diff options
64 files changed, 1577 insertions, 1214 deletions
@@ -4,10 +4,10 @@ load("@bazel_tools//tools/build_defs/repo:git.bzl", "git_repository") # Load go bazel rules and gazelle. http_archive( name = "io_bazel_rules_go", - sha256 = "94f90feaa65c9cdc840cd21f67d967870b5943d684966a47569da8073e42063d", + sha256 = "db2b2d35293f405430f553bc7a865a8749a8ef60c30287e90d2b278c32771afe", urls = [ - "https://mirror.bazel.build/github.com/bazelbuild/rules_go/releases/download/v0.22.0/rules_go-v0.22.0.tar.gz", - "https://github.com/bazelbuild/rules_go/releases/download/v0.22.0/rules_go-v0.22.0.tar.gz", + "https://mirror.bazel.build/github.com/bazelbuild/rules_go/releases/download/v0.22.3/rules_go-v0.22.3.tar.gz", + "https://github.com/bazelbuild/rules_go/releases/download/v0.22.3/rules_go-v0.22.3.tar.gz", ], ) @@ -25,7 +25,7 @@ load("@io_bazel_rules_go//go:deps.bzl", "go_register_toolchains", "go_rules_depe go_rules_dependencies() go_register_toolchains( - go_version = "1.14", + go_version = "1.14.2", nogo = "@//:nogo", ) @@ -99,11 +99,11 @@ pip_install() # See releases at https://releases.bazel.build/bazel-toolchains.html http_archive( name = "bazel_toolchains", - sha256 = "b5a8039df7119d618402472f3adff8a1bd0ae9d5e253f53fcc4c47122e91a3d2", - strip_prefix = "bazel-toolchains-2.1.1", + sha256 = "239a1a673861eabf988e9804f45da3b94da28d1aff05c373b013193c315d9d9e", + strip_prefix = "bazel-toolchains-3.0.1", urls = [ - "https://github.com/bazelbuild/bazel-toolchains/releases/download/2.1.1/bazel-toolchains-2.1.1.tar.gz", - "https://mirror.bazel.build/github.com/bazelbuild/bazel-toolchains/archive/2.1.1.tar.gz", + "https://github.com/bazelbuild/bazel-toolchains/releases/download/3.0.1/bazel-toolchains-3.0.1.tar.gz", + "https://mirror.bazel.build/github.com/bazelbuild/bazel-toolchains/releases/download/3.0.1/bazel-toolchains-3.0.1.tar.gz", ], ) diff --git a/pkg/p9/handlers.go b/pkg/p9/handlers.go index a8b714cf5..1db5797dd 100644 --- a/pkg/p9/handlers.go +++ b/pkg/p9/handlers.go @@ -48,6 +48,8 @@ func ExtractErrno(err error) syscall.Errno { return ExtractErrno(e.Err) case *os.SyscallError: return ExtractErrno(e.Err) + case *os.LinkError: + return ExtractErrno(e.Err) } // Default case. diff --git a/pkg/safecopy/memcpy_amd64.s b/pkg/safecopy/memcpy_amd64.s index 129691d68..00b46c18f 100644 --- a/pkg/safecopy/memcpy_amd64.s +++ b/pkg/safecopy/memcpy_amd64.s @@ -55,15 +55,9 @@ TEXT ·memcpy(SB), NOSPLIT, $0-36 MOVQ from+8(FP), SI MOVQ n+16(FP), BX - // REP instructions have a high startup cost, so we handle small sizes - // with some straightline code. The REP MOVSQ instruction is really fast - // for large sizes. The cutover is approximately 2K. tail: - // move_129through256 or smaller work whether or not the source and the - // destination memory regions overlap because they load all data into - // registers before writing it back. move_256through2048 on the other - // hand can be used only when the memory regions don't overlap or the copy - // direction is forward. + // BSR+branch table make almost all memmove/memclr benchmarks worse. Not + // worth doing. TESTQ BX, BX JEQ move_0 CMPQ BX, $2 @@ -83,31 +77,45 @@ tail: JBE move_65through128 CMPQ BX, $256 JBE move_129through256 - // TODO: use branch table and BSR to make this just a single dispatch -/* - * forward copy loop - */ - CMPQ BX, $2048 - JLS move_256through2048 - - // Check alignment - MOVL SI, AX - ORL DI, AX - TESTL $7, AX - JEQ fwdBy8 - - // Do 1 byte at a time - MOVQ BX, CX - REP; MOVSB - RET - -fwdBy8: - // Do 8 bytes at a time - MOVQ BX, CX - SHRQ $3, CX - ANDQ $7, BX - REP; MOVSQ +move_257plus: + SUBQ $256, BX + MOVOU (SI), X0 + MOVOU X0, (DI) + MOVOU 16(SI), X1 + MOVOU X1, 16(DI) + MOVOU 32(SI), X2 + MOVOU X2, 32(DI) + MOVOU 48(SI), X3 + MOVOU X3, 48(DI) + MOVOU 64(SI), X4 + MOVOU X4, 64(DI) + MOVOU 80(SI), X5 + MOVOU X5, 80(DI) + MOVOU 96(SI), X6 + MOVOU X6, 96(DI) + MOVOU 112(SI), X7 + MOVOU X7, 112(DI) + MOVOU 128(SI), X8 + MOVOU X8, 128(DI) + MOVOU 144(SI), X9 + MOVOU X9, 144(DI) + MOVOU 160(SI), X10 + MOVOU X10, 160(DI) + MOVOU 176(SI), X11 + MOVOU X11, 176(DI) + MOVOU 192(SI), X12 + MOVOU X12, 192(DI) + MOVOU 208(SI), X13 + MOVOU X13, 208(DI) + MOVOU 224(SI), X14 + MOVOU X14, 224(DI) + MOVOU 240(SI), X15 + MOVOU X15, 240(DI) + CMPQ BX, $256 + LEAQ 256(SI), SI + LEAQ 256(DI), DI + JGE move_257plus JMP tail move_1or2: @@ -209,42 +217,3 @@ move_129through256: MOVOU -16(SI)(BX*1), X15 MOVOU X15, -16(DI)(BX*1) RET -move_256through2048: - SUBQ $256, BX - MOVOU (SI), X0 - MOVOU X0, (DI) - MOVOU 16(SI), X1 - MOVOU X1, 16(DI) - MOVOU 32(SI), X2 - MOVOU X2, 32(DI) - MOVOU 48(SI), X3 - MOVOU X3, 48(DI) - MOVOU 64(SI), X4 - MOVOU X4, 64(DI) - MOVOU 80(SI), X5 - MOVOU X5, 80(DI) - MOVOU 96(SI), X6 - MOVOU X6, 96(DI) - MOVOU 112(SI), X7 - MOVOU X7, 112(DI) - MOVOU 128(SI), X8 - MOVOU X8, 128(DI) - MOVOU 144(SI), X9 - MOVOU X9, 144(DI) - MOVOU 160(SI), X10 - MOVOU X10, 160(DI) - MOVOU 176(SI), X11 - MOVOU X11, 176(DI) - MOVOU 192(SI), X12 - MOVOU X12, 192(DI) - MOVOU 208(SI), X13 - MOVOU X13, 208(DI) - MOVOU 224(SI), X14 - MOVOU X14, 224(DI) - MOVOU 240(SI), X15 - MOVOU X15, 240(DI) - CMPQ BX, $256 - LEAQ 256(SI), SI - LEAQ 256(DI), DI - JGE move_256through2048 - JMP tail diff --git a/pkg/sentry/fs/dirent.go b/pkg/sentry/fs/dirent.go index 0266a5287..65be12175 100644 --- a/pkg/sentry/fs/dirent.go +++ b/pkg/sentry/fs/dirent.go @@ -312,9 +312,9 @@ func (d *Dirent) SyncAll(ctx context.Context) { // There is nothing to sync for a read-only filesystem. if !d.Inode.MountSource.Flags.ReadOnly { - // FIXME(b/34856369): This should be a mount traversal, not a - // Dirent traversal, because some Inodes that need to be synced - // may no longer be reachable by name (after sys_unlink). + // NOTE(b/34856369): This should be a mount traversal, not a Dirent + // traversal, because some Inodes that need to be synced may no longer + // be reachable by name (after sys_unlink). // // Write out metadata, dirty page cached pages, and sync disk/remote // caches. diff --git a/pkg/sentry/fs/gofer/inode.go b/pkg/sentry/fs/gofer/inode.go index 1c934981b..811e8ea30 100644 --- a/pkg/sentry/fs/gofer/inode.go +++ b/pkg/sentry/fs/gofer/inode.go @@ -710,13 +710,10 @@ func init() { } // AddLink implements InodeOperations.AddLink, but is currently a noop. -// FIXME(b/63117438): Remove this from InodeOperations altogether. func (*inodeOperations) AddLink() {} // DropLink implements InodeOperations.DropLink, but is currently a noop. -// FIXME(b/63117438): Remove this from InodeOperations altogether. func (*inodeOperations) DropLink() {} // NotifyStatusChange implements fs.InodeOperations.NotifyStatusChange. -// FIXME(b/63117438): Remove this from InodeOperations altogether. func (i *inodeOperations) NotifyStatusChange(ctx context.Context) {} diff --git a/pkg/sentry/fs/gofer/util.go b/pkg/sentry/fs/gofer/util.go index 2d8d3a2ea..47a6c69bf 100644 --- a/pkg/sentry/fs/gofer/util.go +++ b/pkg/sentry/fs/gofer/util.go @@ -20,17 +20,29 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/p9" "gvisor.dev/gvisor/pkg/sentry/fs" + ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" ) func utimes(ctx context.Context, file contextFile, ts fs.TimeSpec) error { if ts.ATimeOmit && ts.MTimeOmit { return nil } + + // Replace requests to use the "system time" with the current time to + // ensure that timestamps remain consistent with the remote + // filesystem. + now := ktime.NowFromContext(ctx) + if ts.ATimeSetSystemTime { + ts.ATime = now + } + if ts.MTimeSetSystemTime { + ts.MTime = now + } mask := p9.SetAttrMask{ ATime: !ts.ATimeOmit, - ATimeNotSystemTime: !ts.ATimeSetSystemTime, + ATimeNotSystemTime: true, MTime: !ts.MTimeOmit, - MTimeNotSystemTime: !ts.MTimeSetSystemTime, + MTimeNotSystemTime: true, } as, ans := ts.ATime.Unix() ms, mns := ts.MTime.Unix() diff --git a/pkg/sentry/fs/host/inode.go b/pkg/sentry/fs/host/inode.go index 1da3c0a17..62f1246aa 100644 --- a/pkg/sentry/fs/host/inode.go +++ b/pkg/sentry/fs/host/inode.go @@ -397,15 +397,12 @@ func (i *inodeOperations) StatFS(context.Context) (fs.Info, error) { } // AddLink implements fs.InodeOperations.AddLink. -// FIXME(b/63117438): Remove this from InodeOperations altogether. func (i *inodeOperations) AddLink() {} // DropLink implements fs.InodeOperations.DropLink. -// FIXME(b/63117438): Remove this from InodeOperations altogether. func (i *inodeOperations) DropLink() {} // NotifyStatusChange implements fs.InodeOperations.NotifyStatusChange. -// FIXME(b/63117438): Remove this from InodeOperations altogether. func (i *inodeOperations) NotifyStatusChange(ctx context.Context) {} // readdirAll returns all of the directory entries in i. diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go index 55fb71c16..73f89abcc 100644 --- a/pkg/sentry/fs/inode.go +++ b/pkg/sentry/fs/inode.go @@ -397,8 +397,6 @@ func (i *Inode) Getlink(ctx context.Context) (*Dirent, error) { // AddLink calls i.InodeOperations.AddLink. func (i *Inode) AddLink() { if i.overlay != nil { - // FIXME(b/63117438): Remove this from InodeOperations altogether. - // // This interface is only used by ramfs to update metadata of // children. These filesystems should _never_ have overlay // Inodes cached as children. So explicitly disallow this diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index d6c5dd2c1..4d42eac83 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -57,6 +57,16 @@ func getTaskMM(t *kernel.Task) (*mm.MemoryManager, error) { 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 +} + // taskDir represents a task-level directory. // // +stateify savable @@ -254,11 +264,12 @@ func newExe(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { } func (e *exe) executable() (file fsbridge.File, err error) { + if err := checkTaskState(e.t); err != nil { + return nil, err + } e.t.WithMuLocked(func(t *kernel.Task) { mm := t.MemoryManager() if mm == nil { - // TODO(b/34851096): Check shouldn't allow Readlink once the - // Task is zombied. err = syserror.EACCES return } @@ -268,7 +279,7 @@ func (e *exe) executable() (file fsbridge.File, err error) { // (with locks held). file = mm.Executable() if file == nil { - err = syserror.ENOENT + err = syserror.ESRCH } }) return @@ -313,11 +324,22 @@ func newNamespaceSymlink(t *kernel.Task, msrc *fs.MountSource, name string) *fs. return newProcInode(t, n, msrc, fs.Symlink, t) } +// Readlink reads the symlink value. +func (n *namespaceSymlink) Readlink(ctx context.Context, inode *fs.Inode) (string, error) { + if err := checkTaskState(n.t); err != nil { + return "", err + } + return n.Symlink.Readlink(ctx, inode) +} + // Getlink implements fs.InodeOperations.Getlink. func (n *namespaceSymlink) Getlink(ctx context.Context, inode *fs.Inode) (*fs.Dirent, error) { if !kernel.ContextCanTrace(ctx, n.t, false) { return nil, syserror.EACCES } + if err := checkTaskState(n.t); err != nil { + return nil, err + } // Create a new regular file to fake the namespace file. iops := fsutil.NewNoReadWriteFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0777), linux.PROC_SUPER_MAGIC) diff --git a/pkg/sentry/fs/tmpfs/fs.go b/pkg/sentry/fs/tmpfs/fs.go index d5be56c3f..bc117ca6a 100644 --- a/pkg/sentry/fs/tmpfs/fs.go +++ b/pkg/sentry/fs/tmpfs/fs.go @@ -44,9 +44,6 @@ const ( // lookup. cacheRevalidate = "revalidate" - // TODO(edahlgren/mpratt): support a tmpfs size limit. - // size = "size" - // Permissions that exceed modeMask will be rejected. modeMask = 01777 diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 16a3c18ae..4433071aa 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -682,7 +682,7 @@ func (fs *Filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu if err != nil { return linux.Statfs{}, err } - // TODO: actually implement statfs + // TODO(gvisor.dev/issue/1193): actually implement statfs. return linux.Statfs{}, syserror.ENOSYS } diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index aee2a4392..888afc0fd 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -214,22 +214,6 @@ func newIO(t *kernel.Task, isThreadGroup bool) *ioData { return &ioData{ioUsage: t} } -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 := &kernfs.StaticSymlink{} - // 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 -} - // 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: diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index 9c8656b28..046265eca 100644 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -30,34 +30,35 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) -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 -} - -func (i *fdDir) lookup(name string) (*vfs.FileDescription, kernel.FDFlags, error) { - fd, err := strconv.ParseUint(name, 10, 64) - if err != nil { - return nil, kernel.FDFlags{}, syserror.ENOENT - } - +func getTaskFD(t *kernel.Task, fd int32) (*vfs.FileDescription, kernel.FDFlags) { var ( file *vfs.FileDescription flags kernel.FDFlags ) - i.task.WithMuLocked(func(t *kernel.Task) { - if fdTable := t.FDTable(); fdTable != nil { - file, flags = fdTable.GetVFS2(int32(fd)) + 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 nil, kernel.FDFlags{}, syserror.ENOENT + return false } - return file, flags, nil + 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. @@ -128,11 +129,15 @@ func newFDDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { // Lookup implements kernfs.inodeDynamicLookup. func (i *fdDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { - file, _, err := i.lookup(name) + fdInt, err := strconv.ParseInt(name, 10, 32) if err != nil { - return nil, err + return nil, syserror.ENOENT + } + fd := int32(fdInt) + if !taskFDExists(i.task, fd) { + return nil, syserror.ENOENT } - taskDentry := newFDSymlink(i.task.Credentials(), file, i.inoGen.NextIno()) + taskDentry := newFDSymlink(i.task, fd, i.inoGen.NextIno()) return taskDentry.VFSDentry(), nil } @@ -169,19 +174,22 @@ func (i *fdDirInode) CheckPermissions(ctx context.Context, creds *auth.Credentia // // +stateify savable type fdSymlink struct { - refs.AtomicRefCount kernfs.InodeAttrs + kernfs.InodeNoopRefCount kernfs.InodeSymlink - file *vfs.FileDescription + task *kernel.Task + fd int32 } var _ kernfs.Inode = (*fdSymlink)(nil) -func newFDSymlink(creds *auth.Credentials, file *vfs.FileDescription, ino uint64) *kernfs.Dentry { - file.IncRef() - inode := &fdSymlink{file: file} - inode.Init(creds, ino, linux.ModeSymlink|0777) +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) @@ -189,29 +197,27 @@ func newFDSymlink(creds *auth.Credentials, file *vfs.FileDescription, ino uint64 } 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() - - vfsObj := s.file.VirtualDentry().Mount().Filesystem().VirtualFilesystem() - return