Head().ToSlice(), offset) - return uint64(n), err - } - // Buffer the write since p9.File.WriteAt() takes []byte. - buf := make([]byte, srcs.NumBytes()) - cp, cperr := safemem.CopySeq(safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), srcs) - if cp == 0 { - return 0, cperr - } - n, err := h.file.writeAt(ctx, buf[:cp], offset) - if err != nil { - return uint64(n), err - } - return cp, cperr -} - -func (h *handle) sync(ctx context.Context) error { - if h.fd >= 0 { - ctx.UninterruptibleSleepStart(false) - err := syscall.Fsync(int(h.fd)) - ctx.UninterruptibleSleepFinish(false) - return err - } - return h.file.fsync(ctx) -} diff --git a/pkg/sentry/fsimpl/gofer/handle_unsafe.go b/pkg/sentry/fsimpl/gofer/handle_unsafe.go deleted file mode 100644 index 19560ab26..000000000 --- a/pkg/sentry/fsimpl/gofer/handle_unsafe.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 gofer - -import ( - "syscall" - "unsafe" - - "gvisor.dev/gvisor/pkg/safemem" -) - -// Preconditions: !dsts.IsEmpty(). -func hostPreadv(fd int32, dsts safemem.BlockSeq, off int64) (uint64, error) { - // No buffering is necessary regardless of safecopy; host syscalls will - // return EFAULT if appropriate, instead of raising SIGBUS. - if dsts.NumBlocks() == 1 { - // Use pread() instead of preadv() to avoid iovec allocation and - // copying. - dst := dsts.Head() - n, _, e := syscall.Syscall6(syscall.SYS_PREAD64, uintptr(fd), dst.Addr(), uintptr(dst.Len()), uintptr(off), 0, 0) - if e != 0 { - return 0, e - } - return uint64(n), nil - } - iovs := safemem.IovecsFromBlockSeq(dsts) - n, _, e := syscall.Syscall6(syscall.SYS_PREADV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(off), 0, 0) - if e != 0 { - return 0, e - } - return uint64(n), nil -} - -// Preconditions: !srcs.IsEmpty(). -func hostPwritev(fd int32, srcs safemem.BlockSeq, off int64) (uint64, error) { - // No buffering is necessary regardless of safecopy; host syscalls will - // return EFAULT if appropriate, instead of raising SIGBUS. - if srcs.NumBlocks() == 1 { - // Use pwrite() instead of pwritev() to avoid iovec allocation and - // copying. - src := srcs.Head() - n, _, e := syscall.Syscall6(syscall.SYS_PWRITE64, uintptr(fd), src.Addr(), uintptr(src.Len()), uintptr(off), 0, 0) - if e != 0 { - return 0, e - } - return uint64(n), nil - } - iovs := safemem.IovecsFromBlockSeq(srcs) - n, _, e := syscall.Syscall6(syscall.SYS_PWRITEV, uintptr(fd), uintptr((unsafe.Pointer)(&iovs[0])), uintptr(len(iovs)), uintptr(off), 0, 0) - if e != 0 { - return 0, e - } - return uint64(n), nil -} diff --git a/pkg/sentry/fsimpl/gofer/p9file.go b/pkg/sentry/fsimpl/gofer/p9file.go deleted file mode 100644 index 755ac2985..000000000 --- a/pkg/sentry/fsimpl/gofer/p9file.go +++ /dev/null @@ -1,219 +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 gofer - -import ( - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fd" - "gvisor.dev/gvisor/pkg/p9" - "gvisor.dev/gvisor/pkg/syserror" -) - -// p9file is a wrapper around p9.File that provides methods that are -// Context-aware. -type p9file struct { - file p9.File -} - -func (f p9file) isNil() bool { - return f.file == nil -} - -func (f p9file) walk(ctx context.Context, names []string) ([]p9.QID, p9file, error) { - ctx.UninterruptibleSleepStart(false) - qids, newfile, err := f.file.Walk(names) - ctx.UninterruptibleSleepFinish(false) - return qids, p9file{newfile}, err -} - -func (f p9file) walkGetAttr(ctx context.Context, names []string) ([]p9.QID, p9file, p9.AttrMask, p9.Attr, error) { - ctx.UninterruptibleSleepStart(false) - qids, newfile, attrMask, attr, err := f.file.WalkGetAttr(names) - ctx.UninterruptibleSleepFinish(false) - return qids, p9file{newfile}, attrMask, attr, err -} - -// walkGetAttrOne is a wrapper around p9.File.WalkGetAttr that takes a single -// path component and returns a single qid. -func (f p9file) walkGetAttrOne(ctx context.Context, name string) (p9.QID, p9file, p9.AttrMask, p9.Attr, error) { - ctx.UninterruptibleSleepStart(false) - qids, newfile, attrMask, attr, err := f.file.WalkGetAttr([]string{name}) - ctx.UninterruptibleSleepFinish(false) - if err != nil { - return p9.QID{}, p9file{}, p9.AttrMask{}, p9.Attr{}, err - } - if len(qids) != 1 { - ctx.Warningf("p9.File.WalkGetAttr returned %d qids (%v), wanted 1", len(qids), qids) - if newfile != nil { - p9file{newfile}.close(ctx) - } - return p9.QID{}, p9file{}, p9.AttrMask{}, p9.Attr{}, syserror.EIO - } - return qids[0], p9file{newfile}, attrMask, attr, nil -} - -func (f p9file) statFS(ctx context.Context) (p9.FSStat, error) { - ctx.UninterruptibleSleepStart(false) - fsstat, err := f.file.StatFS() - ctx.UninterruptibleSleepFinish(false) - return fsstat, err -} - -func (f p9file) getAttr(ctx context.Context, req p9.AttrMask) (p9.QID, p9.AttrMask, p9.Attr, error) { - ctx.UninterruptibleSleepStart(false) - qid, attrMask, attr, err := f.file.GetAttr(req) - ctx.UninterruptibleSleepFinish(false) - return qid, attrMask, attr, err -} - -func (f p9file) setAttr(ctx context.Context, valid p9.SetAttrMask, attr p9.SetAttr) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.SetAttr(valid, attr) - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) getXattr(ctx context.Context, name string, size uint64) (string, error) { - ctx.UninterruptibleSleepStart(false) - val, err := f.file.GetXattr(name, size) - ctx.UninterruptibleSleepFinish(false) - return val, err -} - -func (f p9file) setXattr(ctx context.Context, name, value string, flags uint32) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.SetXattr(name, value, flags) - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) allocate(ctx context.Context, mode p9.AllocateMode, offset, length uint64) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.Allocate(mode, offset, length) - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) close(ctx context.Context) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.Close() - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) open(ctx context.Context, flags p9.OpenFlags) (*fd.FD, p9.QID, uint32, error) { - ctx.UninterruptibleSleepStart(false) - fdobj, qid, iounit, err := f.file.Open(flags) - ctx.UninterruptibleSleepFinish(false) - return fdobj, qid, iounit, err -} - -func (f p9file) readAt(ctx context.Context, p []byte, offset uint64) (int, error) { - ctx.UninterruptibleSleepStart(false) - n, err := f.file.ReadAt(p, offset) - ctx.UninterruptibleSleepFinish(false) - return n, err -} - -func (f p9file) writeAt(ctx context.Context, p []byte, offset uint64) (int, error) { - ctx.UninterruptibleSleepStart(false) - n, err := f.file.WriteAt(p, offset) - ctx.UninterruptibleSleepFinish(false) - return n, err -} - -func (f p9file) fsync(ctx context.Context) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.FSync() - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) create(ctx context.Context, name string, flags p9.OpenFlags, permissions p9.FileMode, uid p9.UID, gid p9.GID) (*fd.FD, p9file, p9.QID, uint32, error) { - ctx.UninterruptibleSleepStart(false) - fdobj, newfile, qid, iounit, err := f.file.Create(name, flags, permissions, uid, gid) - ctx.UninterruptibleSleepFinish(false) - return fdobj, p9file{newfile}, qid, iounit, err -} - -func (f p9file) mkdir(ctx context.Context, name string, permissions p9.FileMode, uid p9.UID, gid p9.GID) (p9.QID, error) { - ctx.UninterruptibleSleepStart(false) - qid, err := f.file.Mkdir(name, permissions, uid, gid) - ctx.UninterruptibleSleepFinish(false) - return qid, err -} - -func (f p9file) symlink(ctx context.Context, oldName string, newName string, uid p9.UID, gid p9.GID) (p9.QID, error) { - ctx.UninterruptibleSleepStart(false) - qid, err := f.file.Symlink(oldName, newName, uid, gid) - ctx.UninterruptibleSleepFinish(false) - return qid, err -} - -func (f p9file) link(ctx context.Context, target p9file, newName string) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.Link(target.file, newName) - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) mknod(ctx context.Context, name string, mode p9.FileMode, major uint32, minor uint32, uid p9.UID, gid p9.GID) (p9.QID, error) { - ctx.UninterruptibleSleepStart(false) - qid, err := f.file.Mknod(name, mode, major, minor, uid, gid) - ctx.UninterruptibleSleepFinish(false) - return qid, err -} - -func (f p9file) rename(ctx context.Context, newDir p9file, newName string) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.Rename(newDir.file, newName) - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) unlinkAt(ctx context.Context, name string, flags uint32) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.UnlinkAt(name, flags) - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) readdir(ctx context.Context, offset uint64, count uint32) ([]p9.Dirent, error) { - ctx.UninterruptibleSleepStart(false) - dirents, err := f.file.Readdir(offset, count) - ctx.UninterruptibleSleepFinish(false) - return dirents, err -} - -func (f p9file) readlink(ctx context.Context) (string, error) { - ctx.UninterruptibleSleepStart(false) - target, err := f.file.Readlink() - ctx.UninterruptibleSleepFinish(false) - return target, err -} - -func (f p9file) flush(ctx context.Context) error { - ctx.UninterruptibleSleepStart(false) - err := f.file.Flush() - ctx.UninterruptibleSleepFinish(false) - return err -} - -func (f p9file) connect(ctx context.Context, flags p9.ConnectFlags) (*fd.FD, error) { - ctx.UninterruptibleSleepStart(false) - fdobj, err := f.file.Connect(flags) - ctx.UninterruptibleSleepFinish(false) - return fdobj, err -} diff --git a/pkg/sentry/fsimpl/gofer/pagemath.go b/pkg/sentry/fsimpl/gofer/pagemath.go deleted file mode 100644 index 847cb0784..000000000 --- a/pkg/sentry/fsimpl/gofer/pagemath.go +++ /dev/null @@ -1,31 +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 gofer - -import ( - "gvisor.dev/gvisor/pkg/usermem" -) - -// This are equivalent to usermem.Addr.RoundDown/Up, but without the -// potentially truncating conversion to usermem.Addr. This is necessary because -// there is no way to define generic "PageRoundDown/Up" functions in Go. - -func pageRoundDown(x uint64) uint64 { - return x &^ (usermem.PageSize - 1) -} - -func pageRoundUp(x uint64) uint64 { - return pageRoundDown(x + usermem.PageSize - 1) -} diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go deleted file mode 100644 index 857f7c74e..000000000 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ /dev/null @@ -1,874 +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 gofer - -import ( - "fmt" - "io" - "math" - "sync" - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -func (d *dentry) isRegularFile() bool { - return d.fileType() == linux.S_IFREG -} - -type regularFileFD struct { - fileDescription - - // off is the file offset. off is protected by mu. - mu sync.Mutex - off int64 -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *regularFileFD) Release() { -} - -// OnClose implements vfs.FileDescriptionImpl.OnClose. -func (fd *regularFileFD) OnClose(ctx context.Context) error { - if !fd.vfsfd.IsWritable() { - return nil - } - // Skip flushing if writes may be buffered by the client, since (as with - // the VFS1 client) we don't flush buffered writes on close anyway. - d := fd.dentry() - if d.fs.opts.interop == InteropModeExclusive { - return nil - } - d.handleMu.RLock() - defer d.handleMu.RUnlock() - return d.handle.file.flush(ctx) -} - -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - if offset < 0 { - return 0, syserror.EINVAL - } - if opts.Flags != 0 { - return 0, syserror.EOPNOTSUPP - } - - // Check for reading at EOF before calling into MM (but not under - // InteropModeShared, which makes d.size unreliable). - d := fd.dentry() - if d.fs.opts.interop != InteropModeShared && uint64(offset) >= atomic.LoadUint64(&d.size) { - return 0, io.EOF - } - - if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { - // Lock d.metadataMu for the rest of the read to prevent d.size from - // changing. - d.metadataMu.Lock() - defer d.metadataMu.Unlock() - // Write dirty cached pages that will be touched by the read back to - // the remote file. - if err := d.writeback(ctx, offset, dst.NumBytes()); err != nil { - return 0, err - } - } - - rw := getDentryReadWriter(ctx, d, offset) - if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { - // Require the read to go to the remote file. - rw.direct = true - } - n, err := dst.CopyOutFrom(ctx, rw) - putDentryReadWriter(rw) - if d.fs.opts.interop != InteropModeShared { - // Compare Linux's mm/filemap.c:do_generic_file_read() => file_accessed(). - d.touchAtime(fd.vfsfd.Mount()) - } - return n, err -} - -// Read implements vfs.FileDescriptionImpl.Read. -func (fd *regularFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - fd.mu.Lock() - n, err := fd.PRead(ctx, dst, fd.off, opts) - fd.off += n - fd.mu.Unlock() - return n, err -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - if offset < 0 { - return 0, syserror.EINVAL - } - if opts.Flags != 0 { - return 0, syserror.EOPNOTSUPP - } - limit, err := vfs.CheckLimit(ctx, offset, src.NumBytes()) - if err != nil { - return 0, err - } - src = src.TakeFirst64(limit) - - d := fd.dentry() - d.metadataMu.Lock() - defer d.metadataMu.Unlock() - if d.fs.opts.interop != InteropModeShared { - // Compare Linux's mm/filemap.c:__generic_file_write_iter() => - // file_update_time(). This is d.touchCMtime(), but without locking - // d.metadataMu (recursively). - d.touchCMtimeLocked() - } - if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { - // Write dirty cached pages that will be touched by the write back to - // the remote file. - if err := d.writeback(ctx, offset, src.NumBytes()); err != nil { - return 0, err - } - // Remove touched pages from the cache. - pgstart := pageRoundDown(uint64(offset)) - pgend := pageRoundUp(uint64(offset + src.NumBytes())) - if pgend < pgstart { - return 0, syserror.EINVAL - } - mr := memmap.MappableRange{pgstart, pgend} - var freed []platform.FileRange - d.dataMu.Lock() - cseg := d.cache.LowerBoundSegment(mr.Start) - for cseg.Ok() && cseg.Start() < mr.End { - cseg = d.cache.Isolate(cseg, mr) - freed = append(freed, platform.FileRange{cseg.Value(), cseg.Value() + cseg.Range().Length()}) - cseg = d.cache.Remove(cseg).NextSegment() - } - d.dataMu.Unlock() - // Invalidate mappings of removed pages. - d.mapsMu.Lock() - d.mappings.Invalidate(mr, memmap.InvalidateOpts{}) - d.mapsMu.Unlock() - // Finally free pages removed from the cache. - mf := d.fs.mfp.MemoryFile() - for _, freedFR := range freed { - mf.DecRef(freedFR) - } - } - rw := getDentryReadWriter(ctx, d, offset) - if fd.vfsfd.StatusFlags()&linux.O_DIRECT != 0 { - // Require the write to go to the remote file. - rw.direct = true - } - n, err := src.CopyInTo(ctx, rw) - putDentryReadWriter(rw) - if n != 0 && fd.vfsfd.StatusFlags()&(linux.O_DSYNC|linux.O_SYNC) != 0 { - // Write dirty cached pages touched by the write back to the remote - // file. - if err := d.writeback(ctx, offset, src.NumBytes()); err != nil { - return 0, err - } - // Request the remote filesystem to sync the remote file. - if err := d.handle.file.fsync(ctx); err != nil { - return 0, err - } - } - return n, err -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *regularFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - fd.mu.Lock() - n, err := fd.PWrite(ctx, src, fd.off, opts) - fd.off += n - fd.mu.Unlock() - return n, err -} - -type dentryReadWriter struct { - ctx context.Context - d *dentry - off uint64 - direct bool -} - -var dentryReadWriterPool = sync.Pool{ - New: func() interface{} { - return &dentryReadWriter{} - }, -} - -func getDentryReadWriter(ctx context.Context, d *dentry, offset int64) *dentryReadWriter { - rw := dentryReadWriterPool.Get().(*dentryReadWriter) - rw.ctx = ctx - rw.d = d - rw.off = uint64(offset) - rw.direct = false - return rw -} - -func putDentryReadWriter(rw *dentryReadWriter) { - rw.ctx = nil - rw.d = nil - dentryReadWriterPool.Put(rw) -} - -// ReadToBlocks implements safemem.Reader.ReadToBlocks. -func (rw *dentryReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { - if dsts.IsEmpty() { - return 0, nil - } - - // If we have a mmappable host FD (which must be used here to ensure - // coherence with memory-mapped I/O), or if InteropModeShared is in effect - // (which prevents us from caching file contents and makes dentry.size - // unreliable), or if the file was opened O_DIRECT, read directly from - // dentry.handle without locking dentry.dataMu. - rw.d.handleMu.RLock() - if (rw.d.handle.fd >= 0 && !rw.d.fs.opts.forcePageCache) || rw.d.fs.opts.interop == InteropModeShared || rw.direct { - n, err := rw.d.handle.readToBlocksAt(rw.ctx, dsts, rw.off) - rw.d.handleMu.RUnlock() - rw.off += n - return n, err - } - - // Otherwise read from/through the cache. - mf := rw.d.fs.mfp.MemoryFile() - fillCache := mf.ShouldCacheEvictable() - var dataMuUnlock func() - if fillCache { - rw.d.dataMu.Lock() - dataMuUnlock = rw.d.dataMu.Unlock - } else { - rw.d.dataMu.RLock() - dataMuUnlock = rw.d.dataMu.RUnlock - } - - // Compute the range to read (limited by file size and overflow-checked). - if rw.off >= rw.d.size { - dataMuUnlock() - rw.d.handleMu.RUnlock() - return 0, io.EOF - } - end := rw.d.size - if rend := rw.off + dsts.NumBytes(); rend > rw.off && rend < end { - end = rend - } - - var done uint64 - seg, gap := rw.d.cache.Find(rw.off) - for rw.off < end { - mr := memmap.MappableRange{rw.off, end} - switch { - case seg.Ok(): - // Get internal mappings from the cache. - ims, err := mf.MapInternal(seg.FileRangeOf(seg.Range().Intersect(mr)), usermem.Read) - if err != nil { - dataMuUnlock() - rw.d.handleMu.RUnlock() - return done, err - } - - // Copy from internal mappings. - n, err := safemem.CopySeq(dsts, ims) - done += n - rw.off += n - dsts = dsts.DropFirst64(n) - if err != nil { - dataMuUnlock() - rw.d.handleMu.RUnlock() - return done, err - } - - // Continue. - seg, gap = seg.NextNonEmpty() - - case gap.Ok(): - gapMR := gap.Range().Intersect(mr) - if fillCache { - // Read into the cache, then re-enter the loop to read from the - // cache. - reqMR := memmap.MappableRange{ - Start: pageRoundDown(gapMR.Start), - End: pageRoundUp(gapMR.End), - } - optMR := gap.Range() - err := rw.d.cache.Fill(rw.ctx, reqMR, maxFillRange(reqMR, optMR), mf, usage.PageCache, rw.d.handle.readToBlocksAt) - mf.MarkEvictable(rw.d, pgalloc.EvictableRange{optMR.Start, optMR.End}) - seg, gap = rw.d.cache.Find(rw.off) - if !seg.Ok() { - dataMuUnlock() - rw.d.handleMu.RUnlock() - return done, err - } - // err might have occurred in part of gap.Range() outside - // gapMR. Forget about it for now; if the error matters and - // persists, we'll run into it again in a later iteration of - // this loop. - } else { - // Read directly from the file. - gapDsts := dsts.TakeFirst64(gapMR.Length()) - n, err := rw.d.handle.readToBlocksAt(rw.ctx, gapDsts, gapMR.Start) - done += n - rw.off += n - dsts = dsts.DropFirst64(n) - // Partial reads are fine. But we must stop reading. - if n != gapDsts.NumBytes() || err != nil { - dataMuUnlock() - rw.d.handleMu.RUnlock() - return done, err - } - - // Continue. - seg, gap = gap.NextSegment(), fsutil.FileRangeGapIterator{} - } - } - } - dataMuUnlock() - rw.d.handleMu.RUnlock() - return done, nil -} - -// WriteFromBlocks implements safemem.Writer.WriteFromBlocks. -// -// Preconditions: rw.d.metadataMu must be locked. -func (rw *dentryReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { - if srcs.IsEmpty() { - return 0, nil - } - - // If we have a mmappable host FD (which must be used here to ensure - // coherence with memory-mapped I/O), or if InteropModeShared is in effect - // (which prevents us from caching file contents), or if the file was - // opened with O_DIRECT, write directly to dentry.handle without locking - // dentry.dataMu. - rw.d.handleMu.RLock() - if (rw.d.handle.fd >= 0 && !rw.d.fs.opts.forcePageCache) || rw.d.fs.opts.interop == InteropModeShared || rw.direct { - n, err := rw.d.handle.writeFromBlocksAt(rw.ctx, srcs, rw.off) - rw.off += n - rw.d.dataMu.Lock() - if rw.off > rw.d.size { - atomic.StoreUint64(&rw.d.size, rw.off) - // The remote file's size will implicitly be extended to the correct - // value when we write back to it. - } - rw.d.dataMu.Unlock() - rw.d.handleMu.RUnlock() - return n, err - } - - // Otherwise write to/through the cache. - mf := rw.d.fs.mfp.MemoryFile() - rw.d.dataMu.Lock() - - // Compute the range to write (overflow-checked). - start := rw.off - end := rw.off + srcs.NumBytes() - if end <= rw.off { - end = math.MaxInt64 - } - - var ( - done uint64 - retErr error - ) - seg, gap := rw.d.cache.Find(rw.off) - for rw.off < end { - mr := memmap.MappableRange{rw.off, end} - switch { - case seg.Ok(): - // Get internal mappings from the cache. - segMR := seg.Range().Intersect(mr) - ims, err := mf.MapInternal(seg.FileRangeOf(segMR), usermem.Write) - if err != nil { - retErr = err - goto exitLoop - } - - // Copy to internal mappings. - n, err := safemem.CopySeq(ims, srcs) - done += n - rw.off += n - srcs = srcs.DropFirst64(n) - rw.d.dirty.MarkDirty(segMR) - if err != nil { - retErr = err - goto exitLoop - } - - // Continue. - seg, gap = seg.NextNonEmpty() - - case gap.Ok(): - // Write directly to the file. At present, we never fill the cache - // when writing, since doing so can convert small writes into - // inefficient read-modify-write cycles, and we have no mechanism - // for detecting or avoiding this. - gapMR := gap.Range().Intersect(mr) - gapSrcs := srcs.TakeFirst64(gapMR.Length()) - n, err := rw.d.handle.writeFromBlocksAt(rw.ctx, gapSrcs, gapMR.Start) - done += n - rw.off += n - srcs = srcs.DropFirst64(n) - // Partial writes are fine. But we must stop writing. - if n != gapSrcs.NumBytes() || err != nil { - retErr = err - goto exitLoop - } - - // Continue. - seg, gap = gap.NextSegment(), fsutil.FileRangeGapIterator{} - } - } -exitLoop: - if rw.off > rw.d.size { - atomic.StoreUint64(&rw.d.size, rw.off) - // The remote file's size will implicitly be extended to the correct - // value when we write back to it. - } - // If InteropModeWritethrough is in effect, flush written data back to the - // remote filesystem. - if rw.d.fs.opts.interop == InteropModeWritethrough && done != 0 { - if err := fsutil.SyncDirty(rw.ctx, memmap.MappableRange{ - Start: start, - End: rw.off, - }, &rw.d.cache, &rw.d.dirty, rw.d.size, mf, rw.d.handle.writeFromBlocksAt); err != nil { - // We have no idea how many bytes were actually flushed. - rw.off = start - done = 0 - retErr = err - } - } - rw.d.dataMu.Unlock() - rw.d.handleMu.RUnlock() - return done, retErr -} - -func (d *dentry) writeback(ctx context.Context, offset, size int64) error { - if size == 0 { - return nil - } - d.handleMu.RLock() - defer d.handleMu.RUnlock() - d.dataMu.Lock() - defer d.dataMu.Unlock() - // Compute the range of valid bytes (overflow-checked). - if uint64(offset) >= d.size { - return nil - } - end := int64(d.size) - if rend := offset + size; rend > offset && rend < end { - end = rend - } - return fsutil.SyncDirty(ctx, memmap.MappableRange{ - Start: uint64(offset), - End: uint64(end), - }, &d.cache, &d.dirty, d.size, d.fs.mfp.MemoryFile(), d.handle.writeFromBlocksAt) -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *regularFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - fd.mu.Lock() - defer fd.mu.Unlock() - switch whence { - case linux.SEEK_SET: - // Use offset as specified. - case linux.SEEK_CUR: - offset += fd.off - case linux.SEEK_END, linux.SEEK_DATA, linux.SEEK_HOLE: - // Ensure file size is up to date. - d := fd.dentry() - if fd.filesystem().opts.interop == InteropModeShared { - if err := d.updateFromGetattr(ctx); err != nil { - return 0, err - } - } - size := int64(atomic.LoadUint64(&d.size)) - // For SEEK_DATA and SEEK_HOLE, treat the file as a single contiguous - // block of data. - switch whence { - case linux.SEEK_END: - offset += size - case linux.SEEK_DATA: - if offset > size { - return 0, syserror.ENXIO - } - // Use offset as specified. - case linux.SEEK_HOLE: - if offset > size { - return 0, syserror.ENXIO - } - offset = size - } - default: - return 0, syserror.EINVAL - } - if offset < 0 { - return 0, syserror.EINVAL - } - fd.off = offset - return offset, nil -} - -// Sync implements vfs.FileDescriptionImpl.Sync. -func (fd *regularFileFD) Sync(ctx context.Context) error { - return fd.dentry().syncSharedHandle(ctx) -} - -func (d *dentry) syncSharedHandle(ctx context.Context) error { - d.handleMu.RLock() - if !d.handleWritable { - d.handleMu.RUnlock() - return nil - } - d.dataMu.Lock() - // Write dirty cached data to the remote file. - err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, d.fs.mfp.MemoryFile(), d.handle.writeFromBlocksAt) - d.dataMu.Unlock() - if err == nil { - // Sync the remote file. - err = d.handle.sync(ctx) - } - d.handleMu.RUnlock() - return err -} - -// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. -func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { - d := fd.dentry() - switch d.fs.opts.interop { - case InteropModeExclusive: - // Any mapping is fine. - case InteropModeWritethrough: - // Shared writable mappings require a host FD, since otherwise we can't - // synchronously flush memory-mapped writes to the remote file. - if opts.Private || !opts.MaxPerms.Write { - break - } - fallthrough - case InteropModeShared: - // All mappings require a host FD to be coherent with other filesystem - // users. - if d.fs.opts.forcePageCache { - // Whether or not we have a host FD, we're not allowed to use it. - return syserror.ENODEV - } - d.handleMu.RLock() - haveFD := d.handle.fd >= 0 - d.handleMu.RUnlock() - if !haveFD { - return syserror.ENODEV - } - default: - panic(fmt.Sprintf("unknown InteropMode %v", d.fs.opts.interop)) - } - // After this point, d may be used as a memmap.Mappable. - d.pf.hostFileMapperInitOnce.Do(d.pf.hostFileMapper.Init) - return vfs.GenericConfigureMMap(&fd.vfsfd, d, opts) -} - -func (d *dentry) mayCachePages() bool { - if d.fs.opts.interop == InteropModeShared { - return false - } - if d.fs.opts.forcePageCache { - return true - } - d.handleMu.RLock() - haveFD := d.handle.fd >= 0 - d.handleMu.RUnlock() - return haveFD -} - -// AddMapping implements memmap.Mappable.AddMapping. -func (d *dentry) AddMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) error { - d.mapsMu.Lock() - mapped := d.mappings.AddMapping(ms, ar, offset, writable) - // Do this unconditionally since whether we have a host FD can change - // across save/restore. - for _, r := range mapped { - d.pf.hostFileMapper.IncRefOn(r) - } - if d.mayCachePages() { - // d.Evict() will refuse to evict memory-mapped pages, so tell the - // MemoryFile to not bother trying. - mf := d.fs.mfp.MemoryFile() - for _, r := range mapped { - mf.MarkUnevictable(d, pgalloc.EvictableRange{r.Start, r.End}) - } - } - d.mapsMu.Unlock() - return nil -} - -// RemoveMapping implements memmap.Mappable.RemoveMapping. -func (d *dentry) RemoveMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) { - d.mapsMu.Lock() - unmapped := d.mappings.RemoveMapping(ms, ar, offset, writable) - for _, r := range unmapped { - d.pf.hostFileMapper.DecRefOn(r) - } - if d.mayCachePages() { - // Pages that are no longer referenced by any application memory - // mappings are now considered unused; allow MemoryFile to evict them - // when necessary. - mf := d.fs.mfp.MemoryFile() - d.dataMu.Lock() - for _, r := range unmapped { - // Since these pages are no longer mapped, they are no longer - // concurrently dirtyable by a writable memory mapping. - d.dirty.AllowClean(r) - mf.MarkEvictable(d, pgalloc.EvictableRange{r.Start, r.End}) - } - d.dataMu.Unlock() - } - d.mapsMu.Unlock() -} - -// CopyMapping implements memmap.Mappable.CopyMapping. -func (d *dentry) CopyMapping(ctx context.Context, ms memmap.MappingSpace, srcAR, dstAR usermem.AddrRange, offset uint64, writable bool) error { - return d.AddMapping(ctx, ms, dstAR, offset, writable) -} - -// Translate implements memmap.Mappable.Translate. -func (d *dentry) Translate(ctx context.Context, required, optional memmap.MappableRange, at usermem.AccessType) ([]memmap.Translation, error) { - d.handleMu.RLock() - if d.handle.fd >= 0 && !d.fs.opts.forcePageCache { - d.handleMu.RUnlock() - mr := optional - if d.fs.opts.limitHostFDTranslation { - mr = maxFillRange(required, optional) - } - return []memmap.Translation{ - { - Source: mr, - File: &d.pf, - Offset: mr.Start, - Perms: usermem.AnyAccess, - }, - }, nil - } - - d.dataMu.Lock() - - // Constrain translations to d.size (rounded up) to prevent translation to - // pages that may be concurrently truncated. - pgend := pageRoundUp(d.size) - var beyondEOF bool - if required.End > pgend { - if required.Start >= pgend { - d.dataMu.Unlock() - d.handleMu.RUnlock() - return nil, &memmap.BusError{io.EOF} - } - beyondEOF = true - required.End = pgend - } - if optional.End > pgend { - optional.End = pgend - } - - mf := d.fs.mfp.MemoryFile() - cerr := d.cache.Fill(ctx, required, maxFillRange(required, optional), mf, usage.PageCache, d.handle.readToBlocksAt) - - var ts []memmap.Translation - var translatedEnd uint64 - for seg := d.cache.FindSegment(required.Start); seg.Ok() && seg.Start() < required.End; seg, _ = seg.NextNonEmpty() { - segMR := seg.Range().Intersect(optional) - // TODO(jamieliu): Make Translations writable even if writability is - // not required if already kept-dirty by another writable translation. - perms := usermem.AccessType{ - Read: true, - Execute: true, - } - if at.Write { - // From this point forward, this memory can be dirtied through the - // mapping at any time. - d.dirty.KeepDirty(segMR) - perms.Write = true - } - ts = append(ts, memmap.Translation{ - Source: segMR, - File: mf, - Offset: seg.FileRangeOf(segMR).Start, - Perms: perms, - }) - translatedEnd = segMR.End - } - - d.dataMu.Unlock() - d.handleMu.RUnlock() - - // Don't return the error returned by c.cache.Fill if it occurred outside - // of required. - if translatedEnd < required.End && cerr != nil { - return ts, &memmap.BusError{cerr} - } - if beyondEOF { - return ts, &memmap.BusError{io.EOF} - } - return ts, nil -} - -func maxFillRange(required, optional memmap.MappableRange) memmap.MappableRange { - const maxReadahead = 64 << 10 // 64 KB, chosen arbitrarily - if required.Length() >= maxReadahead { - return required - } - if optional.Length() <= maxReadahead { - return optional - } - optional.Start = required.Start - if optional.Length() <= maxReadahead { - return optional - } - optional.End = optional.Start + maxReadahead - return optional -} - -// InvalidateUnsavable implements memmap.Mappable.InvalidateUnsavable. -func (d *dentry) InvalidateUnsavable(ctx context.Context) error { - // Whether we have a host fd (and consequently what platform.File is - // mapped) can change across save/restore, so invalidate all translations - // unconditionally. - d.mapsMu.Lock() - defer d.mapsMu.Unlock() - d.mappings.InvalidateAll(memmap.InvalidateOpts{}) - - // Write the cache's contents back to the remote file so that if we have a - // host fd after restore, the remote file's contents are coherent. - mf := d.fs.mfp.MemoryFile() - d.dataMu.Lock() - defer d.dataMu.Unlock() - if err := fsutil.SyncDirtyAll(ctx, &d.cache, &d.dirty, d.size, mf, d.handle.writeFromBlocksAt); err != nil { - return err - } - - // Discard the cache so that it's not stored in saved state. This is safe - // because per InvalidateUnsavable invariants, no new translations can have - // been returned after we invalidated all existing translations above. - d.cache.DropAll(mf) - d.dirty.RemoveAll() - - return nil -} - -// Evict implements pgalloc.EvictableMemoryUser.Evict. -func (d *dentry) Evict(ctx context.Context, er pgalloc.EvictableRange) { - d.mapsMu.Lock() - defer d.mapsMu.Unlock() - d.dataMu.Lock() - defer d.dataMu.Unlock() - - mr := memmap.MappableRange{er.Start, er.End} - mf := d.fs.mfp.MemoryFile() - // Only allow pages that are no longer memory-mapped to be evicted. - for mgap := d.mappings.LowerBoundGap(mr.Start); mgap.Ok() && mgap.Start() < mr.End; mgap = mgap.NextGap() { - mgapMR := mgap.Range().Intersect(mr) - if mgapMR.Length() == 0 { - continue - } - if err := fsutil.SyncDirty(ctx, mgapMR, &d.cache, &d.dirty, d.size, mf, d.handle.writeFromBlocksAt); err != nil { - log.Warningf("Failed to writeback cached data %v: %v", mgapMR, err) - } - d.cache.Drop(mgapMR, mf) - d.dirty.KeepClean(mgapMR) - } -} - -// dentryPlatformFile implements platform.File. It exists solely because dentry -// cannot implement both vfs.DentryImpl.IncRef and platform.File.IncRef. -// -// dentryPlatformFile is only used when a host FD representing the remote file -// is available (i.e. dentry.handle.fd >= 0), and that FD is used for -// application memory mappings (i.e. !filesystem.opts.forcePageCache). -type dentryPlatformFile struct { - *dentry - - // fdRefs counts references on platform.File offsets. fdRefs is protected - // by dentry.dataMu. - fdRefs fsutil.FrameRefSet - - // If this dentry represents a regular file, and handle.fd >= 0, - // hostFileMapper caches mappings of handle.fd. - hostFileMapper fsutil.HostFileMapper - - // hostFileMapperInitOnce is used to lazily initialize hostFileMapper. - hostFileMapperInitOnce sync.Once -} - -// IncRef implements platform.File.IncRef. -func (d *dentryPlatformFile) IncRef(fr platform.FileRange) { - d.dataMu.Lock() - seg, gap := d.fdRefs.Find(fr.Start) - for { - switch { - case seg.Ok() && seg.Start() < fr.End: - seg = d.fdRefs.Isolate(seg, fr) - seg.SetValue(seg.Value() + 1) - seg, gap = seg.NextNonEmpty() - case gap.Ok() && gap.Start() < fr.End: - newRange := gap.Range().Intersect(fr) - usage.MemoryAccounting.Inc(newRange.Length(), usage.Mapped) - seg, gap = d.fdRefs.InsertWithoutMerging(gap, newRange, 1).NextNonEmpty() - default: - d.fdRefs.MergeAdjacent(fr) - d.dataMu.Unlock() - return - } - } -} - -// DecRef implements platform.File.DecRef. -func (d *dentryPlatformFile) DecRef(fr platform.FileRange) { - d.dataMu.Lock() - seg := d.fdRefs.FindSegment(fr.Start) - - for seg.Ok() && seg.Start() < fr.End { - seg = d.fdRefs.Isolate(seg, fr) - if old := seg.Value(); old == 1 { - usage.MemoryAccounting.Dec(seg.Range().Length(), usage.Mapped) - seg = d.fdRefs.Remove(seg).NextSegment() - } else { - seg.SetValue(old - 1) - seg = seg.NextSegment() - } - } - d.fdRefs.MergeAdjacent(fr) - d.dataMu.Unlock() - -} - -// MapInternal implements platform.File.MapInternal. -func (d *dentryPlatformFile) MapInternal(fr platform.FileRange, at usermem.AccessType) (safemem.BlockSeq, error) { - d.handleMu.RLock() - bs, err := d.hostFileMapper.MapInternal(fr, int(d.handle.fd), at.Write) - d.handleMu.RUnlock() - return bs, err -} - -// FD implements platform.File.FD. -func (d *dentryPlatformFile) FD() int { - d.handleMu.RLock() - fd := d.handle.fd - d.handleMu.RUnlock() - return int(fd) -} diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go deleted file mode 100644 index 507e0e276..000000000 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ /dev/null @@ -1,167 +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 gofer - -import ( - "sync" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// specialFileFD implements vfs.FileDescriptionImpl for files other than -// regular files, directories, and symlinks: pipes, sockets, etc. It is also -// used for regular files when filesystemOptions.specialRegularFiles is in -// effect. specialFileFD differs from regularFileFD by using per-FD handles -// instead of shared per-dentry handles, and never buffering I/O. -type specialFileFD struct { - fileDescription - - // handle is immutable. - handle handle - - // off is the file offset. off is protected by mu. (POSIX 2.9.7 only - // requires operations using the file offset to be atomic for regular files - // and symlinks; however, since specialFileFD may be used for regular - // files, we apply this atomicity unconditionally.) - mu sync.Mutex - off int64 -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *specialFileFD) Release() { - fd.handle.close(context.Background()) - fs := fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) - fs.syncMu.Lock() - delete(fs.specialFileFDs, fd) - fs.syncMu.Unlock() -} - -// OnClose implements vfs.FileDescriptionImpl.OnClose. -func (fd *specialFileFD) OnClose(ctx context.Context) error { - if !fd.vfsfd.IsWritable() { - return nil - } - return fd.handle.file.flush(ctx) -} - -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - if offset < 0 { - return 0, syserror.EINVAL - } - if opts.Flags != 0 { - return 0, syserror.EOPNOTSUPP - } - - // Going through dst.CopyOutFrom() holds MM locks around file operations of - // unknown duration. For regularFileFD, doing so is necessary to support - // mmap due to lock ordering; MM locks precede dentry.dataMu. That doesn't - // hold here since specialFileFD doesn't client-cache data. Just buffer the - // read instead. - if d := fd.dentry(); d.fs.opts.interop != InteropModeShared { - d.touchAtime(fd.vfsfd.Mount()) - } - buf := make([]byte, dst.NumBytes()) - n, err := fd.handle.readToBlocksAt(ctx, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), uint64(offset)) - if n == 0 { - return 0, err - } - if cp, cperr := dst.CopyOut(ctx, buf[:n]); cperr != nil { - return int64(cp), cperr - } - return int64(n), err -} - -// Read implements vfs.FileDescriptionImpl.Read. -func (fd *specialFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - fd.mu.Lock() - n, err := fd.PRead(ctx, dst, fd.off, opts) - fd.off += n - fd.mu.Unlock() - return n, err -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *specialFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - if offset < 0 { - return 0, syserror.EINVAL - } - if opts.Flags != 0 { - return 0, syserror.EOPNOTSUPP - } - - if fd.dentry().fileType() == linux.S_IFREG { - limit, err := vfs.CheckLimit(ctx, offset, src.NumBytes()) - if err != nil { - return 0, err - } - src = src.TakeFirst64(limit) - } - - // Do a buffered write. See rationale in PRead. - if d := fd.dentry(); d.fs.opts.interop != InteropModeShared { - d.touchCMtime() - } - buf := make([]byte, src.NumBytes()) - // Don't do partial writes if we get a partial read from src. - if _, err := src.CopyIn(ctx, buf); err != nil { - return 0, err - } - n, err := fd.handle.writeFromBlocksAt(ctx, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), uint64(offset)) - return int64(n), err -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *specialFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - fd.mu.Lock() - n, err := fd.PWrite(ctx, src, fd.off, opts) - fd.off += n - fd.mu.Unlock() - return n, err -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *specialFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - fd.mu.Lock() - defer fd.mu.Unlock() - switch whence { - case linux.SEEK_SET: - // Use offset as given. - case linux.SEEK_CUR: - offset += fd.off - default: - // SEEK_END, SEEK_DATA, and SEEK_HOLE aren't supported since it's not - // clear that file size is even meaningful for these files. - return 0, syserror.EINVAL - } - if offset < 0 { - return 0, syserror.EINVAL - } - fd.off = offset - return offset, nil -} - -// Sync implements vfs.FileDescriptionImpl.Sync. -func (fd *specialFileFD) Sync(ctx context.Context) error { - if !fd.vfsfd.IsWritable() { - return nil - } - return fd.handle.sync(ctx) -} diff --git a/pkg/sentry/fsimpl/gofer/symlink.go b/pkg/sentry/fsimpl/gofer/symlink.go deleted file mode 100644 index 2ec819f86..000000000 --- a/pkg/sentry/fsimpl/gofer/symlink.go +++ /dev/null @@ -1,47 +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 gofer - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -func (d *dentry) isSymlink() bool { - return d.fileType() == linux.S_IFLNK -} - -// Precondition: d.isSymlink(). -func (d *dentry) readlink(ctx context.Context, mnt *vfs.Mount) (string, error) { - if d.fs.opts.interop != InteropModeShared { - d.touchAtime(mnt) - d.dataMu.Lock() - if d.haveTarget { - target := d.target - d.dataMu.Unlock() - return target, nil - } - } - target, err := d.file.readlink(ctx) - if d.fs.opts.interop != InteropModeShared { - if err == nil { - d.haveTarget = true - d.target = target - } - d.dataMu.Unlock() - } - return target, err -} diff --git a/pkg/sentry/fsimpl/gofer/time.go b/pkg/sentry/fsimpl/gofer/time.go deleted file mode 100644 index 2608e7e1d..000000000 --- a/pkg/sentry/fsimpl/gofer/time.go +++ /dev/null @@ -1,74 +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 gofer - -import ( - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -func dentryTimestampFromP9(s, ns uint64) int64 { - return int64(s*1e9 + ns) -} - -func dentryTimestampFromStatx(ts linux.StatxTimestamp) int64 { - return ts.Sec*1e9 + int64(ts.Nsec) -} - -func statxTimestampFromDentry(ns int64) linux.StatxTimestamp { - return linux.StatxTimestamp{ - Sec: ns / 1e9, - Nsec: uint32(ns % 1e9), - } -} - -// Preconditions: fs.interop != InteropModeShared. -func (d *dentry) touchAtime(mnt *vfs.Mount) { - if err := mnt.CheckBeginWrite(); err != nil { - return - } - now := d.fs.clock.Now().Nanoseconds() - d.metadataMu.Lock() - atomic.StoreInt64(&d.atime, now) - d.metadataMu.Unlock() - mnt.EndWrite() -} - -// Preconditions: fs.interop != InteropModeShared. The caller has successfully -// called vfs.Mount.CheckBeginWrite(). -func (d *dentry) touchCtime() { - now := d.fs.clock.Now().Nanoseconds() - d.metadataMu.Lock() - atomic.StoreInt64(&d.ctime, now) - d.metadataMu.Unlock() -} - -// Preconditions: fs.interop != InteropModeShared. The caller has successfully -// called vfs.Mount.CheckBeginWrite(). -func (d *dentry) touchCMtime() { - now := d.fs.clock.Now().Nanoseconds() - d.metadataMu.Lock() - atomic.StoreInt64(&d.mtime, now) - atomic.StoreInt64(&d.ctime, now) - d.metadataMu.Unlock() -} - -func (d *dentry) touchCMtimeLocked() { - now := d.fs.clock.Now().Nanoseconds() - atomic.StoreInt64(&d.mtime, now) - atomic.StoreInt64(&d.ctime, now) -} diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD deleted file mode 100644 index 82e1fb74b..000000000 --- a/pkg/sentry/fsimpl/host/BUILD +++ /dev/null @@ -1,34 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -licenses(["notice"]) - -go_library( - name = "host", - srcs = [ - "host.go", - "ioctl_unsafe.go", - "tty.go", - "util.go", - "util_unsafe.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fd", - "//pkg/log", - "//pkg/refs", - "//pkg/safemem", - "//pkg/sentry/arch", - "//pkg/sentry/fsimpl/kernfs", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/memmap", - "//pkg/sentry/unimpl", - "//pkg/sentry/vfs", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - "@org_golang_x_sys//unix:go_default_library", - ], -) diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go deleted file mode 100644 index 97fa7f7ab..000000000 --- a/pkg/sentry/fsimpl/host/host.go +++ /dev/null @@ -1,667 +0,0 @@ -// Copyright 2020 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 provides a filesystem implementation for host files imported as -// file descriptors. -package host - -import ( - "errors" - "fmt" - "math" - "syscall" - - "golang.org/x/sys/unix" - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fd" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// filesystemType implements vfs.FilesystemType. -type filesystemType struct{} - -// GetFilesystem implements FilesystemType.GetFilesystem. -func (filesystemType) GetFilesystem(context.Context, *vfs.VirtualFilesystem, *auth.Credentials, string, vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - panic("cannot instaniate a host filesystem") -} - -// Name implements FilesystemType.Name. -func (filesystemType) Name() string { - return "none" -} - -// filesystem implements vfs.FilesystemImpl. -type filesystem struct { - kernfs.Filesystem -} - -// NewMount returns a new disconnected mount in vfsObj that may be passed to ImportFD. -func NewMount(vfsObj *vfs.VirtualFilesystem) (*vfs.Mount, error) { - fs := &filesystem{} - fs.Init(vfsObj, &filesystemType{}) - vfsfs := fs.VFSFilesystem() - // NewDisconnectedMount will take an additional reference on vfsfs. - defer vfsfs.DecRef() - return vfsObj.NewDisconnectedMount(vfsfs, nil, &vfs.MountOptions{}) -} - -// ImportFD sets up and returns a vfs.FileDescription from a donated fd. -func ImportFD(mnt *vfs.Mount, hostFD int, isTTY bool) (*vfs.FileDescription, error) { - fs, ok := mnt.Filesystem().Impl().(*kernfs.Filesystem) - if !ok { - return nil, fmt.Errorf("can't import host FDs into filesystems of type %T", mnt.Filesystem().Impl()) - } - - // Retrieve metadata. - var s unix.Stat_t - if err := unix.Fstat(hostFD, &s); err != nil { - return nil, err - } - - fileMode := linux.FileMode(s.Mode) - fileType := fileMode.FileType() - - // Determine if hostFD is seekable. If not, this syscall will return ESPIPE - // (see fs/read_write.c:llseek), e.g. for pipes, sockets, and some character - // devices. - _, err := unix.Seek(hostFD, 0, linux.SEEK_CUR) - seekable := err != syserror.ESPIPE - - i := &inode{ - hostFD: hostFD, - seekable: seekable, - isTTY: isTTY, - canMap: canMap(uint32(fileType)), - ino: fs.NextIno(), - mode: fileMode, - // For simplicity, set offset to 0. Technically, we should use the existing - // offset on the host if the file is seekable. - offset: 0, - } - - // Non-seekable files can't be memory mapped, assert this. - if !i.seekable && i.canMap { - panic("files that can return EWOULDBLOCK (sockets, pipes, etc.) cannot be memory mapped") - } - - d := &kernfs.Dentry{} - d.Init(i) - // i.open will take a reference on d. - defer d.DecRef() - - return i.open(d.VFSDentry(), mnt) -} - -// inode implements kernfs.Inode. -type inode struct { - kernfs.InodeNotDirectory - kernfs.InodeNotSymlink - - // When the reference count reaches zero, the host fd is closed. - refs.AtomicRefCount - - // hostFD contains the host fd that this file was originally created from, - // which must be available at time of restore. - // - // This field is initialized at creation time and is immutable. - hostFD int - - // seekable is false if the host fd points to a file representing a stream, - // e.g. a socket or a pipe. Such files are not seekable and can return - // EWOULDBLOCK for I/O operations. - // - // This field is initialized at creation time and is immutable. - seekable bool - - // isTTY is true if this file represents a TTY. - // - // This field is initialized at creation time and is immutable. - isTTY bool - - // canMap specifies whether we allow the file to be memory mapped. - // - // This field is initialized at creation time and is immutable. - canMap bool - - // ino is an inode number unique within this filesystem. - // - // This field is initialized at creation time and is immutable. - ino uint64 - - // modeMu protects mode. - modeMu sync.Mutex - - // mode is a cached version of the file mode on the host. Note that it may - // become out of date if the mode is changed on the host, e.g. with chmod. - // - // Generally, it is better to retrieve the mode from the host through an - // fstat syscall. We only use this value in inode.Mode(), which cannot - // return an error, if the syscall to host fails. - // - // FIXME(b/152294168): Plumb error into Inode.Mode() return value so we - // can get rid of this. - mode linux.FileMode - - // offsetMu protects offset. - offsetMu sync.Mutex - - // offset specifies the current file offset. - offset int64 -} - -// Note that these flags may become out of date, since they can be modified -// on the host, e.g. with fcntl. -func fileFlagsFromHostFD(fd int) (int, error) { - flags, err := unix.FcntlInt(uintptr(fd), syscall.F_GETFL, 0) - if err != nil { - log.Warningf("Failed to get file flags for donated FD %d: %v", fd, err) - return 0, err - } - // TODO(gvisor.dev/issue/1672): implement behavior corresponding to these allowed flags. - flags &= syscall.O_ACCMODE | syscall.O_DIRECT | syscall.O_NONBLOCK | syscall.O_DSYNC | syscall.O_SYNC | syscall.O_APPEND - return flags, nil -} - -// CheckPermissions implements kernfs.Inode. -func (i *inode) CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { - mode, uid, gid, err := i.getPermissions() - if err != nil { - return err - } - return vfs.GenericCheckPermissions(creds, ats, mode, uid, gid) -} - -// Mode implements kernfs.Inode. -func (i *inode) Mode() linux.FileMode { - mode, _, _, err := i.getPermissions() - if err != nil { - return i.mode - } - - return linux.FileMode(mode) -} - -func (i *inode) getPermissions() (linux.FileMode, auth.KUID, auth.KGID, error) { - // Retrieve metadata. - var s syscall.Stat_t - if err := syscall.Fstat(i.hostFD, &s); err != nil { - return 0, 0, 0, err - } - - // Update cached mode. - i.modeMu.Lock() - i.mode = linux.FileMode(s.Mode) - i.modeMu.Unlock() - return linux.FileMode(s.Mode), auth.KUID(s.Uid), auth.KGID(s.Gid), nil -} - -// Stat implements kernfs.Inode. -func (i *inode) Stat(_ *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { - if opts.Mask&linux.STATX__RESERVED != 0 { - return linux.Statx{}, syserror.EINVAL - } - if opts.Sync&linux.AT_STATX_SYNC_TYPE == linux.AT_STATX_SYNC_TYPE { - return linux.Statx{}, syserror.EINVAL - } - - // Limit our host call only to known flags. - mask := opts.Mask & linux.STATX_ALL - var s unix.Statx_t - err := unix.Statx(i.hostFD, "", int(unix.AT_EMPTY_PATH|opts.Sync), int(mask), &s) - // Fallback to fstat(2), if statx(2) is not supported on the host. - // - // TODO(b/151263641): Remove fallback. - if err == syserror.ENOSYS { - return i.fstat(opts) - } else if err != nil { - return linux.Statx{}, err - } - - ls := linux.Statx{Mask: mask} - // Unconditionally fill blksize, attributes, and device numbers, as indicated - // by /include/uapi/linux/stat.h. - // - // RdevMajor/RdevMinor are left as zero, so as not to expose host device - // numbers. - // - // TODO(gvisor.dev/issue/1672): Use kernfs-specific, internally defined - // device numbers. If we use the device number from the host, it may collide - // with another sentry-internal device number. We handle device/inode - // numbers without relying on the host to prevent collisions. - ls.Blksize = s.Blksize - ls.Attributes = s.Attributes - ls.AttributesMask = s.Attributes_mask - - if mask&linux.STATX_TYPE != 0 { - ls.Mode |= s.Mode & linux.S_IFMT - } - if mask&linux.STATX_MODE != 0 { - ls.Mode |= s.Mode &^ linux.S_IFMT - } - if mask&linux.STATX_NLINK != 0 { - ls.Nlink = s.Nlink - } - if mask&linux.STATX_UID != 0 { - ls.UID = s.Uid - } - if mask&linux.STATX_GID != 0 { - ls.GID = s.Gid - } - if mask&linux.STATX_ATIME != 0 { - ls.Atime = unixToLinuxStatxTimestamp(s.Atime) - } - if mask&linux.STATX_BTIME != 0 { - ls.Btime = unixToLinuxStatxTimestamp(s.Btime) - } - if mask&linux.STATX_CTIME != 0 { - ls.Ctime = unixToLinuxStatxTimestamp(s.Ctime) - } - if mask&linux.STATX_MTIME != 0 { - ls.Mtime = unixToLinuxStatxTimestamp(s.Mtime) - } - if mask&linux.STATX_SIZE != 0 { - ls.Size = s.Size - } - if mask&linux.STATX_BLOCKS != 0 { - ls.Blocks = s.Blocks - } - - // Use our own internal inode number. - if mask&linux.STATX_INO != 0 { - ls.Ino = i.ino - } - - // Update cached mode. - if (mask&linux.STATX_TYPE != 0) && (mask&linux.STATX_MODE != 0) { - i.modeMu.Lock() - i.mode = linux.FileMode(s.Mode) - i.modeMu.Unlock() - } - return ls, nil -} - -// fstat is a best-effort fallback for inode.Stat() if the host does not -// support statx(2). -// -// We ignore the mask and sync flags in opts and simply supply -// STATX_BASIC_STATS, as fstat(2) itself does not allow the specification -// of a mask or sync flags. fstat(2) does not provide any metadata -// equivalent to Statx.Attributes, Statx.AttributesMask, or Statx.Btime, so -// those fields remain empty. -func (i *inode) fstat(opts vfs.StatOptions) (linux.Statx, error) { - var s unix.Stat_t - if err := unix.Fstat(i.hostFD, &s); err != nil { - return linux.Statx{}, err - } - - // Note that rdev numbers are left as 0; do not expose host device numbers. - ls := linux.Statx{ - Mask: linux.STATX_BASIC_STATS, - Blksize: uint32(s.Blksize), - Nlink: uint32(s.Nlink), - UID: s.Uid, - GID: s.Gid, - Mode: uint16(s.Mode), - Size: uint64(s.Size), - Blocks: uint64(s.Blocks), - Atime: timespecToStatxTimestamp(s.Atim), - Ctime: timespecToStatxTimestamp(s.Ctim), - Mtime: timespecToStatxTimestamp(s.Mtim), - } - - // Use our own internal inode number. - // - // TODO(gvisor.dev/issue/1672): Use a kernfs-specific device number as well. - // If we use the device number from the host, it may collide with another - // sentry-internal device number. We handle device/inode numbers without - // relying on the host to prevent collisions. - ls.Ino = i.ino - - return ls, nil -} - -// SetStat implements kernfs.Inode. -func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { - s := opts.Stat - - m := s.Mask - if m == 0 { - return nil - } - if m&^(linux.STATX_MODE|linux.STATX_SIZE|linux.STATX_ATIME|linux.STATX_MTIME) != 0 { - return syserror.EPERM - } - mode, uid, gid, err := i.getPermissions() - if err != nil { - return err - } - if err := vfs.CheckSetStat(ctx, creds, &s, mode.Permissions(), uid, gid); err != nil { - return err - } - - if m&linux.STATX_MODE != 0 { - if err := syscall.Fchmod(i.hostFD, uint32(s.Mode)); err != nil { - return err - } - i.modeMu.Lock() - i.mode = linux.FileMode(s.Mode) - i.modeMu.Unlock() - } - if m&linux.STATX_SIZE != 0 { - if err := syscall.Ftruncate(i.hostFD, int64(s.Size)); err != nil { - return err - } - } - if m&(linux.STATX_ATIME|linux.STATX_MTIME) != 0 { - ts := [2]syscall.Timespec{ - toTimespec(s.Atime, m&linux.STATX_ATIME == 0), - toTimespec(s.Mtime, m&linux.STATX_MTIME == 0), - } - if err := setTimestamps(i.hostFD, &ts); err != nil { - return err - } - } - return nil -} - -// DecRef implements kernfs.Inode. -func (i *inode) DecRef() { - i.AtomicRefCount.DecRefWithDestructor(i.Destroy) -} - -// Destroy implements kernfs.Inode. -func (i *inode) Destroy() { - if err := unix.Close(i.hostFD); err != nil { - log.Warningf("failed to close host fd %d: %v", i.hostFD, err) - } -} - -// Open implements kernfs.Inode. -func (i *inode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - return i.open(vfsd, rp.Mount()) -} - -func (i *inode) open(d *vfs.Dentry, mnt *vfs.Mount) (*vfs.FileDescription, error) { - mode, _, _, err := i.getPermissions() - if err != nil { - return nil, err - } - fileType := mode.FileType() - if fileType == syscall.S_IFSOCK { - if i.isTTY { - return nil, errors.New("cannot use host socket as TTY") - } - // TODO(gvisor.dev/issue/1672): support importing sockets. - return nil, errors.New("importing host sockets not supported") - } - - // TODO(gvisor.dev/issue/1672): Whitelist specific file types here, so that - // we don't allow importing arbitrary file types without proper support. - var ( - vfsfd *vfs.FileDescription - fdImpl vfs.FileDescriptionImpl - ) - if i.isTTY { - fd := &ttyFD{ - fileDescription: fileDescription{inode: i}, - termios: linux.DefaultSlaveTermios, - } - vfsfd = &fd.vfsfd - fdImpl = fd - } else { - // For simplicity, set offset to 0. Technically, we should - // only set to 0 on files that are not seekable (sockets, pipes, etc.), - // and use the offset from the host fd otherwise. - fd := &fileDescription{inode: i} - vfsfd = &fd.vfsfd - fdImpl = fd - } - - flags, err := fileFlagsFromHostFD(i.hostFD) - if err != nil { - return nil, err - } - - if err := vfsfd.Init(fdImpl, uint32(flags), mnt, d, &vfs.FileDescriptionOptions{}); err != nil { - return nil, err - } - return vfsfd, nil -} - -// fileDescription is embedded by host fd implementations of FileDescriptionImpl. -// -// TODO(gvisor.dev/issue/1672): Implement Waitable interface. -type fileDescription struct { - vfsfd vfs.FileDescription - vfs.FileDescriptionDefaultImpl - - // inode is vfsfd.Dentry().Impl().(*kernfs.Dentry).Inode().(*inode), but - // cached to reduce indirections and casting. fileDescription does not hold - // a reference on the inode through the inode field (since one is already - // held via the Dentry). - // - // inode is immutable after fileDescription creation. - inode *inode -} - -// SetStat implements vfs.FileDescriptionImpl. -func (f *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { - creds := auth.CredentialsFromContext(ctx) - return f.inode.SetStat(ctx, nil, creds, opts) -} - -// Stat implements vfs.FileDescriptionImpl. -func (f *fileDescription) Stat(_ context.Context, opts vfs.StatOptions) (linux.Statx, error) { - return f.inode.Stat(nil, opts) -} - -// Release implements vfs.FileDescriptionImpl. -func (f *fileDescription) Release() { - // noop -} - -// PRead implements FileDescriptionImpl. -func (f *fileDescription) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - i := f.inode - if !i.seekable { - return 0, syserror.ESPIPE - } - - return readFromHostFD(ctx, i.hostFD, dst, offset, opts.Flags) -} - -// Read implements FileDescriptionImpl. -func (f *fileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - i := f.inode - if !i.seekable { - n, err := readFromHostFD(ctx, i.hostFD, dst, -1, opts.Flags) - if isBlockError(err) { - // If we got any data at all, return it as a "completed" partial read - // rather than retrying until complete. - if n != 0 { - err = nil - } else { - err = syserror.ErrWouldBlock - } - } - return n, err - } - // TODO(gvisor.dev/issue/1672): Cache pages, when forced to do so. - i.offsetMu.Lock() - n, err := readFromHostFD(ctx, i.hostFD, dst, i.offset, opts.Flags) - i.offset += n - i.offsetMu.Unlock() - return n, err -} - -func readFromHostFD(ctx context.Context, hostFD int, dst usermem.IOSequence, offset int64, flags uint32) (int64, error) { - // TODO(gvisor.dev/issue/1672): Support select preadv2 flags. - if flags != 0 { - return 0, syserror.EOPNOTSUPP - } - - var reader safemem.Reader - if offset == -1 { - reader = safemem.FromIOReader{fd.NewReadWriter(hostFD)} - } else { - reader = safemem.FromVecReaderFunc{ - func(srcs [][]byte) (int64, error) { - n, err := unix.Preadv(hostFD, srcs, offset) - return int64(n), err - }, - } - } - n, err := dst.CopyOutFrom(ctx, reader) - return int64(n), err -} - -// PWrite implements FileDescriptionImpl. -func (f *fileDescription) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - i := f.inode - if !i.seekable { - return 0, syserror.ESPIPE - } - - return writeToHostFD(ctx, i.hostFD, src, offset, opts.Flags) -} - -// Write implements FileDescriptionImpl. -func (f *fileDescription) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - i := f.inode - if !i.seekable { - n, err := writeToHostFD(ctx, i.hostFD, src, -1, opts.Flags) - if isBlockError(err) { - err = syserror.ErrWouldBlock - } - return n, err - } - // TODO(gvisor.dev/issue/1672): Cache pages, when forced to do so. - // TODO(gvisor.dev/issue/1672): Write to end of file and update offset if O_APPEND is set on this file. - i.offsetMu.Lock() - n, err := writeToHostFD(ctx, i.hostFD, src, i.offset, opts.Flags) - i.offset += n - i.offsetMu.Unlock() - return n, err -} - -func writeToHostFD(ctx context.Context, hostFD int, src usermem.IOSequence, offset int64, flags uint32) (int64, error) { - // TODO(gvisor.dev/issue/1672): Support select pwritev2 flags. - if flags != 0 { - return 0, syserror.EOPNOTSUPP - } - - var writer safemem.Writer - if offset == -1 { - writer = safemem.FromIOWriter{fd.NewReadWriter(hostFD)} - } else { - writer = safemem.FromVecWriterFunc{ - func(srcs [][]byte) (int64, error) { - n, err := unix.Pwritev(hostFD, srcs, offset) - return int64(n), err - }, - } - } - n, err := src.CopyInTo(ctx, writer) - return int64(n), err -} - -// Seek implements FileDescriptionImpl. -// -// Note that we do not support seeking on directories, since we do not even -// allow directory fds to be imported at all. -func (f *fileDescription) Seek(_ context.Context, offset int64, whence int32) (int64, error) { - i := f.inode - if !i.seekable { - return 0, syserror.ESPIPE - } - - i.offsetMu.Lock() - defer i.offsetMu.Unlock() - - switch whence { - case linux.SEEK_SET: - if offset < 0 { - return i.offset, syserror.EINVAL - } - i.offset = offset - - case linux.SEEK_CUR: - // Check for overflow. Note that underflow cannot occur, since i.offset >= 0. - if offset > math.MaxInt64-i.offset { - return i.offset, syserror.EOVERFLOW - } - if i.offset+offset < 0 { - return i.offset, syserror.EINVAL - } - i.offset += offset - - case linux.SEEK_END: - var s syscall.Stat_t - if err := syscall.Fstat(i.hostFD, &s); err != nil { - return i.offset, err - } - size := s.Size - - // Check for overflow. Note that underflow cannot occur, since size >= 0. - if offset > math.MaxInt64-size { - return i.offset, syserror.EOVERFLOW - } - if size+offset < 0 { - return i.offset, syserror.EINVAL - } - i.offset = size + offset - - case linux.SEEK_DATA, linux.SEEK_HOLE: - // Modifying the offset in the host file table should not matter, since - // this is the only place where we use it. - // - // For reading and writing, we always rely on our internal offset. - n, err := unix.Seek(i.hostFD, offset, int(whence)) - if err != nil { - return i.offset, err - } - i.offset = n - - default: - // Invalid whence. - return i.offset, syserror.EINVAL - } - - return i.offset, nil -} - -// Sync implements FileDescriptionImpl. -func (f *fileDescription) Sync(context.Context) error { - // TODO(gvisor.dev/issue/1672): Currently we do not support the SyncData optimization, so we always sync everything. - return unix.Fsync(f.inode.hostFD) -} - -// ConfigureMMap implements FileDescriptionImpl. -func (f *fileDescription) ConfigureMMap(_ context.Context, opts *memmap.MMapOpts) error { - if !f.inode.canMap { - return syserror.ENODEV - } - // TODO(gvisor.dev/issue/1672): Implement ConfigureMMap and Mappable interface. - return syserror.ENODEV -} diff --git a/pkg/sentry/fsimpl/host/ioctl_unsafe.go b/pkg/sentry/fsimpl/host/ioctl_unsafe.go deleted file mode 100644 index 0983bf7d8..000000000 --- a/pkg/sentry/fsimpl/host/ioctl_unsafe.go +++ /dev/null @@ -1,56 +0,0 @@ -// Copyright 2020 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" - "unsafe" - - "gvisor.dev/gvisor/pkg/abi/linux" -) - -func ioctlGetTermios(fd int) (*linux.Termios, error) { - var t linux.Termios - _, _, errno := syscall.Syscall(syscall.SYS_IOCTL, uintptr(fd), linux.TCGETS, uintptr(unsafe.Pointer(&t))) - if errno != 0 { - return nil, errno - } - return &t, nil -} - -func ioctlSetTermios(fd int, req uint64, t *linux.Termios) error { - _, _, errno := syscall.Syscall(syscall.SYS_IOCTL, uintptr(fd), uintptr(req), uintptr(unsafe.Pointer(t))) - if errno != 0 { - return errno - } - return nil -} - -func ioctlGetWinsize(fd int) (*linux.Winsize, error) { - var w linux.Winsize - _, _, errno := syscall.Syscall(syscall.SYS_IOCTL, uintptr(fd), linux.TIOCGWINSZ, uintptr(unsafe.Pointer(&w))) - if errno != 0 { - return nil, errno - } - return &w, nil -} - -func ioctlSetWinsize(fd int, w *linux.Winsize) error { - _, _, errno := syscall.Syscall(syscall.SYS_IOCTL, uintptr(fd), linux.TIOCSWINSZ, uintptr(unsafe.Pointer(w))) - if errno != 0 { - return errno - } - return nil -} diff --git a/pkg/sentry/fsimpl/host/tty.go b/pkg/sentry/fsimpl/host/tty.go deleted file mode 100644 index 8936afb06..000000000 --- a/pkg/sentry/fsimpl/host/tty.go +++ /dev/null @@ -1,379 +0,0 @@ -// Copyright 2020 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 ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/unimpl" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// ttyFD implements vfs.FileDescriptionImpl for a host file descriptor -// that wraps a TTY FD. -type ttyFD struct { - fileDescription - - // mu protects the fields below. - mu sync.Mutex `state:"nosave"` - - // session is the session attached to this ttyFD. - session *kernel.Session - - // fgProcessGroup is the foreground process group that is currently - // connected to this TTY. - fgProcessGroup *kernel.ProcessGroup - - // termios contains the terminal attributes for this TTY. - termios linux.KernelTermios -} - -// InitForegroundProcessGroup sets the foreground process group and session for -// the TTY. This should only be called once, after the foreground process group -// has been created, but before it has started running. -func (t *ttyFD) InitForegroundProcessGroup(pg *kernel.ProcessGroup) { - t.mu.Lock() - defer t.mu.Unlock() - if t.fgProcessGroup != nil { - panic("foreground process group is already set") - } - t.fgProcessGroup = pg - t.session = pg.Session() -} - -// ForegroundProcessGroup returns the foreground process for the TTY. -func (t *ttyFD) ForegroundProcessGroup() *kernel.ProcessGroup { - t.mu.Lock() - defer t.mu.Unlock() - return t.fgProcessGroup -} - -// Release implements fs.FileOperations.Release. -func (t *ttyFD) Release() { - t.mu.Lock() - t.fgProcessGroup = nil - t.mu.Unlock() - - t.fileDescription.Release() -} - -// PRead implements vfs.FileDescriptionImpl. -// -// Reading from a TTY is only allowed for foreground process groups. Background -// process groups will either get EIO or a SIGTTIN. -func (t *ttyFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - t.mu.Lock() - defer t.mu.Unlock() - - // Are we allowed to do the read? - // drivers/tty/n_tty.c:n_tty_read()=>job_control()=>tty_check_change(). - if err := t.checkChange(ctx, linux.SIGTTIN); err != nil { - return 0, err - } - - // Do the read. - return t.fileDescription.PRead(ctx, dst, offset, opts) -} - -// Read implements vfs.FileDescriptionImpl. -// -// Reading from a TTY is only allowed for foreground process groups. Background -// process groups will either get EIO or a SIGTTIN. -func (t *ttyFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - t.mu.Lock() - defer t.mu.Unlock() - - // Are we allowed to do the read? - // drivers/tty/n_tty.c:n_tty_read()=>job_control()=>tty_check_change(). - if err := t.checkChange(ctx, linux.SIGTTIN); err != nil { - return 0, err - } - - // Do the read. - return t.fileDescription.Read(ctx, dst, opts) -} - -// PWrite implements vfs.FileDescriptionImpl. -func (t *ttyFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - t.mu.Lock() - defer t.mu.Unlock() - - // Check whether TOSTOP is enabled. This corresponds to the check in - // drivers/tty/n_tty.c:n_tty_write(). - if t.termios.LEnabled(linux.TOSTOP) { - if err := t.checkChange(ctx, linux.SIGTTOU); err != nil { - return 0, err - } - } - return t.fileDescription.PWrite(ctx, src, offset, opts) -} - -// Write implements vfs.FileDescriptionImpl. -func (t *ttyFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - t.mu.Lock() - defer t.mu.Unlock() - - // Check whether TOSTOP is enabled. This corresponds to the check in - // drivers/tty/n_tty.c:n_tty_write(). - if t.termios.LEnabled(linux.TOSTOP) { - if err := t.checkChange(ctx, linux.SIGTTOU); err != nil { - return 0, err - } - } - return t.fileDescription.Write(ctx, src, opts) -} - -// Ioctl implements vfs.FileDescriptionImpl. -func (t *ttyFD) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { - // Ignore arg[0]. This is the real FD: - fd := t.inode.hostFD - ioctl := args[1].Uint64() - switch ioctl { - case linux.TCGETS: - termios, err := ioctlGetTermios(fd) - if err != nil { - return 0, err - } - _, err = usermem.CopyObjectOut(ctx, io, args[2].Pointer(), termios, usermem.IOOpts{ - AddressSpaceActive: true, - }) - return 0, err - - case linux.TCSETS, linux.TCSETSW, linux.TCSETSF: - t.mu.Lock() - defer t.mu.Unlock() - - if err := t.checkChange(ctx, linux.SIGTTOU); err != nil { - return 0, err - } - - var termios linux.Termios - if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &termios, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { - return 0, err - } - err := ioctlSetTermios(fd, ioctl, &termios) - if err == nil { - t.termios.FromTermios(termios) - } - return 0, err - - case linux.TIOCGPGRP: - // Args: pid_t *argp - // When successful, equivalent to *argp = tcgetpgrp(fd). - // Get the process group ID of the foreground process group on this - // terminal. - - pidns := kernel.PIDNamespaceFromContext(ctx) - if pidns == nil { - return 0, syserror.ENOTTY - } - - t.mu.Lock() - defer t.mu.Unlock() - - // Map the ProcessGroup into a ProcessGroupID in the task's PID namespace. - pgID := pidns.IDOfProcessGroup(t.fgProcessGroup) - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), &pgID, usermem.IOOpts{ - AddressSpaceActive: true, - }) - return 0, err - - case linux.TIOCSPGRP: - // Args: const pid_t *argp - // Equivalent to tcsetpgrp(fd, *argp). - // Set the foreground process group ID of this terminal. - - task := kernel.TaskFromContext(ctx) - if task == nil { - return 0, syserror.ENOTTY - } - - t.mu.Lock() - defer t.mu.Unlock() - - // Check that we are allowed to set the process group. - if err := t.checkChange(ctx, linux.SIGTTOU); err != nil { - // drivers/tty/tty_io.c:tiocspgrp() converts -EIO from tty_check_change() - // to -ENOTTY. - if err == syserror.EIO { - return 0, syserror.ENOTTY - } - return 0, err - } - - // Check that calling task's process group is in the TTY session. - if task.ThreadGroup().Session() != t.session { - return 0, syserror.ENOTTY - } - - var pgID kernel.ProcessGroupID - if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &pgID, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { - return 0, err - } - - // pgID must be non-negative. - if pgID < 0 { - return 0, syserror.EINVAL - } - - // Process group with pgID must exist in this PID namespace. - pidns := task.PIDNamespace() - pg := pidns.ProcessGroupWithID(pgID) - if pg == nil { - return 0, syserror.ESRCH - } - - // Check that new process group is in the TTY session. - if pg.Session() != t.session { - return 0, syserror.EPERM - } - - t.fgProcessGroup = pg - return 0, nil - - case linux.TIOCGWINSZ: - // Args: struct winsize *argp - // Get window size. - winsize, err := ioctlGetWinsize(fd) - if err != nil { - return 0, err - } - _, err = usermem.CopyObjectOut(ctx, io, args[2].Pointer(), winsize, usermem.IOOpts{ - AddressSpaceActive: true, - }) - return 0, err - - case linux.TIOCSWINSZ: - // Args: const struct winsize *argp - // Set window size. - - // Unlike setting the termios, any process group (even background ones) can - // set the winsize. - - var winsize linux.Winsize - if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &winsize, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { - return 0, err - } - err := ioctlSetWinsize(fd, &winsize) - return 0, err - - // Unimplemented commands. - case linux.TIOCSETD, - linux.TIOCSBRK, - linux.TIOCCBRK, - linux.TCSBRK, - linux.TCSBRKP, - linux.TIOCSTI, - linux.TIOCCONS, - linux.FIONBIO, - linux.TIOCEXCL, - linux.TIOCNXCL, - linux.TIOCGEXCL, - linux.TIOCNOTTY, - linux.TIOCSCTTY, - linux.TIOCGSID, - linux.TIOCGETD, - linux.TIOCVHANGUP, - linux.TIOCGDEV, - linux.TIOCMGET, - linux.TIOCMSET, - linux.TIOCMBIC, - linux.TIOCMBIS, - linux.TIOCGICOUNT, - linux.TCFLSH, - linux.TIOCSSERIAL, - linux.TIOCGPTPEER: - - unimpl.EmitUnimplementedEvent(ctx) - fallthrough - default: - return 0, syserror.ENOTTY - } -} - -// checkChange checks that the process group is allowed to read, write, or -// change the state of the TTY. -// -// This corresponds to Linux drivers/tty/tty_io.c:tty_check_change(). The logic -// is a bit convoluted, but documented inline. -// -// Preconditions: t.mu must be held. -func (t *ttyFD) checkChange(ctx context.Context, sig linux.Signal) error { - task := kernel.TaskFromContext(ctx) - if task == nil { - // No task? Linux does not have an analog for this case, but - // tty_check_change is more of a blacklist of cases than a - // whitelist, and is surprisingly permissive. Allowing the - // change seems most appropriate. - return nil - } - - tg := task.ThreadGroup() - pg := tg.ProcessGroup() - - // If the session for the task is different than the session for the - // controlling TTY, then the change is allowed. Seems like a bad idea, - // but that's exactly what linux does. - if tg.Session() != t.fgProcessGroup.Session() { - return nil - } - - // If we are the foreground process group, then the change is allowed. - if pg == t.fgProcessGroup { - return nil - } - - // We are not the foreground process group. - - // Is the provided signal blocked or ignored? - if (task.SignalMask()&linux.SignalSetOf(sig) != 0) || tg.SignalHandlers().IsIgnored(sig) { - // If the signal is SIGTTIN, then we are attempting to read - // from the TTY. Don't send the signal and return EIO. - if sig == linux.SIGTTIN { - return syserror.EIO - } - - // Otherwise, we are writing or changing terminal state. This is allowed. - return nil - } - - // If the process group is an orphan, return EIO. - if pg.IsOrphan() { - return syserror.EIO - } - - // Otherwise, send the signal to the process group and return ERESTARTSYS. - // - // Note that Linux also unconditionally sets TIF_SIGPENDING on current, - // but this isn't necessary in gVisor because the rationale given in - // 040b6362d58f "tty: fix leakage of -ERESTARTSYS to userland" doesn't - // apply: the sentry will handle -ERESTARTSYS in - // kernel.runApp.execute() even if the kernel.Task isn't interrupted. - // - // Linux ignores the result of kill_pgrp(). - _ = pg.SendSignal(kernel.SignalInfoPriv(sig)) - return kernel.ERESTARTSYS -} diff --git a/pkg/sentry/fsimpl/host/util.go b/pkg/sentry/fsimpl/host/util.go deleted file mode 100644 index 2bc757b1a..000000000 --- a/pkg/sentry/fsimpl/host/util.go +++ /dev/null @@ -1,66 +0,0 @@ -// Copyright 2020 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" - - "golang.org/x/sys/unix" - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/syserror" -) - -func toTimespec(ts linux.StatxTimestamp, omit bool) syscall.Timespec { - if omit { - return syscall.Timespec{ - Sec: 0, - Nsec: unix.UTIME_OMIT, - } - } - return syscall.Timespec{ - Sec: ts.Sec, - Nsec: int64(ts.Nsec), - } -} - -func unixToLinuxStatxTimestamp(ts unix.StatxTimestamp) linux.StatxTimestamp { - return linux.StatxTimestamp{Sec: ts.Sec, Nsec: ts.Nsec} -} - -func timespecToStatxTimestamp(ts unix.Timespec) linux.StatxTimestamp { - return linux.StatxTimestamp{Sec: int64(ts.Sec), Nsec: uint32(ts.Nsec)} -} - -// wouldBlock returns true for file types that can return EWOULDBLOCK -// for blocking operations, e.g. pipes, character devices, and sockets. -func wouldBlock(fileType uint32) bool { - return fileType == syscall.S_IFIFO || fileType == syscall.S_IFCHR || fileType == syscall.S_IFSOCK -} - -// canMap returns true if a file with fileType is allowed to be memory mapped. -// This is ported over from VFS1, but it's probably not the best way for us -// to check if a file can be memory mapped. -func canMap(fileType uint32) bool { - // TODO(gvisor.dev/issue/1672): Also allow "special files" to be mapped (see fs/host:canMap()). - // - // TODO(b/38213152): Some obscure character devices can be mapped. - return fileType == syscall.S_IFREG -} - -// isBlockError checks if an error is EAGAIN or EWOULDBLOCK. -// If so, they can be transformed into syserror.ErrWouldBlock. -func isBlockError(err error) bool { - return err == syserror.EAGAIN || err == syserror.EWOULDBLOCK -} diff --git a/pkg/sentry/fsimpl/host/util_unsafe.go b/pkg/sentry/fsimpl/host/util_unsafe.go deleted file mode 100644 index 5136ac844..000000000 --- a/pkg/sentry/fsimpl/host/util_unsafe.go +++ /dev/null @@ -1,34 +0,0 @@ -// Copyright 2020 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" - "unsafe" -) - -func setTimestamps(fd int, ts *[2]syscall.Timespec) error { - _, _, errno := syscall.Syscall6( - syscall.SYS_UTIMENSAT, - uintptr(fd), - 0, /* path */ - uintptr(unsafe.Pointer(ts)), - 0, /* flags */ - 0, 0) - if errno != 0 { - return errno - } - return nil -} diff --git a/pkg/sentry/fsimpl/kernfs/BUILD b/pkg/sentry/fsimpl/kernfs/BUILD deleted file mode 100644 index b3d6299d0..000000000 --- a/pkg/sentry/fsimpl/kernfs/BUILD +++ /dev/null @@ -1,62 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -licenses(["notice"]) - -go_template_instance( - name = "slot_list", - out = "slot_list.go", - package = "kernfs", - prefix = "slot", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*slot", - "Linker": "*slot", - }, -) - -go_library( - name = "kernfs", - srcs = [ - "dynamic_bytes_file.go", - "fd_impl_util.go", - "filesystem.go", - "inode_impl_util.go", - "kernfs.go", - "slot_list.go", - "symlink.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fspath", - "//pkg/log", - "//pkg/refs", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/memmap", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/vfs", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - ], -) - -go_test( - name = "kernfs_test", - size = "small", - srcs = ["kernfs_test.go"], - deps = [ - ":kernfs", - "//pkg/abi/linux", - "//pkg/context", - "//pkg/sentry/contexttest", - "//pkg/sentry/fsimpl/testutil", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - "//pkg/syserror", - "//pkg/usermem", - "@com_github_google_go-cmp//cmp:go_default_library", - ], -) diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index d8bddbafa..d8bddbafa 100644..100755 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index bfa786c88..bfa786c88 100644..100755 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 4433071aa..4433071aa 100644..100755 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 65f09af5d..65f09af5d 100644..100755 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index ad76b9f64..ad76b9f64 100644..100755 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_state_autogen.go b/pkg/sentry/fsimpl/kernfs/kernfs_state_autogen.go new file mode 100755 index 000000000..299ad87fa --- /dev/null +++ b/pkg/sentry/fsimpl/kernfs/kernfs_state_autogen.go @@ -0,0 +1,96 @@ +// automatically generated by stateify. + +package kernfs + +import ( + "gvisor.dev/gvisor/pkg/state" +) + +func (x *DynamicBytesFile) beforeSave() {} +func (x *DynamicBytesFile) save(m state.Map) { + x.beforeSave() + m.Save("InodeAttrs", &x.InodeAttrs) + m.Save("InodeNoopRefCount", &x.InodeNoopRefCount) + m.Save("InodeNotDirectory", &x.InodeNotDirectory) + m.Save("InodeNotSymlink", &x.InodeNotSymlink) + m.Save("data", &x.data) +} + +func (x *DynamicBytesFile) afterLoad() {} +func (x *DynamicBytesFile) load(m state.Map) { + m.Load("InodeAttrs", &x.InodeAttrs) + m.Load("InodeNoopRefCount", &x.InodeNoopRefCount) + m.Load("InodeNotDirectory", &x.InodeNotDirectory) + m.Load("InodeNotSymlink", &x.InodeNotSymlink) + m.Load("data", &x.data) +} + +func (x *DynamicBytesFD) beforeSave() {} +func (x *DynamicBytesFD) save(m state.Map) { + x.beforeSave() + m.Save("FileDescriptionDefaultImpl", &x.FileDescriptionDefaultImpl) + m.Save("DynamicBytesFileDescriptionImpl", &x.DynamicBytesFileDescriptionImpl) + m.Save("vfsfd", &x.vfsfd) + m.Save("inode", &x.inode) +} + +func (x *DynamicBytesFD) afterLoad() {} +func (x *DynamicBytesFD) load(m state.Map) { + m.Load("FileDescriptionDefaultImpl", &x.FileDescriptionDefaultImpl) + m.Load("DynamicBytesFileDescriptionImpl", &x.DynamicBytesFileDescriptionImpl) + m.Load("vfsfd", &x.vfsfd) + m.Load("inode", &x.inode) +} + +func (x *StaticDirectory) beforeSave() {} +func (x *StaticDirectory) save(m state.Map) { + x.beforeSave() + m.Save("InodeNotSymlink", &x.InodeNotSymlink) + m.Save("InodeDirectoryNoNewChildren", &x.InodeDirectoryNoNewChildren) + m.Save("InodeAttrs", &x.InodeAttrs) + m.Save("InodeNoDynamicLookup", &x.InodeNoDynamicLookup) + m.Save("OrderedChildren", &x.OrderedChildren) +} + +func (x *StaticDirectory) afterLoad() {} +func (x *StaticDirectory) load(m state.Map) { + m.Load("InodeNotSymlink", &x.InodeNotSymlink) + m.Load("InodeDirectoryNoNewChildren", &x.InodeDirectoryNoNewChildren) + m.Load("InodeAttrs", &x.InodeAttrs) + m.Load("InodeNoDynamicLookup", &x.InodeNoDynamicLookup) + m.Load("OrderedChildren", &x.OrderedChildren) +} + +func (x *slotList) beforeSave() {} +func (x *slotList) save(m state.Map) { + x.beforeSave() + m.Save("head", &x.head) + m.Save("tail", &x.tail) +} + +func (x *slotList) afterLoad() {} +func (x *slotList) load(m state.Map) { + m.Load("head", &x.head) + m.Load("tail", &x.tail) +} + +func (x *slotEntry) beforeSave() {} +func (x *slotEntry) save(m state.Map) { + x.beforeSave() + m.Save("next", &x.next) + m.Save("prev", &x.prev) +} + +func (x *slotEntry) afterLoad() {} +func (x *slotEntry) load(m state.Map) { + m.Load("next", &x.next) + m.Load("prev", &x.prev) +} + +func init() { + state.Register("pkg/sentry/fsimpl/kernfs.DynamicBytesFile", (*DynamicBytesFile)(nil), state.Fns{Save: (*DynamicBytesFile).save, Load: (*DynamicBytesFile).load}) + state.Register("pkg/sentry/fsimpl/kernfs.DynamicBytesFD", (*DynamicBytesFD)(nil), state.Fns{Save: (*DynamicBytesFD).save, Load: (*DynamicBytesFD).load}) + state.Register("pkg/sentry/fsimpl/kernfs.StaticDirectory", (*StaticDirectory)(nil), state.Fns{Save: (*StaticDirectory).save, Load: (*StaticDirectory).load}) + state.Register("pkg/sentry/fsimpl/kernfs.slotList", (*slotList)(nil), state.Fns{Save: (*slotList).save, Load: (*slotList).load}) + state.Register("pkg/sentry/fsimpl/kernfs.slotEntry", (*slotEntry)(nil), state.Fns{Save: (*slotEntry).save, Load: (*slotEntry).load}) +} diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go deleted file mode 100644 index 465451f35..000000000 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ /dev/null @@ -1,324 +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 kernfs_test - -import ( - "bytes" - "fmt" - "testing" - - "github.com/google/go-cmp/cmp" - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -const defaultMode linux.FileMode = 01777 -const staticFileContent = "This is sample content for a static test file." - -// RootDentryFn is a generator function for creating the root dentry of a test -// filesystem. See newTestSystem. -type RootDentryFn func(*auth.Credentials, *filesystem) *kernfs.Dentry - -// newTestSystem sets up a minimal environment for running a test, including an -// instance of a test filesystem. Tests can control the contents of the -// filesystem by providing an appropriate rootFn, which should return a -// pre-populated root dentry. -func newTestSystem(t *testing.T, rootFn RootDentryFn) *testutil.System { - ctx := contexttest.Context(t) - creds := auth.CredentialsFromContext(ctx) - v := &vfs.VirtualFilesystem{} - if err := v.Init(); err != nil { - t.Fatalf("VFS init: %v", err) - } - v.MustRegisterFilesystemType("testfs", &fsType{rootFn: rootFn}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mns, err := v.NewMountNamespace(ctx, creds, "", "testfs", &vfs.GetFilesystemOptions{}) - if err != nil { - t.Fatalf("Failed to create testfs root mount: %v", err) - } - return testutil.NewSystem(ctx, t, v, mns) -} - -type fsType struct { - rootFn RootDentryFn -} - -type filesystem struct { - kernfs.Filesystem -} - -type file struct { - kernfs.DynamicBytesFile - content string -} - -func (fs *filesystem) newFile(creds *auth.Credentials, content string) *kernfs.Dentry { - f := &file{} - f.content = content - f.DynamicBytesFile.Init(creds, fs.NextIno(), f, 0777) - - d := &kernfs.Dentry{} - d.Init(f) - return d -} - -func (f *file) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "%s", f.content) - return nil -} - -type attrs struct { - kernfs.InodeAttrs -} - -func (*attrs) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { - return syserror.EPERM -} - -type readonlyDir struct { - attrs - kernfs.InodeNotSymlink - kernfs.InodeNoDynamicLookup - kernfs.InodeDirectoryNoNewChildren - - kernfs.OrderedChildren - dentry kernfs.Dentry -} - -func (fs *filesystem) newReadonlyDir(creds *auth.Credentials, mode linux.FileMode, contents map[string]*kernfs.Dentry) *kernfs.Dentry { - dir := &readonlyDir{} - dir.attrs.Init(creds, fs.NextIno(), linux.ModeDirectory|mode) - dir.