vfsObj.PathnameWithDeleted(ctx, root, s.file.VirtualDentry()) + return s.task.Kernel().VFS().PathnameWithDeleted(ctx, root, file.VirtualDentry()) } func (s *fdSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, error) { - vd := s.file.VirtualDentry() + 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 } -func (s *fdSymlink) DecRef() { - s.AtomicRefCount.DecRefWithDestructor(func() { - s.Destroy() - }) -} - -func (s *fdSymlink) Destroy() { - s.file.DecRef() -} - // fdInfoDirInode represents the inode for /proc/[pid]/fdinfo directory. // // +stateify savable @@ -244,12 +250,18 @@ func newFDInfoDirInode(task *kernel.Task, inoGen InoGenerator) *kernfs.Dentry { // Lookup implements kernfs.inodeDynamicLookup. func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { - file, flags, err := i.lookup(name) + fdInt, err := strconv.ParseInt(name, 10, 32) if err != nil { - return nil, err + return nil, syserror.ENOENT + } + fd := int32(fdInt) + if !taskFDExists(i.task, fd) { + return nil, syserror.ENOENT + } + data := &fdInfoData{ + task: i.task, + fd: fd, } - - data := &fdInfoData{file: file, flags: flags} dentry := newTaskOwnedFile(i.task, i.inoGen.NextIno(), 0444, data) return dentry.VFSDentry(), nil } @@ -268,26 +280,23 @@ type fdInfoData struct { kernfs.DynamicBytesFile refs.AtomicRefCount - file *vfs.FileDescription - flags kernel.FDFlags + task *kernel.Task + fd int32 } var _ dynamicInode = (*fdInfoData)(nil) -func (d *fdInfoData) DecRef() { - d.AtomicRefCount.DecRefWithDestructor(d.destroy) -} - -func (d *fdInfoData) destroy() { - d.file.DecRef() -} - // 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(d.file.StatusFlags()) | d.flags.ToLinuxFileFlags() + 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 index 88ea6a6d8..2c6f8bdfc 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -64,6 +64,16 @@ func getMMIncRef(task *kernel.Task) (*mm.MemoryManager, error) { 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 } @@ -628,11 +638,13 @@ func (s *exeSymlink) Getlink(ctx context.Context) (vfs.VirtualDentry, string, er } 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 { - // TODO(b/34851096): Check shouldn't allow Readlink once the - // Task is zombied. err = syserror.EACCES return } @@ -642,7 +654,7 @@ func (s *exeSymlink) executable() (file fsbridge.File, err error) { // (with locks held). file = mm.Executable() if file == nil { - err = syserror.ENOENT + err = syserror.ESRCH } }) return @@ -709,3 +721,41 @@ func (i *mountsData) Generate(ctx context.Context, buf *bytes.Buffer) error { 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/kernel/ptrace.go b/pkg/sentry/kernel/ptrace.go index 35ad97d5d..e23e796ef 100644 --- a/pkg/sentry/kernel/ptrace.go +++ b/pkg/sentry/kernel/ptrace.go @@ -184,7 +184,6 @@ func (t *Task) CanTrace(target *Task, attach bool) bool { if targetCreds.PermittedCaps&^callerCreds.PermittedCaps != 0 { return false } - // TODO: Yama LSM return true } diff --git a/pkg/sentry/mm/address_space.go b/pkg/sentry/mm/address_space.go index 0332fc71c..5c667117c 100644 --- a/pkg/sentry/mm/address_space.go +++ b/pkg/sentry/mm/address_space.go @@ -201,8 +201,10 @@ func (mm *MemoryManager) mapASLocked(pseg pmaIterator, ar usermem.AddrRange, pre if pma.needCOW { perms.Write = false } - if err := mm.as.MapFile(pmaMapAR.Start, pma.file, pseg.fileRangeOf(pmaMapAR), perms, precommit); err != nil { - return err + if perms.Any() { // MapFile precondition + if err := mm.as.MapFile(pmaMapAR.Start, pma.file, pseg.fileRangeOf(pmaMapAR), perms, precommit); err != nil { + return err + } } pseg = pseg.NextSegment() } diff --git a/pkg/sentry/mm/aio_context.go b/pkg/sentry/mm/aio_context.go index cb29d94b0..379148903 100644 --- a/pkg/sentry/mm/aio_context.go +++ b/pkg/sentry/mm/aio_context.go @@ -59,25 +59,27 @@ func (a *aioManager) newAIOContext(events uint32, id uint64) bool { } a.contexts[id] = &AIOContext{ - done: make(chan struct{}, 1), + requestReady: make(chan struct{}, 1), maxOutstanding: events, } return true } -// destroyAIOContext destroys an asynchronous I/O context. +// destroyAIOContext destroys an asynchronous I/O context. It doesn't wait for +// for pending requests to complete. Returns the destroyed AIOContext so it can +// be drained. // -// False is returned if the context does not exist. -func (a *aioManager) destroyAIOContext(id uint64) bool { +// Nil is returned if the context does not exist. +func (a *aioManager) destroyAIOContext(id uint64) *AIOContext { a.mu.Lock() defer a.mu.Unlock() ctx, ok := a.contexts[id] if !ok { - return false + return nil } delete(a.contexts, id) ctx.destroy() - return true + return ctx } // lookupAIOContext looks up the given context. @@ -102,8 +104,8 @@ type ioResult struct { // // +stateify savable type AIOContext struct { - // done is the notification channel used for all requests. - done chan struct{} `state:"nosave"` + // requestReady is the notification channel used for all requests. + requestReady chan struct{} `state:"nosave"` // mu protects below. mu sync.Mutex `state:"nosave"` @@ -129,8 +131,14 @@ func (ctx *AIOContext) destroy() { ctx.mu.Lock() defer ctx.mu.Unlock() ctx.dead = true - if ctx.outstanding == 0 { - close(ctx.done) + ctx.checkForDone() +} + +// Preconditions: ctx.mu must be held by caller. +func (ctx *AIOContext) checkForDone() { + if ctx.dead && ctx.outstanding == 0 { + close(ctx.requestReady) + ctx.requestReady = nil } } @@ -154,11 +162,12 @@ func (ctx *AIOContext) PopRequest() (interface{}, bool) { // Is there anything ready? if e := ctx.results.Front(); e != nil { - ctx.results.Remove(e) - ctx.outstanding-- - if ctx.outstanding == 0 && ctx.dead { - close(ctx.done) + if ctx.outstanding == 0 { + panic("AIOContext outstanding is going negative") } + ctx.outstanding-- + ctx.results.Remove(e) + ctx.checkForDone() return e.data, true } return nil, false @@ -172,26 +181,58 @@ func (ctx *AIOContext) FinishRequest(data interface{}) { // Push to the list and notify opportunistically. The channel notify // here is guaranteed to be safe because outstanding must be non-zero. - // The done channel is only closed when outstanding reaches zero. + // The requestReady channel is only closed when outstanding reaches zero. ctx.results.PushBack(&ioResult{data: data}) select { - case ctx.done <- struct{}{}: + case ctx.requestReady <- struct{}{}: default: } } // WaitChannel returns a channel that is notified when an AIO request is -// completed. -// -// The boolean return value indicates whether or not the context is active. -func (ctx *AIOContext) WaitChannel() (chan struct{}, bool) { +// completed. Returns nil if the context is destroyed and there are no more +// outstanding requests. +func (ctx *AIOContext) WaitChannel() chan struct{} { ctx.mu.Lock() defer ctx.mu.Unlock() - if ctx.outstanding == 0 && ctx.dead { - return nil, false + return ctx.requestReady +} + +// Dead returns true if the context has been destroyed. +func (ctx *AIOContext) Dead() bool { + ctx.mu.Lock() + defer ctx.mu.Unlock() + return ctx.dead +} + +// CancelPendingRequest forgets about a request that hasn't yet completed. +func (ctx *AIOContext) CancelPendingRequest() { + ctx.mu.Lock() + defer ctx.mu.Unlock() + + if ctx.outstanding == 0 { + panic("AIOContext outstanding is going negative") } - return ctx.done, true + ctx.outstanding-- + ctx.checkForDone() +} + +// Drain drops all completed requests. Pending requests remain untouched. +func (ctx *AIOContext) Drain() { + ctx.mu.Lock() + defer ctx.mu.Unlock() + + if ctx.outstanding == 0 { + return + } + size := uint32(ctx.results.Len()) + if ctx.outstanding < size { + panic("AIOContext outstanding is going negative") + } + ctx.outstanding -= size + ctx.results.Reset() + ctx.checkForDone() } // aioMappable implements memmap.MappingIdentity and memmap.Mappable for AIO @@ -332,9 +373,9 @@ func (mm *MemoryManager) NewAIOContext(ctx context.Context, events uint32) (uint Length: aioRingBufferSize, MappingIdentity: m, Mappable: m, - // TODO(fvoznika): Linux does "do_mmap_pgoff(..., PROT_READ | - // PROT_WRITE, ...)" in fs/aio.c:aio_setup_ring(); why do we make this - // mapping read-only? + // Linux uses "do_mmap_pgoff(..., PROT_READ | PROT_WRITE, ...)" in + // fs/aio.c:aio_setup_ring(). Since we don't implement AIO_RING_MAGIC, + // user mode should not write to this page. Perms: usermem.Read, MaxPerms: usermem.Read, }) @@ -349,11 +390,11 @@ func (mm *MemoryManager) NewAIOContext(ctx context.Context, events uint32) (uint return id, nil } -// DestroyAIOContext destroys an asynchronous I/O context. It returns false if -// the context does not exist. -func (mm *MemoryManager) DestroyAIOContext(ctx context.Context, id uint64) bool { +// DestroyAIOContext destroys an asynchronous I/O context. It returns the +// destroyed context. nil if the context does not exist. +func (mm *MemoryManager) DestroyAIOContext(ctx context.Context, id uint64) *AIOContext { if _, ok := mm.LookupAIOContext(ctx, id); !ok { - return false + return nil } // Only unmaps after it assured that the address is a valid aio context to diff --git a/pkg/sentry/mm/aio_context_state.go b/pkg/sentry/mm/aio_context_state.go index c37fc9f7b..3dabac1af 100644 --- a/pkg/sentry/mm/aio_context_state.go +++ b/pkg/sentry/mm/aio_context_state.go @@ -16,5 +16,5 @@ package mm // afterLoad is invoked by stateify. func (a *AIOContext) afterLoad() { - a.done = make(chan struct{}, 1) + a.requestReady = make(chan struct{}, 1) } diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 5d0085462..7ac38764d 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -300,7 +300,7 @@ type SocketOperations struct { // New creates a new endpoint socket. func New(t *kernel.Task, family int, skType linux.SockType, protocol int, queue *waiter.Queue, endpoint tcpip.Endpoint) (*fs.File, *syserr.Error) { if skType == linux.SOCK_STREAM { - if err := endpoint.SetSockOptInt(tcpip.DelayOption, 1); err != nil { + if err := endpoint.SetSockOptBool(tcpip.DelayOption, true); err != nil { return nil, syserr.TranslateNetstackError(err) } } @@ -535,7 +535,7 @@ func (s *SocketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IO } if resCh != nil { - t := ctx.(*kernel.Task) + t := kernel.TaskFromContext(ctx) if err := t.Block(resCh); err != nil { return 0, syserr.FromError(err).ToError() } @@ -608,7 +608,7 @@ func (s *SocketOperations) ReadFrom(ctx context.Context, _ *fs.File, r io.Reader } if resCh != nil { - t := ctx.(*kernel.Task) + t := kernel.TaskFromContext(ctx) if err := t.Block(resCh); err != nil { return 0, syserr.FromError(err).ToError() } @@ -965,6 +965,13 @@ func GetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family in return nil, syserr.ErrProtocolNotAvailable } +func boolToInt32(v bool) int32 { + if v { + return 1 + } + return 0 +} + // getSockOptSocket implements GetSockOpt when level is SOL_SOCKET. func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, name, outLen int) (interface{}, *syserr.Error) { // TODO(b/124056281): Stop rejecting short optLen values in getsockopt. @@ -998,12 +1005,11 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.ErrInvalidArgument } - var v tcpip.PasscredOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.PasscredOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + return boolToInt32(v), nil case linux.SO_SNDBUF: if outLen < sizeOfInt32 { @@ -1042,24 +1048,22 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.ErrInvalidArgument } - var v tcpip.ReuseAddressOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.ReuseAddressOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + return boolToInt32(v), nil case linux.SO_REUSEPORT: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - var v tcpip.ReusePortOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.ReusePortOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + return boolToInt32(v), nil case linux.SO_BINDTODEVICE: var v tcpip.BindToDeviceOption @@ -1089,24 +1093,22 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam return nil, syserr.ErrInvalidArgument } - var v tcpip.BroadcastOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.BroadcastOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + return boolToInt32(v), nil case linux.SO_KEEPALIVE: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - var v tcpip.KeepaliveEnabledOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.KeepaliveEnabledOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + return boolToInt32(v), nil case linux.SO_LINGER: if outLen < linux.SizeOfLinger { @@ -1156,47 +1158,41 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa return nil, syserr.ErrInvalidArgument } - v, err := ep.GetSockOptInt(tcpip.DelayOption) + v, err := ep.GetSockOptBool(tcpip.DelayOption) if err != nil { return nil, syserr.TranslateNetstackError(err) } - - if v == 0 { - return int32(1), nil - } - return int32(0), nil + return boolToInt32(!v), nil case linux.TCP_CORK: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - var v tcpip.CorkOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.CorkOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + return boolToInt32(v), nil case linux.TCP_QUICKACK: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - var v tcpip.QuickAckOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.QuickAckOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + return boolToInt32(v), nil case linux.TCP_MAXSEG: if outLen < sizeOfInt32 { return nil, syserr.ErrInvalidArgument } - var v tcpip.MaxSegOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptInt(tcpip.MaxSegOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } @@ -1328,11 +1324,7 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf if err != nil { return nil, syserr.TranslateNetstackError(err) } - var o int32 - if v { - o = 1 - } - return o, nil + return boolToInt32(v), nil case linux.IPV6_PATHMTU: t.Kernel().EmitUnimplementedEvent(t) @@ -1342,8 +1334,8 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf if outLen == 0 { return make([]byte, 0), nil } - var v tcpip.IPv6TrafficClassOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptInt(tcpip.IPv6TrafficClassOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } @@ -1365,12 +1357,7 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf if err != nil { return nil, syserr.TranslateNetstackError(err) } - - var o int32 - if v { - o = 1 - } - return o, nil + return boolToInt32(v), nil default: emitUnimplementedEventIPv6(t, name) @@ -1386,8 +1373,8 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in return nil, syserr.ErrInvalidArgument } - var v tcpip.TTLOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptInt(tcpip.TTLOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } @@ -1403,8 +1390,8 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in return nil, syserr.ErrInvalidArgument } - var v tcpip.MulticastTTLOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptInt(tcpip.MulticastTTLOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } @@ -1429,23 +1416,19 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in return nil, syserr.ErrInvalidArgument } - var v tcpip.MulticastLoopOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.MulticastLoopOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - if v { - return int32(1), nil - } - return int32(0), nil + return boolToInt32(v), nil case linux.IP_TOS: // Length handling for parity with Linux. if outLen == 0 { return []byte(nil), nil } - var v tcpip.IPv4TOSOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptInt(tcpip.IPv4TOSOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } if outLen < sizeOfInt32 { @@ -1462,11 +1445,7 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in if err != nil { return nil, syserr.TranslateNetstackError(err) } - var o int32 - if v { - o = 1 - } - return o, nil + return boolToInt32(v), nil case linux.IP_PKTINFO: if outLen < sizeOfInt32 { @@ -1477,11 +1456,7 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in if err != nil { return nil, syserr.TranslateNetstackError(err) } - var o int32 - if v { - o = 1 - } - return o, nil + return boolToInt32(v), nil default: emitUnimplementedEventIP(t, name) @@ -1592,7 +1567,7 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.ReuseAddressOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReuseAddressOption, v != 0)) case linux.SO_REUSEPORT: if len(optVal) < sizeOfInt32 { @@ -1600,7 +1575,7 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.ReusePortOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReusePortOption, v != 0)) case linux.SO_BINDTODEVICE: n := bytes.IndexByte(optVal, 0) @@ -1628,7 +1603,7 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.BroadcastOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.BroadcastOption, v != 0)) case linux.SO_PASSCRED: if len(optVal) < sizeOfInt32 { @@ -1636,7 +1611,7 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.PasscredOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.PasscredOption, v != 0)) case linux.SO_KEEPALIVE: if len(optVal) < sizeOfInt32 { @@ -1644,7 +1619,7 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.KeepaliveEnabledOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.KeepaliveEnabledOption, v != 0)) case linux.SO_SNDTIMEO: if len(optVal) < linux.SizeOfTimeval { @@ -1716,11 +1691,7 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * } v := usermem.ByteOrder.Uint32(optVal) - var o int - if v == 0 { - o = 1 - } - return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.DelayOption, o)) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.DelayOption, v == 0)) case linux.TCP_CORK: if len(optVal) < sizeOfInt32 { @@ -1728,7 +1699,7 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.CorkOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.CorkOption, v != 0)) case linux.TCP_QUICKACK: if len(optVal) < sizeOfInt32 { @@ -1736,7 +1707,7 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.QuickAckOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.QuickAckOption, v != 0)) case linux.TCP_MAXSEG: if len(optVal) < sizeOfInt32 { @@ -1744,7 +1715,7 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.MaxSegOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.MaxSegOption, int(v))) case linux.TCP_KEEPIDLE: if len(optVal) < sizeOfInt32 { @@ -1855,7 +1826,7 @@ func setSockOptIPv6(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) if v == -1 { v = 0 } - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.IPv6TrafficClassOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.IPv6TrafficClassOption, int(v))) case linux.IPV6_RECVTCLASS: v, err := parseIntOrChar(optVal) @@ -1940,7 +1911,7 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s if v < 0 || v > 255 { return syserr.ErrInvalidArgument } - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.MulticastTTLOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.MulticastTTLOption, int(v))) case linux.IP_ADD_MEMBERSHIP: req, err := copyInMulticastRequest(optVal, false /* allowAddr */) @@ -1987,9 +1958,7 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s return err } - return syserr.TranslateNetstackError(ep.SetSockOpt( - tcpip.MulticastLoopOption(v != 0), - )) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.MulticastLoopOption, v != 0)) case linux.MCAST_JOIN_GROUP: // FIXME(b/124219304): Implement MCAST_JOIN_GROUP. @@ -2008,7 +1977,7 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s } else if v < 1 || v > 255 { return syserr.ErrInvalidArgument } - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.TTLOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.TTLOption, int(v))) case linux.IP_TOS: if len(optVal) == 0 { @@ -2018,7 +1987,7 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s if err != nil { return err } - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.IPv4TOSOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptInt(tcpip.IPv4TOSOption, int(v))) case linux.IP_RECVTOS: v, err := parseIntOrChar(optVal) diff --git a/pkg/sentry/socket/unix/transport/BUILD b/pkg/sentry/socket/unix/transport/BUILD index 74bcd6300..c708b6030 100644 --- a/pkg/sentry/socket/unix/transport/BUILD +++ b/pkg/sentry/socket/unix/transport/BUILD @@ -30,6 +30,7 @@ go_library( "//pkg/abi/linux", "//pkg/context", "//pkg/ilist", + "//pkg/log", "//pkg/refs", "//pkg/sync", "//pkg/syserr", diff --git a/pkg/sentry/socket/unix/transport/unix.go b/pkg/sentry/socket/unix/transport/unix.go index 2ef654235..1f3880cc5 100644 --- a/pkg/sentry/socket/unix/transport/unix.go +++ b/pkg/sentry/socket/unix/transport/unix.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" @@ -838,24 +839,45 @@ func (e *baseEndpoint) SendMsg(ctx context.Context, data [][]byte, c ControlMess // SetSockOpt sets a socket option. Currently not supported. func (e *baseEndpoint) SetSockOpt(opt interface{}) *tcpip.Error { - switch v := opt.(type) { - case tcpip.PasscredOption: - e.setPasscred(v != 0) - return nil - } return nil } func (e *baseEndpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { + switch opt { + case tcpip.BroadcastOption: + case tcpip.PasscredOption: + e.setPasscred(v) + case tcpip.ReuseAddressOption: + default: + log.Warningf("Unsupported socket option: %d", opt) + return tcpip.ErrUnknownProtocolOption + } return nil } func (e *baseEndpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { + switch opt { + case tcpip.SendBufferSizeOption: + case tcpip.ReceiveBufferSizeOption: + default: + log.Warningf("Unsupported socket option: %d", opt) + return tcpip.ErrUnknownProtocolOption + } return nil } func (e *baseEndpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { - return false, tcpip.ErrUnknownProtocolOption + switch opt { + case tcpip.KeepaliveEnabledOption: + return false, nil + + case tcpip.PasscredOption: + return e.Passcred(), nil + + default: + log.Warningf("Unsupported socket option: %d", opt) + return false, tcpip.ErrUnknownProtocolOption + } } func (e *baseEndpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { @@ -914,29 +936,19 @@ func (e *baseEndpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { return int(v), nil default: + log.Warningf("Unsupported socket option: %d", opt) return -1, tcpip.ErrUnknownProtocolOption } } // GetSockOpt implements tcpip.Endpoint.GetSockOpt. func (e *baseEndpoint) GetSockOpt(opt interface{}) *tcpip.Error { - switch o := opt.(type) { + switch opt.(type) { case tcpip.ErrorOption: return nil - case *tcpip.PasscredOption: - if e.Passcred() { - *o = tcpip.PasscredOption(1) - } else { - *o = tcpip.PasscredOption(0) - } - return nil - - case *tcpip.KeepaliveEnabledOption: - *o = 0 - return nil - default: + log.Warningf("Unsupported socket option: %T", opt) return tcpip.ErrUnknownProtocolOption } } diff --git a/pkg/sentry/syscalls/linux/sys_aio.go b/pkg/sentry/syscalls/linux/sys_aio.go index b401978db..38cbeba5a 100644 --- a/pkg/sentry/syscalls/linux/sys_aio.go +++ b/pkg/sentry/syscalls/linux/sys_aio.go @@ -114,14 +114,28 @@ func IoSetup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca func IoDestroy(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { id := args[0].Uint64() - // Destroy the given context. - if !t.MemoryManager().DestroyAIOContext(t, id) { + ctx := t.MemoryManager().DestroyAIOContext(t, id) + if ctx == nil { // Does not exist. return 0, nil, syserror.EINVAL } - // FIXME(fvoznika): Linux blocks until all AIO to the destroyed context is - // done. - return 0, nil, nil + + // Drain completed requests amd wait for pending requests until there are no + // more. + for { + ctx.Drain() + + ch := ctx.WaitChannel() + if ch == nil { + // No more requests, we're done. + return 0, nil, nil + } + // The task cannot be interrupted during the wait. Equivalent to + // TASK_UNINTERRUPTIBLE in Linux. + t.UninterruptibleSleepStart(true /* deactivate */) + <-ch + t.UninterruptibleSleepFinish(true /* activate */) + } } // IoGetevents implements linux syscall io_getevents(2). @@ -200,13 +214,13 @@ func IoGetevents(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S func waitForRequest(ctx *mm.AIOContext, t *kernel.Task, haveDeadline bool, deadline ktime.Time) (interface{}, error) { for { if v, ok := ctx.PopRequest(); ok { - // Request was readly available. Just return it. + // Request was readily available. Just return it. return v, nil } // Need to wait for request completion. - done, active := ctx.WaitChannel() - if !active { + done := ctx.WaitChannel() + if done == nil { // Context has been destroyed. return nil, syserror.EINVAL } @@ -248,6 +262,10 @@ func memoryFor(t *kernel.Task, cb *ioCallback) (usermem.IOSequence, error) { } func performCallback(t *kernel.Task, file *fs.File, cbAddr usermem.Addr, cb *ioCallback, ioseq usermem.IOSequence, ctx *mm.AIOContext, eventFile *fs.File) { + if ctx.Dead() { + ctx.CancelPendingRequest() + return + } ev := &ioEvent{ Data: cb.Data, Obj: uint64(cbAddr), diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go index fd642834b..fbc6cf15f 100644 --- a/pkg/sentry/syscalls/linux/sys_splice.go +++ b/pkg/sentry/syscalls/linux/sys_splice.go @@ -29,6 +29,10 @@ func doSplice(t *kernel.Task, outFile, inFile *fs.File, opts fs.SpliceOpts, nonB return 0, syserror.EINVAL } + if opts.Length > int64(kernel.MAX_RW_COUNT) { + opts.Length = int64(kernel.MAX_RW_COUNT) + } + var ( total int64 n int64 diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index cd34782ff..bef1bd312 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -497,7 +497,7 @@ type FilesystemImpl interface { // Preconditions: vd.Mount().Filesystem().Impl() == this FilesystemImpl. PrependPath(ctx context.Context, vfsroot, vd VirtualDentry, b *fspath.Builder) error - // TODO: inotify_add_watch() + // TODO(gvisor.dev/issue/1479): inotify_add_watch() } // PrependPathAtVFSRootError is returned by implementations of diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 1b8ecc415..f06946103 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -233,9 +233,9 @@ func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentia } vd.dentry.mu.Lock() } - // TODO: Linux requires that either both the mount point and the mount root - // are directories, or neither are, and returns ENOTDIR if this is not the - // case. + // TODO(gvisor.dev/issue/1035): Linux requires that either both the mount + // point and the mount root are directories, or neither are, and returns + // ENOTDIR if this is not the case. mntns := vd.mount.ns mnt := newMount(vfs, fs, root, mntns, opts) vfs.mounts.seq.BeginWrite() @@ -274,9 +274,9 @@ func (vfs *VirtualFilesystem) UmountAt(ctx context.Context, creds *auth.Credenti } } - // TODO(jamieliu): Linux special-cases umount of the caller's root, which - // we don't implement yet (we'll just fail it since the caller holds a - // reference on it). + // TODO(gvisor.dev/issue/1035): Linux special-cases umount of the caller's + // root, which we don't implement yet (we'll just fail it since the caller + // holds a reference on it). vfs.mounts.seq.BeginWrite() if opts.Flags&linux.MNT_DETACH == 0 { diff --git a/pkg/sentry/vfs/mount_test.go b/pkg/sentry/vfs/mount_test.go index 3b933468d..3335e4057 100644 --- a/pkg/sentry/vfs/mount_test.go +++ b/pkg/sentry/vfs/mount_test.go @@ -55,7 +55,7 @@ func TestMountTableInsertLookup(t *testing.T) { } } -// TODO: concurrent lookup/insertion/removal +// TODO(gvisor.dev/issue/1035): concurrent lookup/insertion/removal. // must be powers of 2 var benchNumMounts = []int{1 << 2, 1 << 5, 1 << 8} diff --git a/pkg/tcpip/checker/checker.go b/pkg/tcpip/checker/checker.go index 8dc0f7c0e..307f1b666 100644 --- a/pkg/tcpip/checker/checker.go +++ b/pkg/tcpip/checker/checker.go @@ -728,7 +728,7 @@ func ICMPv6Code(want byte) TransportChecker { // message for type of ty, with potentially additional checks specified by // checkers. // -// checkers may assume that a valid ICMPv6 is passed to it containing a valid +// Checkers may assume that a valid ICMPv6 is passed to it containing a valid // NDP message as far as the size of the message (minSize) is concerned. The // values within the message are up to checkers to validate. func NDP(msgType header.ICMPv6Type, minSize int, checkers ...TransportChecker) NetworkChecker { @@ -760,9 +760,9 @@ func NDP(msgType header.ICMPv6Type, minSize int, checkers ...TransportChecker) N // Neighbor Solicitation message (as per the raw wire format), with potentially // additional checks specified by checkers. // -// checkers may assume that a valid ICMPv6 is passed to it containing a valid -// NDPNS message as far as the size of the messages concerned. The values within -// the message are up to checkers to validate. +// Checkers may assume that a valid ICMPv6 is passed to it containing a valid +// NDPNS message as far as the size of the message is concerned. The values +// within the message are up to checkers to validate. func NDPNS(checkers ...TransportChecker) NetworkChecker { return NDP(header.ICMPv6NeighborSolicit, header.NDPNSMinimumSize, checkers...) } @@ -780,7 +780,54 @@ func NDPNSTargetAddress(want tcpip.Address) TransportChecker { ns := header.NDPNeighborSolicit(icmp.NDPPayload()) if got := ns.TargetAddress(); got != want { - t.Fatalf("got %T.TargetAddress = %s, want = %s", ns, got, want) + t.Errorf("got %T.TargetAddress() = %s, want = %s", ns, got, want) + } + } +} + +// NDPNA creates a checker that checks that the packet contains a valid NDP +// Neighbor Advertisement message (as per the raw wire format), with potentially +// additional checks specified by checkers. +// +// Checkers may assume that a valid ICMPv6 is passed to it containing a valid +// NDPNA message as far as the size of the message is concerned. The values +// within the message are up to checkers to validate. +func NDPNA(checkers ...TransportChecker) NetworkChecker { + return NDP(header.ICMPv6NeighborAdvert, header.NDPNAMinimumSize, checkers...) +} + +// NDPNATargetAddress creates a checker that checks the Target Address field of +// a header.NDPNeighborAdvert. +// +// The returned TransportChecker assumes that a valid ICMPv6 is passed to it +// containing a valid NDPNA message as far as the size is concerned. +func NDPNATargetAddress(want tcpip.Address) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + icmp := h.