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) - dir.dentry.Init(dir) - - dir.IncLinks(dir.OrderedChildren.Populate(&dir.dentry, contents)) - - return &dir.dentry -} - -func (d *readonlyDir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - if err := fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, &opts); err != nil { - return nil, err - } - return fd.VFSFileDescription(), nil -} - -type dir struct { - attrs - kernfs.InodeNotSymlink - kernfs.InodeNoDynamicLookup - - fs *filesystem - dentry kernfs.Dentry - kernfs.OrderedChildren -} - -func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, contents map[string]*kernfs.Dentry) *kernfs.Dentry { - dir := &dir{} - dir.fs = fs - dir.attrs.Init(creds, fs.NextIno(), linux.ModeDirectory|mode) - dir.OrderedChildren.Init(kernfs.OrderedChildrenOptions{Writable: true}) - dir.dentry.Init(dir) - - dir.IncLinks(dir.OrderedChildren.Populate(&dir.dentry, contents)) - - return &dir.dentry -} - -func (d *dir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, &opts) - return fd.VFSFileDescription(), nil -} - -func (d *dir) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*vfs.Dentry, error) { - creds := auth.CredentialsFromContext(ctx) - dir := d.fs.newDir(creds, opts.Mode, nil) - dirVFSD := dir.VFSDentry() - if err := d.OrderedChildren.Insert(name, dirVFSD); err != nil { - dir.DecRef() - return nil, err - } - d.IncLinks(1) - return dirVFSD, nil -} - -func (d *dir) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*vfs.Dentry, error) { - creds := auth.CredentialsFromContext(ctx) - f := d.fs.newFile(creds, "") - fVFSD := f.VFSDentry() - if err := d.OrderedChildren.Insert(name, fVFSD); err != nil { - f.DecRef() - return nil, err - } - return fVFSD, nil -} - -func (*dir) NewLink(context.Context, string, kernfs.Inode) (*vfs.Dentry, error) { - return nil, syserror.EPERM -} - -func (*dir) NewSymlink(context.Context, string, string) (*vfs.Dentry, error) { - return nil, syserror.EPERM -} - -func (*dir) NewNode(context.Context, string, vfs.MknodOptions) (*vfs.Dentry, error) { - return nil, syserror.EPERM -} - -func (fsType) Name() string { - return "kernfs" -} - -func (fst fsType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opt vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - fs := &filesystem{} - fs.Init(vfsObj, &fst) - root := fst.rootFn(creds, fs) - return fs.VFSFilesystem(), root.VFSDentry(), nil -} - -// -------------------- Remainder of the file are test cases -------------------- - -func TestBasic(t *testing.T) { - sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { - return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ - "file1": fs.newFile(creds, staticFileContent), - }) - }) - defer sys.Destroy() - sys.GetDentryOrDie(sys.PathOpAtRoot("file1")).DecRef() -} - -func TestMkdirGetDentry(t *testing.T) { - sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { - return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ - "dir1": fs.newDir(creds, 0755, nil), - }) - }) - defer sys.Destroy() - - pop := sys.PathOpAtRoot("dir1/a new directory") - if err := sys.VFS.MkdirAt(sys.Ctx, sys.Creds, pop, &vfs.MkdirOptions{Mode: 0755}); err != nil { - t.Fatalf("MkdirAt for PathOperation %+v failed: %v", pop, err) - } - sys.GetDentryOrDie(pop).DecRef() -} - -func TestReadStaticFile(t *testing.T) { - sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { - return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ - "file1": fs.newFile(creds, staticFileContent), - }) - }) - defer sys.Destroy() - - pop := sys.PathOpAtRoot("file1") - fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, pop, &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - }) - if err != nil { - t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) - } - defer fd.DecRef() - - content, err := sys.ReadToEnd(fd) - if err != nil { - t.Fatalf("Read failed: %v", err) - } - if diff := cmp.Diff(staticFileContent, content); diff != "" { - t.Fatalf("Read returned unexpected data:\n--- want\n+++ got\n%v", diff) - } -} - -func TestCreateNewFileInStaticDir(t *testing.T) { - sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { - return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ - "dir1": fs.newDir(creds, 0755, nil), - }) - }) - defer sys.Destroy() - - pop := sys.PathOpAtRoot("dir1/newfile") - opts := &vfs.OpenOptions{Flags: linux.O_CREAT | linux.O_EXCL, Mode: defaultMode} - fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, pop, opts) - if err != nil { - t.Fatalf("OpenAt(pop:%+v, opts:%+v) failed: %v", pop, opts, err) - } - - // Close the file. The file should persist. - fd.DecRef() - - fd, err = sys.VFS.OpenAt(sys.Ctx, sys.Creds, pop, &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - }) - if err != nil { - t.Fatalf("OpenAt(pop:%+v) = %+v failed: %v", pop, fd, err) - } - fd.DecRef() -} - -func TestDirFDReadWrite(t *testing.T) { - sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { - return fs.newReadonlyDir(creds, 0755, nil) - }) - defer sys.Destroy() - - pop := sys.PathOpAtRoot("/") - fd, err := sys.VFS.OpenAt(sys.Ctx, sys.Creds, pop, &vfs.OpenOptions{ - Flags: linux.O_RDONLY, - }) - if err != nil { - t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) - } - defer fd.DecRef() - - // Read/Write should fail for directory FDs. - if _, err := fd.Read(sys.Ctx, usermem.BytesIOSequence([]byte{}), vfs.ReadOptions{}); err != syserror.EISDIR { - t.Fatalf("Read for directory FD failed with unexpected error: %v", err) - } - if _, err := fd.Write(sys.Ctx, usermem.BytesIOSequence([]byte{}), vfs.WriteOptions{}); err != syserror.EBADF { - t.Fatalf("Write for directory FD failed with unexpected error: %v", err) - } -} - -func TestDirFDIterDirents(t *testing.T) { - sys := newTestSystem(t, func(creds *auth.Credentials, fs *filesystem) *kernfs.Dentry { - return fs.newReadonlyDir(creds, 0755, map[string]*kernfs.Dentry{ - // Fill root with nodes backed by various inode implementations. - "dir1": fs.newReadonlyDir(creds, 0755, nil), - "dir2": fs.newDir(creds, 0755, map[string]*kernfs.Dentry{ - "dir3": fs.newDir(creds, 0755, nil), - }), - "file1": fs.newFile(creds, staticFileContent), - }) - }) - defer sys.Destroy() - - pop := sys.PathOpAtRoot("/") - sys.AssertAllDirentTypes(sys.ListDirents(pop), map[string]testutil.DirentType{ - "dir1": linux.DT_DIR, - "dir2": linux.DT_DIR, - "file1": linux.DT_REG, - }) -} diff --git a/pkg/sentry/fsimpl/kernfs/slot_list.go b/pkg/sentry/fsimpl/kernfs/slot_list.go new file mode 100755 index 000000000..09c30bca7 --- /dev/null +++ b/pkg/sentry/fsimpl/kernfs/slot_list.go @@ -0,0 +1,193 @@ +package kernfs + +// 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 slotElementMapper struct{} + +// linkerFor maps an Element to a Linker. +// +// This default implementation should be inlined. +// +//go:nosplit +func (slotElementMapper) linkerFor(elem *slot) *slot { 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 slotList struct { + head *slot + tail *slot +} + +// Reset resets list l to the empty state. +func (l *slotList) Reset() { + l.head = nil + l.tail = nil +} + +// Empty returns true iff the list is empty. +func (l *slotList) Empty() bool { + return l.head == nil +} + +// Front returns the first element of list l or nil. +func (l *slotList) Front() *slot { + return l.head +} + +// Back returns the last element of list l or nil. +func (l *slotList) Back() *slot { + return l.tail +} + +// Len returns the number of elements in the list. +// +// NOTE: This is an O(n) operation. +func (l *slotList) Len() (count int) { + for e := l.Front(); e != nil; e = e.Next() { + count++ + } + return count +} + +// PushFront inserts the element e at the front of list l. +func (l *slotList) PushFront(e *slot) { + linker := slotElementMapper{}.linkerFor(e) + linker.SetNext(l.head) + linker.SetPrev(nil) + if l.head != nil { + slotElementMapper{}.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 *slotList) PushBack(e *slot) { + linker := slotElementMapper{}.linkerFor(e) + linker.SetNext(nil) + linker.SetPrev(l.tail) + if l.tail != nil { + slotElementMapper{}.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 *slotList) PushBackList(m *slotList) { + if l.head == nil { + l.head = m.head + l.tail = m.tail + } else if m.head != nil { + slotElementMapper{}.linkerFor(l.tail).SetNext(m.head) + slotElementMapper{}.linkerFor(m.head).SetPrev(l.tail) + + l.tail = m.tail + } + m.head = nil + m.tail = nil +} + +// InsertAfter inserts e after b. +func (l *slotList) InsertAfter(b, e *slot) { + bLinker := slotElementMapper{}.linkerFor(b) + eLinker := slotElementMapper{}.linkerFor(e) + + a := bLinker.Next() + + eLinker.SetNext(a) + eLinker.SetPrev(b) + bLinker.SetNext(e) + + if a != nil { + slotElementMapper{}.linkerFor(a).SetPrev(e) + } else { + l.tail = e + } +} + +// InsertBefore inserts e before a. +func (l *slotList) InsertBefore(a, e *slot) { + aLinker := slotElementMapper{}.linkerFor(a) + eLinker := slotElementMapper{}.linkerFor(e) + + b := aLinker.Prev() + eLinker.SetNext(a) + eLinker.SetPrev(b) + aLinker.SetPrev(e) + + if b != nil { + slotElementMapper{}.linkerFor(b).SetNext(e) + } else { + l.head = e + } +} + +// Remove removes e from l. +func (l *slotList) Remove(e *slot) { + linker := slotElementMapper{}.linkerFor(e) + prev := linker.Prev() + next := linker.Next() + + if prev != nil { + slotElementMapper{}.linkerFor(prev).SetNext(next) + } else { + l.head = next + } + + if next != nil { + slotElementMapper{}.linkerFor(next).SetPrev(prev) + } else { + l.tail = prev + } + + linker.SetNext(nil) + linker.SetPrev(nil) +} + +// 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 slotEntry struct { + next *slot + prev *slot +} + +// Next returns the entry that follows e in the list. +func (e *slotEntry) Next() *slot { + return e.next +} + +// Prev returns the entry that precedes e in the list. +func (e *slotEntry) Prev() *slot { + return e.prev +} + +// SetNext assigns 'entry' as the entry that follows e in the list. +func (e *slotEntry) SetNext(elem *slot) { + e.next = elem +} + +// SetPrev assigns 'entry' as the entry that precedes e in the list. +func (e *slotEntry) SetPrev(elem *slot) { + e.prev = elem +} diff --git a/pkg/sentry/fsimpl/kernfs/symlink.go b/pkg/sentry/fsimpl/kernfs/symlink.go index 018aa503c..018aa503c 100644..100755 --- a/pkg/sentry/fsimpl/kernfs/symlink.go +++ b/pkg/sentry/fsimpl/kernfs/symlink.go diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD deleted file mode 100644 index 17c1342b5..000000000 --- a/pkg/sentry/fsimpl/proc/BUILD +++ /dev/null @@ -1,66 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -licenses(["notice"]) - -go_library( - name = "proc", - srcs = [ - "filesystem.go", - "subtasks.go", - "task.go", - "task_fds.go", - "task_files.go", - "task_net.go", - "tasks.go", - "tasks_files.go", - "tasks_sys.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/log", - "//pkg/refs", - "//pkg/safemem", - "//pkg/sentry/fsbridge", - "//pkg/sentry/fsimpl/kernfs", - "//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/unix", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usage", - "//pkg/sentry/vfs", - "//pkg/syserror", - "//pkg/tcpip/header", - "//pkg/usermem", - ], -) - -go_test( - name = "proc_test", - size = "small", - srcs = [ - "tasks_sys_test.go", - "tasks_test.go", - ], - library = ":proc", - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fspath", - "//pkg/sentry/contexttest", - "//pkg/sentry/fsimpl/testutil", - "//pkg/sentry/fsimpl/tmpfs", - "//pkg/sentry/inet", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - "//pkg/syserror", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/fsimpl/proc/filesystem.go b/pkg/sentry/fsimpl/proc/filesystem.go deleted file mode 100644 index 104fc9030..000000000 --- a/pkg/sentry/fsimpl/proc/filesystem.go +++ /dev/null @@ -1,100 +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 proc implements a partial in-memory file system for procfs. -package proc - -import ( - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -// Name is the default filesystem name. -const Name = "proc" - -// FilesystemType is the factory class for procfs. -// -// +stateify savable -type FilesystemType struct{} - -var _ vfs.FilesystemType = (*FilesystemType)(nil) - -// Name implements vfs.FilesystemType.Name. -func (FilesystemType) Name() string { - return Name -} - -// GetFilesystem implements vfs.FilesystemType.GetFilesystem. -func (ft FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - k := kernel.KernelFromContext(ctx) - if k == nil { - return nil, nil, fmt.Errorf("procfs requires a kernel") - } - pidns := kernel.PIDNamespaceFromContext(ctx) - if pidns == nil { - return nil, nil, fmt.Errorf("procfs requires a PID namespace") - } - - procfs := &kernfs.Filesystem{} - procfs.VFSFilesystem().Init(vfsObj, &ft, procfs) - - var cgroups map[string]string - if opts.InternalData != nil { - data := opts.InternalData.(*InternalData) - cgroups = data.Cgroups - } - - _, dentry := newTasksInode(procfs, k, pidns, cgroups) - return procfs.VFSFilesystem(), dentry.VFSDentry(), nil -} - -// dynamicInode is an overfitted interface for common Inodes with -// dynamicByteSource types used in procfs. -type dynamicInode interface { - kernfs.Inode - vfs.DynamicBytesSource - - Init(creds *auth.Credentials, ino uint64, data vfs.DynamicBytesSource, perm linux.FileMode) -} - -func newDentry(creds *auth.Credentials, ino uint64, perm linux.FileMode, inode dynamicInode) *kernfs.Dentry { - inode.Init(creds, ino, inode, perm) - - d := &kernfs.Dentry{} - d.Init(inode) - return d -} - -type staticFile struct { - kernfs.DynamicBytesFile - vfs.StaticData -} - -var _ dynamicInode = (*staticFile)(nil) - -func newStaticFile(data string) *staticFile { - return &staticFile{StaticData: vfs.StaticData{Data: data}} -} - -// InternalData contains internal data passed in to the procfs mount via -// vfs.GetFilesystemOptions.InternalData. -type InternalData struct { - Cgroups map[string]string -} diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go deleted file mode 100644 index a21313666..000000000 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ /dev/null @@ -1,135 +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 proc - -import ( - "sort" - "strconv" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// subtasksInode represents the inode for /proc/[pid]/task/ directory. -// -// +stateify savable -type subtasksInode struct { - kernfs.InodeNotSymlink - kernfs.InodeDirectoryNoNewChildren - kernfs.InodeAttrs - kernfs.OrderedChildren - kernfs.AlwaysValid - - task *kernel.Task - pidns *kernel.PIDNamespace - inoGen InoGenerator - cgroupControllers map[string]string -} - -var _ kernfs.Inode = (*subtasksInode)(nil) - -func newSubtasks(task *kernel.Task, pidns *kernel.PIDNamespace, inoGen InoGenerator, cgroupControllers map[string]string) *kernfs.Dentry { - subInode := &subtasksInode{ - task: task, - pidns: pidns, - inoGen: inoGen, - cgroupControllers: cgroupControllers, - } - // Note: credentials are overridden by taskOwnedInode. - subInode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) - subInode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) - - inode := &taskOwnedInode{Inode: subInode, owner: task} - dentry := &kernfs.Dentry{} - dentry.Init(inode) - - return dentry -} - -// Lookup implements kernfs.inodeDynamicLookup. -func (i *subtasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { - tid, err := strconv.ParseUint(name, 10, 32) - if err != nil { - return nil, syserror.ENOENT - } - - subTask := i.pidns.TaskWithID(kernel.ThreadID(tid)) - if subTask == nil { - return nil, syserror.ENOENT - } - if subTask.ThreadGroup() != i.task.ThreadGroup() { - return nil, syserror.ENOENT - } - - subTaskDentry := newTaskInode(i.inoGen, subTask, i.pidns, false, i.cgroupControllers) - return subTaskDentry.VFSDentry(), nil -} - -// IterDirents implements kernfs.inodeDynamicLookup. -func (i *subtasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) { - tasks := i.task.ThreadGroup().MemberIDs(i.pidns) - if len(tasks) == 0 { - return offset, syserror.ENOENT - } - - tids := make([]int, 0, len(tasks)) - for _, tid := range tasks { - tids = append(tids, int(tid)) - } - - sort.Ints(tids) - for _, tid := range tids[relOffset:] { - dirent := vfs.Dirent{ - Name: strconv.FormatUint(uint64(tid), 10), - Type: linux.DT_DIR, - Ino: i.inoGen.NextIno(), - NextOff: offset + 1, - } - if err := cb.Handle(dirent); err != nil { - return offset, err - } - offset++ - } - return offset, nil -} - -// Open implements kernfs.Inode. -func (i *subtasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) - return fd.VFSFileDescription(), nil -} - -// Stat implements kernfs.Inode. -func (i *subtasksInode) Stat(vsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { - stat, err := i.InodeAttrs.Stat(vsfs, opts) - if err != nil { - return linux.Statx{}, err - } - if opts.Mask&linux.STATX_NLINK != 0 { - stat.Nlink += uint32(i.task.ThreadGroup().Count()) - } - return stat, nil -} - -// SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*subtasksInode) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { - return syserror.EPERM -} diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go deleted file mode 100644 index 888afc0fd..000000000 --- a/pkg/sentry/fsimpl/proc/task.go +++ /dev/null @@ -1,234 +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 proc - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/mm" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// taskInode represents the inode for /proc/PID/ directory. -// -// +stateify savable -type taskInode struct { - kernfs.InodeNotSymlink - kernfs.InodeDirectoryNoNewChildren - kernfs.InodeNoDynamicLookup - kernfs.InodeAttrs - kernfs.OrderedChildren - - task *kernel.Task -} - -var _ kernfs.Inode = (*taskInode)(nil) - -func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNamespace, isThreadGroup bool, cgroupControllers map[string]string) *kernfs.Dentry { - contents := map[string]*kernfs.Dentry{ - "auxv": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &auxvData{task: task}), - "cmdline": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: cmdlineDataArg}), - "comm": newComm(task, inoGen.NextIno(), 0444), - "environ": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: environDataArg}), - "exe": newExeSymlink(task, inoGen.NextIno()), - "fd": newFDDirInode(task, inoGen), - "fdinfo": newFDInfoDirInode(task, inoGen), - "gid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: true}), - "io": newTaskOwnedFile(task, inoGen.NextIno(), 0400, newIO(task, isThreadGroup)), - "maps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mapsData{task: task}), - "mountinfo": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mountInfoData{task: task}), - "mounts": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mountsData{task: task}), - "net": newTaskNetDir(task, inoGen), - "ns": newTaskOwnedDir(task, inoGen.NextIno(), 0511, map[string]*kernfs.Dentry{ - "net": newNamespaceSymlink(task, inoGen.NextIno(), "net"), - "pid": newNamespaceSymlink(task, inoGen.NextIno(), "pid"), - "user": newNamespaceSymlink(task, inoGen.NextIno(), "user"), - }), - "oom_score": newTaskOwnedFile(task, inoGen.NextIno(), 0444, newStaticFile("0\n")), - "oom_score_adj": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &oomScoreAdj{task: task}), - "smaps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &smapsData{task: task}), - "stat": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &taskStatData{task: task, pidns: pidns, tgstats: isThreadGroup}), - "statm": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statmData{task: task}), - "status": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statusData{task: task, pidns: pidns}), - "uid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: false}), - } - if isThreadGroup { - contents["task"] = newSubtasks(task, pidns, inoGen, cgroupControllers) - } - if len(cgroupControllers) > 0 { - contents["cgroup"] = newTaskOwnedFile(task, inoGen.NextIno(), 0444, newCgroupData(cgroupControllers)) - } - - taskInode := &taskInode{task: task} - // Note: credentials are overridden by taskOwnedInode. - taskInode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) - - inode := &taskOwnedInode{Inode: taskInode, owner: task} - dentry := &kernfs.Dentry{} - dentry.Init(inode) - - taskInode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) - links := taskInode.OrderedChildren.Populate(dentry, contents) - taskInode.IncLinks(links) - - return dentry -} - -// Valid implements kernfs.inodeDynamicLookup. This inode remains valid as long -// as the task is still running. When it's dead, another tasks with the same -// PID could replace it. -func (i *taskInode) Valid(ctx context.Context) bool { - return i.task.ExitState() != kernel.TaskExitDead -} - -// Open implements kernfs.Inode. -func (i *taskInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) - return fd.VFSFileDescription(), nil -} - -// SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*taskInode) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { - return syserror.EPERM -} - -// taskOwnedInode implements kernfs.Inode and overrides inode owner with task -// effective user and group. -type taskOwnedInode struct { - kernfs.Inode - - // owner is the task that owns this inode. - owner *kernel.Task -} - -var _ kernfs.Inode = (*taskOwnedInode)(nil) - -func newTaskOwnedFile(task *kernel.Task, ino uint64, perm linux.FileMode, inode dynamicInode) *kernfs.Dentry { - // Note: credentials are overridden by taskOwnedInode. - inode.Init(task.Credentials(), ino, inode, perm) - - taskInode := &taskOwnedInode{Inode: inode, owner: task} - d := &kernfs.Dentry{} - d.Init(taskInode) - return d -} - -func newTaskOwnedDir(task *kernel.Task, ino uint64, perm linux.FileMode, children map[string]*kernfs.Dentry) *kernfs.Dentry { - dir := &kernfs.StaticDirectory{} - - // Note: credentials are overridden by taskOwnedInode. - dir.Init(task.Credentials(), ino, perm) - - inode := &taskOwnedInode{Inode: dir, owner: task} - d := &kernfs.Dentry{} - d.Init(inode) - - dir.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) - links := dir.OrderedChildren.Populate(d, children) - dir.IncLinks(links) - - return d -} - -// Stat implements kernfs.Inode. -func (i *taskOwnedInode) Stat(fs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { - stat, err := i.Inode.Stat(fs, opts) - if err != nil { - return linux.Statx{}, err - } - if opts.Mask&(linux.STATX_UID|linux.STATX_GID) != 0 { - uid, gid := i.getOwner(linux.FileMode(stat.Mode)) - if opts.Mask&linux.STATX_UID != 0 { - stat.UID = uint32(uid) - } - if opts.Mask&linux.STATX_GID != 0 { - stat.GID = uint32(gid) - } - } - return stat, nil -} - -// CheckPermissions implements kernfs.Inode. -func (i *taskOwnedInode) CheckPermissions(_ context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { - mode := i.Mode() - uid, gid := i.getOwner(mode) - return vfs.GenericCheckPermissions(creds, ats, mode, uid, gid) -} - -func (i *taskOwnedInode) getOwner(mode linux.FileMode) (auth.KUID, auth.KGID) { - // By default, set the task owner as the file owner. - creds := i.owner.Credentials() - uid := creds.EffectiveKUID - gid := creds.EffectiveKGID - - // Linux doesn't apply dumpability adjustments to world readable/executable - // directories so that applications can stat /proc/PID to determine the - // effective UID of a process. See fs/proc/base.c:task_dump_owner. - if mode.FileType() == linux.ModeDirectory && mode.Permissions() == 0555 { - return uid, gid - } - - // If the task is not dumpable, then root (in the namespace preferred) - // owns the file. - m := getMM(i.owner) - if m == nil { - return auth.RootKUID, auth.RootKGID - } - if m.Dumpability() != mm.UserDumpable { - uid = auth.RootKUID - if kuid := creds.UserNamespace.MapToKUID(auth.RootUID); kuid.Ok() { - uid = kuid - } - gid = auth.RootKGID - if kgid := creds.UserNamespace.MapToKGID(auth.RootGID); kgid.Ok() { - gid = kgid - } - } - return uid, gid -} - -func newIO(t *kernel.Task, isThreadGroup bool) *ioData { - if isThreadGroup { - return &ioData{ioUsage: t.ThreadGroup()} - } - return &ioData{ioUsage: t} -} - -// newCgroupData creates inode that shows cgroup information. -// From man 7 cgroups: "For each cgroup hierarchy of which the process is a -// member, there is one entry containing three colon-separated fields: -// hierarchy-ID:controller-list:cgroup-path" -func newCgroupData(controllers map[string]string) dynamicInode { - var buf bytes.Buffer - - // The hierarchy ids must be positive integers (for cgroup v1), but the - // exact number does not matter, so long as they are unique. We can - // just use a counter, but since linux sorts this file in descending - // order, we must count down to preserve this behavior. - i := len(controllers) - for name, dir := range controllers { - fmt.Fprintf(&buf, "%d:%s:%s\n", i, name, dir) - i-- - } - return newStaticFile(buf.String()) -} diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go deleted file mode 100644 index 046265eca..000000000 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ /dev/null @@ -1,302 +0,0 @@ -// Copyright 2020 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 ( - "bytes" - "fmt" - "sort" - "strconv" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -func getTaskFD(t *kernel.Task, fd int32) (*vfs.FileDescription, kernel.FDFlags) { - var ( - file *vfs.FileDescription - flags kernel.FDFlags - ) - t.WithMuLocked(func(t *kernel.Task) { - if fdt := t.FDTable(); fdt != nil { - file, flags = fdt.GetVFS2(fd) - } - }) - return file, flags -} - -func taskFDExists(t *kernel.Task, fd int32) bool { - file, _ := getTaskFD(t, fd) - if file == nil { - return false - } - file.DecRef() - return true -} - -type fdDir struct { - inoGen InoGenerator - task *kernel.Task - - // When produceSymlinks is set, dirents produces for the FDs are reported - // as symlink. Otherwise, they are reported as regular files. - produceSymlink bool -} - -// IterDirents implements kernfs.inodeDynamicLookup. -func (i *fdDir) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, absOffset, relOffset int64) (int64, error) { - var fds []int32 - i.task.WithMuLocked(func(t *kernel.Task) { - if fdTable := t.FDTable(); fdTable != nil { - fds = fdTable.GetFDs() - } - }) - - offset := absOffset + relOffset - typ := uint8(linux.DT_REG) - if i.produceSymlink { - typ = linux.DT_LNK - } - - // Find the appropriate starting point. - idx := sort.Search(len(fds), func(i int) bool { return fds[i] >= int32(relOffset) }) - if idx >= len(fds) { - return offset, nil - } - for _, fd := range fds[idx:] { - dirent := vfs.Dirent{ - Name: strconv.FormatUint(uint64(fd), 10), - Type: typ, - Ino: i.inoGen.NextIno(), - NextOff: offset + 1, - } - if err := cb.Handle(dirent); err != nil { - return offset, err - } - offset++ - } - return offset, nil -} - -// fdDirInode represents the inode for /proc/[pid]/fd directory. -// -// +stateify savable -type fdDirInode struct { - kernfs.InodeNotSymlink - kernfs.InodeDirectoryNoNewChildren - kernfs.InodeAttrs - kernfs.OrderedChildren - kernfs.AlwaysValid - fdDir -} - -var _ kernfs.Inode = (*fdDirInode)(nil) - -func newFDDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { - inode := &fdDirInode{ - fdDir: fdDir{ - inoGen: inoGen, - task: task, - produceSymlink: true, - }, - } - inode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) - - dentry := &kernfs.Dentry{} - dentry.Init(inode) - inode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) - - return dentry -} - -// Lookup implements kernfs.inodeDynamicLookup. -func (i *fdDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { - fdInt, err := strconv.ParseInt(name, 10, 32) - if err != nil { - return nil, syserror.ENOENT - } - fd := int32(fdInt) - if !taskFDExists(i.task, fd) { - return nil, syserror.ENOENT - } - taskDentry := newFDSymlink(i.task, fd, i.inoGen.NextIno()) - return taskDentry.VFSDentry(), nil -} - -// Open implements kernfs.Inode. -func (i *fdDirInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) - return fd.VFSFileDescription(), nil -} - -// CheckPermissions implements kernfs.Inode. -// -// This is to match Linux, which uses a special permission handler to guarantee -// that a process can still access /proc/self/fd after it has executed -// setuid. See fs/proc/fd.c:proc_fd_permission. -func (i *fdDirInode) CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { - err := i.InodeAttrs.CheckPermissions(ctx, creds, ats) - if err == nil { - // Access granted, no extra check needed. - return nil - } - if t := kernel.TaskFromContext(ctx); t != nil { - // Allow access if the task trying to access it is in the thread group - // corresponding to this directory. - if i.task.ThreadGroup() == t.ThreadGroup() { - // Access granted (overridden). - return nil - } - } - return err -} - -// fdSymlink is an symlink for the /proc/[pid]/fd/[fd] file. -// -// +stateify savable -type fdSymlink struct { - kernfs.InodeAttrs - kernfs.InodeNoopRefCount - kernfs.InodeSymlink - - task *kernel.Task - fd int32 -} - -var _ kernfs.Inode = (*fdSymlink)(nil) - -func newFDSymlink(task *kernel.Task, fd int32, ino uint64) *kernfs.Dentry { - inode := &fdSymlink{ - task: task, - fd: fd, - } - inode.Init(task.Credentials(), ino, linux.ModeSymlink|0777) - - d := &kernfs.Dentry{} - d.Init(inode) - return d -} - -func (s *fdSymlink) Readlink(ctx context.Context) (string, error) { - file, _ := getTaskFD(s.task, s.fd) - if file == nil { - return "", syserror.ENOENT - } - defer file.DecRef() - root := vfs.RootFromContext(ctx) - defer root.DecRef() - return s.task.Kernel().VFS().PathnameWithDeleted(ctx, root, file.VirtualDentry()) -} - -func (s *fdSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { - file, _ := getTaskFD(s.task, s.fd) - if file == nil { - return vfs.VirtualDentry{}, "", syserror.ENOENT - } - defer file.DecRef() - vd := file.VirtualDentry() - vd.IncRef() - return vd, "", nil -} - -// fdInfoDirInode represents the inode for /proc/[pid]/fdinfo directory. -// -// +stateify savable -type fdInfoDirInode struct { - kernfs.InodeNotSymlink - kernfs.InodeDirectoryNoNewChildren - kernfs.InodeAttrs - kernfs.OrderedChildren - kernfs.AlwaysValid - fdDir -} - -var _ kernfs.Inode = (*fdInfoDirInode)(nil) - -func newFDInfoDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { - inode := &fdInfoDirInode{ - fdDir: fdDir{ - inoGen: inoGen, - task: task, - }, - } - inode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) - - dentry := &kernfs.Dentry{} - dentry.Init(inode) - inode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) - - return dentry -} - -// Lookup implements kernfs.inodeDynamicLookup. -func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { - fdInt, err := strconv.ParseInt(name, 10, 32) - if err != nil { - return nil, syserror.ENOENT - } - fd := int32(fdInt) - if !taskFDExists(i.task, fd) { - return nil, syserror.ENOENT - } - data := &fdInfoData{ - task: i.task, - fd: fd, - } - dentry := newTaskOwnedFile(i.task, i.inoGen.NextIno(), 0444, data) - return dentry.VFSDentry(), nil -} - -// Open implements kernfs.Inode. -func (i *fdInfoDirInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) - return fd.VFSFileDescription(), nil -} - -// fdInfoData implements vfs.DynamicBytesSource for /proc/[pid]/fdinfo/[fd]. -// -// +stateify savable -type fdInfoData struct { - kernfs.DynamicBytesFile - refs.AtomicRefCount - - task *kernel.Task - fd int32 -} - -var _ dynamicInode = (*fdInfoData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *fdInfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { - file, descriptorFlags := getTaskFD(d.task, d.fd) - if file == nil { - return syserror.ENOENT - } - defer file.DecRef() - // TODO(b/121266871): Include pos, locks, and other data. For now we only - // have flags. - // See https://www.kernel.org/doc/Documentation/filesystems/proc.txt - flags := uint(file.StatusFlags()) | descriptorFlags.ToLinuxFileFlags() - fmt.Fprintf(buf, "flags:\t0%o\n", flags) - return nil -} diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go deleted file mode 100644 index 2c6f8bdfc..000000000 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ /dev/null @@ -1,761 +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 proc - -import ( - "bytes" - "fmt" - "io" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sentry/fsbridge" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/limits" - "gvisor.dev/gvisor/pkg/sentry/mm" - "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// mm gets the kernel task's MemoryManager. No additional reference is taken on -// mm here. This is safe because MemoryManager.destroy is required to leave the -// MemoryManager in a state where it's still usable as a DynamicBytesSource. -func getMM(task *kernel.Task) *mm.MemoryManager { - var tmm *mm.MemoryManager - task.WithMuLocked(func(t *kernel.Task) { - if mm := t.MemoryManager(); mm != nil { - tmm = mm - } - }) - return tmm -} - -// getMMIncRef returns t's MemoryManager. If getMMIncRef succeeds, the -// MemoryManager's users count is incremented, and must be decremented by the -// caller when it is no longer in use. -func getMMIncRef(task *kernel.Task) (*mm.MemoryManager, error) { - if task.ExitState() == kernel.TaskExitDead { - return nil, syserror.ESRCH - } - var m *mm.MemoryManager - task.WithMuLocked(func(t *kernel.Task) { - m = t.MemoryManager() - }) - if m == nil || !m.IncUsers() { - return nil, io.EOF - } - return m, nil -} - -func checkTaskState(t *kernel.Task) error { - switch t.ExitState() { - case kernel.TaskExitZombie: - return syserror.EACCES - case kernel.TaskExitDead: - return syserror.ESRCH - } - return nil -} - -type bufferWriter struct { - buf *bytes.Buffer -} - -// WriteFromBlocks writes up to srcs.NumBytes() bytes from srcs and returns -// the number of bytes written. It may return a partial write without an -// error (i.e. (n, nil) where 0 < n < srcs.NumBytes()). It should not -// return a full write with an error (i.e. srcs.NumBytes(), err) where err -// != nil). -func (w *bufferWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { - written := srcs.NumBytes() - for !srcs.IsEmpty() { - w.buf.Write(srcs.Head().ToSlice()) - srcs = srcs.Tail() - } - return written, nil -} - -// auxvData implements vfs.DynamicBytesSource for /proc/[pid]/auxv. -// -// +stateify savable -type auxvData struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -var _ dynamicInode = (*auxvData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *auxvData) Generate(ctx context.Context, buf *bytes.Buffer) error { - m, err := getMMIncRef(d.task) - if err != nil { - return err - } - defer m.DecUsers(ctx) - - // Space for buffer with AT_NULL (0) terminator at the end. - auxv := m.Auxv() - buf.Grow((len(auxv) + 1) * 16) - for _, e := range auxv { - var tmp [8]byte - usermem.ByteOrder.PutUint64(tmp[:], e.Key) - buf.Write(tmp[:]) - - usermem.ByteOrder.PutUint64(tmp[:], uint64(e.Value)) - buf.Write(tmp[:]) - } - return nil -} - -// execArgType enumerates the types of exec arguments that are exposed through -// proc. -type execArgType int - -const ( - cmdlineDataArg execArgType = iota - environDataArg -) - -// cmdlineData implements vfs.DynamicBytesSource for /proc/[pid]/cmdline. -// -// +stateify savable -type cmdlineData struct { - kernfs.DynamicBytesFile - - task *kernel.Task - - // arg is the type of exec argument this file contains. - arg execArgType -} - -var _ dynamicInode = (*cmdlineData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *cmdlineData) Generate(ctx context.Context, buf *bytes.Buffer) error { - m, err := getMMIncRef(d.task) - if err != nil { - return err - } - defer m.DecUsers(ctx) - - // Figure out the bounds of the exec arg we are trying to read. - var ar usermem.AddrRange - switch d.arg { - case cmdlineDataArg: - ar = usermem.AddrRange{ - Start: m.ArgvStart(), - End: m.ArgvEnd(), - } - case environDataArg: - ar = usermem.AddrRange{ - Start: m.EnvvStart(), - End: m.EnvvEnd(), - } - default: - panic(fmt.Sprintf("unknown exec arg type %v", d.arg)) - } - if ar.Start == 0 || ar.End == 0 { - // Don't attempt to read before the start/end are set up. - return io.EOF - } - - // N.B. Technically this should be usermem.IOOpts.IgnorePermissions = true - // until Linux 4.9 (272ddc8b3735 "proc: don't use FOLL_FORCE for reading - // cmdline and environment"). - writer := &bufferWriter{buf: buf} - if n, err := m.CopyInTo(ctx, usermem.AddrRangeSeqOf(ar), writer, usermem.IOOpts{}); n == 0 || err != nil { - // Nothing to copy or something went wrong. - return err - } - - // On Linux, if the NULL byte at the end of the argument vector has been - // overwritten, it continues reading the environment vector as part of - // the argument vector. - if d.arg == cmdlineDataArg && buf.Bytes()[buf.Len()-1] != 0 { - if end := bytes.IndexByte(buf.Bytes(), 0); end != -1 { - // If we found a NULL character somewhere else in argv, truncate the - // return up to the NULL terminator (including it). - buf.Truncate(end) - return nil - } - - // There is no NULL terminator in the string, return into envp. - arEnvv := usermem.AddrRange{ - Start: m.EnvvStart(), - End: m.EnvvEnd(), - } - - // Upstream limits the returned amount to one page of slop. - // https://elixir.bootlin.com/linux/v4.20/source/fs/proc/base.c#L208 - // we'll return one page total between argv and envp because of the - // above page restrictions. - if buf.Len() >= usermem.PageSize { - // Returned at least one page already, nothing else to add. - return nil - } - remaining := usermem.PageSize - buf.Len() - if int(arEnvv.Length()) > remaining { - end, ok := arEnvv.Start.AddLength(uint64(remaining)) - if !ok { - return syserror.EFAULT - } - arEnvv.End = end - } - if _, err := m.CopyInTo(ctx, usermem.AddrRangeSeqOf(arEnvv), writer, usermem.IOOpts{}); err != nil { - return err - } - - // Linux will return envp up to and including the first NULL character, - // so find it. - if end := bytes.IndexByte(buf.Bytes()[ar.Length():], 0); end != -1 { - buf.Truncate(end) - } - } - - return nil -} - -// +stateify savable -type commInode struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -func newComm(task *kernel.Task, ino uint64, perm linux.FileMode) *kernfs.Dentry { - inode := &commInode{task: task} - inode.DynamicBytesFile.Init(task.Credentials(), ino, &commData{task: task}, perm) - - d := &kernfs.Dentry{} - d.Init(inode) - return d -} - -func (i *commInode) CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { - // This file can always be read or written by members of the same thread - // group. See fs/proc/base.c:proc_tid_comm_permission. - // - // N.B. This check is currently a no-op as we don't yet support writing and - // this file is world-readable anyways. - t := kernel.TaskFromContext(ctx) - if t != nil && t.ThreadGroup() == i.task.ThreadGroup() && !ats.MayExec() { - return nil - } - - return i.DynamicBytesFile.CheckPermissions(ctx, creds, ats) -} - -// commData implements vfs.DynamicBytesSource for /proc/[pid]/comm. -// -// +stateify savable -type commData struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -var _ dynamicInode = (*commData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *commData) Generate(ctx context.Context, buf *bytes.Buffer) error { - buf.WriteString(d.task.Name()) - buf.WriteString("\n") - return nil -} - -// idMapData implements vfs.DynamicBytesSource for /proc/[pid]/{gid_map|uid_map}. -// -// +stateify savable -type idMapData struct { - kernfs.DynamicBytesFile - - task *kernel.Task - gids bool -} - -var _ dynamicInode = (*idMapData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *idMapData) Generate(ctx context.Context, buf *bytes.Buffer) error { - var entries []auth.IDMapEntry - if d.gids { - entries = d.task.UserNamespace().GIDMap() - } else { - entries = d.task.UserNamespace().UIDMap() - } - for _, e := range entries { - fmt.Fprintf(buf, "%10d %10d %10d\n", e.FirstID, e.FirstParentID, e.Length) - } - return nil -} - -// mapsData implements vfs.DynamicBytesSource for /proc/[pid]/maps. -// -// +stateify savable -type mapsData struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -var _ dynamicInode = (*mapsData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *mapsData) Generate(ctx context.Context, buf *bytes.Buffer) error { - if mm := getMM(d.task); mm != nil { - mm.ReadMapsDataInto(ctx, buf) - } - return nil -} - -// smapsData implements vfs.DynamicBytesSource for /proc/[pid]/smaps. -// -// +stateify savable -type smapsData struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -var _ dynamicInode = (*smapsData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *smapsData) Generate(ctx context.Context, buf *bytes.Buffer) error { - if mm := getMM(d.task); mm != nil { - mm.ReadSmapsDataInto(ctx, buf) - } - return nil -} - -// +stateify savable -type taskStatData struct { - kernfs.DynamicBytesFile - - task *kernel.Task - - // If tgstats is true, accumulate fault stats (not implemented) and CPU - // time across all tasks in t's thread group. - tgstats bool - - // pidns is the PID namespace associated with the proc filesystem that - // includes the file using this statData. - pidns *kernel.PIDNamespace -} - -var _ dynamicInode = (*taskStatData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (s *taskStatData) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "%d ", s.pidns.IDOfTask(s.task)) - fmt.Fprintf(buf, "(%s) ", s.task.Name()) - fmt.Fprintf(buf, "%c ", s.task.StateStatus()[0]) - ppid := kernel.ThreadID(0) - if parent := s.task.Parent(); parent != nil { - ppid = s.pidns.IDOfThreadGroup(parent.ThreadGroup()) - } - fmt.Fprintf(buf, "%d ", ppid) - fmt.Fprintf(buf, "%d ", s.pidns.IDOfProcessGroup(s.task.ThreadGroup().ProcessGroup())) - fmt.Fprintf(buf, "%d ", s.pidns.IDOfSession(s.task.ThreadGroup().Session())) - fmt.Fprintf(buf, "0 0 " /* tty_nr tpgid */) - fmt.Fprintf(buf, "0 " /* flags */) - fmt.Fprintf(buf, "0 0 0 0 " /* minflt cminflt majflt cmajflt */) - var cputime usage.CPUStats - if s.tgstats { - cputime = s.task.ThreadGroup().CPUStats() - } else { - cputime = s.task.CPUStats() - } - fmt.Fprintf(buf, "%d %d ", linux.ClockTFromDuration(cputime.UserTime), linux.ClockTFromDuration(cputime.SysTime)) - cputime = s.task.ThreadGroup().JoinedChildCPUStats() - fmt.Fprintf(buf, "%d %d ", linux.ClockTFromDuration(cputime.UserTime), linux.ClockTFromDuration(cputime.SysTime)) - fmt.Fprintf(buf, "%d %d ", s.task.Priority(), s.task.Niceness()) - fmt.Fprintf(buf, "%d ", s.task.ThreadGroup().Count()) - - // itrealvalue. Since kernel 2.6.17, this field is no longer - // maintained, and is hard coded as 0. - fmt.Fprintf(buf, "0 ") - - // Start time is relative to boot time, expressed in clock ticks. - fmt.Fprintf(buf, "%d ", linux.ClockTFromDuration(s.task.StartTime().Sub(s.task.Kernel().Timekeeper().BootTime()))) - - var vss, rss uint64 - s.task.WithMuLocked(func(t *kernel.Task) { - if mm := t.MemoryManager(); mm != nil { - vss = mm.VirtualMemorySize() - rss = mm.ResidentSetSize() - } - }) - fmt.Fprintf(buf, "%d %d ", vss, rss/usermem.PageSize) - - // rsslim. - fmt.Fprintf(buf, "%d ", s.task.ThreadGroup().Limits().Get(limits.Rss).Cur) - - fmt.Fprintf(buf, "0 0 0 0 0 " /* startcode endcode startstack kstkesp kstkeip */) - fmt.Fprintf(buf, "0 0 0 0 0 " /* signal blocked sigignore sigcatch wchan */) - fmt.Fprintf(buf, "0 0 " /* nswap cnswap */) - terminationSignal := linux.Signal(0) - if s.task == s.task.ThreadGroup().Leader() { - terminationSignal = s.task.ThreadGroup().TerminationSignal() - } - fmt.Fprintf(buf, "%d ", terminationSignal) - fmt.Fprintf(buf, "0 0 0 " /* processor rt_priority policy */) - fmt.Fprintf(buf, "0 0 0 " /* delayacct_blkio_ticks guest_time cguest_time */) - fmt.Fprintf(buf, "0 0 0 0 0 0 0 " /* start_data end_data start_brk arg_start arg_end env_start env_end */) - fmt.Fprintf(buf, "0\n" /* exit_code */) - - return nil -} - -// statmData implements vfs.DynamicBytesSource for /proc/[pid]/statm. -// -// +stateify savable -type statmData struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -var _ dynamicInode = (*statmData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (s *statmData) Generate(ctx context.Context, buf *bytes.Buffer) error { - var vss, rss uint64 - s.task.WithMuLocked(func(t *kernel.Task) { - if mm := t.MemoryManager(); mm != nil { - vss = mm.VirtualMemorySize() - rss = mm.ResidentSetSize() - } - }) - - fmt.Fprintf(buf, "%d %d 0 0 0 0 0\n", vss/usermem.PageSize, rss/usermem.PageSize) - return nil -} - -// statusData implements vfs.DynamicBytesSource for /proc/[pid]/status. -// -// +stateify savable -type statusData struct { - kernfs.DynamicBytesFile - - task *kernel.Task - pidns *kernel.PIDNamespace -} - -var _ dynamicInode = (*statusData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (s *statusData) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "Name:\t%s\n", s.task.Name()) - fmt.Fprintf(buf, "State:\t%s\n", s.task.StateStatus()) - fmt.Fprintf(buf, "Tgid:\t%d\n", s.pidns.IDOfThreadGroup(s.task.ThreadGroup())) - fmt.Fprintf(buf, "Pid:\t%d\n", s.pidns.IDOfTask(s.task)) - ppid := kernel.ThreadID(0) - if parent := s.task.Parent(); parent != nil { - ppid = s.pidns.IDOfThreadGroup(parent.ThreadGroup()) - } - fmt.Fprintf(buf, "PPid:\t%d\n", ppid) - tpid := kernel.ThreadID(0) - if tracer := s.task.Tracer(); tracer != nil { - tpid = s.pidns.IDOfTask(tracer) - } - fmt.Fprintf(buf, "TracerPid:\t%d\n", tpid) - var fds int - var vss, rss, data uint64 - s.task.WithMuLocked(func(t *kernel.Task) { - if fdTable := t.FDTable(); fdTable != nil { - fds = fdTable.Size() - } - if mm := t.MemoryManager(); mm != nil { - vss = mm.VirtualMemorySize() - rss = mm.ResidentSetSize() - data = mm.VirtualDataSize() - } - }) - fmt.Fprintf(buf, "FDSize:\t%d\n", fds) - fmt.Fprintf(buf, "VmSize:\t%d kB\n", vss>>10) - fmt.Fprintf(buf, "VmRSS:\t%d kB\n", rss>>10) - fmt.Fprintf(buf, "VmData:\t%d kB\n", data>>10) - fmt.Fprintf(buf, "Threads:\t%d\n", s.task.ThreadGroup().Count()) - creds := s.task.Credentials() - fmt.Fprintf(buf, "CapInh:\t%016x\n", creds.InheritableCaps) - fmt.Fprintf(buf, "CapPrm:\t%016x\n", creds.PermittedCaps) - fmt.Fprintf(buf, "CapEff:\t%016x\n", creds.EffectiveCaps) - fmt.Fprintf(buf, "CapBnd:\t%016x\n", creds.BoundingCaps) - fmt.Fprintf(buf, "Seccomp:\t%d\n", s.task.SeccompMode()) - // We unconditionally report a single NUMA node. See - // pkg/sentry/syscalls/linux/sys_mempolicy.go. - fmt.Fprintf(buf, "Mems_allowed:\t1\n") - fmt.Fprintf(buf, "Mems_allowed_list:\t0\n") - return nil -} - -// ioUsage is the /proc/[pid]/io and /proc/[pid]/task/[tid]/io data provider. -type ioUsage interface { - // IOUsage returns the io usage data. - IOUsage() *usage.IO -} - -// +stateify savable -type ioData struct { - kernfs.DynamicBytesFile - - ioUsage -} - -var _ dynamicInode = (*ioData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (i *ioData) Generate(ctx context.Context, buf *bytes.Buffer) error { - io := usage.IO{} - io.Accumulate(i.IOUsage()) - - fmt.Fprintf(buf, "char: %d\n", io.CharsRead) - fmt.Fprintf(buf, "wchar: %d\n", io.CharsWritten) - fmt.Fprintf(buf, "syscr: %d\n", io.ReadSyscalls) - fmt.Fprintf(buf, "syscw: %d\n", io.WriteSyscalls) - fmt.Fprintf(buf, "read_bytes: %d\n", io.BytesRead) - fmt.Fprintf(buf, "write_bytes: %d\n", io.BytesWritten) - fmt.Fprintf(buf, "cancelled_write_bytes: %d\n", io.BytesWriteCancelled) - return nil -} - -// oomScoreAdj is a stub of the /proc/<pid>/oom_score_adj file. -// -// +stateify savable -type oomScoreAdj struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -var _ vfs.WritableDynamicBytesSource = (*oomScoreAdj)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (o *oomScoreAdj) Generate(ctx context.Context, buf *bytes.Buffer) error { - if o.task.ExitState() == kernel.TaskExitDead { - return syserror.ESRCH - } - fmt.Fprintf(buf, "%d\n", o.task.OOMScoreAdj()) - return nil -} - -// Write implements vfs.WritableDynamicBytesSource.Write. -func (o *oomScoreAdj) Write(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { - if src.NumBytes() == 0 { - return 0, nil - } - - // Limit input size so as not to impact performance if input size is large. - src = src.TakeFirst(usermem.PageSize - 1) - - var v int32 - n, err := usermem.CopyInt32StringInVec(ctx, src.IO, src.Addrs, &v, src.Opts) - if err != nil { - return 0, err - } - - if o.task.ExitState() == kernel.TaskExitDead { - return 0, syserror.ESRCH - } - if err := o.task.SetOOMScoreAdj(v); err != nil { - return 0, err - } - - return n, nil -} - -// exeSymlink is an symlink for the /proc/[pid]/exe file. -// -// +stateify savable -type exeSymlink struct { - kernfs.InodeAttrs - kernfs.InodeNoopRefCount - kernfs.InodeSymlink - - task *kernel.Task -} - -var _ kernfs.Inode = (*exeSymlink)(nil) - -func newExeSymlink(task *kernel.Task, ino uint64) *kernfs.Dentry { - inode := &exeSymlink{task: task} - inode.Init(task.Credentials(), ino, linux.ModeSymlink|0777) - - d := &kernfs.Dentry{} - d.Init(inode) - return d -} - -// Readlink implements kernfs.Inode. -func (s *exeSymlink) Readlink(ctx context.Context) (string, error) { - if !kernel.ContextCanTrace(ctx, s.task, false) { - return "", syserror.EACCES - } - - // Pull out the executable for /proc/[pid]/exe. - exec, err := s.executable() - if err != nil { - return "", err - } - defer exec.DecRef() - - return exec.PathnameWithDeleted(ctx), nil -} - -// Getlink implements kernfs.Inode.Getlink. -func (s *exeSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { - if !kernel.ContextCanTrace(ctx, s.task, false) { - return vfs.VirtualDentry{}, "", syserror.EACCES - } - - exec, err := s.executable() - if err != nil { - return vfs.VirtualDentry{}, "", err - } - defer exec.DecRef() - - vd := exec.(*fsbridge.VFSFile).FileDescription().VirtualDentry() - vd.IncRef() - return vd, "", nil -} - -func (s *exeSymlink) executable() (file fsbridge.File, err error) { - if err := checkTaskState(s.task); err != nil { - return nil, err - } - - s.task.WithMuLocked(func(t *kernel.Task) { - mm := t.MemoryManager() - if mm == nil { - err = syserror.EACCES - return - } - - // The MemoryManager may be destroyed, in which case - // MemoryManager.destroy will simply set the executable to nil - // (with locks held). - file = mm.Executable() - if file == nil { - err = syserror.ESRCH - } - }) - return -} - -// mountInfoData is used to implement /proc/[pid]/mountinfo. -// -// +stateify savable -type mountInfoData struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -var _ dynamicInode = (*mountInfoData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (i *mountInfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { - var fsctx *kernel.FSContext - i.task.WithMuLocked(func(t *kernel.Task) { - fsctx = t.FSContext() - }) - if fsctx == nil { - // The task has been destroyed. Nothing to show here. - return nil - } - rootDir := fsctx.RootDirectoryVFS2() - if !rootDir.Ok() { - // Root has been destroyed. Don't try to read mounts. - return nil - } - defer rootDir.DecRef() - i.task.Kernel().VFS().GenerateProcMountInfo(ctx, rootDir, buf) - return nil -} - -// mountsData is used to implement /proc/[pid]/mounts. -// -// +stateify savable -type mountsData struct { - kernfs.DynamicBytesFile - - task *kernel.Task -} - -var _ dynamicInode = (*mountsData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (i *mountsData) Generate(ctx context.Context, buf *bytes.Buffer) error { - var fsctx *kernel.FSContext - i.task.WithMuLocked(func(t *kernel.Task) { - fsctx = t.FSContext() - }) - if fsctx == nil { - // The task has been destroyed. Nothing to show here. - return nil - } - rootDir := fsctx.RootDirectoryVFS2() - if !rootDir.Ok() { - // Root has been destroyed. Don't try to read mounts. - return nil - } - defer rootDir.DecRef() - i.task.Kernel().VFS().GenerateProcMounts(ctx, rootDir, buf) - return nil -} - -type namespaceSymlink struct { - kernfs.StaticSymlink - - task *kernel.Task -} - -func newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentry { - // Namespace symlinks should contain the namespace name and the inode number - // for the namespace instance, so for example user:[123456]. We currently fake - // the inode number by sticking the symlink inode in its place. - target := fmt.Sprintf("%s:[%d]", ns, ino) - - inode := &namespaceSymlink{task: task} - // Note: credentials are overridden by taskOwnedInode. - inode.Init(task.Credentials(), ino, target) - - taskInode := &taskOwnedInode{Inode: inode, owner: task} - d := &kernfs.Dentry{} - d.Init(taskInode) - return d -} - -// Readlink implements Inode. -func (s *namespaceSymlink) Readlink(ctx context.Context) (string, error) { - if err := checkTaskState(s.task); err != nil { - return "", err - } - return s.StaticSymlink.Readlink(ctx) -} - -// Getlink implements Inode.Getlink. -func (s *namespaceSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { - if err := checkTaskState(s.task); err != nil { - return vfs.VirtualDentry{}, "", err - } - return s.StaticSymlink.Getlink(ctx) -} diff --git a/pkg/sentry/fsimpl/proc/task_net.go b/pkg/sentry/fsimpl/proc/task_net.go deleted file mode 100644 index 6b2a77328..000000000 --- a/pkg/sentry/fsimpl/proc/task_net.go +++ /dev/null @@ -1,808 +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 proc - -import ( - "bytes" - "fmt" - "io" - "reflect" - "time" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/inet" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/socket" - "gvisor.dev/gvisor/pkg/sentry/socket/unix" - "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/tcpip/header" - "gvisor.dev/gvisor/pkg/usermem" -) - -func newTaskNetDir(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { - k := task.Kernel() - pidns := task.PIDNamespace() - root := auth.NewRootCredentials(pidns.UserNamespace()) - - var contents map[string]*kernfs.Dentry - if stack := task.NetworkNamespace().Stack(); stack != nil { - const ( - arp = "IP address HW type Flags HW address Mask Device\n" - netlink = "sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode\n" - packet = "sk RefCnt Type Proto Iface R Rmem User Inode\n" - protocols = "protocol size sockets memory press maxhdr slab module cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n" - ptype = "Type Device Function\n" - upd6 = " sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode\n" - ) - psched := fmt.Sprintf("%08x %08x %08x %08x\n", uint64(time.Microsecond/time.Nanosecond), 64, 1000000, uint64(time.Second/time.Nanosecond)) - - // TODO(gvisor.dev/issue/1833): Make sure file contents reflect the task - // network namespace. - contents = map[string]*kernfs.Dentry{ - "dev": newDentry(root, inoGen.NextIno(), 0444, &netDevData{stack: stack}), - "snmp": newDentry(root, inoGen.NextIno(), 0444, &netSnmpData{stack: stack}), - - // The following files are simple stubs until they are implemented in - // netstack, if the file contains a header the stub is just the header - // otherwise it is an empty file. - "arp": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(arp)), - "netlink": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(netlink)), - "netstat": newDentry(root, inoGen.NextIno(), 0444, &netStatData{}), - "packet": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(packet)), - "protocols": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(protocols)), - - // Linux sets psched values to: nsec per usec, psched tick in ns, 1000000, - // high res timer ticks per sec (ClockGetres returns 1ns resolution). - "psched": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(psched)), - "ptype": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(ptype)), - "route": newDentry(root, inoGen.NextIno(), 0444, &netRouteData{stack: stack}), - "tcp": newDentry(root, inoGen.NextIno(), 0444, &netTCPData{kernel: k}), - "udp": newDentry(root, inoGen.NextIno(), 0444, &netUDPData{kernel: k}), - "unix": newDentry(root, inoGen.NextIno(), 0444, &netUnixData{kernel: k}), - } - - if stack.SupportsIPv6() { - contents["if_inet6"] = newDentry(root, inoGen.NextIno(), 0444, &ifinet6{stack: stack}) - contents["ipv6_route"] = newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")) - contents["tcp6"] = newDentry(root, inoGen.NextIno(), 0444, &netTCP6Data{kernel: k}) - contents["udp6"] = newDentry(root, inoGen.NextIno(), 0444, newStaticFile(upd6)) - } - } - - return newTaskOwnedDir(task, inoGen.NextIno(), 0555, contents) -} - -// ifinet6 implements vfs.DynamicBytesSource for /proc/net/if_inet6. -// -// +stateify savable -type ifinet6 struct { - kernfs.DynamicBytesFile - - stack inet.Stack -} - -var _ dynamicInode = (*ifinet6)(nil) - -func (n *ifinet6) contents() []string { - var lines []string - nics := n.stack.Interfaces() - for id, naddrs := range n.stack.InterfaceAddrs() { - nic, ok := nics[id] - if !ok { - // NIC was added after NICNames was called. We'll just ignore it. - continue - } - - for _, a := range naddrs { - // IPv6 only. - if a.Family != linux.AF_INET6 { - continue - } - - // Fields: - // IPv6 address displayed in 32 hexadecimal chars without colons - // Netlink device number (interface index) in hexadecimal (use nic id) - // Prefix length in hexadecimal - // Scope value (use 0) - // Interface flags - // Device name - lines = append(lines, fmt.Sprintf("%032x %02x %02x %02x %02x %8s\n", a.Addr, id, a.PrefixLen, 0, a.Flags, nic.Name)) - } - } - return lines -} - -// Generate implements vfs.DynamicBytesSource.Generate. -func (n *ifinet6) Generate(ctx context.Context, buf *bytes.Buffer) error { - for _, l := range n.contents() { - buf.WriteString(l) - } - return nil -} - -// netDevData implements vfs.DynamicBytesSource for /proc/net/dev. -// -// +stateify savable -type netDevData struct { - kernfs.DynamicBytesFile - - stack inet.Stack -} - -var _ dynamicInode = (*netDevData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (n *netDevData) Generate(ctx context.Context, buf *bytes.Buffer) error { - interfaces := n.stack.Interfaces() - buf.WriteString("Inter-| Receive | Transmit\n") - buf.WriteString(" face |bytes packets errs drop fifo frame compressed multicast|bytes packets errs drop fifo colls carrier compressed\n") - - for _, i := range interfaces { - // Implements the same format as - // net/core/net-procfs.c:dev_seq_printf_stats. - var stats inet.StatDev - if err := n.stack.Statistics(&stats, i.Name); err != nil { - log.Warningf("Failed to retrieve interface statistics for %v: %v", i.Name, err) - continue - } - fmt.Fprintf( - buf, - "%6s: %7d %7d %4d %4d %4d %5d %10d %9d %8d %7d %4d %4d %4d %5d %7d %10d\n", - i.Name, - // Received - stats[0], // bytes - stats[1], // packets - stats[2], // errors - stats[3], // dropped - stats[4], // fifo - stats[5], // frame - stats[6], // compressed - stats[7], // multicast - // Transmitted - stats[8], // bytes - stats[9], // packets - stats[10], // errors - stats[11], // dropped - stats[12], // fifo - stats[13], // frame - stats[14], // compressed - stats[15], // multicast - ) - } - - return nil -} - -// netUnixData implements vfs.DynamicBytesSource for /proc/net/unix. -// -// +stateify savable -type netUnixData struct { - kernfs.DynamicBytesFile - - kernel *kernel.Kernel -} - -var _ dynamicInode = (*netUnixData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error { - buf.WriteString("Num RefCount Protocol Flags Type St Inode Path\n") - for _, se := range n.kernel.ListSockets() { - s := se.SockVFS2 - if !s.TryIncRef() { - log.Debugf("Couldn't get reference on %v in socket table, racing with destruction?", s) - continue - } - if family, _, _ := s.Impl().(socket.SocketVFS2).Type(); family != linux.AF_UNIX { - s.DecRef() - // Not a unix socket. - continue - } - sops := s.Impl().(*unix.SocketVFS2) - - addr, err := sops.Endpoint().GetLocalAddress() - if err != nil { - log.Warningf("Failed to retrieve socket name from %+v: %v", s, err) - addr.Addr = "<unknown>" - } - - sockFlags := 0 - if ce, ok := sops.Endpoint().(transport.ConnectingEndpoint); ok { - if ce.Listening() { - // For unix domain sockets, linux reports a single flag - // value if the socket is listening, of __SO_ACCEPTCON. - sockFlags = linux.SO_ACCEPTCON - } - } - - // Get inode number. - var ino uint64 - stat, statErr := s.Stat(ctx, vfs.StatOptions{Mask: linux.STATX_INO}) - if statErr != nil || stat.Mask&linux.STATX_INO == 0 { - log.Warningf("Failed to retrieve ino for socket file: %v", statErr) - } else { - ino = stat.Ino - } - - // In the socket entry below, the value for the 'Num' field requires - // some consideration. Linux prints the address to the struct - // unix_sock representing a socket in the kernel, but may redact the - // value for unprivileged users depending on the kptr_restrict - // sysctl. - // - // One use for this field is to allow a privileged user to - // introspect into the kernel memory to determine information about - // a socket not available through procfs, such as the socket's peer. - // - // In gvisor, returning a pointer to our internal structures would - // be pointless, as it wouldn't match the memory layout for struct - // unix_sock, making introspection difficult. We could populate a - // struct unix_sock with the appropriate data, but even that - // requires consideration for which kernel version to emulate, as - // the definition of this struct changes over time. - // - // For now, we always redact this pointer. - fmt.Fprintf(buf, "%#016p: %08X %08X %08X %04X %02X %8d", - (*unix.SocketOperations)(nil), // Num, pointer to kernel socket struct. - s.Refs()-1, // RefCount, don't count our own ref. - 0, // Protocol, always 0 for UDS. - sockFlags, // Flags. - sops.Endpoint().Type(), // Type. - sops.State(), // State. - ino, // Inode. - ) - - // Path - if len(addr.Addr) != 0 { - if addr.Addr[0] == 0 { - // Abstract path. - fmt.Fprintf(buf, " @%s", string(addr.Addr[1:])) - } else { - fmt.Fprintf(buf, " %s", string(addr.Addr)) - } - } - fmt.Fprintf(buf, "\n") - - s.DecRef() - } - return nil -} - -func networkToHost16(n uint16) uint16 { - // n is in network byte order, so is big-endian. The most-significant byte - // should be stored in the lower address. - // - // We manually inline binary.BigEndian.Uint16() because Go does not support - // non-primitive consts, so binary.BigEndian is a (mutable) var, so calls to - // binary.BigEndian.Uint16() require a read of binary.BigEndian and an - // interface method call, defeating inlining. - buf := [2]byte{byte(n >> 8 & 0xff), byte(n & 0xff)} - return usermem.ByteOrder.Uint16(buf[:]) -} - -func writeInetAddr(w io.Writer, family int, i linux.SockAddr) { - switch family { - case linux.AF_INET: - var a linux.SockAddrInet - if i != nil { - a = *i.(*linux.SockAddrInet) - } - - // linux.SockAddrInet.Port is stored in the network byte order and is - // printed like a number in host byte order. Note that all numbers in host - // byte order are printed with the most-significant byte first when - // formatted with %X. See get_tcp4_sock() and udp4_format_sock() in Linux. - port := networkToHost16(a.Port) - - // linux.SockAddrInet.Addr is stored as a byte slice in big-endian order - // (i.e. most-significant byte in index 0). Linux represents this as a - // __be32 which is a typedef for an unsigned int, and is printed with - // %X. This means that for a little-endian machine, Linux prints the - // least-significant byte of the address first. To emulate this, we first - // invert the byte order for the address using usermem.ByteOrder.Uint32, - // which makes it have the equivalent encoding to a __be32 on a little - // endian machine. Note that this operation is a no-op on a big endian - // machine. Then similar to Linux, we format it with %X, which will print - // the most-significant byte of the __be32 address first, which is now - // actually the least-significant byte of the original address in - // linux.SockAddrInet.Addr on little endian machines, due to the conversion. - addr := usermem.ByteOrder.Uint32(a.Addr[:]) - - fmt.Fprintf(w, "%08X:%04X ", addr, port) - case linux.AF_INET6: - var a linux.SockAddrInet6 - if i != nil { - a = *i.(*linux.SockAddrInet6) - } - - port := networkToHost16(a.Port) - addr0 := usermem.ByteOrder.Uint32(a.Addr[0:4]) - addr1 := usermem.ByteOrder.Uint32(a.Addr[4:8]) - addr2 := usermem.ByteOrder.Uint32(a.Addr[8:12]) - addr3 := usermem.ByteOrder.Uint32(a.Addr[12:16]) - fmt.Fprintf(w, "%08X%08X%08X%08X:%04X ", addr0, addr1, addr2, addr3, port) - } -} - -func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, family int) error { - // t may be nil here if our caller is not part of a task goroutine. This can - // happen for example if we're here for "sentryctl cat". When t is nil, - // degrade gracefully and retrieve what we can. - t := kernel.TaskFromContext(ctx) - - for _, se := range k.ListSockets() { - s := se.SockVFS2 - if !s.TryIncRef() { - log.Debugf("Couldn't get reference on %v in socket table, racing with destruction?", s) - continue - } - sops, ok := s.Impl().(socket.SocketVFS2) - if !ok { - panic(fmt.Sprintf("Found non-socket file in socket table: %+v", s)) - } - if fa, stype, _ := sops.Type(); !(family == fa && stype == linux.SOCK_STREAM) { - s.DecRef() - // Not tcp4 sockets. - continue - } - - // Linux's documentation for the fields below can be found at - // https://www.kernel.org/doc/Documentation/networking/proc_net_tcp.txt. - // For Linux's implementation, see net/ipv4/tcp_ipv4.c:get_tcp4_sock(). - // Note that the header doesn't contain labels for all the fields. - - // Field: sl; entry number. - fmt.Fprintf(buf, "%4d: ", se.ID) - - // Field: local_adddress. - var localAddr linux.SockAddr - if t != nil { - if local, _, err := sops.GetSockName(t); err == nil { - localAddr = local - } - } - writeInetAddr(buf, family, localAddr) - - // Field: rem_address. - var remoteAddr linux.SockAddr - if t != nil { - if remote, _, err := sops.GetPeerName(t); err == nil { - remoteAddr = remote - } - } - writeInetAddr(buf, family, remoteAddr) - - // Field: state; socket state. - fmt.Fprintf(buf, "%02X ", sops.State()) - - // Field: tx_queue, rx_queue; number of packets in the transmit and - // receive queue. Unimplemented. - fmt.Fprintf(buf, "%08X:%08X ", 0, 0) - - // Field: tr, tm->when; timer active state and number of jiffies - // until timer expires. Unimplemented. - fmt.Fprintf(buf, "%02X:%08X ", 0, 0) - - // Field: retrnsmt; number of unrecovered RTO timeouts. - // Unimplemented. - fmt.Fprintf(buf, "%08X ", 0) - - stat, statErr := s.Stat(ctx, vfs.StatOptions{Mask: linux.STATX_UID | linux.STATX_INO}) - - // Field: uid. - if statErr != nil || stat.Mask&linux.STATX_UID == 0 { - log.Warningf("Failed to retrieve uid for socket file: %v", statErr) - fmt.Fprintf(buf, "%5d ", 0) - } else { - creds := auth.CredentialsFromContext(ctx) - fmt.Fprintf(buf, "%5d ", uint32(auth.KUID(stat.UID).In(creds.UserNamespace).OrOverflow())) - } - - // Field: timeout; number of unanswered 0-window probes. - // Unimplemented. - fmt.Fprintf(buf, "%8d ", 0) - - // Field: inode. - if statErr != nil || stat.Mask&linux.STATX_INO == 0 { - log.Warningf("Failed to retrieve inode for socket file: %v", statErr) - fmt.Fprintf(buf, "%8d ", 0) - } else { - fmt.Fprintf(buf, "%8d ", stat.Ino) - } - - // Field: refcount. Don't count the ref we obtain while deferencing - // the weakref to this socket. - fmt.Fprintf(buf, "%d ", s.Refs()-1) - - // Field: Socket struct address. Redacted due to the same reason as - // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. - fmt.Fprintf(buf, "%#016p ", (*socket.Socket)(nil)) - - // Field: retransmit timeout. Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: predicted tick of soft clock (delayed ACK control data). - // Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: (ack.quick<<1)|ack.pingpong, Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: sending congestion window, Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: Slow start size threshold, -1 if threshold >= 0xFFFF. - // Unimplemented, report as large threshold. - fmt.Fprintf(buf, "%d", -1) - - fmt.Fprintf(buf, "\n") - - s.DecRef() - } - - return nil -} - -// netTCPData implements vfs.DynamicBytesSource for /proc/net/tcp. -// -// +stateify savable -type netTCPData struct { - kernfs.DynamicBytesFile - - kernel *kernel.Kernel -} - -var _ dynamicInode = (*netTCPData)(nil) - -func (d *netTCPData) Generate(ctx context.Context, buf *bytes.Buffer) error { - buf.WriteString(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode \n") - return commonGenerateTCP(ctx, buf, d.kernel, linux.AF_INET) -} - -// netTCP6Data implements vfs.DynamicBytesSource for /proc/net/tcp6. -// -// +stateify savable -type netTCP6Data struct { - kernfs.DynamicBytesFile - - kernel *kernel.Kernel -} - -var _ dynamicInode = (*netTCP6Data)(nil) - -func (d *netTCP6Data) Generate(ctx context.Context, buf *bytes.Buffer) error { - buf.WriteString(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode\n") - return commonGenerateTCP(ctx, buf, d.kernel, linux.AF_INET6) -} - -// netUDPData implements vfs.DynamicBytesSource for /proc/net/udp. -// -// +stateify savable -type netUDPData struct { - kernfs.DynamicBytesFile - - kernel *kernel.Kernel -} - -var _ dynamicInode = (*netUDPData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error { - // t may be nil here if our caller is not part of a task goroutine. This can - // happen for example if we're here for "sentryctl cat". When t is nil, - // degrade gracefully and retrieve what we can. - t := kernel.TaskFromContext(ctx) - - for _, se := range d.kernel.ListSockets() { - s := se.SockVFS2 - if !s.TryIncRef() { - log.Debugf("Couldn't get reference on %v in socket table, racing with destruction?", s) - continue - } - sops, ok := s.Impl().(socket.SocketVFS2) - if !ok { - panic(fmt.Sprintf("Found non-socket file in socket table: %+v", s)) - } - if family, stype, _ := sops.Type(); family != linux.AF_INET || stype != linux.SOCK_DGRAM { - s.DecRef() - // Not udp4 socket. - continue - } - - // For Linux's implementation, see net/ipv4/udp.c:udp4_format_sock(). - - // Field: sl; entry number. - fmt.Fprintf(buf, "%5d: ", se.ID) - - // Field: local_adddress. - var localAddr linux.SockAddrInet - if t != nil { - if local, _, err := sops.GetSockName(t); err == nil { - localAddr = *local.(*linux.SockAddrInet) - } - } - writeInetAddr(buf, linux.AF_INET, &localAddr) - - // Field: rem_address. - var remoteAddr linux.SockAddrInet - if t != nil { - if remote, _, err := sops.GetPeerName(t); err == nil { - remoteAddr = *remote.(*linux.SockAddrInet) - } - } - writeInetAddr(buf, linux.AF_INET, &remoteAddr) - - // Field: state; socket state. - fmt.Fprintf(buf, "%02X ", sops.State()) - - // Field: tx_queue, rx_queue; number of packets in the transmit and - // receive queue. Unimplemented. - fmt.Fprintf(buf, "%08X:%08X ", 0, 0) - - // Field: tr, tm->when. Always 0 for UDP. - fmt.Fprintf(buf, "%02X:%08X ", 0, 0) - - // Field: retrnsmt. Always 0 for UDP. - fmt.Fprintf(buf, "%08X ", 0) - - stat, statErr := s.Stat(ctx, vfs.StatOptions{Mask: linux.STATX_UID | linux.STATX_INO}) - - // Field: uid. - if statErr != nil || stat.Mask&linux.STATX_UID == 0 { - log.Warningf("Failed to retrieve uid for socket file: %v", statErr) - fmt.Fprintf(buf, "%5d ", 0) - } else { - creds := auth.CredentialsFromContext(ctx) - fmt.Fprintf(buf, "%5d ", uint32(auth.KUID(stat.UID).In(creds.UserNamespace).OrOverflow())) - } - - // Field: timeout. Always 0 for UDP. - fmt.Fprintf(buf, "%8d ", 0) - - // Field: inode. - if statErr != nil || stat.Mask&linux.STATX_INO == 0 { - log.Warningf("Failed to retrieve inode for socket file: %v", statErr) - fmt.Fprintf(buf, "%8d ", 0) - } else { - fmt.Fprintf(buf, "%8d ", stat.Ino) - } - - // Field: ref; reference count on the socket inode. Don't count the ref - // we obtain while deferencing the weakref to this socket. - fmt.Fprintf(buf, "%d ", s.Refs()-1) - - // Field: Socket struct address. Redacted due to the same reason as - // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. - fmt.Fprintf(buf, "%#016p ", (*socket.Socket)(nil)) - - // Field: drops; number of dropped packets. Unimplemented. - fmt.Fprintf(buf, "%d", 0) - - fmt.Fprintf(buf, "\n") - - s.DecRef() - } - return nil -} - -// netSnmpData implements vfs.DynamicBytesSource for /proc/net/snmp. -// -// +stateify savable -type netSnmpData struct { - kernfs.DynamicBytesFile - - stack inet.Stack -} - -var _ dynamicInode = (*netSnmpData)(nil) - -type snmpLine struct { - prefix string - header string -} - -var snmp = []snmpLine{ - { - prefix: "Ip", - header: "Forwarding DefaultTTL InReceives InHdrErrors InAddrErrors ForwDatagrams InUnknownProtos InDiscards InDelivers OutRequests OutDiscards OutNoRoutes ReasmTimeout ReasmReqds ReasmOKs ReasmFails FragOKs FragFails FragCreates", - }, - { - prefix: "Icmp", - header: "InMsgs InErrors InCsumErrors InDestUnreachs InTimeExcds InParmProbs InSrcQuenchs InRedirects InEchos InEchoReps InTimestamps InTimestampReps InAddrMasks InAddrMaskReps OutMsgs OutErrors OutDestUnreachs OutTimeExcds OutParmProbs OutSrcQuenchs OutRedirects OutEchos OutEchoReps OutTimestamps OutTimestampReps OutAddrMasks OutAddrMaskReps", - }, - { - prefix: "IcmpMsg", - }, - { - prefix: "Tcp", - header: "RtoAlgorithm RtoMin RtoMax MaxConn ActiveOpens PassiveOpens AttemptFails EstabResets CurrEstab InSegs OutSegs RetransSegs InErrs OutRsts InCsumErrors", - }, - { - prefix: "Udp", - header: "InDatagrams NoPorts InErrors OutDatagrams RcvbufErrors SndbufErrors InCsumErrors IgnoredMulti", - }, - { - prefix: "UdpLite", - header: "InDatagrams NoPorts InErrors OutDatagrams RcvbufErrors SndbufErrors InCsumErrors IgnoredMulti", - }, -} - -func toSlice(a interface{}) []uint64 { - v := reflect.Indirect(reflect.ValueOf(a)) - return v.Slice(0, v.Len()).Interface().([]uint64) -} - -func sprintSlice(s []uint64) string { - if len(s) == 0 { - return "" - } - r := fmt.Sprint(s) - return r[1 : len(r)-1] // Remove "[]" introduced by fmt of slice. -} - -// Generate implements vfs.DynamicBytesSource. -func (d *netSnmpData) Generate(ctx context.Context, buf *bytes.Buffer) error { - types := []interface{}{ - &inet.StatSNMPIP{}, - &inet.StatSNMPICMP{}, - nil, // TODO(gvisor.dev/issue/628): Support IcmpMsg stats. - &inet.StatSNMPTCP{}, - &inet.StatSNMPUDP{}, - &inet.StatSNMPUDPLite{}, - } - for i, stat := range types { - line := snmp[i] - if stat == nil { - fmt.Fprintf(buf, "%s:\n", line.prefix) - fmt.Fprintf(buf, "%s:\n", line.prefix) - continue - } - if err := d.stack.Statistics(stat, line.prefix); err != nil { - if err == syserror.EOPNOTSUPP { - log.Infof("Failed to retrieve %s of /proc/net/snmp: %v", line.prefix, err) - } else { - log.Warningf("Failed to retrieve %s of /proc/net/snmp: %v", line.prefix, err) - } - } - - fmt.Fprintf(buf, "%s: %s\n", line.prefix, line.header) - - if line.prefix == "Tcp" { - tcp := stat.