(header.ICMPv6) + na := header.NDPNeighborAdvert(icmp.NDPPayload()) + + if got := na.TargetAddress(); got != want { + t.Errorf("got %T.TargetAddress() = %s, want = %s", na, got, want) + } + } +} + +// NDPNASolicitedFlag creates a checker that checks the Solicited field of +// a header.NDPNeighborAdvert. +// +// The returned TransportChecker assumes that a valid ICMPv6 is passed to it +// containing a valid NDPNA message as far as the size is concerned. +func NDPNASolicitedFlag(want bool) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + icmp := h.(header.ICMPv6) + na := header.NDPNeighborAdvert(icmp.NDPPayload()) + + if got := na.SolicitedFlag(); got != want { + t.Errorf("got %T.SolicitedFlag = %t, want = %t", na, got, want) } } } @@ -819,6 +866,13 @@ func ndpOptions(t *testing.T, optsBuf header.NDPOptions, opts []header.NDPOption } else if got, want := gotOpt.EthernetAddress(), wantOpt.EthernetAddress(); got != want { t.Errorf("got EthernetAddress() = %s at index %d, want = %s", got, i, want) } + case header.NDPTargetLinkLayerAddressOption: + gotOpt, ok := opt.(header.NDPTargetLinkLayerAddressOption) + if !ok { + t.Errorf("got type = %T at index = %d; want = %T", opt, i, wantOpt) + } else if got, want := gotOpt.EthernetAddress(), wantOpt.EthernetAddress(); got != want { + t.Errorf("got EthernetAddress() = %s at index %d, want = %s", got, i, want) + } default: t.Fatalf("checker not implemented for expected NDP option: %T", wantOpt) } @@ -831,6 +885,21 @@ func ndpOptions(t *testing.T, optsBuf header.NDPOptions, opts []header.NDPOption } } +// NDPNAOptions creates a checker that checks that the packet contains the +// provided NDP options within an NDP Neighbor Solicitation message. +// +// The returned TransportChecker assumes that a valid ICMPv6 is passed to it +// containing a valid NDPNA message as far as the size is concerned. +func NDPNAOptions(opts []header.NDPOption) TransportChecker { + return func(t *testing.T, h header.Transport) { + t.Helper() + + icmp := h.(header.ICMPv6) + na := header.NDPNeighborAdvert(icmp.NDPPayload()) + ndpOptions(t, na.Options(), opts) + } +} + // NDPNSOptions creates a checker that checks that the packet contains the // provided NDP options within an NDP Neighbor Solicitation message. // @@ -849,7 +918,7 @@ func NDPNSOptions(opts []header.NDPOption) TransportChecker { // NDPRS creates a checker that checks that the packet contains a valid NDP // Router Solicitation message (as per the raw wire format). // -// checkers may assume that a valid ICMPv6 is passed to it containing a valid +// Checkers may assume that a valid ICMPv6 is passed to it containing a valid // NDPRS as far as the size of the message is concerned. The values within the // message are up to checkers to validate. func NDPRS(checkers ...TransportChecker) NetworkChecker { diff --git a/pkg/tcpip/network/ipv6/BUILD b/pkg/tcpip/network/ipv6/BUILD index a93a7621a..3f71fc520 100644 --- a/pkg/tcpip/network/ipv6/BUILD +++ b/pkg/tcpip/network/ipv6/BUILD @@ -31,6 +31,7 @@ go_test( deps = [ "//pkg/tcpip", "//pkg/tcpip/buffer", + "//pkg/tcpip/checker", "//pkg/tcpip/header", "//pkg/tcpip/link/channel", "//pkg/tcpip/link/sniffer", diff --git a/pkg/tcpip/network/ipv6/icmp.go b/pkg/tcpip/network/ipv6/icmp.go index f91180aa3..dc0369156 100644 --- a/pkg/tcpip/network/ipv6/icmp.go +++ b/pkg/tcpip/network/ipv6/icmp.go @@ -138,53 +138,48 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P targetAddr := ns.TargetAddress() s := r.Stack() - rxNICID := r.NICID() - if isTentative, err := s.IsAddrTentative(rxNICID, targetAddr); err != nil { - // We will only get an error if rxNICID is unrecognized, - // which should not happen. For now short-circuit this - // packet. + if isTentative, err := s.IsAddrTentative(e.nicID, targetAddr); err != nil { + // We will only get an error if the NIC is unrecognized, which should not + // happen. For now, drop this packet. // // TODO(b/141002840): Handle this better? return } else if isTentative { - // If the target address is tentative and the source - // of the packet is a unicast (specified) address, then - // the source of the packet is attempting to perform - // address resolution on the target. In this case, the - // solicitation is silently ignored, as per RFC 4862 - // section 5.4.3. + // If the target address is tentative and the source of the packet is a + // unicast (specified) address, then the source of the packet is + // attempting to perform address resolution on the target. In this case, + // the solicitation is silently ignored, as per RFC 4862 section 5.4.3. // - // If the target address is tentative and the source of - // the packet is the unspecified address (::), then we - // know another node is also performing DAD for the - // same address (since targetAddr is tentative for us, - // we know we are also performing DAD on it). In this - // case we let the stack know so it can handle such a - // scenario and do nothing further with the NDP NS. - if iph.SourceAddress() == header.IPv6Any { - s.DupTentativeAddrDetected(rxNICID, targetAddr) + // If the target address is tentative and the source of the packet is the + // unspecified address (::), then we know another node is also performing + // DAD for the same address (since the target address is tentative for us, + // we know we are also performing DAD on it). In this case we let the + // stack know so it can handle such a scenario and do nothing further with + // the NS. + if r.RemoteAddress == header.IPv6Any { + s.DupTentativeAddrDetected(e.nicID, targetAddr) } - // Do not handle neighbor solicitations targeted - // to an address that is tentative on the received - // NIC any further. + // Do not handle neighbor solicitations targeted to an address that is + // tentative on the NIC any further. return } - // At this point we know that targetAddr is not tentative on - // rxNICID so the packet is processed as defined in RFC 4861, - // as per RFC 4862 section 5.4.3. + // At this point we know that the target address is not tentative on the NIC + // so the packet is processed as defined in RFC 4861, as per RFC 4862 + // section 5.4.3. + // Is the NS targetting us? if e.linkAddrCache.CheckLocalAddress(e.nicID, ProtocolNumber, targetAddr) == 0 { - // We don't have a useful answer; the best we can do is ignore the request. return } - // If the NS message has the source link layer option, update the link - // address cache with the link address for the sender of the message. + // If the NS message contains the Source Link-Layer Address option, update + // the link address cache with the value of the option. // // TODO(b/148429853): Properly process the NS message and do Neighbor // Unreachability Detection. + var sourceLinkAddr tcpip.LinkAddress for { opt, done, err := it.Next() if err != nil { @@ -197,22 +192,36 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P switch opt := opt.(type) { case header.NDPSourceLinkLayerAddressOption: - e.linkAddrCache.AddLinkAddress(e.nicID, r.RemoteAddress, opt.EthernetAddress()) + // No RFCs define what to do when an NS message has multiple Source + // Link-Layer Address options. Since no interface can have multiple + // link-layer addresses, we consider such messages invalid. + if len(sourceLinkAddr) != 0 { + received.Invalid.Increment() + return + } + + sourceLinkAddr = opt.EthernetAddress() } } - optsSerializer := header.NDPOptionsSerializer{ - header.NDPTargetLinkLayerAddressOption(r.LocalLinkAddress[:]), + unspecifiedSource := r.RemoteAddress == header.IPv6Any + + // As per RFC 4861 section 4.3, the Source Link-Layer Address Option MUST + // NOT be included when the source IP address is the unspecified address. + // Otherwise, on link layers that have addresses this option MUST be + // included in multicast solicitations and SHOULD be included in unicast + // solicitations. + if len(sourceLinkAddr) == 0 { + if header.IsV6MulticastAddress(r.LocalAddress) && !unspecifiedSource { + received.Invalid.Increment() + return + } + } else if unspecifiedSource { + received.Invalid.Increment() + return + } else { + e.linkAddrCache.AddLinkAddress(e.nicID, r.RemoteAddress, sourceLinkAddr) } - hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6NeighborAdvertMinimumSize + int(optsSerializer.Length())) - packet := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborAdvertSize)) - packet.SetType(header.ICMPv6NeighborAdvert) - na := header.NDPNeighborAdvert(packet.NDPPayload()) - na.SetSolicitedFlag(true) - na.SetOverrideFlag(true) - na.SetTargetAddress(targetAddr) - opts := na.Options() - opts.Serialize(optsSerializer) // ICMPv6 Neighbor Solicit messages are always sent to // specially crafted IPv6 multicast addresses. As a result, the @@ -225,6 +234,40 @@ func (e *endpoint) handleICMP(r *stack.Route, netHeader buffer.View, pkt stack.P r := r.Clone() defer r.Release() r.LocalAddress = targetAddr + + // As per RFC 4861 section 7.2.4, if the the source of the solicitation is + // the unspecified address, the node MUST set the Solicited flag to zero and + // multicast the advertisement to the all-nodes address. + solicited := true + if unspecifiedSource { + solicited = false + r.RemoteAddress = header.IPv6AllNodesMulticastAddress + } + + // If the NS has a source link-layer option, use the link address it + // specifies as the remote link address for the response instead of the + // source link address of the packet. + // + // TODO(#2401): As per RFC 4861 section 7.2.4 we should consult our link + // address cache for the right destination link address instead of manually + // patching the route with the remote link address if one is specified in a + // Source Link-Layer Address option. + if len(sourceLinkAddr) != 0 { + r.RemoteLinkAddress = sourceLinkAddr + } + + optsSerializer := header.NDPOptionsSerializer{ + header.NDPTargetLinkLayerAddressOption(r.LocalLinkAddress), + } + hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6NeighborAdvertMinimumSize + int(optsSerializer.Length())) + packet := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborAdvertSize)) + packet.SetType(header.ICMPv6NeighborAdvert) + na := header.NDPNeighborAdvert(packet.NDPPayload()) + na.SetSolicitedFlag(solicited) + na.SetOverrideFlag(true) + na.SetTargetAddress(targetAddr) + opts := na.Options() + opts.Serialize(optsSerializer) packet.SetChecksum(header.ICMPv6Checksum(packet, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) // RFC 4861 Neighbor Discovery for IP version 6 (IPv6) diff --git a/pkg/tcpip/network/ipv6/icmp_test.go b/pkg/tcpip/network/ipv6/icmp_test.go index bae09ed94..bd099a7f8 100644 --- a/pkg/tcpip/network/ipv6/icmp_test.go +++ b/pkg/tcpip/network/ipv6/icmp_test.go @@ -32,7 +32,8 @@ import ( const ( linkAddr0 = tcpip.LinkAddress("\x02\x02\x03\x04\x05\x06") - linkAddr1 = tcpip.LinkAddress("\x0a\x0b\x0c\x0d\x0e\x0f") + linkAddr1 = tcpip.LinkAddress("\x0a\x0b\x0c\x0d\x0e\x0e") + linkAddr2 = tcpip.LinkAddress("\x0a\x0b\x0c\x0d\x0e\x0f") ) var ( diff --git a/pkg/tcpip/network/ipv6/ndp_test.go b/pkg/tcpip/network/ipv6/ndp_test.go index b113aaacc..8db51da96 100644 --- a/pkg/tcpip/network/ipv6/ndp_test.go +++ b/pkg/tcpip/network/ipv6/ndp_test.go @@ -20,6 +20,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" + "gvisor.dev/gvisor/pkg/tcpip/checker" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/link/channel" "gvisor.dev/gvisor/pkg/tcpip/stack" @@ -173,6 +174,257 @@ func TestNeighorSolicitationWithSourceLinkLayerOption(t *testing.T) { } } +func TestNeighorSolicitationResponse(t *testing.T) { + const nicID = 1 + nicAddr := lladdr0 + remoteAddr := lladdr1 + nicAddrSNMC := header.SolicitedNodeAddr(nicAddr) + nicLinkAddr := linkAddr0 + remoteLinkAddr0 := linkAddr1 + remoteLinkAddr1 := linkAddr2 + + tests := []struct { + name string + nsOpts header.NDPOptionsSerializer + nsSrcLinkAddr tcpip.LinkAddress + nsSrc tcpip.Address + nsDst tcpip.Address + nsInvalid bool + naDstLinkAddr tcpip.LinkAddress + naSolicited bool + naSrc tcpip.Address + naDst tcpip.Address + }{ + { + name: "Unspecified source to multicast destination", + nsOpts: nil, + nsSrcLinkAddr: remoteLinkAddr0, + nsSrc: header.IPv6Any, + nsDst: nicAddrSNMC, + nsInvalid: false, + naDstLinkAddr: remoteLinkAddr0, + naSolicited: false, + naSrc: nicAddr, + naDst: header.IPv6AllNodesMulticastAddress, + }, + { + name: "Unspecified source with source ll option to multicast destination", + nsOpts: header.NDPOptionsSerializer{ + header.NDPSourceLinkLayerAddressOption(remoteLinkAddr0[:]), + }, + nsSrcLinkAddr: remoteLinkAddr0, + nsSrc: header.IPv6Any, + nsDst: nicAddrSNMC, + nsInvalid: true, + }, + { + name: "Unspecified source to unicast destination", + nsOpts: nil, + nsSrcLinkAddr: remoteLinkAddr0, + nsSrc: header.IPv6Any, + nsDst: nicAddr, + nsInvalid: false, + naDstLinkAddr: remoteLinkAddr0, + naSolicited: false, + naSrc: nicAddr, + naDst: header.IPv6AllNodesMulticastAddress, + }, + { + name: "Unspecified source with source ll option to unicast destination", + nsOpts: header.NDPOptionsSerializer{ + header.NDPSourceLinkLayerAddressOption(remoteLinkAddr0[:]), + }, + nsSrcLinkAddr: remoteLinkAddr0, + nsSrc: header.IPv6Any, + nsDst: nicAddr, + nsInvalid: true, + }, + + { + name: "Specified source with 1 source ll to multicast destination", + nsOpts: header.NDPOptionsSerializer{ + header.NDPSourceLinkLayerAddressOption(remoteLinkAddr0[:]), + }, + nsSrcLinkAddr: remoteLinkAddr0, + nsSrc: remoteAddr, + nsDst: nicAddrSNMC, + nsInvalid: false, + naDstLinkAddr: remoteLinkAddr0, + naSolicited: true, + naSrc: nicAddr, + naDst: remoteAddr, + }, + { + name: "Specified source with 1 source ll different from route to multicast destination", + nsOpts: header.NDPOptionsSerializer{ + header.NDPSourceLinkLayerAddressOption(remoteLinkAddr1[:]), + }, + nsSrcLinkAddr: remoteLinkAddr0, + nsSrc: remoteAddr, + nsDst: nicAddrSNMC, + nsInvalid: false, + naDstLinkAddr: remoteLinkAddr1, + naSolicited: true, + naSrc: nicAddr, + naDst: remoteAddr, + }, + { + name: "Specified source to multicast destination", + nsOpts: nil, + nsSrcLinkAddr: remoteLinkAddr0, + nsSrc: remoteAddr, + nsDst: nicAddrSNMC, + nsInvalid: true, + }, + { + name: "Specified source with 2 source ll to multicast destination", + nsOpts: header.NDPOptionsSerializer{ + header.NDPSourceLinkLayerAddressOption(remoteLinkAddr0[:]), + header.NDPSourceLinkLayerAddressOption(remoteLinkAddr1[:]), + }, + nsSrcLinkAddr: remoteLinkAddr0, + nsSrc: remoteAddr, + nsDst: nicAddrSNMC, + nsInvalid: true, + }, + + { + name: "Specified source to unicast destination", + nsOpts: nil, + nsSrcLinkAddr: remoteLinkAddr0, + nsSrc: remoteAddr, + nsDst: nicAddr, + nsInvalid: false, + naDstLinkAddr: remoteLinkAddr0, + naSolicited: true, + naSrc: nicAddr, + naDst: remoteAddr, + }, + { + name: "Specified source with 1 source ll to unicast destination", + nsOpts: header.NDPOptionsSerializer{ + header.NDPSourceLinkLayerAddressOption(remoteLinkAddr0[:]), + }, + nsSrcLinkAddr: remoteLinkAddr0, + nsSrc: remoteAddr, + nsDst: nicAddr, + nsInvalid: false, + naDstLinkAddr: remoteLinkAddr0, + naSolicited: true, + naSrc: nicAddr, + naDst: remoteAddr, + }, + { + name: "Specified source with 1 source ll different from route to unicast destination", + nsOpts: header.NDPOptionsSerializer{ + header.NDPSourceLinkLayerAddressOption(remoteLinkAddr1[:]), + }, + nsSrcLinkAddr: remoteLinkAddr0, + nsSrc: remoteAddr, + nsDst: nicAddr, + nsInvalid: false, + naDstLinkAddr: remoteLinkAddr1, + naSolicited: true, + naSrc: nicAddr, + naDst: remoteAddr, + }, + { + name: "Specified source with 2 source ll to unicast destination", + nsOpts: header.NDPOptionsSerializer{ + header.NDPSourceLinkLayerAddressOption(remoteLinkAddr0[:]), + header.NDPSourceLinkLayerAddressOption(remoteLinkAddr1[:]), + }, + nsSrcLinkAddr: remoteLinkAddr0, + nsSrc: remoteAddr, + nsDst: nicAddr, + nsInvalid: true, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{NewProtocol()}, + }) + e := channel.New(1, 1280, nicLinkAddr) + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + } + if err := s.AddAddress(nicID, ProtocolNumber, nicAddr); err != nil { + t.Fatalf("AddAddress(%d, %d, %s) = %s", nicID, ProtocolNumber, nicAddr, err) + } + + ndpNSSize := header.ICMPv6NeighborSolicitMinimumSize + test.nsOpts.Length() + hdr := buffer.NewPrependable(header.IPv6MinimumSize + ndpNSSize) + pkt := header.ICMPv6(hdr.Prepend(ndpNSSize)) + pkt.SetType(header.ICMPv6NeighborSolicit) + ns := header.NDPNeighborSolicit(pkt.NDPPayload()) + ns.SetTargetAddress(nicAddr) + opts := ns.Options() + opts.Serialize(test.nsOpts) + pkt.SetChecksum(header.ICMPv6Checksum(pkt, test.nsSrc, test.nsDst, buffer.VectorisedView{})) + payloadLength := hdr.UsedLength() + ip := header.IPv6(hdr.Prepend(header.IPv6MinimumSize)) + ip.Encode(&header.IPv6Fields{ + PayloadLength: uint16(payloadLength), + NextHeader: uint8(header.ICMPv6ProtocolNumber), + HopLimit: 255, + SrcAddr: test.nsSrc, + DstAddr: test.nsDst, + }) + + invalid := s.Stats().ICMP.V6PacketsReceived.Invalid + + // Invalid count should initially be 0. + if got := invalid.Value(); got != 0 { + t.Fatalf("got invalid = %d, want = 0", got) + } + + e.InjectLinkAddr(ProtocolNumber, test.nsSrcLinkAddr, stack.PacketBuffer{ + Data: hdr.View().ToVectorisedView(), + }) + + if test.nsInvalid { + if got := invalid.Value(); got != 1 { + t.Fatalf("got invalid = %d, want = 1", got) + } + + if p, got := e.Read(); got { + t.Fatalf("unexpected response to an invalid NS = %+v", p.Pkt) + } + + // If we expected the NS to be invalid, we have nothing else to check. + return + } + + if got := invalid.Value(); got != 0 { + t.Fatalf("got invalid = %d, want = 0", got) + } + + p, got := e.Read() + if !got { + t.Fatal("expected an NDP NA response") + } + + if p.Route.RemoteLinkAddress != test.naDstLinkAddr { + t.Errorf("got p.Route.RemoteLinkAddress = %s, want = %s", p.Route.RemoteLinkAddress, test.naDstLinkAddr) + } + + checker.IPv6(t, p.Pkt.Header.View(), + checker.SrcAddr(test.naSrc), + checker.DstAddr(test.naDst), + checker.TTL(header.NDPHopLimit), + checker.NDPNA( + checker.NDPNASolicitedFlag(test.naSolicited), + checker.NDPNATargetAddress(nicAddr), + checker.NDPNAOptions([]header.NDPOption{ + header.NDPTargetLinkLayerAddressOption(nicLinkAddr[:]), + }), + )) + }) + } +} + // TestNeighorAdvertisementWithTargetLinkLayerOption tests that receiving a // valid NDP NA message with the Target Link Layer Address option results in a // new entry in the link address cache for the target of the message. diff --git a/pkg/tcpip/stack/transport_demuxer_test.go b/pkg/tcpip/stack/transport_demuxer_test.go index c65b0c632..2474a7db3 100644 --- a/pkg/tcpip/stack/transport_demuxer_test.go +++ b/pkg/tcpip/stack/transport_demuxer_test.go @@ -206,7 +206,7 @@ func TestTransportDemuxerRegister(t *testing.T) { // the distribution of packets received matches expectations. func TestBindToDeviceDistribution(t *testing.T) { type endpointSockopts struct { - reuse int + reuse bool bindToDevice tcpip.NICID } for _, test := range []struct { @@ -221,11 +221,11 @@ func TestBindToDeviceDistribution(t *testing.T) { "BindPortReuse", // 5 endpoints that all have reuse set. []endpointSockopts{ - {reuse: 1, bindToDevice: 0}, - {reuse: 1, bindToDevice: 0}, - {reuse: 1, bindToDevice: 0}, - {reuse: 1, bindToDevice: 0}, - {reuse: 1, bindToDevice: 0}, + {reuse: true, bindToDevice: 0}, + {reuse: true, bindToDevice: 0}, + {reuse: true, bindToDevice: 0}, + {reuse: true, bindToDevice: 0}, + {reuse: true, bindToDevice: 0}, }, map[tcpip.NICID][]float64{ // Injected packets on dev0 get distributed evenly. @@ -236,9 +236,9 @@ func TestBindToDeviceDistribution(t *testing.T) { "BindToDevice", // 3 endpoints with various bindings. []endpointSockopts{ - {reuse: 0, bindToDevice: 1}, - {reuse: 0, bindToDevice: 2}, - {reuse: 0, bindToDevice: 3}, + {reuse: false, bindToDevice: 1}, + {reuse: false, bindToDevice: 2}, + {reuse: false, bindToDevice: 3}, }, map[tcpip.NICID][]float64{ // Injected packets on dev0 go only to the endpoint bound to dev0. @@ -253,12 +253,12 @@ func TestBindToDeviceDistribution(t *testing.T) { "ReuseAndBindToDevice", // 6 endpoints with various bindings. []endpointSockopts{ - {reuse: 1, bindToDevice: 1}, - {reuse: 1, bindToDevice: 1}, - {reuse: 1, bindToDevice: 2}, - {reuse: 1, bindToDevice: 2}, - {reuse: 1, bindToDevice: 2}, - {reuse: 1, bindToDevice: 0}, + {reuse: true, bindToDevice: 1}, + {reuse: true, bindToDevice: 1}, + {reuse: true, bindToDevice: 2}, + {reuse: true, bindToDevice: 2}, + {reuse: true, bindToDevice: 2}, + {reuse: true, bindToDevice: 0}, }, map[tcpip.NICID][]float64{ // Injected packets on dev0 get distributed among endpoints bound to @@ -309,9 +309,8 @@ func TestBindToDeviceDistribution(t *testing.T) { }(ep) defer ep.Close() - reusePortOption := tcpip.ReusePortOption(endpoint.reuse) - if err := ep.SetSockOpt(reusePortOption); err != nil { - t.Fatalf("SetSockOpt(%#v) on endpoint %d failed: %s", reusePortOption, i, err) + if err := ep.SetSockOptBool(tcpip.ReusePortOption, endpoint.reuse); err != nil { + t.Fatalf("SetSockOptBool(ReusePortOption, %t) on endpoint %d failed: %s", endpoint.reuse, i, err) } bindToDeviceOption := tcpip.BindToDeviceOption(endpoint.bindToDevice) if err := ep.SetSockOpt(bindToDeviceOption); err != nil { diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index 2ef3271f1..aec7126ff 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -520,34 +520,90 @@ type WriteOptions struct { type SockOptBool int const ( + // BroadcastOption is used by SetSockOpt/GetSockOpt to specify whether + // datagram sockets are allowed to send packets to a broadcast address. + BroadcastOption SockOptBool = iota + + // CorkOption is used by SetSockOpt/GetSockOpt to specify if data should be + // held until segments are full by the TCP transport protocol. + CorkOption + + // DelayOption is used by SetSockOpt/GetSockOpt to specify if data + // should be sent out immediately by the transport protocol. For TCP, + // it determines if the Nagle algorithm is on or off. + DelayOption + + // KeepaliveEnabledOption is used by SetSockOpt/GetSockOpt to specify whether + // TCP keepalive is enabled for this socket. + KeepaliveEnabledOption + + // MulticastLoopOption is used by SetSockOpt/GetSockOpt to specify whether + // multicast packets sent over a non-loopback interface will be looped back. + MulticastLoopOption + + // PasscredOption is used by SetSockOpt/GetSockOpt to specify whether + // SCM_CREDENTIALS socket control messages are enabled. + // + // Only supported on Unix sockets. + PasscredOption + + // QuickAckOption is stubbed out in SetSockOpt/GetSockOpt. + QuickAckOption + // ReceiveTClassOption is used by SetSockOpt/GetSockOpt to specify if the // IPV6_TCLASS ancillary message is passed with incoming packets. - ReceiveTClassOption SockOptBool = iota + ReceiveTClassOption // ReceiveTOSOption is used by SetSockOpt/GetSockOpt to specify if the TOS // ancillary message is passed with incoming packets. ReceiveTOSOption - // V6OnlyOption is used by {G,S}etSockOptBool to specify whether an IPv6 - // socket is to be restricted to sending and receiving IPv6 packets only. - V6OnlyOption - // ReceiveIPPacketInfoOption is used by {G,S}etSockOptBool to specify // if more inforamtion is provided with incoming packets such // as interface index and address. ReceiveIPPacketInfoOption - // TODO(b/146901447): convert existing bool socket options to be handled via - // Get/SetSockOptBool + // ReuseAddressOption is used by SetSockOpt/GetSockOpt to specify whether Bind() + // should allow reuse of local address. + ReuseAddressOption + + // ReusePortOption is used by SetSockOpt/GetSockOpt to permit multiple sockets + // to be bound to an identical socket address. + ReusePortOption + + // V6OnlyOption is used by {G,S}etSockOptBool to specify whether an IPv6 + // socket is to be restricted to sending and receiving IPv6 packets only. + V6OnlyOption ) // SockOptInt represents socket options which values have the int type. type SockOptInt int const ( + // KeepaliveCountOption is used by SetSockOpt/GetSockOpt to specify the number + // of un-ACKed TCP keepalives that will be sent before the connection is + // closed. + KeepaliveCountOption SockOptInt = iota + + // IPv4TOSOption is used by SetSockOpt/GetSockOpt to specify TOS + // for all subsequent outgoing IPv4 packets from the endpoint. + IPv4TOSOption + + // IPv6TrafficClassOption is used by SetSockOpt/GetSockOpt to specify TOS + // for all subsequent outgoing IPv6 packets from the endpoint. + IPv6TrafficClassOption + + // MaxSegOption is used by SetSockOpt/GetSockOpt to set/get the current + // Maximum Segment Size(MSS) value as specified using the TCP_MAXSEG option. + MaxSegOption + + // MulticastTTLOption is used by SetSockOpt/GetSockOpt to control the default + // TTL value for multicast messages. The default is 1. + MulticastTTLOption + // ReceiveQueueSizeOption is used in GetSockOptInt to specify that the // number of unread bytes in the input buffer should be returned. - ReceiveQueueSizeOption SockOptInt = iota + ReceiveQueueSizeOption // SendBufferSizeOption is used by SetSockOptInt/GetSockOptInt to // specify the send buffer size option. @@ -561,44 +617,21 @@ const ( // number of unread bytes in the output buffer should be returned. SendQueueSizeOption - // DelayOption is used by SetSockOpt/GetSockOpt to specify if data - // should be sent out immediately by the transport protocol. For TCP, - // it determines if the Nagle algorithm is on or off. - DelayOption - - // TODO(b/137664753): convert all int socket options to be handled via - // GetSockOptInt. + // TTLOption is used by SetSockOpt/GetSockOpt to control the default TTL/hop + // limit value for unicast messages. The default is protocol specific. + // + // A zero value indicates the default. + TTLOption ) // ErrorOption is used in GetSockOpt to specify that the last error reported by // the endpoint should be cleared and returned. type ErrorOption struct{} -// CorkOption is used by SetSockOpt/GetSockOpt to specify if data should be -// held until segments are full by the TCP transport protocol. -type CorkOption int - -// ReuseAddressOption is used by SetSockOpt/GetSockOpt to specify whether Bind() -// should allow reuse of local address. -type ReuseAddressOption int - -// ReusePortOption is used by SetSockOpt/GetSockOpt to permit multiple sockets -// to be bound to an identical socket address. -type ReusePortOption int - // BindToDeviceOption is used by SetSockOpt/GetSockOpt to specify that sockets // should bind only on a specific NIC. type BindToDeviceOption NICID -// QuickAckOption is stubbed out in SetSockOpt/GetSockOpt. -type QuickAckOption int - -// PasscredOption is used by SetSockOpt/GetSockOpt to specify whether -// SCM_CREDENTIALS socket control messages are enabled. -// -// Only supported on Unix sockets. -type PasscredOption int - // TCPInfoOption is used by GetSockOpt to expose TCP statistics. // // TODO(b/64800844): Add and populate stat fields. @@ -607,10 +640,6 @@ type TCPInfoOption struct { RTTVar time.Duration } -// KeepaliveEnabledOption is used by SetSockOpt/GetSockOpt to specify whether -// TCP keepalive is enabled for this socket. -type KeepaliveEnabledOption int - // KeepaliveIdleOption is used by SetSockOpt/GetSockOpt to specify the time a // connection must remain idle before the first TCP keepalive packet is sent. // Once this time is reached, KeepaliveIntervalOption is used instead. @@ -620,11 +649,6 @@ type KeepaliveIdleOption time.Duration // interval between sending TCP keepalive packets. type KeepaliveIntervalOption time.Duration -// KeepaliveCountOption is used by SetSockOpt/GetSockOpt to specify the number -// of un-ACKed TCP keepalives that will be sent before the connection is -// closed. -type KeepaliveCountOption int - // TCPUserTimeoutOption is used by SetSockOpt/GetSockOpt to specify a user // specified timeout for a given TCP connection. // See: RFC5482 for details. @@ -638,20 +662,9 @@ type CongestionControlOption string // control algorithms. type AvailableCongestionControlOption string -// ModerateReceiveBufferOption allows the caller to enable/disable TCP receive // buffer moderation. type ModerateReceiveBufferOption bool -// MaxSegOption is used by SetSockOpt/GetSockOpt to set/get the current -// Maximum Segment Size(MSS) value as specified using the TCP_MAXSEG option. -type MaxSegOption int - -// TTLOption is used by SetSockOpt/GetSockOpt to control the default TTL/hop -// limit value for unicast messages. The default is protocol specific. -// -// A zero value indicates the default. -type TTLOption uint8 - // TCPLingerTimeoutOption is used by SetSockOpt/GetSockOpt to set/get the // maximum duration for which a socket lingers in the TCP_FIN_WAIT_2 state // before being marked closed. @@ -668,10 +681,6 @@ type TCPTimeWaitTimeoutOption time.Duration // for a handshake till the specified timeout until a segment with data arrives. type TCPDeferAcceptOption time.Duration -// MulticastTTLOption is used by SetSockOpt/GetSockOpt to control the default -// TTL value for multicast messages. The default is 1. -type MulticastTTLOption uint8 - // MulticastInterfaceOption is used by SetSockOpt/GetSockOpt to specify a // default interface for multicast. type MulticastInterfaceOption struct { @@ -679,10 +688,6 @@ type MulticastInterfaceOption struct { InterfaceAddr Address } -// MulticastLoopOption is used by SetSockOpt/GetSockOpt to specify whether -// multicast packets sent over a non-loopback interface will be looped back. -type MulticastLoopOption bool - // MembershipOption is used by SetSockOpt/GetSockOpt as an argument to // AddMembershipOption and RemoveMembershipOption. type MembershipOption struct { @@ -705,22 +710,10 @@ type RemoveMembershipOption MembershipOption // TCP out-of-band data is delivered along with the normal in-band data. type OutOfBandInlineOption int -// BroadcastOption is used by SetSockOpt/GetSockOpt to specify whether -// datagram sockets are allowed to send packets to a broadcast address. -type BroadcastOption int - // DefaultTTLOption is used by stack.