(*inet.StatSNMPTCP) - // "Tcp" needs special processing because MaxConn is signed. RFC 2012. - fmt.Sprintf("%s: %s %d %s\n", line.prefix, sprintSlice(tcp[:3]), int64(tcp[3]), sprintSlice(tcp[4:])) - } else { - fmt.Sprintf("%s: %s\n", line.prefix, sprintSlice(toSlice(stat))) - } - } - return nil -} - -// netRouteData implements vfs.DynamicBytesSource for /proc/net/route. -// -// +stateify savable -type netRouteData struct { - kernfs.DynamicBytesFile - - stack inet.Stack -} - -var _ dynamicInode = (*netRouteData)(nil) - -// Generate implements vfs.DynamicBytesSource. -// See Linux's net/ipv4/fib_trie.c:fib_route_seq_show. -func (d *netRouteData) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "%-127s\n", "Iface\tDestination\tGateway\tFlags\tRefCnt\tUse\tMetric\tMask\tMTU\tWindow\tIRTT") - - interfaces := d.stack.Interfaces() - for _, rt := range d.stack.RouteTable() { - // /proc/net/route only includes ipv4 routes. - if rt.Family != linux.AF_INET { - continue - } - - // /proc/net/route does not include broadcast or multicast routes. - if rt.Type == linux.RTN_BROADCAST || rt.Type == linux.RTN_MULTICAST { - continue - } - - iface, ok := interfaces[rt.OutputInterface] - if !ok || iface.Name == "lo" { - continue - } - - var ( - gw uint32 - prefix uint32 - flags = linux.RTF_UP - ) - if len(rt.GatewayAddr) == header.IPv4AddressSize { - flags |= linux.RTF_GATEWAY - gw = usermem.ByteOrder.Uint32(rt.GatewayAddr) - } - if len(rt.DstAddr) == header.IPv4AddressSize { - prefix = usermem.ByteOrder.Uint32(rt.DstAddr) - } - l := fmt.Sprintf( - "%s\t%08X\t%08X\t%04X\t%d\t%d\t%d\t%08X\t%d\t%d\t%d", - iface.Name, - prefix, - gw, - flags, - 0, // RefCnt. - 0, // Use. - 0, // Metric. - (uint32(1)<<rt.DstLen)-1, - 0, // MTU. - 0, // Window. - 0, // RTT. - ) - fmt.Fprintf(buf, "%-127s\n", l) - } - return nil -} - -// netStatData implements vfs.DynamicBytesSource for /proc/net/netstat. -// -// +stateify savable -type netStatData struct { - kernfs.DynamicBytesFile - - stack inet.Stack -} - -var _ dynamicInode = (*netStatData)(nil) - -// Generate implements vfs.DynamicBytesSource. -// See Linux's net/ipv4/fib_trie.c:fib_route_seq_show. -func (d *netStatData) Generate(ctx context.Context, buf *bytes.Buffer) error { - buf.WriteString("TcpExt: SyncookiesSent SyncookiesRecv SyncookiesFailed " + - "EmbryonicRsts PruneCalled RcvPruned OfoPruned OutOfWindowIcmps " + - "LockDroppedIcmps ArpFilter TW TWRecycled TWKilled PAWSPassive " + - "PAWSActive PAWSEstab DelayedACKs DelayedACKLocked DelayedACKLost " + - "ListenOverflows ListenDrops TCPPrequeued TCPDirectCopyFromBacklog " + - "TCPDirectCopyFromPrequeue TCPPrequeueDropped TCPHPHits TCPHPHitsToUser " + - "TCPPureAcks TCPHPAcks TCPRenoRecovery TCPSackRecovery TCPSACKReneging " + - "TCPFACKReorder TCPSACKReorder TCPRenoReorder TCPTSReorder TCPFullUndo " + - "TCPPartialUndo TCPDSACKUndo TCPLossUndo TCPLostRetransmit " + - "TCPRenoFailures TCPSackFailures TCPLossFailures TCPFastRetrans " + - "TCPForwardRetrans TCPSlowStartRetrans TCPTimeouts TCPLossProbes " + - "TCPLossProbeRecovery TCPRenoRecoveryFail TCPSackRecoveryFail " + - "TCPSchedulerFailed TCPRcvCollapsed TCPDSACKOldSent TCPDSACKOfoSent " + - "TCPDSACKRecv TCPDSACKOfoRecv TCPAbortOnData TCPAbortOnClose " + - "TCPAbortOnMemory TCPAbortOnTimeout TCPAbortOnLinger TCPAbortFailed " + - "TCPMemoryPressures TCPSACKDiscard TCPDSACKIgnoredOld " + - "TCPDSACKIgnoredNoUndo TCPSpuriousRTOs TCPMD5NotFound TCPMD5Unexpected " + - "TCPMD5Failure TCPSackShifted TCPSackMerged TCPSackShiftFallback " + - "TCPBacklogDrop TCPMinTTLDrop TCPDeferAcceptDrop IPReversePathFilter " + - "TCPTimeWaitOverflow TCPReqQFullDoCookies TCPReqQFullDrop TCPRetransFail " + - "TCPRcvCoalesce TCPOFOQueue TCPOFODrop TCPOFOMerge TCPChallengeACK " + - "TCPSYNChallenge TCPFastOpenActive TCPFastOpenActiveFail " + - "TCPFastOpenPassive TCPFastOpenPassiveFail TCPFastOpenListenOverflow " + - "TCPFastOpenCookieReqd TCPSpuriousRtxHostQueues BusyPollRxPackets " + - "TCPAutoCorking TCPFromZeroWindowAdv TCPToZeroWindowAdv " + - "TCPWantZeroWindowAdv TCPSynRetrans TCPOrigDataSent TCPHystartTrainDetect " + - "TCPHystartTrainCwnd TCPHystartDelayDetect TCPHystartDelayCwnd " + - "TCPACKSkippedSynRecv TCPACKSkippedPAWS TCPACKSkippedSeq " + - "TCPACKSkippedFinWait2 TCPACKSkippedTimeWait TCPACKSkippedChallenge " + - "TCPWinProbe TCPKeepAlive TCPMTUPFail TCPMTUPSuccess\n") - return nil -} diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go deleted file mode 100644 index 9f2ef8200..000000000 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ /dev/null @@ -1,257 +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 proc - -import ( - "bytes" - "sort" - "strconv" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -const ( - selfName = "self" - threadSelfName = "thread-self" -) - -// InoGenerator generates unique inode numbers for a given filesystem. -type InoGenerator interface { - NextIno() uint64 -} - -// tasksInode represents the inode for /proc/ directory. -// -// +stateify savable -type tasksInode struct { - kernfs.InodeNotSymlink - kernfs.InodeDirectoryNoNewChildren - kernfs.InodeAttrs - kernfs.OrderedChildren - kernfs.AlwaysValid - - inoGen InoGenerator - pidns *kernel.PIDNamespace - - // '/proc/self' and '/proc/thread-self' have custom directory offsets in - // Linux. So handle them outside of OrderedChildren. - selfSymlink *vfs.Dentry - threadSelfSymlink *vfs.Dentry - - // cgroupControllers is a map of controller name to directory in the - // cgroup hierarchy. These controllers are immutable and will be listed - // in /proc/pid/cgroup if not nil. - cgroupControllers map[string]string -} - -var _ kernfs.Inode = (*tasksInode)(nil) - -func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNamespace, cgroupControllers map[string]string) (*tasksInode, *kernfs.Dentry) { - root := auth.NewRootCredentials(pidns.UserNamespace()) - contents := map[string]*kernfs.Dentry{ - "cpuinfo": newDentry(root, inoGen.NextIno(), 0444, newStaticFileSetStat(cpuInfoData(k))), - "filesystems": newDentry(root, inoGen.NextIno(), 0444, &filesystemsData{}), - "loadavg": newDentry(root, inoGen.NextIno(), 0444, &loadavgData{}), - "sys": newSysDir(root, inoGen, k), - "meminfo": newDentry(root, inoGen.NextIno(), 0444, &meminfoData{}), - "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/mounts"), - "net": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/net"), - "stat": newDentry(root, inoGen.NextIno(), 0444, &statData{}), - "uptime": newDentry(root, inoGen.NextIno(), 0444, &uptimeData{}), - "version": newDentry(root, inoGen.NextIno(), 0444, &versionData{}), - } - - inode := &tasksInode{ - pidns: pidns, - inoGen: inoGen, - selfSymlink: newSelfSymlink(root, inoGen.NextIno(), pidns).VFSDentry(), - threadSelfSymlink: newThreadSelfSymlink(root, inoGen.NextIno(), pidns).VFSDentry(), - cgroupControllers: cgroupControllers, - } - inode.InodeAttrs.Init(root, inoGen.NextIno(), linux.ModeDirectory|0555) - - dentry := &kernfs.Dentry{} - dentry.Init(inode) - - inode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) - links := inode.OrderedChildren.Populate(dentry, contents) - inode.IncLinks(links) - - return inode, dentry -} - -// Lookup implements kernfs.inodeDynamicLookup. -func (i *tasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { - // Try to lookup a corresponding task. - tid, err := strconv.ParseUint(name, 10, 64) - if err != nil { - // If it failed to parse, check if it's one of the special handled files. - switch name { - case selfName: - return i.selfSymlink, nil - case threadSelfName: - return i.threadSelfSymlink, nil - } - return nil, syserror.ENOENT - } - - task := i.pidns.TaskWithID(kernel.ThreadID(tid)) - if task == nil { - return nil, syserror.ENOENT - } - - taskDentry := newTaskInode(i.inoGen, task, i.pidns, true, i.cgroupControllers) - return taskDentry.VFSDentry(), nil -} - -// IterDirents implements kernfs.inodeDynamicLookup. -func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, offset, _ int64) (int64, error) { - // fs/proc/internal.h: #define FIRST_PROCESS_ENTRY 256 - const FIRST_PROCESS_ENTRY = 256 - - // Use maxTaskID to shortcut searches that will result in 0 entries. - const maxTaskID = kernel.TasksLimit + 1 - if offset >= maxTaskID { - return offset, nil - } - - // According to Linux (fs/proc/base.c:proc_pid_readdir()), process directories - // start at offset FIRST_PROCESS_ENTRY with '/proc/self', followed by - // '/proc/thread-self' and then '/proc/[pid]'. - if offset < FIRST_PROCESS_ENTRY { - offset = FIRST_PROCESS_ENTRY - } - - if offset == FIRST_PROCESS_ENTRY { - dirent := vfs.Dirent{ - Name: selfName, - Type: linux.DT_LNK, - Ino: i.inoGen.NextIno(), - NextOff: offset + 1, - } - if err := cb.Handle(dirent); err != nil { - return offset, err - } - offset++ - } - if offset == FIRST_PROCESS_ENTRY+1 { - dirent := vfs.Dirent{ - Name: threadSelfName, - Type: linux.DT_LNK, - Ino: i.inoGen.NextIno(), - NextOff: offset + 1, - } - if err := cb.Handle(dirent); err != nil { - return offset, err - } - offset++ - } - - // Collect all tasks that TGIDs are greater than the offset specified. Per - // Linux we only include in directory listings if it's the leader. But for - // whatever crazy reason, you can still walk to the given node. - var tids []int - startTid := offset - FIRST_PROCESS_ENTRY - 2 - for _, tg := range i.pidns.ThreadGroups() { - tid := i.pidns.IDOfThreadGroup(tg) - if int64(tid) < startTid { - continue - } - if leader := tg.Leader(); leader != nil { - tids = append(tids, int(tid)) - } - } - - if len(tids) == 0 { - return offset, nil - } - - sort.Ints(tids) - for _, tid := range tids { - dirent := vfs.Dirent{ - Name: strconv.FormatUint(uint64(tid), 10), - Type: linux.DT_DIR, - Ino: i.inoGen.NextIno(), - NextOff: FIRST_PROCESS_ENTRY + 2 + int64(tid) + 1, - } - if err := cb.Handle(dirent); err != nil { - return offset, err - } - offset++ - } - return maxTaskID, nil -} - -// Open implements kernfs.Inode. -func (i *tasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, &opts) - return fd.VFSFileDescription(), nil -} - -func (i *tasksInode) Stat(vsfs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { - stat, err := i.InodeAttrs.Stat(vsfs, opts) - if err != nil { - return linux.Statx{}, err - } - - if opts.Mask&linux.STATX_NLINK != 0 { - // Add dynamic children to link count. - for _, tg := range i.pidns.ThreadGroups() { - if leader := tg.Leader(); leader != nil { - stat.Nlink++ - } - } - } - - return stat, nil -} - -// staticFileSetStat implements a special static file that allows inode -// attributes to be set. This is to support /proc files that are readonly, but -// allow attributes to be set. -type staticFileSetStat struct { - dynamicBytesFileSetAttr - vfs.StaticData -} - -var _ dynamicInode = (*staticFileSetStat)(nil) - -func newStaticFileSetStat(data string) *staticFileSetStat { - return &staticFileSetStat{StaticData: vfs.StaticData{Data: data}} -} - -func cpuInfoData(k *kernel.Kernel) string { - features := k.FeatureSet() - if features == nil { - // Kernel is always initialized with a FeatureSet. - panic("cpuinfo read with nil FeatureSet") - } - var buf bytes.Buffer - for i, max := uint(0), k.ApplicationCores(); i < max; i++ { - features.WriteCPUInfoTo(i, &buf) - } - return buf.String() -} - -func shmData(v uint64) dynamicInode { - return newStaticFile(strconv.FormatUint(v, 10)) -} diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go deleted file mode 100644 index 4621e2de0..000000000 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ /dev/null @@ -1,380 +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 proc - -import ( - "bytes" - "fmt" - "strconv" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -type selfSymlink struct { - kernfs.InodeAttrs - kernfs.InodeNoopRefCount - kernfs.InodeSymlink - - pidns *kernel.PIDNamespace -} - -var _ kernfs.Inode = (*selfSymlink)(nil) - -func newSelfSymlink(creds *auth.Credentials, ino uint64, pidns *kernel.PIDNamespace) *kernfs.Dentry { - inode := &selfSymlink{pidns: pidns} - inode.Init(creds, ino, linux.ModeSymlink|0777) - - d := &kernfs.Dentry{} - d.Init(inode) - return d -} - -func (s *selfSymlink) Readlink(ctx context.Context) (string, error) { - t := kernel.TaskFromContext(ctx) - if t == nil { - // Who is reading this link? - return "", syserror.EINVAL - } - tgid := s.pidns.IDOfThreadGroup(t.ThreadGroup()) - if tgid == 0 { - return "", syserror.ENOENT - } - return strconv.FormatUint(uint64(tgid), 10), nil -} - -func (s *selfSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { - target, err := s.Readlink(ctx) - return vfs.VirtualDentry{}, target, err -} - -// SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*selfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { - return syserror.EPERM -} - -type threadSelfSymlink struct { - kernfs.InodeAttrs - kernfs.InodeNoopRefCount - kernfs.InodeSymlink - - pidns *kernel.PIDNamespace -} - -var _ kernfs.Inode = (*threadSelfSymlink)(nil) - -func newThreadSelfSymlink(creds *auth.Credentials, ino uint64, pidns *kernel.PIDNamespace) *kernfs.Dentry { - inode := &threadSelfSymlink{pidns: pidns} - inode.Init(creds, ino, linux.ModeSymlink|0777) - - d := &kernfs.Dentry{} - d.Init(inode) - return d -} - -func (s *threadSelfSymlink) Readlink(ctx context.Context) (string, error) { - t := kernel.TaskFromContext(ctx) - if t == nil { - // Who is reading this link? - return "", syserror.EINVAL - } - tgid := s.pidns.IDOfThreadGroup(t.ThreadGroup()) - tid := s.pidns.IDOfTask(t) - if tid == 0 || tgid == 0 { - return "", syserror.ENOENT - } - return fmt.Sprintf("%d/task/%d", tgid, tid), nil -} - -func (s *threadSelfSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { - target, err := s.Readlink(ctx) - return vfs.VirtualDentry{}, target, err -} - -// SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*threadSelfSymlink) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { - return syserror.EPERM -} - -// dynamicBytesFileSetAttr implements a special file that allows inode -// attributes to be set. This is to support /proc files that are readonly, but -// allow attributes to be set. -type dynamicBytesFileSetAttr struct { - kernfs.DynamicBytesFile -} - -// SetStat implements Inode.SetStat. -func (d *dynamicBytesFileSetAttr) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { - return d.DynamicBytesFile.InodeAttrs.SetStat(ctx, fs, creds, opts) -} - -// cpuStats contains the breakdown of CPU time for /proc/stat. -type cpuStats struct { - // user is time spent in userspace tasks with non-positive niceness. - user uint64 - - // nice is time spent in userspace tasks with positive niceness. - nice uint64 - - // system is time spent in non-interrupt kernel context. - system uint64 - - // idle is time spent idle. - idle uint64 - - // ioWait is time spent waiting for IO. - ioWait uint64 - - // irq is time spent in interrupt context. - irq uint64 - - // softirq is time spent in software interrupt context. - softirq uint64 - - // steal is involuntary wait time. - steal uint64 - - // guest is time spent in guests with non-positive niceness. - guest uint64 - - // guestNice is time spent in guests with positive niceness. - guestNice uint64 -} - -// String implements fmt.Stringer. -func (c cpuStats) String() string { - return fmt.Sprintf("%d %d %d %d %d %d %d %d %d %d", c.user, c.nice, c.system, c.idle, c.ioWait, c.irq, c.softirq, c.steal, c.guest, c.guestNice) -} - -// statData implements vfs.DynamicBytesSource for /proc/stat. -// -// +stateify savable -type statData struct { - dynamicBytesFileSetAttr -} - -var _ dynamicInode = (*statData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (*statData) Generate(ctx context.Context, buf *bytes.Buffer) error { - // TODO(b/37226836): We currently export only zero CPU stats. We could - // at least provide some aggregate stats. - var cpu cpuStats - fmt.Fprintf(buf, "cpu %s\n", cpu) - - k := kernel.KernelFromContext(ctx) - for c, max := uint(0), k.ApplicationCores(); c < max; c++ { - fmt.Fprintf(buf, "cpu%d %s\n", c, cpu) - } - - // The total number of interrupts is dependent on the CPUs and PCI - // devices on the system. See arch_probe_nr_irqs. - // - // Since we don't report real interrupt stats, just choose an arbitrary - // value from a representative VM. - const numInterrupts = 256 - - // The Kernel doesn't handle real interrupts, so report all zeroes. - // TODO(b/37226836): We could count page faults as #PF. - fmt.Fprintf(buf, "intr 0") // total - for i := 0; i < numInterrupts; i++ { - fmt.Fprintf(buf, " 0") - } - fmt.Fprintf(buf, "\n") - - // Total number of context switches. - // TODO(b/37226836): Count this. - fmt.Fprintf(buf, "ctxt 0\n") - - // CLOCK_REALTIME timestamp from boot, in seconds. - fmt.Fprintf(buf, "btime %d\n", k.Timekeeper().BootTime().Seconds()) - - // Total number of clones. - // TODO(b/37226836): Count this. - fmt.Fprintf(buf, "processes 0\n") - - // Number of runnable tasks. - // TODO(b/37226836): Count this. - fmt.Fprintf(buf, "procs_running 0\n") - - // Number of tasks waiting on IO. - // TODO(b/37226836): Count this. - fmt.Fprintf(buf, "procs_blocked 0\n") - - // Number of each softirq handled. - fmt.Fprintf(buf, "softirq 0") // total - for i := 0; i < linux.NumSoftIRQ; i++ { - fmt.Fprintf(buf, " 0") - } - fmt.Fprintf(buf, "\n") - return nil -} - -// loadavgData backs /proc/loadavg. -// -// +stateify savable -type loadavgData struct { - dynamicBytesFileSetAttr -} - -var _ dynamicInode = (*loadavgData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (*loadavgData) Generate(ctx context.Context, buf *bytes.Buffer) error { - // TODO(b/62345059): Include real data in fields. - // Column 1-3: CPU and IO utilization of the last 1, 5, and 10 minute periods. - // Column 4-5: currently running processes and the total number of processes. - // Column 6: the last process ID used. - fmt.Fprintf(buf, "%.2f %.2f %.2f %d/%d %d\n", 0.00, 0.00, 0.00, 0, 0, 0) - return nil -} - -// meminfoData implements vfs.DynamicBytesSource for /proc/meminfo. -// -// +stateify savable -type meminfoData struct { - dynamicBytesFileSetAttr -} - -var _ dynamicInode = (*meminfoData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (*meminfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { - k := kernel.KernelFromContext(ctx) - mf := k.MemoryFile() - mf.UpdateUsage() - snapshot, totalUsage := usage.MemoryAccounting.Copy() - totalSize := usage.TotalMemory(mf.TotalSize(), totalUsage) - anon := snapshot.Anonymous + snapshot.Tmpfs - file := snapshot.PageCache + snapshot.Mapped - // We don't actually have active/inactive LRUs, so just make up numbers. - activeFile := (file / 2) &^ (usermem.PageSize - 1) - inactiveFile := file - activeFile - - fmt.Fprintf(buf, "MemTotal: %8d kB\n", totalSize/1024) - memFree := (totalSize - totalUsage) / 1024 - // We use MemFree as MemAvailable because we don't swap. - // TODO(rahat): When reclaim is implemented the value of MemAvailable - // should change. - fmt.Fprintf(buf, "MemFree: %8d kB\n", memFree) - fmt.Fprintf(buf, "MemAvailable: %8d kB\n", memFree) - fmt.Fprintf(buf, "Buffers: 0 kB\n") // memory usage by block devices - fmt.Fprintf(buf, "Cached: %8d kB\n", (file+snapshot.Tmpfs)/1024) - // Emulate a system with no swap, which disables inactivation of anon pages. - fmt.Fprintf(buf, "SwapCache: 0 kB\n") - fmt.Fprintf(buf, "Active: %8d kB\n", (anon+activeFile)/1024) - fmt.Fprintf(buf, "Inactive: %8d kB\n", inactiveFile/1024) - fmt.Fprintf(buf, "Active(anon): %8d kB\n", anon/1024) - fmt.Fprintf(buf, "Inactive(anon): 0 kB\n") - fmt.Fprintf(buf, "Active(file): %8d kB\n", activeFile/1024) - fmt.Fprintf(buf, "Inactive(file): %8d kB\n", inactiveFile/1024) - fmt.Fprintf(buf, "Unevictable: 0 kB\n") // TODO(b/31823263) - fmt.Fprintf(buf, "Mlocked: 0 kB\n") // TODO(b/31823263) - fmt.Fprintf(buf, "SwapTotal: 0 kB\n") - fmt.Fprintf(buf, "SwapFree: 0 kB\n") - fmt.Fprintf(buf, "Dirty: 0 kB\n") - fmt.Fprintf(buf, "Writeback: 0 kB\n") - fmt.Fprintf(buf, "AnonPages: %8d kB\n", anon/1024) - fmt.Fprintf(buf, "Mapped: %8d kB\n", file/1024) // doesn't count mapped tmpfs, which we don't know - fmt.Fprintf(buf, "Shmem: %8d kB\n", snapshot.Tmpfs/1024) - return nil -} - -// uptimeData implements vfs.DynamicBytesSource for /proc/uptime. -// -// +stateify savable -type uptimeData struct { - dynamicBytesFileSetAttr -} - -var _ dynamicInode = (*uptimeData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (*uptimeData) Generate(ctx context.Context, buf *bytes.Buffer) error { - k := kernel.KernelFromContext(ctx) - now := time.NowFromContext(ctx) - - // Pretend that we've spent zero time sleeping (second number). - fmt.Fprintf(buf, "%.2f 0.00\n", now.Sub(k.Timekeeper().BootTime()).Seconds()) - return nil -} - -// versionData implements vfs.DynamicBytesSource for /proc/version. -// -// +stateify savable -type versionData struct { - dynamicBytesFileSetAttr -} - -var _ dynamicInode = (*versionData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (*versionData) Generate(ctx context.Context, buf *bytes.Buffer) error { - k := kernel.KernelFromContext(ctx) - init := k.GlobalInit() - if init == nil { - // Attempted to read before the init Task is created. This can - // only occur during startup, which should never need to read - // this file. - panic("Attempted to read version before initial Task is available") - } - - // /proc/version takes the form: - // - // "SYSNAME version RELEASE (COMPILE_USER@COMPILE_HOST) - // (COMPILER_VERSION) VERSION" - // - // where: - // - SYSNAME, RELEASE, and VERSION are the same as returned by - // sys_utsname - // - COMPILE_USER is the user that build the kernel - // - COMPILE_HOST is the hostname of the machine on which the kernel - // was built - // - COMPILER_VERSION is the version reported by the building compiler - // - // Since we don't really want to expose build information to - // applications, those fields are omitted. - // - // FIXME(mpratt): Using Version from the init task SyscallTable - // disregards the different version a task may have (e.g., in a uts - // namespace). - ver := init.Leader().SyscallTable().Version - fmt.Fprintf(buf, "%s version %s %s\n", ver.Sysname, ver.Release, ver.Version) - return nil -} - -// filesystemsData backs /proc/filesystems. -// -// +stateify savable -type filesystemsData struct { - kernfs.DynamicBytesFile -} - -var _ dynamicInode = (*filesystemsData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *filesystemsData) Generate(ctx context.Context, buf *bytes.Buffer) error { - k := kernel.KernelFromContext(ctx) - k.VFS().GenerateProcFilesystems(buf) - return nil -} diff --git a/pkg/sentry/fsimpl/proc/tasks_sys.go b/pkg/sentry/fsimpl/proc/tasks_sys.go deleted file mode 100644 index 3d5dc463c..000000000 --- a/pkg/sentry/fsimpl/proc/tasks_sys.go +++ /dev/null @@ -1,211 +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 proc - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/inet" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// newSysDir returns the dentry corresponding to /proc/sys directory. -func newSysDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry { - return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "kernel": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "hostname": newDentry(root, inoGen.NextIno(), 0444, &hostnameData{}), - "shmall": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMALL)), - "shmmax": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMMAX)), - "shmmni": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMMNI)), - }), - "vm": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "mmap_min_addr": newDentry(root, inoGen.NextIno(), 0444, &mmapMinAddrData{}), - "overcommit_memory": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0\n")), - }), - "net": newSysNetDir(root, inoGen, k), - }) -} - -// newSysNetDir returns the dentry corresponding to /proc/sys/net directory. -func newSysNetDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry { - var contents map[string]*kernfs.Dentry - - // TODO(gvisor.dev/issue/1833): Support for using the network stack in the - // network namespace of the calling process. - if stack := k.RootNetworkNamespace().Stack(); stack != nil { - contents = map[string]*kernfs.Dentry{ - "ipv4": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "tcp_sack": newDentry(root, inoGen.NextIno(), 0644, &tcpSackData{stack: stack}), - - // The following files are simple stubs until they are implemented in - // netstack, most of these files are configuration related. We use the - // value closest to the actual netstack behavior or any empty file, all - // of these files will have mode 0444 (read-only for all users). - "ip_local_port_range": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("16000 65535")), - "ip_local_reserved_ports": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), - "ipfrag_time": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("30")), - "ip_nonlocal_bind": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "ip_no_pmtu_disc": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), - - // tcp_allowed_congestion_control tell the user what they are able to - // do as an unprivledged process so we leave it empty. - "tcp_allowed_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), - "tcp_available_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("reno")), - "tcp_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("reno")), - - // Many of the following stub files are features netstack doesn't - // support. The unsupported features return "0" to indicate they are - // disabled. - "tcp_base_mss": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1280")), - "tcp_dsack": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_early_retrans": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_fack": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_fastopen": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_fastopen_key": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), - "tcp_invalid_ratelimit": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_keepalive_intvl": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_keepalive_probes": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_keepalive_time": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("7200")), - "tcp_mtu_probing": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_no_metrics_save": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), - "tcp_probe_interval": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_probe_threshold": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "tcp_retries1": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("3")), - "tcp_retries2": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("15")), - "tcp_rfc1337": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), - "tcp_slow_start_after_idle": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), - "tcp_synack_retries": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("5")), - "tcp_syn_retries": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("3")), - "tcp_timestamps": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), - }), - "core": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "default_qdisc": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("pfifo_fast")), - "message_burst": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("10")), - "message_cost": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("5")), - "optmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), - "rmem_default": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), - "rmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), - "somaxconn": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("128")), - "wmem_default": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), - "wmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), - }), - } - } - - return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ - "net": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, contents), - }) -} - -// mmapMinAddrData implements vfs.DynamicBytesSource for -// /proc/sys/vm/mmap_min_addr. -// -// +stateify savable -type mmapMinAddrData struct { - kernfs.DynamicBytesFile - - k *kernel.Kernel -} - -var _ dynamicInode = (*mmapMinAddrData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *mmapMinAddrData) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "%d\n", d.k.Platform.MinUserAddress()) - return nil -} - -// hostnameData implements vfs.DynamicBytesSource for /proc/sys/kernel/hostname. -// -// +stateify savable -type hostnameData struct { - kernfs.DynamicBytesFile -} - -var _ dynamicInode = (*hostnameData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (*hostnameData) Generate(ctx context.Context, buf *bytes.Buffer) error { - utsns := kernel.UTSNamespaceFromContext(ctx) - buf.WriteString(utsns.HostName()) - buf.WriteString("\n") - return nil -} - -// tcpSackData implements vfs.WritableDynamicBytesSource for -// /proc/sys/net/tcp_sack. -// -// +stateify savable -type tcpSackData struct { - kernfs.DynamicBytesFile - - stack inet.Stack `state:"wait"` - enabled *bool -} - -var _ vfs.WritableDynamicBytesSource = (*tcpSackData)(nil) - -// Generate implements vfs.DynamicBytesSource. -func (d *tcpSackData) Generate(ctx context.Context, buf *bytes.Buffer) error { - if d.enabled == nil { - sack, err := d.stack.TCPSACKEnabled() - if err != nil { - return err - } - d.enabled = &sack - } - - val := "0\n" - if *d.enabled { - // Technically, this is not quite compatible with Linux. Linux stores these - // as an integer, so if you write "2" into tcp_sack, you should get 2 back. - // Tough luck. - val = "1\n" - } - buf.WriteString(val) - return nil -} - -func (d *tcpSackData) Write(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { - if offset != 0 { - // No need to handle partial writes thus far. - return 0, syserror.EINVAL - } - if src.NumBytes() == 0 { - return 0, nil - } - - // Limit the amount of memory allocated. - src = src.TakeFirst(usermem.PageSize - 1) - - var v int32 - n, err := usermem.CopyInt32StringInVec(ctx, src.IO, src.Addrs, &v, src.Opts) - if err != nil { - return n, err - } - if d.enabled == nil { - d.enabled = new(bool) - } - *d.enabled = v != 0 - return n, d.stack.SetTCPSACKEnabled(*d.enabled) -} diff --git a/pkg/sentry/fsimpl/proc/tasks_sys_test.go b/pkg/sentry/fsimpl/proc/tasks_sys_test.go deleted file mode 100644 index be54897bb..000000000 --- a/pkg/sentry/fsimpl/proc/tasks_sys_test.go +++ /dev/null @@ -1,78 +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 proc - -import ( - "bytes" - "reflect" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "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{stack: newIPv6TestStack()} - var buf bytes.Buffer - n.Generate(contexttest.Context(t), &buf) - if buf.Len() > 0 { - t.Errorf("n.Generate() generated = %v, want = %v", buf.Bytes(), []byte{}) - } -} - -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{stack: 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/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go deleted file mode 100644 index d0f97c137..000000000 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ /dev/null @@ -1,505 +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 proc - -import ( - "fmt" - "math" - "path" - "strconv" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -var ( - // Next offset 256 by convention. Adds 1 for the next offset. - selfLink = vfs.Dirent{Type: linux.DT_LNK, NextOff: 256 + 0 + 1} - threadSelfLink = vfs.Dirent{Type: linux.DT_LNK, NextOff: 256 + 1 + 1} - - // /proc/[pid] next offset starts at 256+2 (files above), then adds the - // PID, and adds 1 for the next offset. - proc1 = vfs.Dirent{Type: linux.DT_DIR, NextOff: 258 + 1 + 1} - proc2 = vfs.Dirent{Type: linux.DT_DIR, NextOff: 258 + 2 + 1} - proc3 = vfs.Dirent{Type: linux.DT_DIR, NextOff: 258 + 3 + 1} -) - -var ( - tasksStaticFiles = map[string]testutil.DirentType{ - "cpuinfo": linux.DT_REG, - "filesystems": linux.DT_REG, - "loadavg": linux.DT_REG, - "meminfo": linux.DT_REG, - "mounts": linux.DT_LNK, - "net": linux.DT_LNK, - "self": linux.DT_LNK, - "stat": linux.DT_REG, - "sys": linux.DT_DIR, - "thread-self": linux.DT_LNK, - "uptime": linux.DT_REG, - "version": linux.DT_REG, - } - tasksStaticFilesNextOffs = map[string]int64{ - "self": selfLink.NextOff, - "thread-self": threadSelfLink.NextOff, - } - taskStaticFiles = map[string]testutil.DirentType{ - "auxv": linux.DT_REG, - "cgroup": linux.DT_REG, - "cmdline": linux.DT_REG, - "comm": linux.DT_REG, - "environ": linux.DT_REG, - "exe": linux.DT_LNK, - "fd": linux.DT_DIR, - "fdinfo": linux.DT_DIR, - "gid_map": linux.DT_REG, - "io": linux.DT_REG, - "maps": linux.DT_REG, - "mountinfo": linux.DT_REG, - "mounts": linux.DT_REG, - "net": linux.DT_DIR, - "ns": linux.DT_DIR, - "oom_score": linux.DT_REG, - "oom_score_adj": linux.DT_REG, - "smaps": linux.DT_REG, - "stat": linux.DT_REG, - "statm": linux.DT_REG, - "status": linux.DT_REG, - "task": linux.DT_DIR, - "uid_map": linux.DT_REG, - } -) - -func setup(t *testing.T) *testutil.System { - k, err := testutil.Boot() - if err != nil { - t.Fatalf("Error creating kernel: %v", err) - } - - ctx := k.SupervisorContext() - creds := auth.CredentialsFromContext(ctx) - - k.VFS().MustRegisterFilesystemType(Name, &FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - - mntns, err := k.VFS().NewMountNamespace(ctx, creds, "", tmpfs.Name, &vfs.GetFilesystemOptions{}) - if err != nil { - t.Fatalf("NewMountNamespace(): %v", err) - } - pop := &vfs.PathOperation{ - Root: mntns.Root(), - Start: mntns.Root(), - Path: fspath.Parse("/proc"), - } - if err := k.VFS().MkdirAt(ctx, creds, pop, &vfs.MkdirOptions{Mode: 0777}); err != nil { - t.Fatalf("MkDir(/proc): %v", err) - } - - pop = &vfs.PathOperation{ - Root: mntns.Root(), - Start: mntns.Root(), - Path: fspath.Parse("/proc"), - } - mntOpts := &vfs.MountOptions{ - GetFilesystemOptions: vfs.GetFilesystemOptions{ - InternalData: &InternalData{ - Cgroups: map[string]string{ - "cpuset": "/foo/cpuset", - "memory": "/foo/memory", - }, - }, - }, - } - if err := k.VFS().MountAt(ctx, creds, "", pop, Name, mntOpts); err != nil { - t.Fatalf("MountAt(/proc): %v", err) - } - return testutil.NewSystem(ctx, t, k.VFS(), mntns) -} - -func TestTasksEmpty(t *testing.T) { - s := setup(t) - defer s.Destroy() - - collector := s.ListDirents(s.PathOpAtRoot("/proc")) - s.AssertAllDirentTypes(collector, tasksStaticFiles) - s.AssertDirentOffsets(collector, tasksStaticFilesNextOffs) -} - -func TestTasks(t *testing.T) { - s := setup(t) - defer s.Destroy() - - expectedDirents := make(map[string]testutil.DirentType) - for n, d := range tasksStaticFiles { - expectedDirents[n] = d - } - - k := kernel.KernelFromContext(s.Ctx) - var tasks []*kernel.Task - for i := 0; i < 5; i++ { - tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc, s.MntNs, s.Root, s.Root) - if err != nil { - t.Fatalf("CreateTask(): %v", err) - } - tasks = append(tasks, task) - expectedDirents[fmt.Sprintf("%d", i+1)] = linux.DT_DIR - } - - collector := s.ListDirents(s.PathOpAtRoot("/proc")) - s.AssertAllDirentTypes(collector, expectedDirents) - s.AssertDirentOffsets(collector, tasksStaticFilesNextOffs) - - lastPid := 0 - dirents := collector.OrderedDirents() - doneSkippingNonTaskDirs := false - for _, d := range dirents { - pid, err := strconv.Atoi(d.Name) - if err != nil { - if !doneSkippingNonTaskDirs { - // We haven't gotten to the task dirs yet. - continue - } - t.Fatalf("Invalid process directory %q", d.Name) - } - doneSkippingNonTaskDirs = true - if lastPid > pid { - t.Errorf("pids not in order: %v", dirents) - } - found := false - for _, t := range tasks { - if k.TaskSet().Root.IDOfTask(t) == kernel.ThreadID(pid) { - found = true - } - } - if !found { - t.Errorf("Additional task ID %d listed: %v", pid, tasks) - } - // Next offset starts at 256+2 ('self' and 'thread-self'), then adds the - // PID, and adds 1 for the next offset. - if want := int64(256 + 2 + pid + 1); d.NextOff != want { - t.Errorf("Wrong dirent offset want: %d got: %d: %+v", want, d.NextOff, d) - } - } - if !doneSkippingNonTaskDirs { - t.Fatalf("Never found any process directories.") - } - - // Test lookup. - for _, path := range []string{"/proc/1", "/proc/2"} { - fd, err := s.VFS.OpenAt( - s.Ctx, - s.Creds, - s.PathOpAtRoot(path), - &vfs.OpenOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.OpenAt(%q) failed: %v", path, err) - } - defer fd.DecRef() - buf := make([]byte, 1) - bufIOSeq := usermem.BytesIOSequence(buf) - if _, err := fd.Read(s.Ctx, bufIOSeq, vfs.ReadOptions{}); err != syserror.EISDIR { - t.Errorf("wrong error reading directory: %v", err) - } - } - - if _, err := s.VFS.OpenAt( - s.Ctx, - s.Creds, - s.PathOpAtRoot("/proc/9999"), - &vfs.OpenOptions{}, - ); err != syserror.ENOENT { - t.Fatalf("wrong error from vfsfs.OpenAt(/proc/9999): %v", err) - } -} - -func TestTasksOffset(t *testing.T) { - s := setup(t) - defer s.Destroy() - - k := kernel.KernelFromContext(s.Ctx) - for i := 0; i < 3; i++ { - tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - if _, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc, s.MntNs, s.Root, s.Root); err != nil { - t.Fatalf("CreateTask(): %v", err) - } - } - - for _, tc := range []struct { - name string - offset int64 - wants map[string]vfs.Dirent - }{ - { - name: "small offset", - offset: 100, - wants: map[string]vfs.Dirent{ - "self": selfLink, - "thread-self": threadSelfLink, - "1": proc1, - "2": proc2, - "3": proc3, - }, - }, - { - name: "offset at start", - offset: 256, - wants: map[string]vfs.Dirent{ - "self": selfLink, - "thread-self": threadSelfLink, - "1": proc1, - "2": proc2, - "3": proc3, - }, - }, - { - name: "skip /proc/self", - offset: 257, - wants: map[string]vfs.Dirent{ - "thread-self": threadSelfLink, - "1": proc1, - "2": proc2, - "3": proc3, - }, - }, - { - name: "skip symlinks", - offset: 258, - wants: map[string]vfs.Dirent{ - "1": proc1, - "2": proc2, - "3": proc3, - }, - }, - { - name: "skip first process", - offset: 260, - wants: map[string]vfs.Dirent{ - "2": proc2, - "3": proc3, - }, - }, - { - name: "last process", - offset: 261, - wants: map[string]vfs.Dirent{ - "3": proc3, - }, - }, - { - name: "after last", - offset: 262, - wants: nil, - }, - { - name: "TaskLimit+1", - offset: kernel.TasksLimit + 1, - wants: nil, - }, - { - name: "max", - offset: math.MaxInt64, - wants: nil, - }, - } { - t.Run(tc.name, func(t *testing.T) { - s := s.WithSubtest(t) - fd, err := s.VFS.OpenAt( - s.Ctx, - s.Creds, - s.PathOpAtRoot("/proc"), - &vfs.OpenOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.OpenAt(/) failed: %v", err) - } - defer fd.DecRef() - if _, err := fd.Seek(s.Ctx, tc.offset, linux.SEEK_SET); err != nil { - t.Fatalf("Seek(%d, SEEK_SET): %v", tc.offset, err) - } - - var collector testutil.DirentCollector - if err := fd.IterDirents(s.Ctx, &collector); err != nil { - t.Fatalf("IterDirent(): %v", err) - } - - expectedTypes := make(map[string]testutil.DirentType) - expectedOffsets := make(map[string]int64) - for name, want := range tc.wants { - expectedTypes[name] = want.Type - if want.NextOff != 0 { - expectedOffsets[name] = want.NextOff - } - } - - collector.SkipDotsChecks(true) // We seek()ed past the dots. - s.AssertAllDirentTypes(&collector, expectedTypes) - s.AssertDirentOffsets(&collector, expectedOffsets) - }) - } -} - -func TestTask(t *testing.T) { - s := setup(t) - defer s.Destroy() - - k := kernel.KernelFromContext(s.Ctx) - tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - _, err := testutil.CreateTask(s.Ctx, "name", tc, s.MntNs, s.Root, s.Root) - if err != nil { - t.Fatalf("CreateTask(): %v", err) - } - - collector := s.ListDirents(s.PathOpAtRoot("/proc/1")) - s.AssertAllDirentTypes(collector, taskStaticFiles) -} - -func TestProcSelf(t *testing.T) { - s := setup(t) - defer s.Destroy() - - k := kernel.KernelFromContext(s.Ctx) - tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := testutil.CreateTask(s.Ctx, "name", tc, s.MntNs, s.Root, s.Root) - if err != nil { - t.Fatalf("CreateTask(): %v", err) - } - - collector := s.WithTemporaryContext(task).ListDirents(&vfs.PathOperation{ - Root: s.Root, - Start: s.Root, - Path: fspath.Parse("/proc/self/"), - FollowFinalSymlink: true, - }) - s.AssertAllDirentTypes(collector, taskStaticFiles) -} - -func iterateDir(ctx context.Context, t *testing.T, s *testutil.System, fd *vfs.FileDescription) { - t.Logf("Iterating: %s", fd.MappedName(ctx)) - - var collector testutil.DirentCollector - if err := fd.IterDirents(ctx, &collector); err != nil { - t.Fatalf("IterDirents(): %v", err) - } - if err := collector.Contains(".", linux.DT_DIR); err != nil { - t.Error(err.Error()) - } - if err := collector.Contains("..", linux.DT_DIR); err != nil { - t.Error(err.Error()) - } - - for _, d := range collector.Dirents() { - if d.Name == "." || d.Name == ".." { - continue - } - childPath := path.Join(fd.MappedName(ctx), d.Name) - if d.Type == linux.DT_LNK { - link, err := s.VFS.ReadlinkAt( - ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: s.Root, Start: s.Root, Path: fspath.Parse(childPath)}, - ) - if err != nil { - t.Errorf("vfsfs.ReadlinkAt(%v) failed: %v", childPath, err) - } else { - t.Logf("Skipping symlink: /proc%s => %s", childPath, link) - } - continue - } - - t.Logf("Opening: /proc%s", childPath) - child, err := s.VFS.OpenAt( - ctx, - auth.CredentialsFromContext(ctx), - &vfs.PathOperation{Root: s.Root, Start: s.Root, Path: fspath.Parse(childPath)}, - &vfs.OpenOptions{}, - ) - if err != nil { - t.Errorf("vfsfs.OpenAt(%v) failed: %v", childPath, err) - continue - } - defer child.DecRef() - stat, err := child.Stat(ctx, vfs.StatOptions{}) - if err != nil { - t.Errorf("Stat(%v) failed: %v", childPath, err) - } - if got := linux.FileMode(stat.Mode).DirentType(); got != d.Type { - t.Errorf("wrong file mode, stat: %v, dirent: %v", got, d.Type) - } - if d.Type == linux.DT_DIR { - // Found another dir, let's do it again! - iterateDir(ctx, t, s, child) - } - } -} - -// TestTree iterates all directories and stats every file. -func TestTree(t *testing.T) { - s := setup(t) - defer s.Destroy() - - k := kernel.KernelFromContext(s.Ctx) - - pop := &vfs.PathOperation{ - Root: s.Root, - Start: s.Root, - Path: fspath.Parse("test-file"), - } - opts := &vfs.OpenOptions{ - Flags: linux.O_RDONLY | linux.O_CREAT, - Mode: 0777, - } - file, err := s.VFS.OpenAt(s.Ctx, s.Creds, pop, opts) - if err != nil { - t.Fatalf("failed to create test file: %v", err) - } - defer file.DecRef() - - var tasks []*kernel.Task - for i := 0; i < 5; i++ { - tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc, s.MntNs, s.Root, s.Root) - if err != nil { - t.Fatalf("CreateTask(): %v", err) - } - // Add file to populate /proc/[pid]/fd and fdinfo directories. - task.FDTable().NewFDVFS2(task, 0, file, kernel.FDFlags{}) - tasks = append(tasks, task) - } - - ctx := tasks[0] - fd, err := s.VFS.OpenAt( - ctx, - auth.CredentialsFromContext(s.Ctx), - &vfs.PathOperation{Root: s.Root, Start: s.Root, Path: fspath.Parse("/proc")}, - &vfs.OpenOptions{}, - ) - if err != nil { - t.Fatalf("vfsfs.OpenAt(/proc) failed: %v", err) - } - iterateDir(ctx, t, s, fd) - fd.DecRef() -} diff --git a/pkg/sentry/fsimpl/sockfs/BUILD b/pkg/sentry/fsimpl/sockfs/BUILD deleted file mode 100644 index 52084ddb5..000000000 --- a/pkg/sentry/fsimpl/sockfs/BUILD +++ /dev/null @@ -1,17 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -licenses(["notice"]) - -go_library( - name = "sockfs", - srcs = ["sockfs.go"], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/sentry/fsimpl/kernfs", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - "//pkg/syserror", - ], -) diff --git a/pkg/sentry/fsimpl/sockfs/sockfs.go b/pkg/sentry/fsimpl/sockfs/sockfs.go index 3f7ad1d65..3f7ad1d65 100644..100755 --- a/pkg/sentry/fsimpl/sockfs/sockfs.go +++ b/pkg/sentry/fsimpl/sockfs/sockfs.go diff --git a/pkg/sentry/fsimpl/sockfs/sockfs_state_autogen.go b/pkg/sentry/fsimpl/sockfs/sockfs_state_autogen.go new file mode 100755 index 000000000..8d5b85e2c --- /dev/null +++ b/pkg/sentry/fsimpl/sockfs/sockfs_state_autogen.go @@ -0,0 +1,3 @@ +// automatically generated by stateify. + +package sockfs diff --git a/pkg/sentry/fsimpl/sys/BUILD b/pkg/sentry/fsimpl/sys/BUILD deleted file mode 100644 index a741e2bb6..000000000 --- a/pkg/sentry/fsimpl/sys/BUILD +++ /dev/null @@ -1,34 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") - -licenses(["notice"]) - -go_library( - name = "sys", - srcs = [ - "sys.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/sentry/fsimpl/kernfs", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - "//pkg/syserror", - ], -) - -go_test( - name = "sys_test", - srcs = ["sys_test.go"], - deps = [ - ":sys", - "//pkg/abi/linux", - "//pkg/sentry/fsimpl/testutil", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - "@com_github_google_go-cmp//cmp:go_default_library", - ], -) diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go deleted file mode 100644 index 5c617270e..000000000 --- a/pkg/sentry/fsimpl/sys/sys.go +++ /dev/null @@ -1,134 +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 sys implements sysfs. -package sys - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// Name is the default filesystem name. -const Name = "sysfs" - -// FilesystemType implements vfs.FilesystemType. -type FilesystemType struct{} - -// filesystem implements vfs.FilesystemImpl. -type filesystem struct { - kernfs.Filesystem -} - -// Name implements vfs.FilesystemType.Name. -func (FilesystemType) Name() string { - return Name -} - -// GetFilesystem implements vfs.FilesystemType.GetFilesystem. -func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - fs := &filesystem{} - fs.Filesystem.Init(vfsObj, &fsType) - k := kernel.KernelFromContext(ctx) - maxCPUCores := k.ApplicationCores() - defaultSysDirMode := linux.FileMode(0755) - - root := fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ - "block": fs.newDir(creds, defaultSysDirMode, nil), - "bus": fs.newDir(creds, defaultSysDirMode, nil), - "class": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ - "power_supply": fs.newDir(creds, defaultSysDirMode, nil), - }), - "dev": fs.newDir(creds, defaultSysDirMode, nil), - "devices": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ - "system": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ - "cpu": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ - "online": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), - "possible": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), - "present": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), - }), - }), - }), - "firmware": fs.newDir(creds, defaultSysDirMode, nil), - "fs": fs.newDir(creds, defaultSysDirMode, nil), - "kernel": fs.newDir(creds, defaultSysDirMode, nil), - "module": fs.newDir(creds, defaultSysDirMode, nil), - "power": fs.newDir(creds, defaultSysDirMode, nil), - }) - return fs.VFSFilesystem(), root.VFSDentry(), nil -} - -// dir implements kernfs.Inode. -type dir struct { - kernfs.InodeAttrs - kernfs.InodeNoDynamicLookup - kernfs.InodeNotSymlink - kernfs.InodeDirectoryNoNewChildren - - kernfs.OrderedChildren - dentry kernfs.Dentry -} - -func (fs *filesystem) newDir(creds *auth.Credentials, mode linux.FileMode, contents map[string]*kernfs.Dentry) *kernfs.Dentry { - d := &dir{} - d.InodeAttrs.Init(creds, fs.NextIno(), linux.ModeDirectory|0755) - d.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) - d.dentry.Init(d) - - d.IncLinks(d.OrderedChildren.Populate(&d.dentry, contents)) - - return &d.dentry -} - -// SetStat implements Inode.SetStat not allowing inode attributes to be changed. -func (*dir) SetStat(context.Context, *vfs.Filesystem, *auth.Credentials, vfs.SetStatOptions) error { - return syserror.EPERM -} - -// Open implements kernfs.Inode.Open. -func (d *dir) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - fd := &kernfs.GenericDirectoryFD{} - if err := fd.Init(rp.Mount(), vfsd, &d.OrderedChildren, &opts); err != nil { - return nil, err - } - return fd.VFSFileDescription(), nil -} - -// cpuFile implements kernfs.Inode. -type cpuFile struct { - kernfs.DynamicBytesFile - maxCores uint -} - -// Generate implements vfs.DynamicBytesSource.Generate. -func (c *cpuFile) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "0-%d", c.maxCores-1) - return nil -} - -func (fs *filesystem) newCPUFile(creds *auth.Credentials, maxCores uint, mode linux.FileMode) *kernfs.Dentry { - c := &cpuFile{maxCores: maxCores} - c.DynamicBytesFile.Init(creds, fs.NextIno(), c, mode) - d := &kernfs.Dentry{} - d.Init(c) - return d -} diff --git a/pkg/sentry/fsimpl/sys/sys_test.go b/pkg/sentry/fsimpl/sys/sys_test.go deleted file mode 100644 index 4b3602d47..000000000 --- a/pkg/sentry/fsimpl/sys/sys_test.go +++ /dev/null @@ -1,89 +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 sys_test - -import ( - "fmt" - "testing" - - "github.com/google/go-cmp/cmp" - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/sys" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -func newTestSystem(t *testing.T) *testutil.System { - k, err := testutil.Boot() - if err != nil { - t.Fatalf("Failed to create test kernel: %v", err) - } - ctx := k.SupervisorContext() - creds := auth.CredentialsFromContext(ctx) - k.VFS().MustRegisterFilesystemType(sys.Name, sys.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - - mns, err := k.VFS().NewMountNamespace(ctx, creds, "", sys.Name, &vfs.GetFilesystemOptions{}) - if err != nil { - t.Fatalf("Failed to create new mount namespace: %v", err) - } - return testutil.NewSystem(ctx, t, k.VFS(), mns) -} - -func TestReadCPUFile(t *testing.T) { - s := newTestSystem(t) - defer s.Destroy() - k := kernel.KernelFromContext(s.Ctx) - maxCPUCores := k.ApplicationCores() - - expected := fmt.Sprintf("0-%d", maxCPUCores-1) - - for _, fname := range []string{"online", "possible", "present"} { - pop := s.PathOpAtRoot(fmt.Sprintf("devices/system/cpu/%s", fname)) - fd, err := s.VFS.OpenAt(s.Ctx, s.Creds, pop, &vfs.OpenOptions{}) - if err != nil { - t.Fatalf("OpenAt(pop:%+v) = %+v failed: %v", pop, fd, err) - } - defer fd.DecRef() - content, err := s.ReadToEnd(fd) - if err != nil { - t.Fatalf("Read failed: %v", err) - } - if diff := cmp.Diff(expected, content); diff != "" { - t.Fatalf("Read returned unexpected data:\n--- want\n+++ got\n%v", diff) - } - } -} - -func TestSysRootContainsExpectedEntries(t *testing.T) { - s := newTestSystem(t) - defer s.Destroy() - pop := s.PathOpAtRoot("/") - s.AssertAllDirentTypes(s.ListDirents(pop), map[string]testutil.DirentType{ - "block": linux.DT_DIR, - "bus": linux.DT_DIR, - "class": linux.DT_DIR, - "dev": linux.DT_DIR, - "devices": linux.DT_DIR, - "firmware": linux.DT_DIR, - "fs": linux.DT_DIR, - "kernel": linux.DT_DIR, - "module": linux.DT_DIR, - "power": linux.DT_DIR, - }) -} diff --git a/pkg/sentry/fsimpl/testutil/BUILD b/pkg/sentry/fsimpl/testutil/BUILD deleted file mode 100644 index 0e4053a46..000000000 --- a/pkg/sentry/fsimpl/testutil/BUILD +++ /dev/null @@ -1,37 +0,0 @@ -load("//tools:defs.bzl", "go_library") - -licenses(["notice"]) - -go_library( - name = "testutil", - testonly = 1, - srcs = [ - "kernel.go", - "testutil.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/cpuid", - "//pkg/fspath", - "//pkg/memutil", - "//pkg/sentry/fsbridge", - "//pkg/sentry/fsimpl/tmpfs", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/sched", - "//pkg/sentry/limits", - "//pkg/sentry/loader", - "//pkg/sentry/mm", - "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", - "//pkg/sentry/platform/kvm", - "//pkg/sentry/platform/ptrace", - "//pkg/sentry/time", - "//pkg/sentry/vfs", - "//pkg/sync", - "//pkg/usermem", - "@com_github_google_go-cmp//cmp:go_default_library", - ], -) diff --git a/pkg/sentry/fsimpl/testutil/kernel.go b/pkg/sentry/fsimpl/testutil/kernel.go deleted file mode 100644 index c16a36cdb..000000000 --- a/pkg/sentry/fsimpl/testutil/kernel.go +++ /dev/null @@ -1,180 +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 testutil - -import ( - "flag" - "fmt" - "os" - "runtime" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/cpuid" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/memutil" - "gvisor.dev/gvisor/pkg/sentry/fsbridge" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/sched" - "gvisor.dev/gvisor/pkg/sentry/limits" - "gvisor.dev/gvisor/pkg/sentry/loader" - "gvisor.dev/gvisor/pkg/sentry/mm" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/time" - "gvisor.dev/gvisor/pkg/sentry/vfs" - - // Platforms are plugable. - _ "gvisor.dev/gvisor/pkg/sentry/platform/kvm" - _ "gvisor.dev/gvisor/pkg/sentry/platform/ptrace" -) - -var ( - platformFlag = flag.String("platform", "ptrace", "specify which platform to use") -) - -// Boot initializes a new bare bones kernel for test. -func Boot() (*kernel.Kernel, error) { - platformCtr, err := platform.Lookup(*platformFlag) - if err != nil { - return nil, fmt.Errorf("platform not found: %v", err) - } - deviceFile, err := platformCtr.OpenDevice() - if err != nil { - return nil, fmt.Errorf("creating platform: %v", err) - } - plat, err := platformCtr.New(deviceFile) - if err != nil { - return nil, fmt.Errorf("creating platform: %v", err) - } - - k := &kernel.Kernel{ - Platform: plat, - } - - mf, err := createMemoryFile() - if err != nil { - return nil, err - } - k.SetMemoryFile(mf) - - // Pass k as the platform since it is savable, unlike the actual platform. - vdso, err := loader.PrepareVDSO(nil, k) - if err != nil { - return nil, fmt.Errorf("creating vdso: %v", err) - } - - // Create timekeeper. - tk, err := kernel.NewTimekeeper(k, vdso.ParamPage.FileRange()) - if err != nil { - return nil, fmt.Errorf("creating timekeeper: %v", err) - } - tk.SetClocks(time.NewCalibratedClocks()) - - creds := auth.NewRootCredentials(auth.NewRootUserNamespace()) - - // Initiate the Kernel object, which is required by the Context passed - // to createVFS in order to mount (among other things) procfs. - if err = k.Init(kernel.InitKernelArgs{ - ApplicationCores: uint(runtime.GOMAXPROCS(-1)), - FeatureSet: cpuid.HostFeatureSet(), - Timekeeper: tk, - RootUserNamespace: creds.UserNamespace, - Vdso: vdso, - RootUTSNamespace: kernel.NewUTSNamespace("hostname", "domain", creds.UserNamespace), - RootIPCNamespace: kernel.NewIPCNamespace(creds.UserNamespace), - RootAbstractSocketNamespace: kernel.NewAbstractSocketNamespace(), - PIDNamespace: kernel.NewRootPIDNamespace(creds.UserNamespace), - }); err != nil { - return nil, fmt.Errorf("initializing kernel: %v", err) - } - - kernel.VFS2Enabled = true - - if err := k.VFS().Init(); err != nil { - return nil, fmt.Errorf("VFS init: %v", err) - } - k.VFS().MustRegisterFilesystemType(tmpfs.Name, &tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - AllowUserList: true, - }) - - ls, err := limits.NewLinuxLimitSet() - if err != nil { - return nil, err - } - tg := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, ls) - k.TestOnly_SetGlobalInit(tg) - - return k, nil -} - -// CreateTask creates a new bare bones task for tests. -func CreateTask(ctx context.Context, name string, tc *kernel.ThreadGroup, mntns *vfs.MountNamespace, root, cwd vfs.VirtualDentry) (*kernel.Task, error) { - k := kernel.KernelFromContext(ctx) - exe, err := newFakeExecutable(ctx, k.VFS(), auth.CredentialsFromContext(ctx), root) - if err != nil { - return nil, err - } - m := mm.NewMemoryManager(k, k, k.SleepForAddressSpaceActivation) - m.SetExecutable(fsbridge.NewVFSFile(exe)) - - config := &kernel.TaskConfig{ - Kernel: k, - ThreadGroup: tc, - TaskContext: &kernel.TaskContext{Name: name, MemoryManager: m}, - Credentials: auth.CredentialsFromContext(ctx), - NetworkNamespace: k.RootNetworkNamespace(), - AllowedCPUMask: sched.NewFullCPUSet(k.ApplicationCores()), - UTSNamespace: kernel.UTSNamespaceFromContext(ctx), - IPCNamespace: kernel.IPCNamespaceFromContext(ctx), - AbstractSocketNamespace: kernel.NewAbstractSocketNamespace(), - MountNamespaceVFS2: mntns, - FSContext: kernel.NewFSContextVFS2(root, cwd, 0022), - FDTable: k.NewFDTable(), - } - return k.TaskSet().NewTask(config) -} - -func newFakeExecutable(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, root vfs.VirtualDentry) (*vfs.FileDescription, error) { - const name = "executable" - pop := &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(name), - } - opts := &vfs.OpenOptions{ - Flags: linux.O_RDONLY | linux.O_CREAT, - Mode: 0777, - } - return vfsObj.OpenAt(ctx, creds, pop, opts) -} - -func createMemoryFile() (*pgalloc.MemoryFile, error) { - const memfileName = "test-memory" - memfd, err := memutil.CreateMemFD(memfileName, 0) - if err != nil { - return nil, fmt.Errorf("error creating memfd: %v", err) - } - memfile := os.NewFile(uintptr(memfd), memfileName) - mf, err := pgalloc.NewMemoryFile(memfile, pgalloc.MemoryFileOpts{}) - if err != nil { - memfile.Close() - return nil, fmt.Errorf("error creating pgalloc.MemoryFile: %v", err) - } - return mf, nil -} diff --git a/pkg/sentry/fsimpl/testutil/testutil.go b/pkg/sentry/fsimpl/testutil/testutil.go deleted file mode 100644 index 0556af877..000000000 --- a/pkg/sentry/fsimpl/testutil/testutil.go +++ /dev/null @@ -1,284 +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 testutil provides common test utilities for kernfs-based -// filesystems. -package testutil - -import ( - "fmt" - "io" - "strings" - "testing" - - "github.com/google/go-cmp/cmp" - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/usermem" -) - -// System represents the context for a single test. -// -// Test systems must be explicitly destroyed with System.Destroy. -type System struct { - t *testing.T - Ctx context.Context - Creds *auth.Credentials - VFS *vfs.VirtualFilesystem - Root vfs.VirtualDentry - MntNs *vfs.MountNamespace -} - -// NewSystem constructs a System. -// -// Precondition: Caller must hold a reference on MntNs, whose ownership -// is transferred to the new System. -func NewSystem(ctx context.Context, t *testing.T, v *vfs.VirtualFilesystem, mns *vfs.MountNamespace) *System { - s := &System{ - t: t, - Ctx: ctx, - Creds: auth.CredentialsFromContext(ctx), - VFS: v, - MntNs: mns, - Root: mns.Root(), - } - return s -} - -// WithSubtest creates a temporary test system with a new test harness, -// referencing all other resources from the original system. This is useful when -// a system is reused for multiple subtests, and the T needs to change for each -// case. Note that this is safe when test cases run in parallel, as all -// resources referenced by the system are immutable, or handle interior -// mutations in a thread-safe manner. -// -// The returned system must not outlive the original and should not be destroyed -// via System.Destroy. -func (s *System) WithSubtest(t *testing.T) *System { - return &System{ - t: t, - Ctx: s.Ctx, - Creds: s.Creds, - VFS: s.VFS, - MntNs: s.MntNs, - Root: s.Root, - } -} - -// WithTemporaryContext constructs a temporary test system with a new context -// ctx. The temporary system borrows all resources and references from the -// original system. The returned temporary system must not outlive the original -// system, and should not be destroyed via System.Destroy. -func (s *System) WithTemporaryContext(ctx context.Context) *System { - return &System{ - t: s.t, - Ctx: ctx, - Creds: s.Creds, - VFS: s.VFS, - MntNs: s.MntNs, - Root: s.Root, - } -} - -// Destroy release resources associated with a test system. -func (s *System) Destroy() { - s.Root.DecRef() - s.MntNs.DecRef() // Reference on MntNs passed to NewSystem. -} - -// ReadToEnd reads the contents of fd until EOF to a string. -func (s *System) ReadToEnd(fd *vfs.FileDescription) (string, error) { - buf := make([]byte, usermem.PageSize) - bufIOSeq := usermem.BytesIOSequence(buf) - opts := vfs.ReadOptions{} - - var content strings.Builder - for { - n, err := fd.Read(s.Ctx, bufIOSeq, opts) - if n == 0 || err != nil { - if err == io.EOF { - err = nil - } - return content.String(), err - } - content.Write(buf[:n]) - } -} - -// PathOpAtRoot constructs a PathOperation with the given path from -// the root of the filesystem. -func (s *System) PathOpAtRoot(path string) *vfs.PathOperation { - return &vfs.PathOperation{ - Root: s.Root, - Start: s.Root, - Path: fspath.Parse(path), - } -} - -// GetDentryOrDie attempts to resolve a dentry referred to by the -// provided path operation. If unsuccessful, the test fails. -func (s *System) GetDentryOrDie(pop *vfs.PathOperation) vfs.VirtualDentry { - vd, err := s.VFS.GetDentryAt(s.Ctx, s.Creds, pop, &vfs.GetDentryOptions{}) - if err != nil { - s.t.Fatalf("GetDentryAt(pop:%+v) failed: %v", pop, err) - } - return vd -} - -// DirentType is an alias for values for linux_dirent64.d_type. -type DirentType = uint8 - -// ListDirents lists the Dirents for a directory at pop. -func (s *System) ListDirents(pop *vfs.PathOperation) *DirentCollector { - fd, err := s.VFS.OpenAt(s.Ctx, s.Creds, pop, &vfs.OpenOptions{Flags: linux.O_RDONLY}) - if err != nil { - s.t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) - } - defer fd.DecRef() - - collector := &DirentCollector{} - if err := fd.IterDirents(s.Ctx, collector); err != nil { - s.t.Fatalf("IterDirent failed: %v", err) - } - return collector -} - -// AssertAllDirentTypes verifies that the set of dirents in collector contains -// exactly the specified set of expected entries. AssertAllDirentTypes respects -// collector.skipDots, and implicitly checks for "." and ".." accordingly. -func (s *System) AssertAllDirentTypes(collector *DirentCollector, expected map[string]DirentType) { - if expected == nil { - expected = make(map[string]DirentType) - } - // Also implicitly check for "." and "..", if enabled. - if !collector.skipDots { - expected["."] = linux.DT_DIR - expected[".."] = linux.DT_DIR - } - - dentryTypes := make(map[string]DirentType) - collector.mu.Lock() - for _, dirent := range collector.dirents { - dentryTypes[dirent.Name] = dirent.Type - } - collector.mu.Unlock() - if diff := cmp.Diff(expected, dentryTypes); diff != "" { - s.t.Fatalf("IterDirent had unexpected results:\n--- want\n+++ got\n%v", diff) - } -} - -// AssertDirentOffsets verifies that collector contains at least the entries -// specified in expected, with the given NextOff field. Entries specified in -// expected but missing from collector result in failure. Extra entries in -// collector are ignored. AssertDirentOffsets respects collector.skipDots, and -// implicitly checks for "." and ".." accordingly. -func (s *System) AssertDirentOffsets(collector *DirentCollector, expected map[string]int64) { - // Also implicitly check for "." and "..", if enabled. - if !collector.skipDots { - expected["."] = 1 - expected[".."] = 2 - } - - dentryNextOffs := make(map[string]int64) - collector.mu.Lock() - for _, dirent := range collector.dirents { - // Ignore extra entries in dentries that are not in expected. - if _, ok := expected[dirent.Name]; ok { - dentryNextOffs[dirent.Name] = dirent.NextOff - } - } - collector.mu.Unlock() - if diff := cmp.Diff(expected, dentryNextOffs); diff != "" { - s.t.Fatalf("IterDirent had unexpected results:\n--- want\n+++ got\n%v", diff) - } -} - -// DirentCollector provides an implementation for vfs.IterDirentsCallback for -// testing. It simply iterates to the end of a given directory FD and collects -// all dirents emitted by the callback. -type DirentCollector struct { - mu sync.Mutex - order []*vfs.Dirent - dirents map[string]*vfs.Dirent - // When the collector is used in various Assert* functions, should "." and - // ".." be implicitly checked? - skipDots bool -} - -// SkipDotsChecks enables or disables the implicit checks on "." and ".." when -// the collector is used in various Assert* functions. Note that "." and ".." -// are still collected if passed to d.Handle, so the caller should only disable -// the checks when they aren't expected. -func (d *DirentCollector) SkipDotsChecks(value bool) { - d.skipDots = value -} - -// Handle implements vfs.IterDirentsCallback.Handle. -func (d *DirentCollector) Handle(dirent vfs.Dirent) error { - d.mu.Lock() - if d.dirents == nil { - d.dirents = make(map[string]*vfs.Dirent) - } - d.order = append(d.order, &dirent) - d.dirents[dirent.Name] = &dirent - d.mu.Unlock() - return nil -} - -// Count returns the number of dirents currently in the collector. -func (d *DirentCollector) Count() int { - d.mu.Lock() - defer d.mu.Unlock() - return len(d.dirents) -} - -// Contains checks whether the collector has a dirent with the given name and -// type. -func (d *DirentCollector) Contains(name string, typ uint8) error { - d.mu.Lock() - defer d.mu.Unlock() - dirent, ok := d.dirents[name] - if !ok { - return fmt.Errorf("No dirent named %q found", name) - } - if dirent.Type != typ { - return fmt.Errorf("Dirent named %q found, but was expecting type %s, got: %+v", name, linux.DirentType.Parse(uint64(typ)), dirent) - } - return nil -} - -// Dirents returns all dirents discovered by this collector. -func (d *DirentCollector) Dirents() map[string]*vfs.Dirent { - d.mu.Lock() - dirents := make(map[string]*vfs.Dirent) - for n, d := range d.dirents { - dirents[n] = d - } - d.mu.Unlock() - return dirents -} - -// OrderedDirents returns an ordered list of dirents as discovered by this -// collector. -func (d *DirentCollector) OrderedDirents() []*vfs.Dirent { - d.mu.Lock() - dirents := make([]*vfs.Dirent, len(d.order)) - copy(dirents, d.order) - d.mu.Unlock() - return dirents -} diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD deleted file mode 100644 index f2ac23c88..000000000 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ /dev/null @@ -1,100 +0,0 @@ -load("//tools:defs.bzl", "go_library", "go_test") -load("//tools/go_generics:defs.bzl", "go_template_instance") - -licenses(["notice"]) - -go_template_instance( - name = "dentry_list", - out = "dentry_list.go", - package = "tmpfs", - prefix = "dentry", - template = "//pkg/ilist:generic_list", - types = { - "Element": "*dentry", - "Linker": "*dentry", - }, -) - -go_library( - name = "tmpfs", - srcs = [ - "dentry_list.go", - "device_file.go", - "directory.go", - "filesystem.go", - "named_pipe.go", - "regular_file.go", - "socket_file.go", - "symlink.go", - "tmpfs.go", - ], - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/abi/linux", - "//pkg/amutex", - "//pkg/context", - "//pkg/fspath", - "//pkg/log", - "//pkg/safemem", - "//pkg/sentry/arch", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/fs/lock", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/pipe", - "//pkg/sentry/kernel/time", - "//pkg/sentry/memmap", - "//pkg/sentry/pgalloc", - "//pkg/sentry/platform", - "//pkg/sentry/socket/unix/transport", - "//pkg/sentry/usage", - "//pkg/sentry/vfs", - "//pkg/sentry/vfs/lock", - "//pkg/sync", - "//pkg/syserror", - "//pkg/usermem", - ], -) - -go_test( - name = "benchmark_test", - size = "small", - srcs = ["benchmark_test.go"], - deps = [ - ":tmpfs", - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fspath", - "//pkg/refs", - "//pkg/sentry/contexttest", - "//pkg/sentry/fs", - "//pkg/sentry/fs/tmpfs", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/vfs", - "//pkg/syserror", - ], -) - -go_test( - name = "tmpfs_test", - size = "small", - srcs = [ - "pipe_test.go", - "regular_file_test.go", - "stat_test.go", - ], - library = ":tmpfs", - deps = [ - "//pkg/abi/linux", - "//pkg/context", - "//pkg/fspath", - "//pkg/sentry/contexttest", - "//pkg/sentry/fs/lock", - "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/contexttest", - "//pkg/sentry/vfs", - "//pkg/syserror", - "//pkg/usermem", - ], -) diff --git a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go deleted file mode 100644 index 383133e44..000000000 --- a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go +++ /dev/null @@ -1,493 +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 benchmark_test - -import ( - "fmt" - "runtime" - "strings" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/fs" - _ "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// Differences from stat_benchmark: -// -// - Syscall interception, CopyInPath, copyOutStat, and overlayfs overheads are -// not included. -// -// - *MountStat benchmarks use a tmpfs root mount and a tmpfs submount at /tmp. -// Non-MountStat benchmarks use a tmpfs root mount and no submounts. -// stat_benchmark uses a varying root mount, a tmpfs submount at /tmp, and a -// subdirectory /tmp/<top_dir> (assuming TEST_TMPDIR == "/tmp"). Thus -// stat_benchmark at depth 1 does a comparable amount of work to *MountStat -// benchmarks at depth 2, and non-MountStat benchmarks at depth 3. -var depths = []int{1, 2, 3, 8, 64, 100} - -const ( - mountPointName = "tmp" - filename = "gvisor_test_temp_0_1557494568" -) - -// This is copied from syscalls/linux/sys_file.go, with the dependency on -// kernel.Task stripped out. -func fileOpOn(ctx context.Context, mntns *fs.MountNamespace, root, wd *fs.Dirent, dirFD int32, path string, resolve bool, fn func(root *fs.Dirent, d *fs.Dirent) error) error { - var ( - d *fs.Dirent // The file. - rel *fs.Dirent // The relative directory for search (if required.) - err error - ) - - // Extract the working directory (maybe). - if len(path) > 0 && path[0] == '/' { - // Absolute path; rel can be nil. - } else if dirFD == linux.AT_FDCWD { - // Need to reference the working directory. - rel = wd - } else { - // Need to extract the given FD. - return syserror.EBADF - } - - // Lookup the node. - remainingTraversals := uint(linux.MaxSymlinkTraversals) - if resolve { - d, err = mntns.FindInode(ctx, root, rel, path, &remainingTraversals) - } else { - d, err = mntns.FindLink(ctx, root, rel, path, &remainingTraversals) - } - if err != nil { - return err - } - - err = fn(root, d) - d.DecRef() - return err -} - -func BenchmarkVFS1TmpfsStat(b *testing.B) { - for _, depth := range depths { - b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { - ctx := contexttest.Context(b) - - // Create VFS. - tmpfsFS, ok := fs.FindFilesystem("tmpfs") - if !ok { - b.Fatalf("failed to find tmpfs filesystem type") - } - rootInode, err := tmpfsFS.Mount(ctx, "tmpfs", fs.MountSourceFlags{}, "", nil) - if err != nil { - b.Fatalf("failed to create tmpfs root mount: %v", err) - } - mntns, err := fs.NewMountNamespace(ctx, rootInode) - if err != nil { - b.Fatalf("failed to create mount namespace: %v", err) - } - defer mntns.DecRef() - - var filePathBuilder strings.Builder - filePathBuilder.WriteByte('/') - - // Create nested directories with given depth. - root := mntns.Root() - defer root.DecRef() - d := root - d.IncRef() - defer d.DecRef() - for i := depth; i > 0; i-- { - name := fmt.Sprintf("%d", i) - if err := d.Inode.CreateDirectory(ctx, d, name, fs.FilePermsFromMode(0755)); err != nil { - b.Fatalf("failed to create directory %q: %v", name, err) - } - next, err := d.Walk(ctx, root, name) - if err != nil { - b.Fatalf("failed to walk to directory %q: %v", name, err) - } - d.DecRef() - d = next - filePathBuilder.WriteString(name) - filePathBuilder.WriteByte('/') - } - - // Create the file that will be stat'd. - file, err := d.Inode.Create(ctx, d, filename, fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0644)) - if err != nil { - b.Fatalf("failed to create file %q: %v", filename, err) - } - file.DecRef() - filePathBuilder.WriteString(filename) - filePath := filePathBuilder.String() - - dirPath := false - runtime.GC() - b.ResetTimer() - for i := 0; i < b.N; i++ { - err := fileOpOn(ctx, mntns, root, root, linux.AT_FDCWD, filePath, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent) error { - if dirPath && !fs.IsDir(d.Inode.StableAttr) { - return syserror.ENOTDIR - } - uattr, err := d.Inode.UnstableAttr(ctx) - if err != nil { - return err - } - // Sanity check. - if uattr.Perms.User.Execute { - b.Fatalf("got wrong permissions (%0o)", uattr.Perms.LinuxMode()) - } - return nil - }) - if err != nil { - b.Fatalf("stat(%q) failed: %v", filePath, err) - } - } - // Don't include deferred cleanup in benchmark time. - b.StopTimer() - }) - } -} - -func BenchmarkVFS2MemfsStat(b *testing.B) { - for _, depth := range depths { - b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { - ctx := contexttest.Context(b) - creds := auth.CredentialsFromContext(ctx) - - // Create VFS. - vfsObj := vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - b.Fatalf("VFS init: %v", err) - } - vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{}) - if err != nil { - b.Fatalf("failed to create tmpfs root mount: %v", err) - } - defer mntns.DecRef() - - var filePathBuilder strings.Builder - filePathBuilder.WriteByte('/') - - // Create nested directories with given depth. - root := mntns.Root() - defer root.DecRef() - vd := root - vd.IncRef() - for i := depth; i > 0; i-- { - name := fmt.Sprintf("%d", i) - pop := vfs.PathOperation{ - Root: root, - Start: vd, - Path: fspath.Parse(name), - } - if err := vfsObj.MkdirAt(ctx, creds, &pop, &vfs.MkdirOptions{ - Mode: 0755, - }); err != nil { - b.Fatalf("failed to create directory %q: %v", name, err) - } - nextVD, err := vfsObj.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) - if err != nil { - b.Fatalf("failed to walk to directory %q: %v", name, err) - } - vd.DecRef() - vd = nextVD - filePathBuilder.WriteString(name) - filePathBuilder.WriteByte('/') - } - - // Create the file that will be stat'd. - fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: vd, - Path: fspath.Parse(filename), - FollowFinalSymlink: true, - }, &vfs.OpenOptions{ - Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, - Mode: 0644, - }) - vd.DecRef() - vd = vfs.VirtualDentry{} - if err != nil { - b.Fatalf("failed to create file %q: %v", filename, err) - } - defer fd.DecRef() - filePathBuilder.WriteString(filename) - filePath := filePathBuilder.String() - - runtime.GC() - b.ResetTimer() - for i := 0; i < b.N; i++ { - stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(filePath), - FollowFinalSymlink: true, - }, &vfs.StatOptions{}) - if err != nil { - b.Fatalf("stat(%q) failed: %v", filePath, err) - } - // Sanity check. - if stat.Mode&^linux.S_IFMT != 0644 { - b.Fatalf("got wrong permissions (%0o)", stat.Mode) - } - } - // Don't include deferred cleanup in benchmark time. - b.StopTimer() - }) - } -} - -func BenchmarkVFS1TmpfsMountStat(b *testing.B) { - for _, depth := range depths { - b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { - ctx := contexttest.Context(b) - - // Create VFS. - tmpfsFS, ok := fs.FindFilesystem("tmpfs") - if !ok { - b.Fatalf("failed to find tmpfs filesystem type") - } - rootInode, err := tmpfsFS.Mount(ctx, "tmpfs", fs.MountSourceFlags{}, "", nil) - if err != nil { - b.Fatalf("failed to create tmpfs root mount: %v", err) - } - mntns, err := fs.NewMountNamespace(ctx, rootInode) - if err != nil { - b.Fatalf("failed to create mount namespace: %v", err) - } - defer mntns.DecRef() - - var filePathBuilder strings.Builder - filePathBuilder.WriteByte('/') - - // Create and mount the submount. - root := mntns.Root() - defer root.DecRef() - if err := root.Inode.CreateDirectory(ctx, root, mountPointName, fs.FilePermsFromMode(0755)); err != nil { - b.Fatalf("failed to create mount point: %v", err) - } - mountPoint, err := root.Walk(ctx, root, mountPointName) - if err != nil { - b.Fatalf("failed to walk to mount point: %v", err) - } - defer mountPoint.DecRef() - submountInode, err := tmpfsFS.Mount(ctx, "tmpfs", fs.MountSourceFlags{}, "", nil) - if err != nil { - b.Fatalf("failed to create tmpfs submount: %v", err) - } - if err := mntns.Mount(ctx, mountPoint, submountInode); err != nil { - b.Fatalf("failed to mount tmpfs submount: %v", err) - } - filePathBuilder.WriteString(mountPointName) - filePathBuilder.WriteByte('/') - - // Create nested directories with given depth. - d, err := root.Walk(ctx, root, mountPointName) - if err != nil { - b.Fatalf("failed to walk to mount root: %v", err) - } - defer d.DecRef() - for i := depth; i > 0; i-- { - name := fmt.Sprintf("%d", i) - if err := d.Inode.CreateDirectory(ctx, d, name, fs.FilePermsFromMode(0755)); err != nil { - b.Fatalf("failed to create directory %q: %v", name, err) - } - next, err := d.Walk(ctx, root, name) - if err != nil { - b.Fatalf("failed to walk to directory %q: %v", name, err) - } - d.DecRef() - d = next - filePathBuilder.WriteString(name) - filePathBuilder.WriteByte('/') - } - - // Create the file that will be stat'd. - file, err := d.Inode.Create(ctx, d, filename, fs.FileFlags{Read: true, Write: true}, fs.FilePermsFromMode(0644)) - if err != nil { - b.Fatalf("failed to create file %q: %v", filename, err) - } - file.DecRef() - filePathBuilder.WriteString(filename) - filePath := filePathBuilder.String() - - dirPath := false - runtime.GC() - b.ResetTimer() - for i := 0; i < b.N; i++ { - err := fileOpOn(ctx, mntns, root, root, linux.AT_FDCWD, filePath, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent) error { - if dirPath && !fs.IsDir(d.Inode.StableAttr) { - return syserror.ENOTDIR - } - uattr, err := d.Inode.UnstableAttr(ctx) - if err != nil { - return err - } - // Sanity check. - if uattr.Perms.User.Execute { - b.Fatalf("got wrong permissions (%0o)", uattr.Perms.LinuxMode()) - } - return nil - }) - if err != nil { - b.Fatalf("stat(%q) failed: %v", filePath, err) - } - } - // Don't include deferred cleanup in benchmark time. - b.StopTimer() - }) - } -} - -func BenchmarkVFS2MemfsMountStat(b *testing.B) { - for _, depth := range depths { - b.Run(fmt.Sprintf("%d", depth), func(b *testing.B) { - ctx := contexttest.Context(b) - creds := auth.CredentialsFromContext(ctx) - - // Create VFS. - vfsObj := vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - b.Fatalf("VFS init: %v", err) - } - vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{}) - if err != nil { - b.Fatalf("failed to create tmpfs root mount: %v", err) - } - defer mntns.DecRef() - - var filePathBuilder strings.Builder - filePathBuilder.WriteByte('/') - - // Create the mount point. - root := mntns.Root() - defer root.DecRef() - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(mountPointName), - } - if err := vfsObj.MkdirAt(ctx, creds, &pop, &vfs.MkdirOptions{ - Mode: 0755, - }); err != nil { - b.Fatalf("failed to create mount point: %v", err) - } - // Save the mount point for later use. - mountPoint, err := vfsObj.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) - if err != nil { - b.Fatalf("failed to walk to mount point: %v", err) - } - defer mountPoint.DecRef() - // Create and mount the submount. - if err := vfsObj.MountAt(ctx, creds, "", &pop, "tmpfs", &vfs.MountOptions{}); err != nil { - b.Fatalf("failed to mount tmpfs submount: %v", err) - } - filePathBuilder.WriteString(mountPointName) - filePathBuilder.WriteByte('/') - - // Create nested directories with given depth. - vd, err := vfsObj.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) - if err != nil { - b.Fatalf("failed to walk to mount root: %v", err) - } - for i := depth; i > 0; i-- { - name := fmt.Sprintf("%d", i) - pop := vfs.PathOperation{ - Root: root, - Start: vd, - Path: fspath.Parse(name), - } - if err := vfsObj.MkdirAt(ctx, creds, &pop, &vfs.MkdirOptions{ - Mode: 0755, - }); err != nil { - b.Fatalf("failed to create directory %q: %v", name, err) - } - nextVD, err := vfsObj.GetDentryAt(ctx, creds, &pop, &vfs.GetDentryOptions{}) - if err != nil { - b.Fatalf("failed to walk to directory %q: %v", name, err) - } - vd.DecRef() - vd = nextVD - filePathBuilder.WriteString(name) - filePathBuilder.WriteByte('/') - } - - // Verify that we didn't create any directories under the mount - // point (i.e. they were all created on the submount). - firstDirName := fmt.Sprintf("%d", depth) - if child := mountPoint.Dentry().Child(firstDirName); child != nil { - b.Fatalf("created directory %q under root mount, not submount", firstDirName) - } - - // Create the file that will be stat'd. - fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: vd, - Path: fspath.Parse(filename), - FollowFinalSymlink: true, - }, &vfs.OpenOptions{ - Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, - Mode: 0644, - }) - vd.DecRef() - if err != nil { - b.Fatalf("failed to create file %q: %v", filename, err) - } - fd.DecRef() - filePathBuilder.WriteString(filename) - filePath := filePathBuilder.String() - - runtime.GC() - b.ResetTimer() - for i := 0; i < b.N; i++ { - stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(filePath), - FollowFinalSymlink: true, - }, &vfs.StatOptions{}) - if err != nil { - b.Fatalf("stat(%q) failed: %v", filePath, err) - } - // Sanity check. - if stat.Mode&^linux.S_IFMT != 0644 { - b.Fatalf("got wrong permissions (%0o)", stat.Mode) - } - } - // Don't include deferred cleanup in benchmark time. - b.StopTimer() - }) - } -} - -func init() { - // Turn off reference leak checking for a fair comparison between vfs1 and - // vfs2. - refs.SetLeakMode(refs.NoLeakChecking) -} diff --git a/pkg/sentry/fsimpl/tmpfs/device_file.go b/pkg/sentry/fsimpl/tmpfs/device_file.go deleted file mode 100644 index 83bf885ee..000000000 --- a/pkg/sentry/fsimpl/tmpfs/device_file.go +++ /dev/null @@ -1,49 +0,0 @@ -// Copyright 2020 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 ( - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -type deviceFile struct { - inode inode - kind vfs.DeviceKind - major uint32 - minor uint32 -} - -func (fs *filesystem) newDeviceFile(creds *auth.Credentials, mode linux.FileMode, kind vfs.DeviceKind, major, minor uint32) *inode { - file := &deviceFile{ - kind: kind, - major: major, - minor: minor, - } - switch kind { - case vfs.BlockDevice: - mode |= linux.S_IFBLK - case vfs.CharDevice: - mode |= linux.S_IFCHR - default: - panic(fmt.Sprintf("invalid DeviceKind: %v", kind)) - } - file.inode.init(file, fs, creds, mode) - file.inode.nlink = 1 // from parent directory - return &file.inode -} diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go deleted file mode 100644 index 45712c9b9..000000000 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ /dev/null @@ -1,184 +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 tmpfs - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -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 protected by filesystem.mu. - childList dentryList -} - -func (fs *filesystem) newDirectory(creds *auth.Credentials, mode linux.FileMode) *inode { - dir := &directory{} - dir.inode.init(dir, fs, creds, linux.S_IFDIR|mode) - dir.inode.nlink = 2 // from "." and parent directory or ".." for root - return &dir.inode -} - -type directoryFD struct { - fileDescription - vfs.DirectoryFileDescriptionDefaultImpl - - // Protected by filesystem.mu. - iter *dentry - off int64 -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *directoryFD) Release() { - if fd.iter != nil { - fs := fd.filesystem() - dir := fd.inode().impl.(*directory) - fs.mu.Lock() - dir.childList.Remove(fd.iter) - fs.mu.Unlock() - fd.iter = nil - } -} - -// IterDirents implements vfs.FileDescriptionImpl.IterDirents. -func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback) error { - fs := fd.filesystem() - vfsd := fd.vfsfd.VirtualDentry().Dentry() - - fs.mu.Lock() - defer fs.mu.Unlock() - - fd.inode().touchAtime(fd.vfsfd.Mount()) - - if fd.off == 0 { - if err := cb.Handle(vfs.Dirent{ - Name: ".", - Type: linux.DT_DIR, - Ino: vfsd.Impl().(*dentry).inode.ino, - NextOff: 1, - }); err != nil { - return err - } - fd.off++ - } - if fd.off == 1 { - parentInode := vfsd.ParentOrSelf().Impl().(*dentry).inode - if err := cb.Handle(vfs.Dirent{ - Name: "..", - Type: parentInode.direntType(), - Ino: parentInode.ino, - NextOff: 2, - }); err != nil { - return err - } - fd.off++ - } - - dir := vfsd.Impl().(*dentry).inode.impl.(*directory) - var child *dentry - if fd.iter == nil { - // Start iteration at the beginning of dir. - child = dir.childList.Front() - fd.iter = &dentry{} - } else { - // Continue iteration from where we left off. - child = fd.iter.Next() - dir.childList.Remove(fd.iter) - } - for child != nil { - // Skip other directoryFD iterators. - if child.inode != nil { - if err := cb.Handle(vfs.Dirent{ - Name: child.vfsd.Name(), - Type: child.inode.direntType(), - Ino: child.inode.ino, - NextOff: fd.off + 1, - }); err != nil { - dir.childList.InsertBefore(child, fd.iter) - return err - } - fd.off++ - } - child = child.Next() - } - dir.childList.PushBack(fd.iter) - return nil -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *directoryFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - fs := fd.filesystem() - fs.mu.Lock() - defer fs.mu.Unlock() - - switch whence { - case linux.SEEK_SET: - // Use offset as given. - case linux.SEEK_CUR: - offset += fd.off - default: - return 0, syserror.EINVAL - } - if offset < 0 { - return 0, syserror.EINVAL - } - - // If the offset isn't changing (e.g. due to lseek(0, SEEK_CUR)), don't - // seek even if doing so might reposition the iterator due to concurrent - // mutation of the directory. Compare fs/libfs.c:dcache_dir_lseek(). - if fd.off == offset { - return offset, nil - } - - fd.off = offset - // Compensate for "." and "..". - remChildren := int64(0) - if offset >= 2 { - remChildren = offset - 2 - } - - dir := fd.inode().impl.(*directory) - - // Ensure that fd.iter exists and is not linked into dir.childList. - if fd.iter == nil { - fd.iter = &dentry{} - } else { - dir.childList.Remove(fd.iter) - } - // Insert fd.iter before the remChildren'th child, or at the end of the - // list if remChildren >= number of children. - child := dir.childList.Front() - for child != nil { - // Skip other directoryFD iterators. - if child.inode != nil { - if remChildren == 0 { - dir.childList.InsertBefore(child, fd.iter) - return offset, nil - } - remChildren-- - } - child = child.Next() - } - dir.childList.PushBack(fd.iter) - return offset, nil -} diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go deleted file mode 100644 index 5339d7072..000000000 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ /dev/null @@ -1,751 +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 tmpfs - -import ( - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" -) - -// Sync implements vfs.FilesystemImpl.Sync. -func (fs *filesystem) Sync(ctx context.Context) error { - // All filesystem state is in-memory. - return nil -} - -// stepLocked resolves rp.Component() to an existing file, starting from the -// given directory. -// -// stepLocked is loosely analogous to fs/namei.c:walk_component(). -// -// Preconditions: filesystem.mu must be locked. !rp.Done(). -func stepLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { - if !d.inode.isDir() { - return nil, syserror.ENOTDIR - } - if err := d.inode.