(*Stack).NetworkProtocolOption to specify // a default TTL. type DefaultTTLOption uint8 -// IPv4TOSOption is used by SetSockOpt/GetSockOpt to specify TOS -// for all subsequent outgoing IPv4 packets from the endpoint. -type IPv4TOSOption uint8 - -// IPv6TrafficClassOption is used by SetSockOpt/GetSockOpt to specify TOS -// for all subsequent outgoing IPv6 packets from the endpoint. -type IPv6TrafficClassOption uint8 - // IPPacketInfo is the message struture for IP_PKTINFO. // // +stateify savable diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index b007302fb..3a133eef9 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -348,29 +348,37 @@ func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { // SetSockOpt sets a socket option. func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { - switch o := opt.(type) { - case tcpip.TTLOption: - e.mu.Lock() - e.ttl = uint8(o) - e.mu.Unlock() - } - return nil } // SetSockOptBool sets a socket option. Currently not supported. func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { - return nil + return tcpip.ErrUnknownProtocolOption } // SetSockOptInt sets a socket option. Currently not supported. func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { + switch opt { + case tcpip.TTLOption: + e.mu.Lock() + e.ttl = uint8(v) + e.mu.Unlock() + + default: + return tcpip.ErrUnknownProtocolOption + } return nil } // GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { - return false, tcpip.ErrUnknownProtocolOption + switch opt { + case tcpip.KeepaliveEnabledOption: + return false, nil + + default: + return false, tcpip.ErrUnknownProtocolOption + } } // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. @@ -397,26 +405,23 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { e.rcvMu.Unlock() return v, nil + case tcpip.TTLOption: + e.rcvMu.Lock() + v := int(e.ttl) + e.rcvMu.Unlock() + return v, nil + + default: + return -1, tcpip.ErrUnknownProtocolOption } - return -1, tcpip.ErrUnknownProtocolOption } // GetSockOpt implements tcpip.Endpoint.GetSockOpt. func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { - switch o := opt.(type) { + switch opt.(type) { case tcpip.ErrorOption: return nil - case *tcpip.KeepaliveEnabledOption: - *o = 0 - return nil - - case *tcpip.TTLOption: - e.rcvMu.Lock() - *o = tcpip.TTLOption(e.ttl) - e.rcvMu.Unlock() - return nil - default: return tcpip.ErrUnknownProtocolOption } diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go index 337bc1c71..eee754a5a 100644 --- a/pkg/tcpip/transport/raw/endpoint.go +++ b/pkg/tcpip/transport/raw/endpoint.go @@ -533,14 +533,10 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { // GetSockOpt implements tcpip.Endpoint.GetSockOpt. func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { - switch o := opt.(type) { + switch opt.(type) { case tcpip.ErrorOption: return nil - case *tcpip.KeepaliveEnabledOption: - *o = 0 - return nil - default: return tcpip.ErrUnknownProtocolOption } @@ -548,7 +544,13 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { // GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { - return false, tcpip.ErrUnknownProtocolOption + switch opt { + case tcpip.KeepaliveEnabledOption: + return false, nil + + default: + return false, tcpip.ErrUnknownProtocolOption + } } // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. @@ -576,9 +578,9 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { e.rcvMu.Unlock() return v, nil + default: + return -1, tcpip.ErrUnknownProtocolOption } - - return -1, tcpip.ErrUnknownProtocolOption } // HandlePacket implements stack.RawTransportEndpoint.HandlePacket. diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 9b123e968..a8d443f73 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -821,7 +821,7 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, waiterQue var de DelayEnabled if err := s.TransportProtocolOption(ProtocolNumber, &de); err == nil && de { - e.SetSockOptInt(tcpip.DelayOption, 1) + e.SetSockOptBool(tcpip.DelayOption, true) } var tcpLT tcpip.TCPLingerTimeoutOption @@ -1409,10 +1409,60 @@ func (e *endpoint) windowCrossedACKThresholdLocked(deltaBefore int) (crossed boo // SetSockOptBool sets a socket option. func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { - e.LockUser() - defer e.UnlockUser() - switch opt { + + case tcpip.BroadcastOption: + e.LockUser() + e.broadcast = v + e.UnlockUser() + + case tcpip.CorkOption: + e.LockUser() + if !v { + atomic.StoreUint32(&e.cork, 0) + + // Handle the corked data. + e.sndWaker.Assert() + } else { + atomic.StoreUint32(&e.cork, 1) + } + e.UnlockUser() + + case tcpip.DelayOption: + if v { + atomic.StoreUint32(&e.delay, 1) + } else { + atomic.StoreUint32(&e.delay, 0) + + // Handle delayed data. + e.sndWaker.Assert() + } + + case tcpip.KeepaliveEnabledOption: + e.keepalive.Lock() + e.keepalive.enabled = v + e.keepalive.Unlock() + e.notifyProtocolGoroutine(notifyKeepaliveChanged) + + case tcpip.QuickAckOption: + o := uint32(1) + if v { + o = 0 + } + atomic.StoreUint32(&e.slowAck, o) + + case tcpip.ReuseAddressOption: + e.LockUser() + e.reuseAddr = v + e.UnlockUser() + return nil + + case tcpip.ReusePortOption: + e.LockUser() + e.reusePort = v + e.UnlockUser() + return nil + case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. if e.NetProto != header.IPv6ProtocolNumber { @@ -1424,7 +1474,11 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { return tcpip.ErrInvalidEndpointState } + e.LockUser() e.v6only = v + e.UnlockUser() + default: + return tcpip.ErrUnknownProtocolOption } return nil @@ -1432,7 +1486,40 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { // SetSockOptInt sets a socket option. func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { + // Lower 2 bits represents ECN bits. RFC 3168, section 23.1 + const inetECNMask = 3 + switch opt { + case tcpip.KeepaliveCountOption: + e.keepalive.Lock() + e.keepalive.count = int(v) + e.keepalive.Unlock() + e.notifyProtocolGoroutine(notifyKeepaliveChanged) + + case tcpip.IPv4TOSOption: + e.LockUser() + // TODO(gvisor.dev/issue/995): ECN is not currently supported, + // ignore the bits for now. + e.sendTOS = uint8(v) & ^uint8(inetECNMask) + e.UnlockUser() + + case tcpip.IPv6TrafficClassOption: + e.LockUser() + // TODO(gvisor.dev/issue/995): ECN is not currently supported, + // ignore the bits for now. + e.sendTOS = uint8(v) & ^uint8(inetECNMask) + e.UnlockUser() + + case tcpip.MaxSegOption: + userMSS := v + if userMSS < header.TCPMinimumMSS || userMSS > header.TCPMaximumMSS { + return tcpip.ErrInvalidOptionValue + } + e.LockUser() + e.userMSS = uint16(userMSS) + e.UnlockUser() + e.notifyProtocolGoroutine(notifyMSSChanged) + case tcpip.ReceiveBufferSizeOption: // Make sure the receive buffer size is within the min and max // allowed. @@ -1483,7 +1570,6 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { e.rcvListMu.Unlock() e.UnlockUser() e.notifyProtocolGoroutine(mask) - return nil case tcpip.SendBufferSizeOption: // Make sure the send buffer size is within the min and max @@ -1502,52 +1588,21 @@ func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { e.sndBufMu.Lock() e.sndBufSize = size e.sndBufMu.Unlock() - return nil - case tcpip.DelayOption: - if v == 0 { - atomic.StoreUint32(&e.delay, 0) - - // Handle delayed data. - e.sndWaker.Assert() - } else { - atomic.StoreUint32(&e.delay, 1) - } - return nil + case tcpip.TTLOption: + e.LockUser() + e.ttl = uint8(v) + e.UnlockUser() default: - return nil + return tcpip.ErrUnknownProtocolOption } + return nil } // SetSockOpt sets a socket option. func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { - // Lower 2 bits represents ECN bits. RFC 3168, section 23.1 - const inetECNMask = 3 switch v := opt.(type) { - case tcpip.CorkOption: - if v == 0 { - atomic.StoreUint32(&e.cork, 0) - - // Handle the corked data. - e.sndWaker.Assert() - } else { - atomic.StoreUint32(&e.cork, 1) - } - return nil - - case tcpip.ReuseAddressOption: - e.LockUser() - e.reuseAddr = v != 0 - e.UnlockUser() - return nil - - case tcpip.ReusePortOption: - e.LockUser() - e.reusePort = v != 0 - e.UnlockUser() - return nil - case tcpip.BindToDeviceOption: id := tcpip.NICID(v) if id != 0 && !e.stack.HasNIC(id) { @@ -1556,72 +1611,26 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.LockUser() e.bindToDevice = id e.UnlockUser() - return nil - - case tcpip.QuickAckOption: - if v == 0 { - atomic.StoreUint32(&e.slowAck, 1) - } else { - atomic.StoreUint32(&e.slowAck, 0) - } - return nil - - case tcpip.MaxSegOption: - userMSS := v - if userMSS < header.TCPMinimumMSS || userMSS > header.TCPMaximumMSS { - return tcpip.ErrInvalidOptionValue - } - e.LockUser() - e.userMSS = uint16(userMSS) - e.UnlockUser() - e.notifyProtocolGoroutine(notifyMSSChanged) - return nil - - case tcpip.TTLOption: - e.LockUser() - e.ttl = uint8(v) - e.UnlockUser() - return nil - - case tcpip.KeepaliveEnabledOption: - e.keepalive.Lock() - e.keepalive.enabled = v != 0 - e.keepalive.Unlock() - e.notifyProtocolGoroutine(notifyKeepaliveChanged) - return nil case tcpip.KeepaliveIdleOption: e.keepalive.Lock() e.keepalive.idle = time.Duration(v) e.keepalive.Unlock() e.notifyProtocolGoroutine(notifyKeepaliveChanged) - return nil case tcpip.KeepaliveIntervalOption: e.keepalive.Lock() e.keepalive.interval = time.Duration(v) e.keepalive.Unlock() e.notifyProtocolGoroutine(notifyKeepaliveChanged) - return nil - case tcpip.KeepaliveCountOption: - e.keepalive.Lock() - e.keepalive.count = int(v) - e.keepalive.Unlock() - e.notifyProtocolGoroutine(notifyKeepaliveChanged) - return nil + case tcpip.OutOfBandInlineOption: + // We don't currently support disabling this option. case tcpip.TCPUserTimeoutOption: e.LockUser() e.userTimeout = time.Duration(v) e.UnlockUser() - return nil - - case tcpip.BroadcastOption: - e.LockUser() - e.broadcast = v != 0 - e.UnlockUser() - return nil case tcpip.CongestionControlOption: // Query the available cc algorithms in the stack and @@ -1652,22 +1661,6 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { // control algorithm is specified. return tcpip.ErrNoSuchFile - case tcpip.IPv4TOSOption: - e.LockUser() - // TODO(gvisor.dev/issue/995): ECN is not currently supported, - // ignore the bits for now. - e.sendTOS = uint8(v) & ^uint8(inetECNMask) - e.UnlockUser() - return nil - - case tcpip.IPv6TrafficClassOption: - e.LockUser() - // TODO(gvisor.dev/issue/995): ECN is not currently supported, - // ignore the bits for now. - e.sendTOS = uint8(v) & ^uint8(inetECNMask) - e.UnlockUser() - return nil - case tcpip.TCPLingerTimeoutOption: e.LockUser() if v < 0 { @@ -1688,7 +1681,6 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { } e.tcpLingerTimeout = time.Duration(v) e.UnlockUser() - return nil case tcpip.TCPDeferAcceptOption: e.LockUser() @@ -1697,11 +1689,11 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { } e.deferAccept = time.Duration(v) e.UnlockUser() - return nil default: return nil } + return nil } // readyReceiveSize returns the number of bytes ready to be received. @@ -1723,6 +1715,43 @@ func (e *endpoint) readyReceiveSize() (int, *tcpip.Error) { // GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { switch opt { + case tcpip.BroadcastOption: + e.LockUser() + v := e.broadcast + e.UnlockUser() + return v, nil + + case tcpip.CorkOption: + return atomic.LoadUint32(&e.cork) != 0, nil + + case tcpip.DelayOption: + return atomic.LoadUint32(&e.delay) != 0, nil + + case tcpip.KeepaliveEnabledOption: + e.keepalive.Lock() + v := e.keepalive.enabled + e.keepalive.Unlock() + + return v, nil + + case tcpip.QuickAckOption: + v := atomic.LoadUint32(&e.slowAck) == 0 + return v, nil + + case tcpip.ReuseAddressOption: + e.LockUser() + v := e.reuseAddr + e.UnlockUser() + + return v, nil + + case tcpip.ReusePortOption: + e.LockUser() + v := e.reusePort + e.UnlockUser() + + return v, nil + case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. if e.NetProto != header.IPv6ProtocolNumber { @@ -1734,14 +1763,41 @@ func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { e.UnlockUser() return v, nil - } - return false, tcpip.ErrUnknownProtocolOption + default: + return false, tcpip.ErrUnknownProtocolOption + } } // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { + case tcpip.KeepaliveCountOption: + e.keepalive.Lock() + v := e.keepalive.count + e.keepalive.Unlock() + return v, nil + + case tcpip.IPv4TOSOption: + e.LockUser() + v := int(e.sendTOS) + e.UnlockUser() + return v, nil + + case tcpip.IPv6TrafficClassOption: + e.LockUser() + v := int(e.sendTOS) + e.UnlockUser() + return v, nil + + case tcpip.MaxSegOption: + // This is just stubbed out. Linux never returns the user_mss + // value as it either returns the defaultMSS or returns the + // actual current MSS. Netstack just returns the defaultMSS + // always for now. + v := header.TCPDefaultMSS + return v, nil + case tcpip.ReceiveQueueSizeOption: return e.readyReceiveSize() @@ -1757,12 +1813,11 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { e.rcvListMu.Unlock() return v, nil - case tcpip.DelayOption: - var o int - if