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { - return nil, err - } -afterSymlink: - if len(rp.Component()) > linux.NAME_MAX { - return nil, syserror.ENAMETOOLONG - } - nextVFSD, err := rp.ResolveComponent(&d.vfsd) - if err != nil { - return nil, err - } - if nextVFSD == nil { - // Since the Dentry tree is the sole source of truth for tmpfs, if it's - // not in the Dentry tree, it doesn't exist. - return nil, syserror.ENOENT - } - next := nextVFSD.Impl().(*dentry) - if symlink, ok := next.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { - // TODO(gvisor.dev/issue/1197): Symlink traversals updates - // access time. - if err := rp.HandleSymlink(symlink.target); err != nil { - return nil, err - } - goto afterSymlink // don't check the current directory again - } - rp.Advance() - return next, nil -} - -// walkParentDirLocked resolves all but the last path component of rp to an -// existing directory, starting from the given directory (which is usually -// rp.Start().Impl().(*dentry)). It does not check that the returned directory -// is searchable by the provider of rp. -// -// walkParentDirLocked is loosely analogous to Linux's -// fs/namei.c:path_parentat(). -// -// Preconditions: filesystem.mu must be locked. !rp.Done(). -func walkParentDirLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { - for !rp.Final() { - next, err := stepLocked(rp, d) - if err != nil { - return nil, err - } - d = next - } - if !d.inode.isDir() { - return nil, syserror.ENOTDIR - } - return d, nil -} - -// resolveLocked resolves rp to an existing file. -// -// resolveLocked is loosely analogous to Linux's fs/namei.c:path_lookupat(). -// -// Preconditions: filesystem.mu must be locked. -func resolveLocked(rp *vfs.ResolvingPath) (*dentry, error) { - d := rp.Start().Impl().(*dentry) - for !rp.Done() { - next, err := stepLocked(rp, d) - if err != nil { - return nil, err - } - d = next - } - if rp.MustBeDir() && !d.inode.isDir() { - return nil, syserror.ENOTDIR - } - return d, nil -} - -// doCreateAt checks that creating a file at rp is permitted, then invokes -// create to do so. -// -// doCreateAt is loosely analogous to a conjunction of Linux's -// fs/namei.c:filename_create() and done_path_create(). -// -// Preconditions: !rp.Done(). For the final path component in rp, -// !rp.ShouldFollowSymlink(). -func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(parent *dentry, name string) error) error { - fs.mu.Lock() - defer fs.mu.Unlock() - parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) - if err != nil { - return err - } - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - name := rp.Component() - if name == "." || name == ".." { - return syserror.EEXIST - } - if len(name) > linux.NAME_MAX { - return syserror.ENAMETOOLONG - } - // Call parent.vfsd.Child() instead of stepLocked() or rp.ResolveChild(), - // because if the child exists we want to return EEXIST immediately instead - // of attempting symlink/mount traversal. - if parent.vfsd.Child(name) != nil { - return syserror.EEXIST - } - if !dir && rp.MustBeDir() { - return syserror.ENOENT - } - // In memfs, the only way to cause a dentry to be disowned is by removing - // it from the filesystem, so this check is equivalent to checking if - // parent has been removed. - if parent.vfsd.IsDisowned() { - return syserror.ENOENT - } - mnt := rp.Mount() - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - if err := create(parent, name); err != nil { - return err - } - parent.inode.touchCMtime() - return nil -} - -// AccessAt implements vfs.Filesystem.Impl.AccessAt. -func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := resolveLocked(rp) - if err != nil { - return err - } - return d.inode.checkPermissions(creds, ats) -} - -// GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. -func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := resolveLocked(rp) - if err != nil { - return nil, err - } - if opts.CheckSearchable { - if !d.inode.isDir() { - return nil, syserror.ENOTDIR - } - if err := d.inode.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { - return nil, err - } - } - d.IncRef() - return &d.vfsd, nil -} - -// GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. -func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) - if err != nil { - return nil, err - } - d.IncRef() - return &d.vfsd, nil -} - -// LinkAt implements vfs.FilesystemImpl.LinkAt. -func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { - return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { - if rp.Mount() != vd.Mount() { - return syserror.EXDEV - } - d := vd.Dentry().Impl().(*dentry) - if d.inode.isDir() { - return syserror.EPERM - } - if d.inode.nlink == 0 { - return syserror.ENOENT - } - if d.inode.nlink == maxLinks { - return syserror.EMLINK - } - d.inode.incLinksLocked() - child := fs.newDentry(d.inode) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - return nil - }) -} - -// MkdirAt implements vfs.FilesystemImpl.MkdirAt. -func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { - return fs.doCreateAt(rp, true /* dir */, func(parent *dentry, name string) error { - if parent.inode.nlink == maxLinks { - return syserror.EMLINK - } - parent.inode.incLinksLocked() // from child's ".." - child := fs.newDentry(fs.newDirectory(rp.Credentials(), opts.Mode)) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - return nil - }) -} - -// MknodAt implements vfs.FilesystemImpl.MknodAt. -func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { - return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { - var childInode *inode - switch opts.Mode.FileType() { - case 0, linux.S_IFREG: - childInode = fs.newRegularFile(rp.Credentials(), opts.Mode) - case linux.S_IFIFO: - childInode = fs.newNamedPipe(rp.Credentials(), opts.Mode) - case linux.S_IFBLK: - childInode = fs.newDeviceFile(rp.Credentials(), opts.Mode, vfs.BlockDevice, opts.DevMajor, opts.DevMinor) - case linux.S_IFCHR: - childInode = fs.newDeviceFile(rp.Credentials(), opts.Mode, vfs.CharDevice, opts.DevMajor, opts.DevMinor) - case linux.S_IFSOCK: - childInode = fs.newSocketFile(rp.Credentials(), opts.Mode, opts.Endpoint) - default: - return syserror.EINVAL - } - child := fs.newDentry(childInode) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - return nil - }) -} - -// OpenAt implements vfs.FilesystemImpl.OpenAt. -func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - if opts.Flags&linux.O_TMPFILE != 0 { - // Not yet supported. - return nil, syserror.EOPNOTSUPP - } - - // Handle O_CREAT and !O_CREAT separately, since in the latter case we - // don't need fs.mu for writing. - if opts.Flags&linux.O_CREAT == 0 { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := resolveLocked(rp) - if err != nil { - return nil, err - } - return d.open(ctx, rp, &opts, false /* afterCreate */) - } - - mustCreate := opts.Flags&linux.O_EXCL != 0 - start := rp.Start().Impl().(*dentry) - fs.mu.Lock() - defer fs.mu.Unlock() - if rp.Done() { - // Reject attempts to open directories with O_CREAT. - if rp.MustBeDir() { - return nil, syserror.EISDIR - } - if mustCreate { - return nil, syserror.EEXIST - } - return start.open(ctx, rp, &opts, false /* afterCreate */) - } -afterTrailingSymlink: - parent, err := walkParentDirLocked(rp, start) - if err != nil { - return nil, err - } - // Check for search permission in the parent directory. - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayExec); err != nil { - return nil, err - } - // Reject attempts to open directories with O_CREAT. - if rp.MustBeDir() { - return nil, syserror.EISDIR - } - name := rp.Component() - if name == "." || name == ".." { - return nil, syserror.EISDIR - } - // Determine whether or not we need to create a file. - child, err := stepLocked(rp, parent) - if err == syserror.ENOENT { - // Already checked for searchability above; now check for writability. - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { - return nil, err - } - if err := rp.Mount().CheckBeginWrite(); err != nil { - return nil, err - } - defer rp.Mount().EndWrite() - // Create and open the child. - child := fs.newDentry(fs.newRegularFile(rp.Credentials(), opts.Mode)) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - fd, err := child.open(ctx, rp, &opts, true) - if err != nil { - return nil, err - } - parent.inode.touchCMtime() - return fd, nil - } - if err != nil { - return nil, err - } - // Do we need to resolve a trailing symlink? - if !rp.Done() { - start = parent - goto afterTrailingSymlink - } - // Open existing file. - if mustCreate { - return nil, syserror.EEXIST - } - return child.open(ctx, rp, &opts, false) -} - -func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.OpenOptions, afterCreate bool) (*vfs.FileDescription, error) { - ats := vfs.AccessTypesForOpenFlags(opts) - if !afterCreate { - if err := d.inode.checkPermissions(rp.Credentials(), ats); err != nil { - return nil, err - } - } - switch impl := d.inode.impl.(type) { - case *regularFile: - var fd regularFileFD - if err := fd.vfsfd.Init(&fd, opts.Flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { - return nil, err - } - if opts.Flags&linux.O_TRUNC != 0 { - if _, err := impl.truncate(0); err != nil { - return nil, err - } - } - return &fd.vfsfd, nil - case *directory: - // Can't open directories writably. - if ats&vfs.MayWrite != 0 { - return nil, syserror.EISDIR - } - var fd directoryFD - if err := fd.vfsfd.Init(&fd, opts.Flags, rp.Mount(), &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil { - return nil, err - } - return &fd.vfsfd, nil - case *symlink: - // Can't open symlinks without O_PATH (which is unimplemented). - return nil, syserror.ELOOP - case *namedPipe: - return newNamedPipeFD(ctx, impl, rp, &d.vfsd, opts.Flags) - case *deviceFile: - return rp.VirtualFilesystem().OpenDeviceSpecialFile(ctx, rp.Mount(), &d.vfsd, impl.kind, impl.major, impl.minor, opts) - case *socketFile: - return nil, syserror.ENXIO - default: - panic(fmt.Sprintf("unknown inode type: %T", d.inode.impl)) - } -} - -// ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. -func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := resolveLocked(rp) - if err != nil { - return "", err - } - symlink, ok := d.inode.impl.(*symlink) - if !ok { - return "", syserror.EINVAL - } - symlink.inode.touchAtime(rp.Mount()) - return symlink.target, nil -} - -// RenameAt implements vfs.FilesystemImpl.RenameAt. -func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldParentVD vfs.VirtualDentry, oldName string, opts vfs.RenameOptions) error { - if opts.Flags != 0 { - // TODO(b/145974740): Support renameat2 flags. - return syserror.EINVAL - } - - // Resolve newParent first to verify that it's on this Mount. - fs.mu.Lock() - defer fs.mu.Unlock() - newParent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) - if err != nil { - return err - } - newName := rp.Component() - if newName == "." || newName == ".." { - return syserror.EBUSY - } - mnt := rp.Mount() - if mnt != oldParentVD.Mount() { - return syserror.EXDEV - } - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - - oldParent := oldParentVD.Dentry().Impl().(*dentry) - if err := oldParent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - // Call vfs.Dentry.Child() instead of stepLocked() or rp.ResolveChild(), - // because if the existing child is a symlink or mount point then we want - // to rename over it rather than follow it. - renamedVFSD := oldParent.vfsd.Child(oldName) - if renamedVFSD == nil { - return syserror.ENOENT - } - renamed := renamedVFSD.Impl().(*dentry) - if renamed.inode.isDir() { - if renamed == newParent || renamedVFSD.IsAncestorOf(&newParent.vfsd) { - return syserror.EINVAL - } - if oldParent != newParent { - // Writability is needed to change renamed's "..". - if err := renamed.inode.checkPermissions(rp.Credentials(), vfs.MayWrite); err != nil { - return err - } - } - } else { - if opts.MustBeDir || rp.MustBeDir() { - return syserror.ENOTDIR - } - } - - if err := newParent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - replacedVFSD := newParent.vfsd.Child(newName) - var replaced *dentry - if replacedVFSD != nil { - replaced = replacedVFSD.Impl().(*dentry) - if replaced.inode.isDir() { - if !renamed.inode.isDir() { - return syserror.EISDIR - } - if replaced.vfsd.HasChildren() { - return syserror.ENOTEMPTY - } - } else { - if rp.MustBeDir() { - return syserror.ENOTDIR - } - if renamed.inode.isDir() { - return syserror.ENOTDIR - } - } - } else { - if renamed.inode.isDir() && newParent.inode.nlink == maxLinks { - return syserror.EMLINK - } - } - if newParent.vfsd.IsDisowned() { - return syserror.ENOENT - } - - // Linux places this check before some of those above; we do it here for - // simplicity, under the assumption that applications are not intentionally - // doing noop renames expecting them to succeed where non-noop renames - // would fail. - if renamedVFSD == replacedVFSD { - return nil - } - vfsObj := rp.VirtualFilesystem() - oldParentDir := oldParent.inode.impl.(*directory) - newParentDir := newParent.inode.impl.(*directory) - mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() - if err := vfsObj.PrepareRenameDentry(mntns, renamedVFSD, replacedVFSD); err != nil { - return err - } - if replaced != nil { - newParentDir.childList.Remove(replaced) - if replaced.inode.isDir() { - newParent.inode.decLinksLocked() // from replaced's ".." - } - replaced.inode.decLinksLocked() - } - oldParentDir.childList.Remove(renamed) - newParentDir.childList.PushBack(renamed) - if renamed.inode.isDir() { - oldParent.inode.decLinksLocked() - newParent.inode.incLinksLocked() - } - oldParent.inode.touchCMtime() - newParent.inode.touchCMtime() - renamed.inode.touchCtime() - // TODO(gvisor.dev/issue/1197): Update timestamps and parent directory - // sizes. - vfsObj.CommitRenameReplaceDentry(renamedVFSD, &newParent.vfsd, newName, replacedVFSD) - return nil -} - -// RmdirAt implements vfs.FilesystemImpl.RmdirAt. -func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { - fs.mu.Lock() - defer fs.mu.Unlock() - parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) - if err != nil { - return err - } - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - name := rp.Component() - if name == "." { - return syserror.EINVAL - } - if name == ".." { - return syserror.ENOTEMPTY - } - childVFSD := parent.vfsd.Child(name) - if childVFSD == nil { - return syserror.ENOENT - } - child := childVFSD.Impl().(*dentry) - if !child.inode.isDir() { - return syserror.ENOTDIR - } - if childVFSD.HasChildren() { - return syserror.ENOTEMPTY - } - mnt := rp.Mount() - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - vfsObj := rp.VirtualFilesystem() - mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() - if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { - return err - } - parent.inode.impl.(*directory).childList.Remove(child) - parent.inode.decLinksLocked() // from child's ".." - child.inode.decLinksLocked() - vfsObj.CommitDeleteDentry(childVFSD) - parent.inode.touchCMtime() - return nil -} - -// SetStatAt implements vfs.FilesystemImpl.SetStatAt. -func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := resolveLocked(rp) - if err != nil { - return err - } - return d.inode.setStat(ctx, rp.Credentials(), &opts.Stat) -} - -// StatAt implements vfs.FilesystemImpl.StatAt. -func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := resolveLocked(rp) - if err != nil { - return linux.Statx{}, err - } - var stat linux.Statx - d.inode.statTo(&stat) - return stat, nil -} - -// StatFSAt implements vfs.FilesystemImpl.StatFSAt. -func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - _, err := resolveLocked(rp) - if err != nil { - return linux.Statfs{}, err - } - // TODO(gvisor.dev/issue/1197): Actually implement statfs. - return linux.Statfs{}, syserror.ENOSYS -} - -// SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. -func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { - return fs.doCreateAt(rp, false /* dir */, func(parent *dentry, name string) error { - child := fs.newDentry(fs.newSymlink(rp.Credentials(), target)) - parent.vfsd.InsertChild(&child.vfsd, name) - parent.inode.impl.(*directory).childList.PushBack(child) - return nil - }) -} - -// UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. -func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { - fs.mu.Lock() - defer fs.mu.Unlock() - parent, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) - if err != nil { - return err - } - if err := parent.inode.checkPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - name := rp.Component() - if name == "." || name == ".." { - return syserror.EISDIR - } - childVFSD := parent.vfsd.Child(name) - if childVFSD == nil { - return syserror.ENOENT - } - child := childVFSD.Impl().(*dentry) - if child.inode.isDir() { - return syserror.EISDIR - } - if rp.MustBeDir() { - return syserror.ENOTDIR - } - mnt := rp.Mount() - if err := mnt.CheckBeginWrite(); err != nil { - return err - } - defer mnt.EndWrite() - vfsObj := rp.VirtualFilesystem() - mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() - if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { - return err - } - parent.inode.impl.(*directory).childList.Remove(child) - child.inode.decLinksLocked() - vfsObj.CommitDeleteDentry(childVFSD) - parent.inode.touchCMtime() - return nil -} - -// BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. -func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath) (transport.BoundEndpoint, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - d, err := resolveLocked(rp) - if err != nil { - return nil, err - } - switch impl := d.inode.impl.(type) { - case *socketFile: - return impl.ep, nil - default: - return nil, syserror.ECONNREFUSED - } -} - -// ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. -func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath) ([]string, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - _, err := resolveLocked(rp) - if err != nil { - return nil, err - } - // TODO(b/127675828): support extended attributes - return nil, syserror.ENOTSUP -} - -// GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. -func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) (string, error) { - fs.mu.RLock() - defer fs.mu.RUnlock() - _, err := resolveLocked(rp) - if err != nil { - return "", err - } - // TODO(b/127675828): support extended attributes - return "", syserror.ENOTSUP -} - -// SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. -func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - _, err := resolveLocked(rp) - if err != nil { - return err - } - // TODO(b/127675828): support extended attributes - return syserror.ENOTSUP -} - -// RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. -func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - _, err := resolveLocked(rp) - if err != nil { - return err - } - // TODO(b/127675828): support extended attributes - return syserror.ENOTSUP -} - -// PrependPath implements vfs.FilesystemImpl.PrependPath. -func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { - fs.mu.RLock() - defer fs.mu.RUnlock() - return vfs.GenericPrependPath(vfsroot, vd, b) -} diff --git a/pkg/sentry/fsimpl/tmpfs/named_pipe.go b/pkg/sentry/fsimpl/tmpfs/named_pipe.go deleted file mode 100644 index 2c5c739df..000000000 --- a/pkg/sentry/fsimpl/tmpfs/named_pipe.go +++ /dev/null @@ -1,59 +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 tmpfs - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/usermem" -) - -type namedPipe struct { - inode inode - - pipe *pipe.VFSPipe -} - -// Preconditions: -// * fs.mu must be locked. -// * rp.Mount().CheckBeginWrite() has been called successfully. -func (fs *filesystem) newNamedPipe(creds *auth.Credentials, mode linux.FileMode) *inode { - file := &namedPipe{pipe: pipe.NewVFSPipe(pipe.DefaultPipeSize, usermem.PageSize)} - file.inode.init(file, fs, creds, linux.S_IFIFO|mode) - file.inode.nlink = 1 // Only the parent has a link. - return &file.inode -} - -// namedPipeFD implements vfs.FileDescriptionImpl. Methods are implemented -// entirely via struct embedding. -type namedPipeFD struct { - fileDescription - - *pipe.VFSPipeFD -} - -func newNamedPipeFD(ctx context.Context, np *namedPipe, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { - var err error - var fd namedPipeFD - fd.VFSPipeFD, err = np.pipe.NewVFSPipeFD(ctx, vfsd, &fd.vfsfd, flags) - if err != nil { - return nil, err - } - fd.vfsfd.Init(&fd, flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}) - return &fd.vfsfd, nil -} diff --git a/pkg/sentry/fsimpl/tmpfs/pipe_test.go b/pkg/sentry/fsimpl/tmpfs/pipe_test.go deleted file mode 100644 index 1614f2c39..000000000 --- a/pkg/sentry/fsimpl/tmpfs/pipe_test.go +++ /dev/null @@ -1,238 +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 tmpfs - -import ( - "bytes" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/contexttest" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -const fileName = "mypipe" - -func TestSeparateFDs(t *testing.T) { - ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() - - // Open the read side. This is done in a concurrently because opening - // One end the pipe blocks until the other end is opened. - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - FollowFinalSymlink: true, - } - rfdchan := make(chan *vfs.FileDescription) - go func() { - openOpts := vfs.OpenOptions{Flags: linux.O_RDONLY} - rfd, _ := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - rfdchan <- rfd - }() - - // Open the write side. - openOpts := vfs.OpenOptions{Flags: linux.O_WRONLY} - wfd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - if err != nil { - t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) - } - defer wfd.DecRef() - - rfd, ok := <-rfdchan - if !ok { - t.Fatalf("failed to open pipe for reading %q", fileName) - } - defer rfd.DecRef() - - const msg = "vamos azul" - checkEmpty(ctx, t, rfd) - checkWrite(ctx, t, wfd, msg) - checkRead(ctx, t, rfd, msg) -} - -func TestNonblockingRead(t *testing.T) { - ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() - - // Open the read side as nonblocking. - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - FollowFinalSymlink: true, - } - openOpts := vfs.OpenOptions{Flags: linux.O_RDONLY | linux.O_NONBLOCK} - rfd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - if err != nil { - t.Fatalf("failed to open pipe for reading %q: %v", fileName, err) - } - defer rfd.DecRef() - - // Open the write side. - openOpts = vfs.OpenOptions{Flags: linux.O_WRONLY} - wfd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - if err != nil { - t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) - } - defer wfd.DecRef() - - const msg = "geh blau" - checkEmpty(ctx, t, rfd) - checkWrite(ctx, t, wfd, msg) - checkRead(ctx, t, rfd, msg) -} - -func TestNonblockingWriteError(t *testing.T) { - ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() - - // Open the write side as nonblocking, which should return ENXIO. - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - FollowFinalSymlink: true, - } - openOpts := vfs.OpenOptions{Flags: linux.O_WRONLY | linux.O_NONBLOCK} - _, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - if err != syserror.ENXIO { - t.Fatalf("expected ENXIO, but got error: %v", err) - } -} - -func TestSingleFD(t *testing.T) { - ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() - - // Open the pipe as readable and writable. - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - FollowFinalSymlink: true, - } - openOpts := vfs.OpenOptions{Flags: linux.O_RDWR} - fd, err := vfsObj.OpenAt(ctx, creds, &pop, &openOpts) - if err != nil { - t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) - } - defer fd.DecRef() - - const msg = "forza blu" - checkEmpty(ctx, t, fd) - checkWrite(ctx, t, fd, msg) - checkRead(ctx, t, fd, msg) -} - -// setup creates a VFS with a pipe in the root directory at path fileName. The -// returned VirtualDentry must be DecRef()'d be the caller. It calls t.Fatal -// upon failure. -func setup(t *testing.T) (context.Context, *auth.Credentials, *vfs.VirtualFilesystem, vfs.VirtualDentry) { - ctx := contexttest.Context(t) - creds := auth.CredentialsFromContext(ctx) - - // Create VFS. - vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - t.Fatalf("VFS init: %v", err) - } - vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{}) - if err != nil { - t.Fatalf("failed to create tmpfs root mount: %v", err) - } - - // Create the pipe. - root := mntns.Root() - pop := vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - } - mknodOpts := vfs.MknodOptions{Mode: linux.ModeNamedPipe | 0644} - if err := vfsObj.MknodAt(ctx, creds, &pop, &mknodOpts); err != nil { - t.Fatalf("failed to create file %q: %v", fileName, err) - } - - // Sanity check: the file pipe exists and has the correct mode. - stat, err := vfsObj.StatAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(fileName), - FollowFinalSymlink: true, - }, &vfs.StatOptions{}) - if err != nil { - t.Fatalf("stat(%q) failed: %v", fileName, err) - } - if stat.Mode&^linux.S_IFMT != 0644 { - t.Errorf("got wrong permissions (%0o)", stat.Mode) - } - if stat.Mode&linux.S_IFMT != linux.ModeNamedPipe { - t.Errorf("got wrong file type (%0o)", stat.Mode) - } - - return ctx, creds, vfsObj, root -} - -// checkEmpty calls t.Fatal if the pipe in fd is not empty. -func checkEmpty(ctx context.Context, t *testing.T, fd *vfs.FileDescription) { - readData := make([]byte, 1) - dst := usermem.BytesIOSequence(readData) - bytesRead, err := fd.Read(ctx, dst, vfs.ReadOptions{}) - if err != syserror.ErrWouldBlock { - t.Fatalf("expected ErrWouldBlock reading from empty pipe %q, but got: %v", fileName, err) - } - if bytesRead != 0 { - t.Fatalf("expected to read 0 bytes, but got %d", bytesRead) - } -} - -// checkWrite calls t.Fatal if it fails to write all of msg to fd. -func checkWrite(ctx context.Context, t *testing.T, fd *vfs.FileDescription, msg string) { - writeData := []byte(msg) - src := usermem.BytesIOSequence(writeData) - bytesWritten, err := fd.Write(ctx, src, vfs.WriteOptions{}) - if err != nil { - t.Fatalf("error writing to pipe %q: %v", fileName, err) - } - if bytesWritten != int64(len(writeData)) { - t.Fatalf("expected to write %d bytes, but wrote %d", len(writeData), bytesWritten) - } -} - -// checkRead calls t.Fatal if it fails to read msg from fd. -func checkRead(ctx context.Context, t *testing.T, fd *vfs.FileDescription, msg string) { - readData := make([]byte, len(msg)) - dst := usermem.BytesIOSequence(readData) - bytesRead, err := fd.Read(ctx, dst, vfs.ReadOptions{}) - if err != nil { - t.Fatalf("error reading from pipe %q: %v", fileName, err) - } - if bytesRead != int64(len(msg)) { - t.Fatalf("expected to read %d bytes, but got %d", len(msg), bytesRead) - } - if !bytes.Equal(readData, []byte(msg)) { - t.Fatalf("expected to read %q from pipe, but got %q", msg, string(readData)) - } -} diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go deleted file mode 100644 index 57e5e28ec..000000000 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ /dev/null @@ -1,579 +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 tmpfs - -import ( - "fmt" - "io" - "math" - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/safemem" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/fs/lock" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// regularFile is a regular (=S_IFREG) tmpfs file. -type regularFile struct { - inode inode - - // memFile is a platform.File used to allocate pages to this regularFile. - memFile *pgalloc.MemoryFile - - // mapsMu protects mappings. - mapsMu sync.Mutex `state:"nosave"` - - // mappings tracks mappings of the file into memmap.MappingSpaces. - // - // Protected by mapsMu. - mappings memmap.MappingSet - - // writableMappingPages tracks how many pages of virtual memory are mapped - // as potentially writable from this file. If a page has multiple mappings, - // each mapping is counted separately. - // - // This counter is susceptible to overflow as we can potentially count - // mappings from many VMAs. We count pages rather than bytes to slightly - // mitigate this. - // - // Protected by mapsMu. - writableMappingPages uint64 - - // dataMu protects the fields below. - dataMu sync.RWMutex - - // data maps offsets into the file to offsets into memFile that store - // the file's data. - // - // Protected by dataMu. - data fsutil.FileRangeSet - - // seals represents file seals on this inode. - // - // Protected by dataMu. - seals uint32 - - // size is the size of data. - // - // Protected by both dataMu and inode.mu; reading it requires holding - // either mutex, while writing requires holding both AND using atomics. - // Readers that do not require consistency (like Stat) may read the - // value atomically without holding either lock. - size uint64 -} - -func (fs *filesystem) newRegularFile(creds *auth.Credentials, mode linux.FileMode) *inode { - file := ®ularFile{ - memFile: fs.memFile, - } - file.inode.init(file, fs, creds, linux.S_IFREG|mode) - file.inode.nlink = 1 // from parent directory - return &file.inode -} - -// truncate grows or shrinks the file to the given size. It returns true if the -// file size was updated. -func (rf *regularFile) truncate(newSize uint64) (bool, error) { - rf.inode.mu.Lock() - defer rf.inode.mu.Unlock() - return rf.truncateLocked(newSize) -} - -// Preconditions: rf.inode.mu must be held. -func (rf *regularFile) truncateLocked(newSize uint64) (bool, error) { - oldSize := rf.size - if newSize == oldSize { - // Nothing to do. - return false, nil - } - - // Need to hold inode.mu and dataMu while modifying size. - rf.dataMu.Lock() - if newSize > oldSize { - // Can we grow the file? - if rf.seals&linux.F_SEAL_GROW != 0 { - rf.dataMu.Unlock() - return false, syserror.EPERM - } - // We only need to update the file size. - atomic.StoreUint64(&rf.size, newSize) - rf.dataMu.Unlock() - return true, nil - } - - // We are shrinking the file. First check if this is allowed. - if rf.seals&linux.F_SEAL_SHRINK != 0 { - rf.dataMu.Unlock() - return false, syserror.EPERM - } - - // Update the file size. - atomic.StoreUint64(&rf.size, newSize) - rf.dataMu.Unlock() - - // Invalidate past translations of truncated pages. - oldpgend := fs.OffsetPageEnd(int64(oldSize)) - newpgend := fs.OffsetPageEnd(int64(newSize)) - if newpgend < oldpgend { - rf.mapsMu.Lock() - rf.mappings.Invalidate(memmap.MappableRange{newpgend, oldpgend}, memmap.InvalidateOpts{ - // Compare Linux's mm/shmem.c:shmem_setattr() => - // mm/memory.c:unmap_mapping_range(evencows=1). - InvalidatePrivate: true, - }) - rf.mapsMu.Unlock() - } - - // We are now guaranteed that there are no translations of truncated pages, - // and can remove them. - rf.dataMu.Lock() - rf.data.Truncate(newSize, rf.memFile) - rf.dataMu.Unlock() - return true, nil -} - -// AddMapping implements memmap.Mappable.AddMapping. -func (rf *regularFile) AddMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) error { - rf.mapsMu.Lock() - defer rf.mapsMu.Unlock() - rf.dataMu.RLock() - defer rf.dataMu.RUnlock() - - // Reject writable mapping if F_SEAL_WRITE is set. - if rf.seals&linux.F_SEAL_WRITE != 0 && writable { - return syserror.EPERM - } - - rf.mappings.AddMapping(ms, ar, offset, writable) - if writable { - pagesBefore := rf.writableMappingPages - - // ar is guaranteed to be page aligned per memmap.Mappable. - rf.writableMappingPages += uint64(ar.Length() / usermem.PageSize) - - if rf.writableMappingPages < pagesBefore { - panic(fmt.Sprintf("Overflow while mapping potentially writable pages pointing to a tmpfs file. Before %v, after %v", pagesBefore, rf.writableMappingPages)) - } - } - - return nil -} - -// RemoveMapping implements memmap.Mappable.RemoveMapping. -func (rf *regularFile) RemoveMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) { - rf.mapsMu.Lock() - defer rf.mapsMu.Unlock() - - rf.mappings.RemoveMapping(ms, ar, offset, writable) - - if writable { - pagesBefore := rf.writableMappingPages - - // ar is guaranteed to be page aligned per memmap.Mappable. - rf.writableMappingPages -= uint64(ar.Length() / usermem.PageSize) - - if rf.writableMappingPages > pagesBefore { - panic(fmt.Sprintf("Underflow while unmapping potentially writable pages pointing to a tmpfs file. Before %v, after %v", pagesBefore, rf.writableMappingPages)) - } - } -} - -// CopyMapping implements memmap.Mappable.CopyMapping. -func (rf *regularFile) CopyMapping(ctx context.Context, ms memmap.MappingSpace, srcAR, dstAR usermem.AddrRange, offset uint64, writable bool) error { - return rf.AddMapping(ctx, ms, dstAR, offset, writable) -} - -// Translate implements memmap.Mappable.Translate. -func (rf *regularFile) Translate(ctx context.Context, required, optional memmap.MappableRange, at usermem.AccessType) ([]memmap.Translation, error) { - rf.dataMu.Lock() - defer rf.dataMu.Unlock() - - // Constrain translations to f.attr.Size (rounded up) to prevent - // translation to pages that may be concurrently truncated. - pgend := fs.OffsetPageEnd(int64(rf.size)) - var beyondEOF bool - if required.End > pgend { - if required.Start >= pgend { - return nil, &memmap.BusError{io.EOF} - } - beyondEOF = true - required.End = pgend - } - if optional.End > pgend { - optional.End = pgend - } - - cerr := rf.data.Fill(ctx, required, optional, rf.memFile, usage.Tmpfs, func(_ context.Context, dsts safemem.BlockSeq, _ uint64) (uint64, error) { - // Newly-allocated pages are zeroed, so we don't need to do anything. - return dsts.NumBytes(), nil - }) - - var ts []memmap.Translation - var translatedEnd uint64 - for seg := rf.data.FindSegment(required.Start); seg.Ok() && seg.Start() < required.End; seg, _ = seg.NextNonEmpty() { - segMR := seg.Range().Intersect(optional) - ts = append(ts, memmap.Translation{ - Source: segMR, - File: rf.memFile, - Offset: seg.FileRangeOf(segMR).Start, - Perms: usermem.AnyAccess, - }) - translatedEnd = segMR.End - } - - // Don't return the error returned by f.data.Fill if it occurred outside of - // required. - if translatedEnd < required.End && cerr != nil { - return ts, &memmap.BusError{cerr} - } - if beyondEOF { - return ts, &memmap.BusError{io.EOF} - } - return ts, nil -} - -// InvalidateUnsavable implements memmap.Mappable.InvalidateUnsavable. -func (*regularFile) InvalidateUnsavable(context.Context) error { - return nil -} - -type regularFileFD struct { - fileDescription - - // off is the file offset. off is accessed using atomic memory operations. - // offMu serializes operations that may mutate off. - off int64 - offMu sync.Mutex -} - -// Release implements vfs.FileDescriptionImpl.Release. -func (fd *regularFileFD) Release() { - // noop -} - -// PRead implements vfs.FileDescriptionImpl.PRead. -func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { - if offset < 0 { - return 0, syserror.EINVAL - } - if dst.NumBytes() == 0 { - return 0, nil - } - f := fd.inode().impl.(*regularFile) - rw := getRegularFileReadWriter(f, offset) - n, err := dst.CopyOutFrom(ctx, rw) - putRegularFileReadWriter(rw) - fd.inode().touchAtime(fd.vfsfd.Mount()) - return n, err -} - -// Read implements vfs.FileDescriptionImpl.Read. -func (fd *regularFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - fd.offMu.Lock() - n, err := fd.PRead(ctx, dst, fd.off, opts) - fd.off += n - fd.offMu.Unlock() - return n, err -} - -// PWrite implements vfs.FileDescriptionImpl.PWrite. -func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { - if offset < 0 { - return 0, syserror.EINVAL - } - srclen := src.NumBytes() - if srclen == 0 { - return 0, nil - } - f := fd.inode().impl.(*regularFile) - if end := offset + srclen; end < offset { - // Overflow. - return 0, syserror.EFBIG - } - - var err error - srclen, err = vfs.CheckLimit(ctx, offset, srclen) - if err != nil { - return 0, err - } - src = src.TakeFirst64(srclen) - - f.inode.mu.Lock() - rw := getRegularFileReadWriter(f, offset) - n, err := src.CopyInTo(ctx, rw) - fd.inode().touchCMtimeLocked() - f.inode.mu.Unlock() - putRegularFileReadWriter(rw) - return n, err -} - -// Write implements vfs.FileDescriptionImpl.Write. -func (fd *regularFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { - fd.offMu.Lock() - n, err := fd.PWrite(ctx, src, fd.off, opts) - fd.off += n - fd.offMu.Unlock() - return n, err -} - -// Seek implements vfs.FileDescriptionImpl.Seek. -func (fd *regularFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { - fd.offMu.Lock() - defer fd.offMu.Unlock() - switch whence { - case linux.SEEK_SET: - // use offset as specified - case linux.SEEK_CUR: - offset += fd.off - case linux.SEEK_END: - offset += int64(atomic.LoadUint64(&fd.inode().impl.(*regularFile).size)) - default: - return 0, syserror.EINVAL - } - if offset < 0 { - return 0, syserror.EINVAL - } - fd.off = offset - return offset, nil -} - -// Sync implements vfs.FileDescriptionImpl.Sync. -func (fd *regularFileFD) Sync(ctx context.Context) error { - return nil -} - -// LockBSD implements vfs.FileDescriptionImpl.LockBSD. -func (fd *regularFileFD) LockBSD(ctx context.Context, uid lock.UniqueID, t lock.LockType, block lock.Blocker) error { - return fd.inode().lockBSD(uid, t, block) -} - -// UnlockBSD implements vfs.FileDescriptionImpl.UnlockBSD. -func (fd *regularFileFD) UnlockBSD(ctx context.Context, uid lock.UniqueID) error { - fd.inode().unlockBSD(uid) - return nil -} - -// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. -func (fd *regularFileFD) LockPOSIX(ctx context.Context, uid lock.UniqueID, t lock.LockType, rng lock.LockRange, block lock.Blocker) error { - return fd.inode().lockPOSIX(uid, t, rng, block) -} - -// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX. -func (fd *regularFileFD) UnlockPOSIX(ctx context.Context, uid lock.UniqueID, rng lock.LockRange) error { - fd.inode().unlockPOSIX(uid, rng) - return nil -} - -// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. -func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { - file := fd.inode().impl.(*regularFile) - return vfs.GenericConfigureMMap(&fd.vfsfd, file, opts) -} - -// regularFileReadWriter implements safemem.Reader and Safemem.Writer. -type regularFileReadWriter struct { - file *regularFile - - // Offset into the file to read/write at. Note that this may be - // different from the FD offset if PRead/PWrite is used. - off uint64 -} - -var regularFileReadWriterPool = sync.Pool{ - New: func() interface{} { - return ®ularFileReadWriter{} - }, -} - -func getRegularFileReadWriter(file *regularFile, offset int64) *regularFileReadWriter { - rw := regularFileReadWriterPool.Get().(*regularFileReadWriter) - rw.file = file - rw.off = uint64(offset) - return rw -} - -func putRegularFileReadWriter(rw *regularFileReadWriter) { - rw.file = nil - regularFileReadWriterPool.Put(rw) -} - -// ReadToBlocks implements safemem.Reader.ReadToBlocks. -func (rw *regularFileReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { - rw.file.dataMu.RLock() - defer rw.file.dataMu.RUnlock() - size := rw.file.size - - // Compute the range to read (limited by file size and overflow-checked). - if rw.off >= size { - return 0, io.EOF - } - end := size - if rend := rw.off + dsts.NumBytes(); rend > rw.off && rend < end { - end = rend - } - - var done uint64 - seg, gap := rw.file.data.Find(uint64(rw.off)) - for rw.off < end { - mr := memmap.MappableRange{uint64(rw.off), uint64(end)} - switch { - case seg.Ok(): - // Get internal mappings. - ims, err := rw.file.memFile.MapInternal(seg.FileRangeOf(seg.Range().Intersect(mr)), usermem.Read) - if err != nil { - return done, err - } - - // Copy from internal mappings. - n, err := safemem.CopySeq(dsts, ims) - done += n - rw.off += uint64(n) - dsts = dsts.DropFirst64(n) - if err != nil { - return done, err - } - - // Continue. - seg, gap = seg.NextNonEmpty() - - case gap.Ok(): - // Tmpfs holes are zero-filled. - gapmr := gap.Range().Intersect(mr) - dst := dsts.TakeFirst64(gapmr.Length()) - n, err := safemem.ZeroSeq(dst) - done += n - rw.off += uint64(n) - dsts = dsts.DropFirst64(n) - if err != nil { - return done, err - } - - // Continue. - seg, gap = gap.NextSegment(), fsutil.FileRangeGapIterator{} - } - } - return done, nil -} - -// WriteFromBlocks implements safemem.Writer.WriteFromBlocks. -// -// Preconditions: inode.mu must be held. -func (rw *regularFileReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { - // Hold dataMu so we can modify size. - rw.file.dataMu.Lock() - defer rw.file.dataMu.Unlock() - - // Compute the range to write (overflow-checked). - end := rw.off + srcs.NumBytes() - if end <= rw.off { - end = math.MaxInt64 - } - - // Check if seals prevent either file growth or all writes. - switch { - case rw.file.seals&linux.F_SEAL_WRITE != 0: // Write sealed - return 0, syserror.EPERM - case end > rw.file.size && rw.file.seals&linux.F_SEAL_GROW != 0: // Grow sealed - // When growth is sealed, Linux effectively allows writes which would - // normally grow the file to partially succeed up to the current EOF, - // rounded down to the page boundary before the EOF. - // - // This happens because writes (and thus the growth check) for tmpfs - // files proceed page-by-page on Linux, and the final write to the page - // containing EOF fails, resulting in a partial write up to the start of - // that page. - // - // To emulate this behaviour, artifically truncate the write to the - // start of the page containing the current EOF. - // - // See Linux, mm/filemap.c:generic_perform_write() and - // mm/shmem.c:shmem_write_begin(). - if pgstart := uint64(usermem.Addr(rw.file.size).RoundDown()); end > pgstart { - end = pgstart - } - if end <= rw.off { - // Truncation would result in no data being written. - return 0, syserror.EPERM - } - } - - // Page-aligned mr for when we need to allocate memory. RoundUp can't - // overflow since end is an int64. - pgstartaddr := usermem.Addr(rw.off).RoundDown() - pgendaddr, _ := usermem.Addr(end).RoundUp() - pgMR := memmap.MappableRange{uint64(pgstartaddr), uint64(pgendaddr)} - - var ( - done uint64 - retErr error - ) - seg, gap := rw.file.data.Find(uint64(rw.off)) - for rw.off < end { - mr := memmap.MappableRange{uint64(rw.off), uint64(end)} - switch { - case seg.Ok(): - // Get internal mappings. - ims, err := rw.file.memFile.MapInternal(seg.FileRangeOf(seg.Range().Intersect(mr)), usermem.Write) - if err != nil { - retErr = err - goto exitLoop - } - - // Copy to internal mappings. - n, err := safemem.CopySeq(ims, srcs) - done += n - rw.off += uint64(n) - srcs = srcs.DropFirst64(n) - if err != nil { - retErr = err - goto exitLoop - } - - // Continue. - seg, gap = seg.NextNonEmpty() - - case gap.Ok(): - // Allocate memory for the write. - gapMR := gap.Range().Intersect(pgMR) - fr, err := rw.file.memFile.Allocate(gapMR.Length(), usage.Tmpfs) - if err != nil { - retErr = err - goto exitLoop - } - - // Write to that memory as usual. - seg, gap = rw.file.data.Insert(gap, gapMR, fr.Start), fsutil.FileRangeGapIterator{} - } - } -exitLoop: - // If the write ends beyond the file's previous size, it causes the - // file to grow. - if rw.off > rw.file.size { - rw.file.size = rw.off - } - - return done, retErr -} diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go deleted file mode 100644 index 0399725cf..000000000 --- a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go +++ /dev/null @@ -1,496 +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 tmpfs - -import ( - "bytes" - "fmt" - "io" - "sync/atomic" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - "gvisor.dev/gvisor/pkg/fspath" - "gvisor.dev/gvisor/pkg/sentry/fs/lock" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" -) - -// nextFileID is used to generate unique file names. -var nextFileID int64 - -// newTmpfsRoot creates a new tmpfs mount, and returns the root. If the error -// is not nil, then cleanup should be called when the root is no longer needed. -func newTmpfsRoot(ctx context.Context) (*vfs.VirtualFilesystem, vfs.VirtualDentry, func(), error) { - creds := auth.CredentialsFromContext(ctx) - - vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { - return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("VFS init: %v", err) - } - - vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserMount: true, - }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{}) - if err != nil { - return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("failed to create tmpfs root mount: %v", err) - } - root := mntns.Root() - return vfsObj, root, func() { - root.DecRef() - mntns.DecRef() - }, nil -} - -// newFileFD creates a new file in a new tmpfs mount, and returns the FD. If -// the returned err is not nil, then cleanup should be called when the FD is no -// longer needed. -func newFileFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { - creds := auth.CredentialsFromContext(ctx) - vfsObj, root, cleanup, err := newTmpfsRoot(ctx) - if err != nil { - return nil, nil, err - } - - filename := fmt.Sprintf("tmpfs-test-file-%d", atomic.AddInt64(&nextFileID, 1)) - - // Create the file that will be write/read. - fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(filename), - }, &vfs.OpenOptions{ - Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, - Mode: linux.ModeRegular | mode, - }) - if err != nil { - cleanup() - return nil, nil, fmt.Errorf("failed to create file %q: %v", filename, err) - } - - return fd, cleanup, nil -} - -// newDirFD is like newFileFD, but for directories. -func newDirFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { - creds := auth.CredentialsFromContext(ctx) - vfsObj, root, cleanup, err := newTmpfsRoot(ctx) - if err != nil { - return nil, nil, err - } - - dirname := fmt.Sprintf("tmpfs-test-dir-%d", atomic.AddInt64(&nextFileID, 1)) - - // Create the dir. - if err := vfsObj.MkdirAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(dirname), - }, &vfs.MkdirOptions{ - Mode: linux.ModeDirectory | mode, - }); err != nil { - cleanup() - return nil, nil, fmt.Errorf("failed to create directory %q: %v", dirname, err) - } - - // Open the dir and return it. - fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(dirname), - }, &vfs.OpenOptions{ - Flags: linux.O_RDONLY | linux.O_DIRECTORY, - }) - if err != nil { - cleanup() - return nil, nil, fmt.Errorf("failed to open directory %q: %v", dirname, err) - } - - return fd, cleanup, nil -} - -// newPipeFD is like newFileFD, but for pipes. -func newPipeFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { - creds := auth.CredentialsFromContext(ctx) - vfsObj, root, cleanup, err := newTmpfsRoot(ctx) - if err != nil { - return nil, nil, err - } - - pipename := fmt.Sprintf("tmpfs-test-pipe-%d", atomic.AddInt64(&nextFileID, 1)) - - // Create the pipe. - if err := vfsObj.MknodAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(pipename), - }, &vfs.MknodOptions{ - Mode: linux.ModeNamedPipe | mode, - }); err != nil { - cleanup() - return nil, nil, fmt.Errorf("failed to create pipe %q: %v", pipename, err) - } - - // Open the pipe and return it. - fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(pipename), - }, &vfs.OpenOptions{ - Flags: linux.O_RDWR, - }) - if err != nil { - cleanup() - return nil, nil, fmt.Errorf("failed to open pipe %q: %v", pipename, err) - } - - return fd, cleanup, nil -} - -// Test that we can write some data to a file and read it back.` -func TestSimpleWriteRead(t *testing.T) { - ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, 0644) - if err != nil { - t.Fatal(err) - } - defer cleanup() - - // Write. - data := []byte("foobarbaz") - n, err := fd.Write(ctx, usermem.BytesIOSequence(data), vfs.WriteOptions{}) - if err != nil { - t.Fatalf("fd.Write failed: %v", err) - } - if n != int64(len(data)) { - t.Errorf("fd.Write got short write length %d, want %d", n, len(data)) - } - if got, want := fd.Impl().(*regularFileFD).off, int64(len(data)); got != want { - t.Errorf("fd.Write left offset at %d, want %d", got, want) - } - - // Seek back to beginning. - if _, err := fd.Seek(ctx, 0, linux.SEEK_SET); err != nil { - t.Fatalf("fd.Seek failed: %v", err) - } - if got, want := fd.Impl().(*regularFileFD).off, int64(0); got != want { - t.Errorf("fd.Seek(0) left offset at %d, want %d", got, want) - } - - // Read. - buf := make([]byte, len(data)) - n, err = fd.Read(ctx, usermem.BytesIOSequence(buf), vfs.ReadOptions{}) - if err != nil && err != io.EOF { - t.Fatalf("fd.Read failed: %v", err) - } - if n != int64(len(data)) { - t.Errorf("fd.Read got short read length %d, want %d", n, len(data)) - } - if got, want := string(buf), string(data); got != want { - t.Errorf("Read got %q want %s", got, want) - } - if got, want := fd.Impl().(*regularFileFD).off, int64(len(data)); got != want { - t.Errorf("fd.Write left offset at %d, want %d", got, want) - } -} - -func TestPWrite(t *testing.T) { - ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, 0644) - if err != nil { - t.Fatal(err) - } - defer cleanup() - - // Fill file with 1k 'a's. - data := bytes.Repeat([]byte{'a'}, 1000) - n, err := fd.Write(ctx, usermem.BytesIOSequence(data), vfs.WriteOptions{}) - if err != nil { - t.Fatalf("fd.Write failed: %v", err) - } - if n != int64(len(data)) { - t.Errorf("fd.Write got short write length %d, want %d", n, len(data)) - } - - // Write "gVisor is awesome" at various offsets. - buf := []byte("gVisor is awesome") - offsets := []int{0, 1, 2, 10, 20, 50, 100, len(data) - 100, len(data) - 1, len(data), len(data) + 1} - for _, offset := range offsets { - name := fmt.Sprintf("PWrite offset=%d", offset) - t.Run(name, func(t *testing.T) { - n, err := fd.PWrite(ctx, usermem.BytesIOSequence(buf), int64(offset), vfs.WriteOptions{}) - if err != nil { - t.Errorf("fd.PWrite got err %v want nil", err) - } - if n != int64(len(buf)) { - t.Errorf("fd.PWrite got %d bytes want %d", n, len(buf)) - } - - // Update data to reflect expected file contents. - if len(data) < offset+len(buf) { - data = append(data, make([]byte, (offset+len(buf))-len(data))...) - } - copy(data[offset:], buf) - - // Read the whole file and compare with data. - readBuf := make([]byte, len(data)) - n, err = fd.PRead(ctx, usermem.BytesIOSequence(readBuf), 0, vfs.ReadOptions{}) - if err != nil { - t.Fatalf("fd.PRead failed: %v", err) - } - if n != int64(len(data)) { - t.Errorf("fd.PRead got short read length %d, want %d", n, len(data)) - } - if got, want := string(readBuf), string(data); got != want { - t.Errorf("PRead got %q want %s", got, want) - } - - }) - } -} - -func TestLocks(t *testing.T) { - ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, 0644) - if err != nil { - t.Fatal(err) - } - defer cleanup() - - var ( - uid1 lock.UniqueID - uid2 lock.UniqueID - // Non-blocking. - block lock.Blocker - ) - - uid1 = 123 - uid2 = 456 - - if err := fd.Impl().LockBSD(ctx, uid1, lock.ReadLock, block); err != nil { - t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) - } - if err := fd.Impl().LockBSD(ctx, uid2, lock.ReadLock, block); err != nil { - t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) - } - if got, want := fd.Impl().LockBSD(ctx, uid2, lock.WriteLock, block), syserror.ErrWouldBlock; got != want { - t.Fatalf("fd.Impl().LockBSD failed: got = %v, want = %v", got, want) - } - if err := fd.Impl().UnlockBSD(ctx, uid1); err != nil { - t.Fatalf("fd.Impl().UnlockBSD failed: err = %v", err) - } - if err := fd.Impl().LockBSD(ctx, uid2, lock.WriteLock, block); err != nil { - t.Fatalf("fd.Impl().LockBSD failed: err = %v", err) - } - - rng1 := lock.LockRange{0, 1} - rng2 := lock.LockRange{1, 2} - - if err := fd.Impl().LockPOSIX(ctx, uid1, lock.ReadLock, rng1, block); err != nil { - t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) - } - if err := fd.Impl().LockPOSIX(ctx, uid2, lock.ReadLock, rng2, block); err != nil { - t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) - } - if err := fd.Impl().LockPOSIX(ctx, uid1, lock.WriteLock, rng1, block); err != nil { - t.Fatalf("fd.Impl().LockPOSIX failed: err = %v", err) - } - if got, want := fd.Impl().LockPOSIX(ctx, uid2, lock.ReadLock, rng1, block), syserror.ErrWouldBlock; got != want { - t.Fatalf("fd.Impl().LockPOSIX failed: got = %v, want = %v", got, want) - } - if err := fd.Impl().UnlockPOSIX(ctx, uid1, rng1); err != nil { - t.Fatalf("fd.Impl().UnlockPOSIX failed: err = %v", err) - } -} - -func TestPRead(t *testing.T) { - ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, 0644) - if err != nil { - t.Fatal(err) - } - defer cleanup() - - // Write 100 sequences of 'gVisor is awesome'. - data := bytes.Repeat([]byte("gVisor is awsome"), 100) - n, err := fd.Write(ctx, usermem.BytesIOSequence(data), vfs.WriteOptions{}) - if err != nil { - t.Fatalf("fd.Write failed: %v", err) - } - if n != int64(len(data)) { - t.Errorf("fd.Write got short write length %d, want %d", n, len(data)) - } - - // Read various sizes from various offsets. - sizes := []int{0, 1, 2, 10, 20, 50, 100, 1000} - offsets := []int{0, 1, 2, 10, 20, 50, 100, 1000, len(data) - 100, len(data) - 1, len(data), len(data) + 1} - - for _, size := range sizes { - for _, offset := range offsets { - name := fmt.Sprintf("PRead offset=%d size=%d", offset, size) - t.Run(name, func(t *testing.T) { - var ( - wantRead []byte - wantErr error - ) - if offset < len(data) { - wantRead = data[offset:] - } else if size > 0 { - wantErr = io.EOF - } - if offset+size < len(data) { - wantRead = wantRead[:size] - } - buf := make([]byte, size) - n, err := fd.PRead(ctx, usermem.BytesIOSequence(buf), int64(offset), vfs.ReadOptions{}) - if err != wantErr { - t.Errorf("fd.PRead got err %v want %v", err, wantErr) - } - if n != int64(len(wantRead)) { - t.Errorf("fd.PRead got %d bytes want %d", n, len(wantRead)) - } - if got := string(buf[:n]); got != string(wantRead) { - t.Errorf("fd.PRead got %q want %q", got, string(wantRead)) - } - }) - } - } -} - -func TestTruncate(t *testing.T) { - ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, 0644) - if err != nil { - t.Fatal(err) - } - defer cleanup() - - // Fill the file with some data. - data := bytes.Repeat([]byte("gVisor is awsome"), 100) - written, err := fd.Write(ctx, usermem.BytesIOSequence(data), vfs.WriteOptions{}) - if err != nil { - t.Fatalf("fd.Write failed: %v", err) - } - - // Size should be same as written. - sizeStatOpts := vfs.StatOptions{Mask: linux.STATX_SIZE} - stat, err := fd.Stat(ctx, sizeStatOpts) - if err != nil { - t.Fatalf("fd.Stat failed: %v", err) - } - if got, want := int64(stat.Size), written; got != want { - t.Errorf("fd.Stat got size %d, want %d", got, want) - } - - // Truncate down. - newSize := uint64(10) - if err := fd.SetStat(ctx, vfs.SetStatOptions{ - Stat: linux.Statx{ - Mask: linux.STATX_SIZE, - Size: newSize, - }, - }); err != nil { - t.Errorf("fd.Truncate failed: %v", err) - } - // Size should be updated. - statAfterTruncateDown, err := fd.Stat(ctx, sizeStatOpts) - if err != nil { - t.Fatalf("fd.Stat failed: %v", err) - } - if got, want := statAfterTruncateDown.Size, newSize; got != want { - t.Errorf("fd.Stat got size %d, want %d", got, want) - } - // We should only read newSize worth of data. - buf := make([]byte, 1000) - if n, err := fd.PRead(ctx, usermem.BytesIOSequence(buf), 0, vfs.ReadOptions{}); err != nil && err != io.EOF { - t.Fatalf("fd.PRead failed: %v", err) - } else if uint64(n) != newSize { - t.Errorf("fd.PRead got size %d, want %d", n, newSize) - } - // Mtime and Ctime should be bumped. - if got := statAfterTruncateDown.Mtime.ToNsec(); got <= stat.Mtime.ToNsec() { - t.Errorf("fd.Stat got Mtime %v, want > %v", got, stat.Mtime) - } - if got := statAfterTruncateDown.Ctime.ToNsec(); got <= stat.Ctime.ToNsec() { - t.Errorf("fd.Stat got Ctime %v, want > %v", got, stat.Ctime) - } - - // Truncate up. - newSize = 100 - if err := fd.SetStat(ctx, vfs.SetStatOptions{ - Stat: linux.Statx{ - Mask: linux.STATX_SIZE, - Size: newSize, - }, - }); err != nil { - t.Errorf("fd.Truncate failed: %v", err) - } - // Size should be updated. - statAfterTruncateUp, err := fd.Stat(ctx, sizeStatOpts) - if err != nil { - t.Fatalf("fd.Stat failed: %v", err) - } - if got, want := statAfterTruncateUp.Size, newSize; got != want { - t.Errorf("fd.Stat got size %d, want %d", got, want) - } - // We should read newSize worth of data. - buf = make([]byte, 1000) - if n, err := fd.PRead(ctx, usermem.BytesIOSequence(buf), 0, vfs.ReadOptions{}); err != nil && err != io.EOF { - t.Fatalf("fd.PRead failed: %v", err) - } else if uint64(n) != newSize { - t.Errorf("fd.PRead got size %d, want %d", n, newSize) - } - // Bytes should be null after 10, since we previously truncated to 10. - for i := uint64(10); i < newSize; i++ { - if buf[i] != 0 { - t.Errorf("fd.PRead got byte %d=%x, want 0", i, buf[i]) - break - } - } - // Mtime and Ctime should be bumped. - if got := statAfterTruncateUp.Mtime.ToNsec(); got <= statAfterTruncateDown.Mtime.ToNsec() { - t.Errorf("fd.Stat got Mtime %v, want > %v", got, statAfterTruncateDown.Mtime) - } - if got := statAfterTruncateUp.Ctime.ToNsec(); got <= statAfterTruncateDown.Ctime.ToNsec() { - t.Errorf("fd.Stat got Ctime %v, want > %v", got, stat.Ctime) - } - - // Truncate to the current size. - newSize = statAfterTruncateUp.Size - if err := fd.SetStat(ctx, vfs.SetStatOptions{ - Stat: linux.Statx{ - Mask: linux.STATX_SIZE, - Size: newSize, - }, - }); err != nil { - t.Errorf("fd.Truncate failed: %v", err) - } - statAfterTruncateNoop, err := fd.Stat(ctx, sizeStatOpts) - if err != nil { - t.Fatalf("fd.Stat failed: %v", err) - } - // Mtime and Ctime should not be bumped, since operation is a noop. - if got := statAfterTruncateNoop.Mtime.ToNsec(); got != statAfterTruncateUp.Mtime.ToNsec() { - t.Errorf("fd.Stat got Mtime %v, want %v", got, statAfterTruncateUp.Mtime) - } - if got := statAfterTruncateNoop.Ctime.ToNsec(); got != statAfterTruncateUp.Ctime.ToNsec() { - t.Errorf("fd.Stat got Ctime %v, want %v", got, statAfterTruncateUp.Ctime) - } -} diff --git a/pkg/sentry/fsimpl/tmpfs/socket_file.go b/pkg/sentry/fsimpl/tmpfs/socket_file.go deleted file mode 100644 index 25c2321af..000000000 --- a/pkg/sentry/fsimpl/tmpfs/socket_file.go +++ /dev/null @@ -1,34 +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 tmpfs - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" -) - -// socketFile is a socket (=S_IFSOCK) tmpfs file. -type socketFile struct { - inode inode - ep transport.BoundEndpoint -} - -func (fs *filesystem) newSocketFile(creds *auth.Credentials, mode linux.FileMode, ep transport.BoundEndpoint) *inode { - file := &socketFile{ep: ep} - file.inode.init(file, fs, creds, mode) - file.inode.nlink = 1 // from parent directory - return &file.inode -} diff --git a/pkg/sentry/fsimpl/tmpfs/stat_test.go b/pkg/sentry/fsimpl/tmpfs/stat_test.go deleted file mode 100644 index 3e02e7190..000000000 --- a/pkg/sentry/fsimpl/tmpfs/stat_test.go +++ /dev/null @@ -1,232 +0,0 @@ -// Copyright 2020 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 ( - "fmt" - "testing" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -func TestStatAfterCreate(t *testing.T) { - ctx := contexttest.Context(t) - mode := linux.FileMode(0644) - - // Run with different file types. - // TODO(gvisor.dev/issue/1197): Also test symlinks and sockets. - for _, typ := range []string{"file", "dir", "pipe"} { - t.Run(fmt.Sprintf("type=%q", typ), func(t *testing.T) { - var ( - fd *vfs.FileDescription - cleanup func() - err error - ) - switch typ { - case "file": - fd, cleanup, err = newFileFD(ctx, mode) - case "dir": - fd, cleanup, err = newDirFD(ctx, mode) - case "pipe": - fd, cleanup, err = newPipeFD(ctx, mode) - default: - panic(fmt.Sprintf("unknown typ %q", typ)) - } - if err != nil { - t.Fatal(err) - } - defer cleanup() - - got, err := fd.Stat(ctx, vfs.StatOptions{}) - if err != nil { - t.Fatalf("Stat failed: %v", err) - } - - // Atime, Ctime, Mtime should all be current time (non-zero). - atime, ctime, mtime := got.Atime.ToNsec(), got.Ctime.ToNsec(), got.Mtime.ToNsec() - if atime != ctime || ctime != mtime { - t.Errorf("got atime=%d ctime=%d mtime=%d, wanted equal values", atime, ctime, mtime) - } - if atime == 0 { - t.Errorf("got atime=%d, want non-zero", atime) - } - - // Btime should be 0, as it is not set by tmpfs. - if btime := got.Btime.ToNsec(); btime != 0 { - t.Errorf("got btime %d, want 0", got.Btime.ToNsec()) - } - - // Size should be 0. - if got.Size != 0 { - t.Errorf("got size %d, want 0", got.Size) - } - - // Nlink should be 1 for files, 2 for dirs. - wantNlink := uint32(1) - if typ == "dir" { - wantNlink = 2 - } - if got.Nlink != wantNlink { - t.Errorf("got nlink %d, want %d", got.Nlink, wantNlink) - } - - // UID and GID are set from context creds. - creds := auth.CredentialsFromContext(ctx) - if got.UID != uint32(creds.EffectiveKUID) { - t.Errorf("got uid %d, want %d", got.UID, uint32(creds.EffectiveKUID)) - } - if got.GID != uint32(creds.EffectiveKGID) { - t.Errorf("got gid %d, want %d", got.GID, uint32(creds.EffectiveKGID)) - } - - // Mode. - wantMode := uint16(mode) - switch typ { - case "file": - wantMode |= linux.S_IFREG - case "dir": - wantMode |= linux.S_IFDIR - case "pipe": - wantMode |= linux.S_IFIFO - default: - panic(fmt.Sprintf("unknown typ %q", typ)) - } - - if got.Mode != wantMode { - t.Errorf("got mode %x, want %x", got.Mode, wantMode) - } - - // Ino. - if got.Ino == 0 { - t.Errorf("got ino %d, want not 0", got.Ino) - } - }) - } -} - -func TestSetStatAtime(t *testing.T) { - ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, 0644) - if err != nil { - t.Fatal(err) - } - defer cleanup() - - allStatOptions := vfs.StatOptions{Mask: linux.STATX_ALL} - - // Get initial stat. - initialStat, err := fd.Stat(ctx, allStatOptions) - if err != nil { - t.Fatalf("Stat failed: %v", err) - } - - // Set atime, but without the mask. - if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: linux.Statx{ - Mask: 0, - Atime: linux.NsecToStatxTimestamp(100), - }}); err != nil { - t.Errorf("SetStat atime without mask failed: %v") - } - // Atime should be unchanged. - if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { - t.Errorf("Stat got error: %v", err) - } else if gotStat.Atime != initialStat.Atime { - t.Errorf("Stat got atime %d, want %d", gotStat.Atime, initialStat.Atime) - } - - // Set atime, this time included in the mask. - setStat := linux.Statx{ - Mask: linux.STATX_ATIME, - Atime: linux.NsecToStatxTimestamp(100), - } - if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: setStat}); err != nil { - t.Errorf("SetStat atime with mask failed: %v") - } - if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { - t.Errorf("Stat got error: %v", err) - } else if gotStat.Atime != setStat.Atime { - t.Errorf("Stat got atime %d, want %d", gotStat.Atime, setStat.Atime) - } -} - -func TestSetStat(t *testing.T) { - ctx := contexttest.Context(t) - mode := linux.FileMode(0644) - - // Run with different file types. - // TODO(gvisor.dev/issue/1197): Also test symlinks and sockets. - for _, typ := range []string{"file", "dir", "pipe"} { - t.Run(fmt.Sprintf("type=%q", typ), func(t *testing.T) { - var ( - fd *vfs.FileDescription - cleanup func() - err error - ) - switch typ { - case "file": - fd, cleanup, err = newFileFD(ctx, mode) - case "dir": - fd, cleanup, err = newDirFD(ctx, mode) - case "pipe": - fd, cleanup, err = newPipeFD(ctx, mode) - default: - panic(fmt.Sprintf("unknown typ %q", typ)) - } - if err != nil { - t.Fatal(err) - } - defer cleanup() - - allStatOptions := vfs.StatOptions{Mask: linux.STATX_ALL} - - // Get initial stat. - initialStat, err := fd.Stat(ctx, allStatOptions) - if err != nil { - t.Fatalf("Stat failed: %v", err) - } - - // Set atime, but without the mask. - if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: linux.Statx{ - Mask: 0, - Atime: linux.NsecToStatxTimestamp(100), - }}); err != nil { - t.Errorf("SetStat atime without mask failed: %v") - } - // Atime should be unchanged. - if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { - t.Errorf("Stat got error: %v", err) - } else if gotStat.Atime != initialStat.Atime { - t.Errorf("Stat got atime %d, want %d", gotStat.Atime, initialStat.Atime) - } - - // Set atime, this time included in the mask. - setStat := linux.Statx{ - Mask: linux.STATX_ATIME, - Atime: linux.NsecToStatxTimestamp(100), - } - if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: setStat}); err != nil { - t.Errorf("SetStat atime with mask failed: %v") - } - if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { - t.Errorf("Stat got error: %v", err) - } else if gotStat.Atime != setStat.Atime { - t.Errorf("Stat got atime %d, want %d", gotStat.Atime, setStat.Atime) - } - }) - } -} diff --git a/pkg/sentry/fsimpl/tmpfs/symlink.go b/pkg/sentry/fsimpl/tmpfs/symlink.go deleted file mode 100644 index 47e075ed4..000000000 --- a/pkg/sentry/fsimpl/tmpfs/symlink.go +++ /dev/null @@ -1,37 +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 tmpfs - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" -) - -type symlink struct { - inode inode - target string // immutable -} - -func (fs *filesystem) newSymlink(creds *auth.Credentials, target string) *inode { - link := &symlink{ - target: target, - } - link.inode.init(link, fs, creds, linux.S_IFLNK|0777) - link.inode.nlink = 1 // from parent directory - return &link.inode -} - -// O_PATH is unimplemented, so there's no way to get a FileDescription -// representing a symlink yet. diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go deleted file mode 100644 index 654e788e3..000000000 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ /dev/null @@ -1,564 +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 tmpfs provides a filesystem implementation that behaves like tmpfs: -// the Dentry tree is the sole source of truth for the state of the filesystem. -// -// Lock order: -// -// filesystem.mu -// inode.mu -// regularFileFD.offMu -// regularFile.mapsMu -// regularFile.dataMu -package tmpfs - -import ( - "fmt" - "math" - "sync/atomic" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/context" - fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/vfs" - "gvisor.dev/gvisor/pkg/sentry/vfs/lock" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserror" -) - -// Name is the default filesystem name. -const Name = "tmpfs" - -// FilesystemType implements vfs.FilesystemType. -type FilesystemType struct{} - -// filesystem implements vfs.FilesystemImpl. -type filesystem struct { - vfsfs vfs.Filesystem - - // memFile is used to allocate pages to for regular files. - memFile *pgalloc.MemoryFile - - // clock is a realtime clock used to set timestamps in file operations. - clock time.Clock - - // mu serializes changes to the Dentry tree. - mu sync.RWMutex - - nextInoMinusOne uint64 // accessed using atomic memory operations -} - -// Name implements vfs.FilesystemType.Name. -func (FilesystemType) Name() string { - return Name -} - -// FilesystemOpts is used to pass configuration data to tmpfs. -type FilesystemOpts struct { - // RootFileType is the FileType of the filesystem root. Valid values - // are: S_IFDIR, S_IFREG, and S_IFLNK. Defaults to S_IFDIR. - RootFileType uint16 - - // RootSymlinkTarget is the target of the root symlink. Only valid if - // RootFileType == S_IFLNK. - RootSymlinkTarget string - - // FilesystemType allows setting a different FilesystemType for this - // tmpfs filesystem. This allows tmpfs to "impersonate" other - // filesystems, like ramdiskfs and cgroupfs. - FilesystemType vfs.FilesystemType -} - -// GetFilesystem implements vfs.FilesystemType.GetFilesystem. -func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { - memFileProvider := pgalloc.MemoryFileProviderFromContext(ctx) - if memFileProvider == nil { - panic("MemoryFileProviderFromContext returned nil") - } - clock := time.RealtimeClockFromContext(ctx) - fs := filesystem{ - memFile: memFileProvider.MemoryFile(), - clock: clock, - } - - rootFileType := uint16(linux.S_IFDIR) - newFSType := vfs.FilesystemType(&fstype) - tmpfsOpts, ok := opts.InternalData.(FilesystemOpts) - if ok { - if tmpfsOpts.RootFileType != 0 { - rootFileType = tmpfsOpts.RootFileType - } - if tmpfsOpts.FilesystemType != nil { - newFSType = tmpfsOpts.FilesystemType - } - } - - fs.vfsfs.Init(vfsObj, newFSType, &fs) - - var root *inode - switch rootFileType { - case linux.S_IFREG: - root = fs.newRegularFile(creds, 0777) - case linux.S_IFLNK: - root = fs.newSymlink(creds, tmpfsOpts.RootSymlinkTarget) - case linux.S_IFDIR: - root = fs.newDirectory(creds, 01777) - default: - return nil, nil, fmt.Errorf("invalid tmpfs root file type: %#o", rootFileType) - } - return &fs.vfsfs, &fs.newDentry(root).vfsd, nil -} - -// Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { -} - -// 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 - - // tmpfs doesn't count references on dentries; because the dentry tree is - // the sole source of truth, it is by definition always consistent with the - // state of the filesystem. However, it does count references on inodes, - // because inode resources are released when all references are dropped. - // (tmpfs doesn't really have resources to release, but we implement - // reference counting because tmpfs regular files will.) - - // dentryEntry (ugh) links dentries into their parent directory.childList. - dentryEntry -} - -func (fs *filesystem) newDentry(inode *inode) *dentry { - d := &dentry{ - inode: inode, - } - d.vfsd.Init(d) - return d -} - -// IncRef implements vfs.DentryImpl.IncRef. -func (d *dentry) IncRef() { - d.inode.incRef() -} - -// TryIncRef implements vfs.DentryImpl.TryIncRef. -func (d *dentry) TryIncRef() bool { - return d.inode.tryIncRef() -} - -// DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef() { - d.inode.decRef() -} - -// inode represents a filesystem object. -type inode struct { - // clock is a realtime clock used to set timestamps in file operations. - clock time.Clock - - // refs is a reference count. refs is accessed using atomic memory - // operations. - // - // A reference is held on all inodes that are reachable in the filesystem - // tree. For non-directories (which may have multiple hard links), this - // means that a reference is dropped when nlink reaches 0. For directories, - // nlink never reaches 0 due to the "." entry; instead, - // filesystem.RmdirAt() drops the reference. - refs int64 - - // Inode metadata. Writing multiple fields atomically requires holding - // mu, othewise atomic operations can be used. - mu sync.Mutex - mode uint32 // file type and mode - nlink uint32 // protected by filesystem.mu instead of inode.mu - uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic - gid uint32 // auth.KGID, but ... - ino uint64 // immutable - - // Linux's tmpfs has no concept of btime. - atime int64 // nanoseconds - ctime int64 // nanoseconds - mtime int64 // nanoseconds - - // Only meaningful for device special files. - rdevMajor uint32 - rdevMinor uint32 - - // Advisory file locks, which lock at the inode level. - locks lock.FileLocks - - impl interface{} // immutable -} - -const maxLinks = math.MaxUint32 - -func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, mode linux.FileMode) { - if mode.FileType() == 0 { - panic("file type is required in FileMode") - } - i.clock = fs.clock - i.refs = 1 - i.mode = uint32(mode) - i.uid = uint32(creds.EffectiveKUID) - i.gid = uint32(creds.EffectiveKGID) - i.ino = atomic.AddUint64(&fs.nextInoMinusOne, 1) - // Tmpfs creation sets atime, ctime, and mtime to current time. - now := i.clock.Now().Nanoseconds() - i.atime = now - i.ctime = now - i.mtime = now - // i.nlink initialized by caller - i.impl = impl -} - -// incLinksLocked increments i's link count. -// -// Preconditions: filesystem.mu must be locked for writing. i.nlink != 0. -// i.nlink < maxLinks. -func (i *inode) incLinksLocked() { - if i.nlink == 0 { - panic("tmpfs.inode.incLinksLocked() called with no existing links") - } - if i.nlink == maxLinks { - panic("memfs.inode.incLinksLocked() called with maximum link count") - } - atomic.AddUint32(&i.nlink, 1) -} - -// decLinksLocked decrements i's link count. -// -// Preconditions: filesystem.mu must be locked for writing. i.nlink != 0. -func (i *inode) decLinksLocked() { - if i.nlink == 0 { - panic("tmpfs.inode.decLinksLocked() called with no existing links") - } - atomic.AddUint32(&i.nlink, ^uint32(0)) -} - -func (i *inode) incRef() { - if atomic.AddInt64(&i.refs, 1) <= 1 { - panic("tmpfs.inode.incRef() called without holding a reference") - } -} - -func (i *inode) tryIncRef() bool { - for { - refs := atomic.LoadInt64(&i.refs) - if refs == 0 { - return false - } - if atomic.CompareAndSwapInt64(&i.refs, refs, refs+1) { - return true - } - } -} - -func (i *inode) decRef() { - if refs := atomic.AddInt64(&i.refs, -1); refs == 0 { - if regFile, ok := i.impl.(*regularFile); ok { - // Hold inode.mu and regFile.dataMu while mutating - // size. - i.mu.Lock() - regFile.dataMu.Lock() - regFile.data.DropAll(regFile.memFile) - atomic.StoreUint64(®File.size, 0) - regFile.dataMu.Unlock() - i.mu.Unlock() - } - } else if refs < 0 { - panic("tmpfs.inode.decRef() called without holding a reference") - } -} - -func (i *inode) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { - mode := linux.FileMode(atomic.LoadUint32(&i.mode)) - return vfs.GenericCheckPermissions(creds, ats, mode, auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))) -} - -// Go won't inline this function, and returning linux.Statx (which is quite -// big) means spending a lot of time in runtime.duffcopy(), so instead it's an -// output parameter. -// -// Note that Linux does not guarantee to return consistent data (in the case of -// a concurrent modification), so we do not require holding inode.mu. -func (i *inode) statTo(stat *linux.Statx) { - stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | - linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_ATIME | - linux.STATX_BTIME | linux.STATX_CTIME | linux.STATX_MTIME - stat.Blksize = 1 // usermem.PageSize in tmpfs - stat.Nlink = atomic.LoadUint32(&i.nlink) - stat.UID = atomic.LoadUint32(&i.uid) - stat.GID = atomic.LoadUint32(&i.gid) - stat.Mode = uint16(atomic.LoadUint32(&i.mode)) - stat.Ino = i.ino - // Linux's tmpfs has no concept of btime, so zero-value is returned. - stat.Atime = linux.NsecToStatxTimestamp(i.atime) - stat.Ctime = linux.NsecToStatxTimestamp(i.ctime) - stat.Mtime = linux.NsecToStatxTimestamp(i.mtime) - // TODO(gvisor.dev/issue/1197): Device number. - switch impl := i.impl.(type) { - case *regularFile: - stat.Mask |= linux.STATX_SIZE | linux.STATX_BLOCKS - stat.Size = uint64(atomic.LoadUint64(&impl.size)) - // In tmpfs, this will be FileRangeSet.Span() / 512 (but also cached in - // a uint64 accessed using atomic memory operations to avoid taking - // locks). - stat.Blocks = allocatedBlocksForSize(stat.Size) - case *symlink: - stat.Mask |= linux.STATX_SIZE | linux.STATX_BLOCKS - stat.Size = uint64(len(impl.target)) - stat.Blocks = allocatedBlocksForSize(stat.Size) - case *deviceFile: - stat.RdevMajor = impl.major - stat.RdevMinor = impl.minor - case *socketFile, *directory, *namedPipe: - // Nothing to do. - default: - panic(fmt.Sprintf("unknown inode type: %T", i.impl)) - } -} - -func (i *inode) setStat(ctx context.Context, creds *auth.Credentials, stat *linux.Statx) error { - if stat.Mask == 0 { - return nil - } - if stat.Mask&^(linux.STATX_MODE|linux.STATX_UID|linux.STATX_GID|linux.STATX_ATIME|linux.STATX_MTIME|linux.STATX_CTIME|linux.STATX_SIZE) != 0 { - return syserror.EPERM - } - mode := linux.FileMode(atomic.LoadUint32(&i.mode)) - if err := vfs.CheckSetStat(ctx, creds, stat, mode, auth.KUID(atomic.LoadUint32(&i.uid)), auth.KGID(atomic.LoadUint32(&i.gid))); err != nil { - return err - } - i.mu.Lock() - var ( - needsMtimeBump bool - needsCtimeBump bool - ) - mask := stat.Mask - if mask&linux.STATX_MODE != 0 { - ft := atomic.LoadUint32(&i.mode) & linux.S_IFMT - atomic.StoreUint32(&i.mode, ft|uint32(stat.Mode&^linux.S_IFMT)) - needsCtimeBump = true - } - if mask&linux.STATX_UID != 0 { - atomic.StoreUint32(&i.uid, stat.UID) - needsCtimeBump = true - } - if mask&linux.STATX_GID != 0 { - atomic.StoreUint32(&i.gid, stat.GID) - needsCtimeBump = true - } - if mask&linux.STATX_SIZE != 0 { - switch impl := i.impl.(type) { - case *regularFile: - updated, err := impl.truncateLocked(stat.Size) - if err != nil { - return err - } - if updated { - needsMtimeBump = true - needsCtimeBump = true - } - case *directory: - return syserror.EISDIR - default: - return syserror.EINVAL - } - } - now := i.clock.Now().Nanoseconds() - if mask&linux.STATX_ATIME != 0 { - if stat.Atime.Nsec == linux.UTIME_NOW { - atomic.StoreInt64(&i.atime, now) - } else { - atomic.StoreInt64(&i.atime, stat.Atime.ToNsecCapped()) - } - needsCtimeBump = true - } - if mask&linux.STATX_MTIME != 0 { - if stat.Mtime.Nsec == linux.UTIME_NOW { - atomic.StoreInt64(&i.mtime, now) - } else { - atomic.StoreInt64(&i.mtime, stat.Mtime.ToNsecCapped()) - } - needsCtimeBump = true - // Ignore the mtime bump, since we just set it ourselves. - needsMtimeBump = false - } - if mask&linux.STATX_CTIME != 0 { - if stat.Ctime.Nsec == linux.UTIME_NOW { - atomic.StoreInt64(&i.ctime, now) - } else { - atomic.StoreInt64(&i.ctime, stat.Ctime.ToNsecCapped()) - } - // Ignore the ctime bump, since we just set it ourselves. - needsCtimeBump = false - } - if needsMtimeBump { - atomic.StoreInt64(&i.mtime, now) - } - if needsCtimeBump { - atomic.StoreInt64(&i.ctime, now) - } - - i.mu.Unlock() - return nil -} - -// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. -func (i *inode) lockBSD(uid fslock.UniqueID, t fslock.LockType, block fslock.Blocker) error { - switch i.impl.(type) { - case *regularFile: - return i.locks.LockBSD(uid, t, block) - } - return syserror.EBADF -} - -// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. -func (i *inode) unlockBSD(uid fslock.UniqueID) error { - switch i.impl.(type) { - case *regularFile: - i.locks.UnlockBSD(uid) - return nil - } - return syserror.EBADF -} - -// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. -func (i *inode) lockPOSIX(uid fslock.UniqueID, t fslock.LockType, rng fslock.LockRange, block fslock.Blocker) error { - switch i.impl.(type) { - case *regularFile: - return i.locks.LockPOSIX(uid, t, rng, block) - } - return syserror.EBADF -} - -// TODO(gvisor.dev/issue/1480): support file locking for file types other than regular. -func (i *inode) unlockPOSIX(uid fslock.UniqueID, rng fslock.LockRange) error { - switch i.impl.(type) { - case *regularFile: - i.locks.UnlockPOSIX(uid, rng) - return nil - } - return syserror.EBADF -} - -// allocatedBlocksForSize returns the number of 512B blocks needed to -// accommodate the given size in bytes, as appropriate for struct -// stat::st_blocks and struct statx::stx_blocks. (Note that this 512B block -// size is independent of the "preferred block size for I/O", struct -// stat::st_blksize and struct statx::stx_blksize.) -func allocatedBlocksForSize(size uint64) uint64 { - return (size + 511) / 512 -} - -func (i *inode) direntType() uint8 { - switch impl := i.impl.(type) { - case *regularFile: - return linux.DT_REG - case *directory: - return linux.DT_DIR - case *symlink: - return linux.DT_LNK - case *socketFile: - return linux.DT_SOCK - case *deviceFile: - switch impl.kind { - case vfs.BlockDevice: - return linux.DT_BLK - case vfs.CharDevice: - return linux.DT_CHR - default: - panic(fmt.Sprintf("unknown vfs.DeviceKind: %v", impl.kind)) - } - default: - panic(fmt.Sprintf("unknown inode type: %T", i.impl)) - } -} - -func (i *inode) isDir() bool { - return linux.FileMode(i.mode).FileType() == linux.S_IFDIR -} - -func (i *inode) touchAtime(mnt *vfs.Mount) { - if err := mnt.CheckBeginWrite(); err != nil { - return - } - now := i.clock.Now().Nanoseconds() - i.mu.Lock() - atomic.StoreInt64(&i.atime, now) - i.mu.Unlock() - mnt.EndWrite() -} - -// Preconditions: The caller has called vfs.Mount.CheckBeginWrite(). -func (i *inode) touchCtime() { - now := i.clock.Now().Nanoseconds() - i.mu.Lock() - atomic.StoreInt64(&i.ctime, now) - i.mu.Unlock() -} - -// Preconditions: The caller has called vfs.Mount.CheckBeginWrite(). -func (i *inode) touchCMtime() { - now := i.clock.Now().Nanoseconds() - i.mu.Lock() - atomic.StoreInt64(&i.mtime, now) - atomic.StoreInt64(&i.ctime, now) - i.mu.Unlock() -} - -// Preconditions: The caller has called vfs.Mount.CheckBeginWrite() and holds -// inode.mu. -func (i *inode) touchCMtimeLocked() { - now := i.clock.Now().Nanoseconds() - atomic.StoreInt64(&i.mtime, now) - atomic.StoreInt64(&i.ctime, now) -} - -// fileDescription is embedded by tmpfs implementations of -// vfs.FileDescriptionImpl. -type fileDescription struct { - vfsfd vfs.FileDescription - vfs.FileDescriptionDefaultImpl -} - -func (fd *fileDescription) filesystem() *filesystem { - return fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) -} - -func (fd *fileDescription) inode() *inode { - return fd.vfsfd.Dentry().Impl().(*dentry).inode -} - -// Stat implements vfs.FileDescriptionImpl.Stat. -func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { - var stat linux.Statx - fd.inode().statTo(&stat) - return stat, nil -} - -// SetStat implements vfs.FileDescriptionImpl.SetStat. -func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { - creds := auth.CredentialsFromContext(ctx) - return fd.inode().setStat(ctx, creds, &opts.Stat) -} |