v := atomic.LoadUint32(&e.delay); v != 0 { - o = 1 - } - return o, nil + case tcpip.TTLOption: + e.LockUser() + v := int(e.ttl) + e.UnlockUser() + return v, nil default: return -1, tcpip.ErrUnknownProtocolOption @@ -1779,61 +1834,10 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { e.lastErrorMu.Unlock() return err - case *tcpip.MaxSegOption: - // This is just stubbed out. Linux never returns the user_mss - // value as it either returns the defaultMSS or returns the - // actual current MSS. Netstack just returns the defaultMSS - // always for now. - *o = header.TCPDefaultMSS - return nil - - case *tcpip.CorkOption: - *o = 0 - if v := atomic.LoadUint32(&e.cork); v != 0 { - *o = 1 - } - return nil - - case *tcpip.ReuseAddressOption: - e.LockUser() - v := e.reuseAddr - e.UnlockUser() - - *o = 0 - if v { - *o = 1 - } - return nil - - case *tcpip.ReusePortOption: - e.LockUser() - v := e.reusePort - e.UnlockUser() - - *o = 0 - if v { - *o = 1 - } - return nil - case *tcpip.BindToDeviceOption: e.LockUser() *o = tcpip.BindToDeviceOption(e.bindToDevice) e.UnlockUser() - return nil - - case *tcpip.QuickAckOption: - *o = 1 - if v := atomic.LoadUint32(&e.slowAck); v != 0 { - *o = 0 - } - return nil - - case *tcpip.TTLOption: - e.LockUser() - *o = tcpip.TTLOption(e.ttl) - e.UnlockUser() - return nil case *tcpip.TCPInfoOption: *o = tcpip.TCPInfoOption{} @@ -1846,92 +1850,45 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { o.RTTVar = snd.rtt.rttvar snd.rtt.Unlock() } - return nil - - case *tcpip.KeepaliveEnabledOption: - e.keepalive.Lock() - v := e.keepalive.enabled - e.keepalive.Unlock() - - *o = 0 - if v { - *o = 1 - } - return nil case *tcpip.KeepaliveIdleOption: e.keepalive.Lock() *o = tcpip.KeepaliveIdleOption(e.keepalive.idle) e.keepalive.Unlock() - return nil case *tcpip.KeepaliveIntervalOption: e.keepalive.Lock() *o = tcpip.KeepaliveIntervalOption(e.keepalive.interval) e.keepalive.Unlock() - return nil - - case *tcpip.KeepaliveCountOption: - e.keepalive.Lock() - *o = tcpip.KeepaliveCountOption(e.keepalive.count) - e.keepalive.Unlock() - return nil case *tcpip.TCPUserTimeoutOption: e.LockUser() *o = tcpip.TCPUserTimeoutOption(e.userTimeout) e.UnlockUser() - return nil case *tcpip.OutOfBandInlineOption: // We don't currently support disabling this option. *o = 1 - return nil - - case *tcpip.BroadcastOption: - e.LockUser() - v := e.broadcast - e.UnlockUser() - - *o = 0 - if v { - *o = 1 - } - return nil case *tcpip.CongestionControlOption: e.LockUser() *o = e.cc e.UnlockUser() - return nil - - case *tcpip.IPv4TOSOption: - e.LockUser() - *o = tcpip.IPv4TOSOption(e.sendTOS) - e.UnlockUser() - return nil - - case *tcpip.IPv6TrafficClassOption: - e.LockUser() - *o = tcpip.IPv6TrafficClassOption(e.sendTOS) - e.UnlockUser() - return nil case *tcpip.TCPLingerTimeoutOption: e.LockUser() *o = tcpip.TCPLingerTimeoutOption(e.tcpLingerTimeout) e.UnlockUser() - return nil case *tcpip.TCPDeferAcceptOption: e.LockUser() *o = tcpip.TCPDeferAcceptOption(e.deferAccept) e.UnlockUser() - return nil default: return tcpip.ErrUnknownProtocolOption } + return nil } // checkV4MappedLocked determines the effective network protocol and converts diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 8a87fa0a8..29301a45c 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -728,7 +728,7 @@ func TestUserSuppliedMSSOnConnectV4(t *testing.T) { const maxMSS = mtu - header.IPv4MinimumSize - header.TCPMinimumSize tests := []struct { name string - setMSS uint16 + setMSS int expMSS uint16 }{ { @@ -756,15 +756,14 @@ func TestUserSuppliedMSSOnConnectV4(t *testing.T) { c.Create(-1) // Set the MSS socket option. - opt := tcpip.MaxSegOption(test.setMSS) - if err := c.EP.SetSockOpt(opt); err != nil { - t.Fatalf("SetSockOpt(%#v) failed: %s", opt, err) + if err := c.EP.SetSockOptInt(tcpip.MaxSegOption, test.setMSS); err != nil { + t.Fatalf("SetSockOptInt(MaxSegOption, %d) failed: %s", test.setMSS, err) } // Get expected window size. rcvBufSize, err := c.EP.GetSockOptInt(tcpip.ReceiveBufferSizeOption) if err != nil { - t.Fatalf("GetSockOpt(%v) failed: %s", tcpip.ReceiveBufferSizeOption, err) + t.Fatalf("GetSockOptInt(ReceiveBufferSizeOption) failed: %s", err) } ws := tcp.FindWndScale(seqnum.Size(rcvBufSize)) @@ -818,15 +817,14 @@ func TestUserSuppliedMSSOnConnectV6(t *testing.T) { c.CreateV6Endpoint(true) // Set the MSS socket option. - opt := tcpip.MaxSegOption(test.setMSS) - if err := c.EP.SetSockOpt(opt); err != nil { - t.Fatalf("SetSockOpt(%#v) failed: %s", opt, err) + if err := c.EP.SetSockOptInt(tcpip.MaxSegOption, int(test.setMSS)); err != nil { + t.Fatalf("SetSockOptInt(MaxSegOption, %d) failed: %s", test.setMSS, err) } // Get expected window size. rcvBufSize, err := c.EP.GetSockOptInt(tcpip.ReceiveBufferSizeOption) if err != nil { - t.Fatalf("GetSockOpt(%v) failed: %s", tcpip.ReceiveBufferSizeOption, err) + t.Fatalf("GetSockOptInt(ReceiveBufferSizeOption) failed: %s", err) } ws := tcp.FindWndScale(seqnum.Size(rcvBufSize)) @@ -1077,17 +1075,17 @@ func TestTOSV4(t *testing.T) { c.EP = ep const tos = 0xC0 - if err := c.EP.SetSockOpt(tcpip.IPv4TOSOption(tos)); err != nil { - t.Errorf("SetSockOpt(%#v) failed: %s", tcpip.IPv4TOSOption(tos), err) + if err := c.EP.SetSockOptInt(tcpip.IPv4TOSOption, tos); err != nil { + t.Errorf("SetSockOptInt(IPv4TOSOption, %d) failed: %s", tos, err) } - var v tcpip.IPv4TOSOption - if err := c.EP.GetSockOpt(&v); err != nil { - t.Errorf("GetSockopt failed: %s", err) + v, err := c.EP.GetSockOptInt(tcpip.IPv4TOSOption) + if err != nil { + t.Errorf("GetSockoptInt(IPv4TOSOption) failed: %s", err) } - if want := tcpip.IPv4TOSOption(tos); v != want { - t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, want) + if v != tos { + t.Errorf("got GetSockOptInt(IPv4TOSOption) = %d, want = %d", v, tos) } testV4Connect(t, c, checker.TOS(tos, 0)) @@ -1125,17 +1123,17 @@ func TestTrafficClassV6(t *testing.T) { c.CreateV6Endpoint(false) const tos = 0xC0 - if err := c.EP.SetSockOpt(tcpip.IPv6TrafficClassOption(tos)); err != nil { - t.Errorf("SetSockOpt(%#v) failed: %s", tcpip.IPv6TrafficClassOption(tos), err) + if err := c.EP.SetSockOptInt(tcpip.IPv6TrafficClassOption, tos); err != nil { + t.Errorf("SetSockOpInt(IPv6TrafficClassOption, %d) failed: %s", tos, err) } - var v tcpip.IPv6TrafficClassOption - if err := c.EP.GetSockOpt(&v); err != nil { - t.Fatalf("GetSockopt failed: %s", err) + v, err := c.EP.GetSockOptInt(tcpip.IPv6TrafficClassOption) + if err != nil { + t.Fatalf("GetSockoptInt(IPv6TrafficClassOption) failed: %s", err) } - if want := tcpip.IPv6TrafficClassOption(tos); v != want { - t.Errorf("got GetSockOpt(...) = %#v, want = %#v", v, want) + if v != tos { + t.Errorf("got GetSockOptInt(IPv6TrafficClassOption) = %d, want = %d", v, tos) } // Test the connection request. @@ -1711,7 +1709,7 @@ func TestNoWindowShrinking(t *testing.T) { c.CreateConnected(789, 30000, 10) if err := c.EP.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 5); err != nil { - t.Fatalf("SetSockOpt failed: %v", err) + t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, 5) failed: %v", err) } we, ch := waiter.NewChannelEntry(nil) @@ -1984,7 +1982,7 @@ func TestScaledWindowAccept(t *testing.T) { // Set the window size greater than the maximum non-scaled window. if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 65535*3); err != nil { - t.Fatalf("SetSockOpt failed failed: %v", err) + t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, 65535*3) failed failed: %v", err) } if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { @@ -2057,7 +2055,7 @@ func TestNonScaledWindowAccept(t *testing.T) { // Set the window size greater than the maximum non-scaled window. if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 65535*3); err != nil { - t.Fatalf("SetSockOpt failed failed: %v", err) + t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, 65535*3) failed failed: %v", err) } if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { @@ -2221,10 +2219,10 @@ func TestSegmentMerging(t *testing.T) { { "cork", func(ep tcpip.Endpoint) { - ep.SetSockOpt(tcpip.CorkOption(1)) + ep.SetSockOptBool(tcpip.CorkOption, true) }, func(ep tcpip.Endpoint) { - ep.SetSockOpt(tcpip.CorkOption(0)) + ep.SetSockOptBool(tcpip.CorkOption, false) }, }, } @@ -2316,7 +2314,7 @@ func TestDelay(t *testing.T) { c.CreateConnected(789, 30000, -1 /* epRcvBuf */) - c.EP.SetSockOptInt(tcpip.DelayOption, 1) + c.EP.SetSockOptBool(tcpip.DelayOption, true) var allData []byte for i, data := range [][]byte{{0}, {1, 2, 3, 4}, {5, 6, 7}, {8, 9}, {10}, {11}} { @@ -2364,7 +2362,7 @@ func TestUndelay(t *testing.T) { c.CreateConnected(789, 30000, -1 /* epRcvBuf */) - c.EP.SetSockOptInt(tcpip.DelayOption, 1) + c.EP.SetSockOptBool(tcpip.DelayOption, true) allData := [][]byte{{0}, {1, 2, 3}} for i, data := range allData { @@ -2397,7 +2395,7 @@ func TestUndelay(t *testing.T) { // Check that we don't get the second packet yet. c.CheckNoPacketTimeout("delayed second packet transmitted", 100*time.Millisecond) - c.EP.SetSockOptInt(tcpip.DelayOption, 0) + c.EP.SetSockOptBool(tcpip.DelayOption, false) // Check that data is received. second := c.GetPacket() @@ -2434,8 +2432,8 @@ func TestMSSNotDelayed(t *testing.T) { fn func(tcpip.Endpoint) }{ {"no-op", func(tcpip.Endpoint) {}}, - {"delay", func(ep tcpip.Endpoint) { ep.SetSockOptInt(tcpip.DelayOption, 1) }}, - {"cork", func(ep tcpip.Endpoint) { ep.SetSockOpt(tcpip.CorkOption(1)) }}, + {"delay", func(ep tcpip.Endpoint) { ep.SetSockOptBool(tcpip.DelayOption, true) }}, + {"cork", func(ep tcpip.Endpoint) { ep.SetSockOptBool(tcpip.CorkOption, true) }}, } for _, test := range tests { @@ -2576,12 +2574,12 @@ func TestSetTTL(t *testing.T) { t.Fatalf("NewEndpoint failed: %v", err) } - if err := c.EP.SetSockOpt(tcpip.TTLOption(wantTTL)); err != nil { - t.Fatalf("SetSockOpt failed: %v", err) + if err := c.EP.SetSockOptInt(tcpip.TTLOption, int(wantTTL)); err != nil { + t.Fatalf("SetSockOptInt(TTLOption, %d) failed: %s", wantTTL, err) } if err := c.EP.Connect(tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort}); err != tcpip.ErrConnectStarted { - t.Fatalf("Unexpected return value from Connect: %v", err) + t.Fatalf("Unexpected return value from Connect: %s", err) } // Receive SYN packet. @@ -2621,7 +2619,7 @@ func TestPassiveSendMSSLessThanMTU(t *testing.T) { // window scaling option. const rcvBufferSize = 0x20000 if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, rcvBufferSize); err != nil { - t.Fatalf("SetSockOpt failed failed: %v", err) + t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, %d) failed failed: %s", rcvBufferSize, err) } if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil { @@ -2765,7 +2763,7 @@ func TestSynOptionsOnActiveConnect(t *testing.T) { const rcvBufferSize = 0x20000 const wndScale = 2 if err := c.EP.SetSockOptInt(tcpip.ReceiveBufferSizeOption, rcvBufferSize); err != nil { - t.Fatalf("SetSockOpt failed failed: %v", err) + t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, %d) failed failed: %s", rcvBufferSize, err) } // Start connection attempt. @@ -3882,26 +3880,26 @@ func TestMinMaxBufferSizes(t *testing.T) { // Set values below the min. if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 199); err != nil { - t.Fatalf("GetSockOpt failed: %v", err) + t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption, 199) failed: %s", err) } checkRecvBufferSize(t, ep, 200) if err := ep.SetSockOptInt(tcpip.SendBufferSizeOption, 299); err != nil { - t.Fatalf("GetSockOpt failed: %v", err) + t.Fatalf("SetSockOptInt(SendBufferSizeOption, 299) failed: %s", err) } checkSendBufferSize(t, ep, 300) // Set values above the max. if err := ep.SetSockOptInt(tcpip.ReceiveBufferSizeOption, 1+tcp.DefaultReceiveBufferSize*20); err != nil { - t.Fatalf("GetSockOpt failed: %v", err) + t.Fatalf("SetSockOptInt(ReceiveBufferSizeOption) failed: %s", err) } checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize*20) if err := ep.SetSockOptInt(tcpip.SendBufferSizeOption, 1+tcp.DefaultSendBufferSize*30); err != nil { - t.Fatalf("GetSockOpt failed: %v", err) + t.Fatalf("SetSockOptInt(SendBufferSizeOption) failed: %s", err) } checkSendBufferSize(t, ep, tcp.DefaultSendBufferSize*30) @@ -4147,11 +4145,11 @@ func TestConnectAvoidsBoundPorts(t *testing.T) { case "ipv4": case "ipv6": if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil { - t.Fatalf("SetSockOpt(V6OnlyOption(true)) failed: %v", err) + t.Fatalf("SetSockOptBool(V6OnlyOption(true)) failed: %s", err) } case "dual": if err := ep.SetSockOptBool(tcpip.V6OnlyOption, false); err != nil { - t.Fatalf("SetSockOpt(V6OnlyOption(false)) failed: %v", err) + t.Fatalf("SetSockOptBool(V6OnlyOption(false)) failed: %s", err) } default: t.Fatalf("unknown network: '%s'", network) @@ -4477,8 +4475,8 @@ func TestKeepalive(t *testing.T) { const keepAliveInterval = 10 * time.Millisecond c.EP.SetSockOpt(tcpip.KeepaliveIdleOption(10 * time.Millisecond)) c.EP.SetSockOpt(tcpip.KeepaliveIntervalOption(keepAliveInterval)) - c.EP.SetSockOpt(tcpip.KeepaliveCountOption(5)) - c.EP.SetSockOpt(tcpip.KeepaliveEnabledOption(1)) + c.EP.SetSockOptInt(tcpip.KeepaliveCountOption, 5) + c.EP.SetSockOptBool(tcpip.KeepaliveEnabledOption, true) // 5 unacked keepalives are sent. ACK each one, and check that the // connection stays alive after 5. @@ -5770,14 +5768,14 @@ func TestReceiveBufferAutoTuning(t *testing.T) { func TestDelayEnabled(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() - checkDelayOption(t, c, false, 0) // Delay is disabled by default. + checkDelayOption(t, c, false, false) // Delay is disabled by default. for _, v := range []struct { delayEnabled tcp.DelayEnabled - wantDelayOption int + wantDelayOption bool }{ - {delayEnabled: false, wantDelayOption: 0}, - {delayEnabled: true, wantDelayOption: 1}, + {delayEnabled: false, wantDelayOption: false}, + {delayEnabled: true, wantDelayOption: true}, } { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -5788,7 +5786,7 @@ func TestDelayEnabled(t *testing.T) { } } -func checkDelayOption(t *testing.T, c *context.Context, wantDelayEnabled tcp.DelayEnabled, wantDelayOption int) { +func checkDelayOption(t *testing.T, c *context.Context, wantDelayEnabled tcp.DelayEnabled, wantDelayOption bool) { t.Helper() var gotDelayEnabled tcp.DelayEnabled @@ -5803,12 +5801,12 @@ func checkDelayOption(t *testing.T, c *context.Context, wantDelayEnabled tcp.Del if err != nil { t.Fatalf("NewEndPoint(tcp, ipv4, new(waiter.Queue)) failed: %v", err) } - gotDelayOption, err := ep.GetSockOptInt(tcpip.DelayOption) + gotDelayOption, err := ep.GetSockOptBool(tcpip.DelayOption) if err != nil { - t.Fatalf("ep.GetSockOptInt(tcpip.DelayOption) failed: %v", err) + t.Fatalf("ep.GetSockOptBool(tcpip.DelayOption) failed: %s", err) } if gotDelayOption != wantDelayOption { - t.Errorf("ep.GetSockOptInt(tcpip.DelayOption) got: %d, want: %d", gotDelayOption, wantDelayOption) + t.Errorf("ep.GetSockOptBool(tcpip.DelayOption) got: %t, want: %t", gotDelayOption, wantDelayOption) } } @@ -6620,8 +6618,8 @@ func TestKeepaliveWithUserTimeout(t *testing.T) { const keepAliveInterval = 10 * time.Millisecond c.EP.SetSockOpt(tcpip.KeepaliveIdleOption(10 * time.Millisecond)) c.EP.SetSockOpt(tcpip.KeepaliveIntervalOption(keepAliveInterval)) - c.EP.SetSockOpt(tcpip.KeepaliveCountOption(10)) - c.EP.SetSockOpt(tcpip.KeepaliveEnabledOption(1)) + c.EP.SetSockOptInt(tcpip.KeepaliveCountOption, 10) + c.EP.SetSockOptBool(tcpip.KeepaliveEnabledOption, true) // Set userTimeout to be the duration for 3 keepalive probes. userTimeout := 30 * time.Millisecond diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index 120d3baa3..492cc1fcb 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -501,11 +501,20 @@ func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { // SetSockOptBool implements tcpip.Endpoint.SetSockOptBool. func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { switch opt { + case tcpip.BroadcastOption: + e.mu.Lock() + e.broadcast = v + e.mu.Unlock() + + case tcpip.MulticastLoopOption: + e.mu.Lock() + e.multicastLoop = v + e.mu.Unlock() + case tcpip.ReceiveTOSOption: e.mu.Lock() e.receiveTOS = v e.mu.Unlock() - return nil case tcpip.ReceiveTClassOption: // We only support this option on v6 endpoints. @@ -516,7 +525,18 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { e.mu.Lock() e.receiveTClass = v e.mu.Unlock() - return nil + + case tcpip.ReceiveIPPacketInfoOption: + e.mu.Lock() + e.receiveIPPacketInfo = v + e.mu.Unlock() + + case tcpip.ReuseAddressOption: + + case tcpip.ReusePortOption: + e.mu.Lock() + e.reusePort = v + e.mu.Unlock() case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. @@ -533,13 +553,8 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { } e.v6only = v - return nil - - case tcpip.ReceiveIPPacketInfoOption: - e.mu.Lock() - e.receiveIPPacketInfo = v - e.mu.Unlock() - return nil + default: + return tcpip.ErrUnknownProtocolOption } return nil @@ -547,22 +562,40 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt. func (e *endpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { - return nil -} + switch opt { + case tcpip.MulticastTTLOption: + e.mu.Lock() + e.multicastTTL = uint8(v) + e.mu.Unlock() -// SetSockOpt implements tcpip.Endpoint.SetSockOpt. -func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { - switch v := opt.(type) { case tcpip.TTLOption: e.mu.Lock() e.ttl = uint8(v) e.mu.Unlock() - case tcpip.MulticastTTLOption: + case tcpip.IPv4TOSOption: e.mu.Lock() - e.multicastTTL = uint8(v) + e.sendTOS = uint8(v) + e.mu.Unlock() + + case tcpip.IPv6TrafficClassOption: + e.mu.Lock() + e.sendTOS = uint8(v) e.mu.Unlock() + case tcpip.ReceiveBufferSizeOption: + case tcpip.SendBufferSizeOption: + + default: + return tcpip.ErrUnknownProtocolOption + } + + return nil +} + +// SetSockOpt implements tcpip.Endpoint.SetSockOpt. +func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { + switch v := opt.(type) { case tcpip.MulticastInterfaceOption: e.mu.Lock() defer e.mu.Unlock() @@ -686,16 +719,6 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.multicastMemberships[memToRemoveIndex] = e.multicastMemberships[len(e.multicastMemberships)-1] e.multicastMemberships = e.multicastMemberships[:len(e.multicastMemberships)-1] - case tcpip.MulticastLoopOption: - e.mu.Lock() - e.multicastLoop = bool(v) - e.mu.Unlock() - - case tcpip.ReusePortOption: - e.mu.Lock() - e.reusePort = v != 0 - e.mu.Unlock() - case tcpip.BindToDeviceOption: id := tcpip.NICID(v) if id != 0 && !e.stack.HasNIC(id) { @@ -704,26 +727,6 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { e.mu.Lock() e.bindToDevice = id e.mu.Unlock() - return nil - - case tcpip.BroadcastOption: - e.mu.Lock() - e.broadcast = v != 0 - e.mu.Unlock() - - return nil - - case tcpip.IPv4TOSOption: - e.mu.Lock() - e.sendTOS = uint8(v) - e.mu.Unlock() - return nil - - case tcpip.IPv6TrafficClassOption: - e.mu.Lock() - e.sendTOS = uint8(v) - e.mu.Unlock() - return nil } return nil } @@ -731,6 +734,21 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { // GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { switch opt { + case tcpip.BroadcastOption: + e.mu.RLock() + v := e.broadcast + e.mu.RUnlock() + return v, nil + + case tcpip.KeepaliveEnabledOption: + return false, nil + + case tcpip.MulticastLoopOption: + e.mu.RLock() + v := e.multicastLoop + e.mu.RUnlock() + return v, nil + case tcpip.ReceiveTOSOption: e.mu.RLock() v := e.receiveTOS @@ -748,6 +766,22 @@ func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { e.mu.RUnlock() return v, nil + case tcpip.ReceiveIPPacketInfoOption: + e.mu.RLock() + v := e.receiveIPPacketInfo + e.mu.RUnlock() + return v, nil + + case tcpip.ReuseAddressOption: + return false, nil + + case tcpip.ReusePortOption: + e.mu.RLock() + v := e.reusePort + e.mu.RUnlock() + + return v, nil + case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. if e.NetProto != header.IPv6ProtocolNumber { @@ -760,19 +794,32 @@ func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { return v, nil - case tcpip.ReceiveIPPacketInfoOption: - e.mu.RLock() - v := e.receiveIPPacketInfo - e.mu.RUnlock() - return v, nil + default: + return false, tcpip.ErrUnknownProtocolOption } - - return false, tcpip.ErrUnknownProtocolOption } // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt. func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { + case tcpip.IPv4TOSOption: + e.mu.RLock() + v := int(e.sendTOS) + e.mu.RUnlock() + return v, nil + + case tcpip.IPv6TrafficClassOption: + e.mu.RLock() + v := int(e.sendTOS) + e.mu.RUnlock() + return v, nil + + case tcpip.MulticastTTLOption: + e.mu.Lock() + v := int(e.multicastTTL) + e.mu.Unlock() + return v, nil + case tcpip.ReceiveQueueSizeOption: v := 0 e.rcvMu.Lock() @@ -794,29 +841,22 @@ func (e *endpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { v := e.rcvBufSizeMax e.rcvMu.Unlock() return v, nil - } - return -1, tcpip.ErrUnknownProtocolOption + case tcpip.TTLOption: + e.mu.Lock() + v := int(e.ttl) + e.mu.Unlock() + return v, nil + + default: + return -1, tcpip.ErrUnknownProtocolOption + } } // GetSockOpt implements tcpip.Endpoint.GetSockOpt. func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { switch o := opt.(type) { case tcpip.ErrorOption: - return nil - - case *tcpip.TTLOption: - e.mu.Lock() - *o = tcpip.TTLOption(e.ttl) - e.mu.Unlock() - return nil - - case *tcpip.MulticastTTLOption: - e.mu.Lock() - *o = tcpip.MulticastTTLOption(e.multicastTTL) - e.mu.Unlock() - return nil - case *tcpip.MulticastInterfaceOption: e.mu.Lock() *o = tcpip.MulticastInterfaceOption{ @@ -824,67 +864,16 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { e.multicastAddr, } e.mu.Unlock() - return nil - - case *tcpip.MulticastLoopOption: - e.mu.RLock() - v := e.multicastLoop - e.mu.RUnlock() - - *o = tcpip.MulticastLoopOption(v) - return nil - - case *tcpip.ReuseAddressOption: - *o = 0 - return nil - - case *tcpip.ReusePortOption: - e.mu.RLock() - v := e.reusePort - e.mu.RUnlock() - - *o = 0 - if v { - *o = 1 - } - return nil case *tcpip.BindToDeviceOption: e.mu.RLock() *o = tcpip.BindToDeviceOption(e.bindToDevice) e.mu.RUnlock() - return nil - - case *tcpip.KeepaliveEnabledOption: - *o = 0 - return nil - - case *tcpip.BroadcastOption: - e.mu.RLock() - v := e.broadcast - e.mu.RUnlock() - - *o = 0 - if v { - *o = 1 - } - return nil - - case *tcpip.IPv4TOSOption: - e.mu.RLock() - *o = tcpip.IPv4TOSOption(e.sendTOS) - e.mu.RUnlock() - return nil - - case *tcpip.IPv6TrafficClassOption: - e.mu.RLock() - *o = tcpip.IPv6TrafficClassOption(e.sendTOS) - e.mu.RUnlock() - return nil default: return tcpip.ErrUnknownProtocolOption } + return nil } // sendUDP sends a UDP segment via the provided network endpoint and under the diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index fd818243f..8acaa607a 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -343,11 +343,11 @@ func (c *testContext) createEndpointForFlow(flow testFlow) { c.createEndpoint(flow.sockProto()) if flow.isV6Only() { if err := c.ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil { - c.t.Fatalf("SetSockOpt failed: %v", err) + c.t.Fatalf("SetSockOptBool failed: %s", err) } } else if flow.isBroadcast() { - if err := c.ep.SetSockOpt(tcpip.BroadcastOption(1)); err != nil { - c.t.Fatal("SetSockOpt failed:", err) + if err := c.ep.SetSockOptBool(tcpip.BroadcastOption, true); err != nil { + c.t.Fatalf("SetSockOptBool failed: %s", err) } } } @@ -1272,8 +1272,8 @@ func TestTTL(t *testing.T) { c.createEndpointForFlow(flow) const multicastTTL = 42 - if err := c.ep.SetSockOpt(tcpip.MulticastTTLOption(multicastTTL)); err != nil { - c.t.Fatalf("SetSockOpt failed: %v", err) + if err := c.ep.SetSockOptInt(tcpip.MulticastTTLOption, multicastTTL); err != nil { + c.t.Fatalf("SetSockOptInt failed: %s", err) } var wantTTL uint8 @@ -1312,8 +1312,8 @@ func TestSetTTL(t *testing.T) { c.createEndpointForFlow(flow) - if err := c.ep.SetSockOpt(tcpip.TTLOption(wantTTL)); err != nil { - c.t.Fatalf("SetSockOpt failed: %v", err) + if err := c.ep.SetSockOptInt(tcpip.TTLOption, int(wantTTL)); err != nil { + c.t.Fatalf("SetSockOptInt(TTLOption, %d) failed: %s", wantTTL, err) } var p stack.NetworkProtocol @@ -1347,25 +1347,26 @@ func TestSetTOS(t *testing.T) { c.createEndpointForFlow(flow) const tos = testTOS - var v tcpip.IPv4TOSOption - if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt(%T) failed: %s", v, err) + v, err := c.ep.GetSockOptInt(tcpip.IPv4TOSOption) + if err != nil { + c.t.Errorf("GetSockOptInt(IPv4TOSOption) failed: %s", err) } // Test for expected default value. if v != 0 { - c.t.Errorf("got GetSockOpt(%T) = 0x%x, want = 0x%x", v, v, 0) + c.t.Errorf("got GetSockOpt(IPv4TOSOption) = 0x%x, want = 0x%x", v, 0) } - if err := c.ep.SetSockOpt(tcpip.IPv4TOSOption(tos)); err != nil { - c.t.Errorf("SetSockOpt(%T, 0x%x) failed: %s", v, tcpip.IPv4TOSOption(tos), err) + if err := c.ep.SetSockOptInt(tcpip.IPv4TOSOption, tos); err != nil { + c.t.Errorf("SetSockOptInt(IPv4TOSOption, 0x%x) failed: %s", tos, err) } - if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt(%T) failed: %s", v, err) + v, err = c.ep.GetSockOptInt(tcpip.IPv4TOSOption) + if err != nil { + c.t.Errorf("GetSockOptInt(IPv4TOSOption) failed: %s", err) } - if want := tcpip.IPv4TOSOption(tos); v != want { - c.t.Errorf("got GetSockOpt(%T) = 0x%x, want = 0x%x", v, v, want) + if v != tos { + c.t.Errorf("got GetSockOptInt(IPv4TOSOption) = 0x%x, want = 0x%x", v, tos) } testWrite(c, flow, checker.TOS(tos, 0)) @@ -1382,25 +1383,26 @@ func TestSetTClass(t *testing.T) { c.createEndpointForFlow(flow) const tClass = testTOS - var v tcpip.IPv6TrafficClassOption - if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt(%T) failed: %s", v, err) + v, err := c.ep.GetSockOptInt(tcpip.IPv6TrafficClassOption) + if err != nil { + c.t.Errorf("GetSockOptInt(IPv6TrafficClassOption) failed: %s", err) } // Test for expected default value. if v != 0 { - c.t.Errorf("got GetSockOpt(%T) = 0x%x, want = 0x%x", v, v, 0) + c.t.Errorf("got GetSockOptInt(IPv6TrafficClassOption) = 0x%x, want = 0x%x", v, 0) } - if err := c.ep.SetSockOpt(tcpip.IPv6TrafficClassOption(tClass)); err != nil { - c.t.Errorf("SetSockOpt(%T, 0x%x) failed: %s", v, tcpip.IPv6TrafficClassOption(tClass), err) + if err := c.ep.SetSockOptInt(tcpip.IPv6TrafficClassOption, tClass); err != nil { + c.t.Errorf("SetSockOptInt(IPv6TrafficClassOption, 0x%x) failed: %s", tClass, err) } - if err := c.ep.GetSockOpt(&v); err != nil { - c.t.Errorf("GetSockopt(%T) failed: %s", v, err) + v, err = c.ep.GetSockOptInt(tcpip.IPv6TrafficClassOption) + if err != nil { + c.t.Errorf("GetSockOptInt(IPv6TrafficClassOption) failed: %s", err) } - if want := tcpip.IPv6TrafficClassOption(tClass); v != want { - c.t.Errorf("got GetSockOpt(%T) = 0x%x, want = 0x%x", v, v, want) + if v != tClass { + c.t.Errorf("got GetSockOptInt(IPv6TrafficClassOption) = 0x%x, want = 0x%x", v, tClass) } // The header getter for TClass is called TOS, so use that checker. @@ -1431,7 +1433,7 @@ func TestReceiveTosTClass(t *testing.T) { // Verify that setting and reading the option works. v, err := c.ep.GetSockOptBool(option) if err != nil { - c.t.Errorf("GetSockoptBool(%s) failed: %s", name, err) + c.t.Errorf("GetSockOptBool(%s) failed: %s", name, err) } // Test for expected default value. if v != false { @@ -1445,7 +1447,7 @@ func TestReceiveTosTClass(t *testing.T) { got, err := c.ep.GetSockOptBool(option) if err != nil { - c.t.Errorf("GetSockoptBool(%s) failed: %s", name, err) + c.t.Errorf("GetSockOptBool(%s) failed: %s", name, err) } if got != want { diff --git a/runsc/cmd/gofer.go b/runsc/cmd/gofer.go index 02e5af3d3..28f0d54b9 100644 --- a/runsc/cmd/gofer.go +++ b/runsc/cmd/gofer.go @@ -272,9 +272,8 @@ func setupRootFS(spec *specs.Spec, conf *boot.Config) error { root := spec.Root.Path if !conf.TestOnlyAllowRunAsCurrentUserWithoutChroot { - // FIXME: runsc can't be re-executed without - // /proc, so we create a tmpfs mount, mount ./proc and ./root - // there, then move this mount to the root and after + // runsc can't be re-executed without /proc, so we create a tmpfs mount, + // mount ./proc and ./root there, then move this mount to the root and after // setCapsAndCallSelf, runsc will chroot into /root. // // We need a directory to construct a new root and we know that diff --git a/runsc/sandbox/sandbox.go b/runsc/sandbox/sandbox.go index 2d464b1bf..e82bcef6f 100644 --- a/runsc/sandbox/sandbox.go +++ b/runsc/sandbox/sandbox.go @@ -402,8 +402,6 @@ func (s *Sandbox) createSandboxProcess(conf *boot.Config, args *Args, startSyncF nextFD++ } - cmd.Args = append(cmd.Args, "--panic-signal="+strconv.Itoa(int(syscall.SIGTERM))) - // Add the "boot" command to the args. // // All flags after this must be for the boot command diff --git a/scripts/common.sh b/scripts/common.sh index 735a383de..bc6ba71e8 100755 --- a/scripts/common.sh +++ b/scripts/common.sh @@ -89,12 +89,20 @@ function install_runsc() { # be correct, otherwise this may result in a loop that spins until time out. function apt_install() { while true; do - if (sudo apt-get update && sudo apt-get install -y "$@"); then - break - fi - result=$? - if [[ $result -ne 100 ]]; then - return $result - fi + sudo apt-get update && + sudo apt-get install -y "$@" && + true + result="${?}" + case $result in + 0) + break + ;; + 100) + # 100 is the error code that apt-get returns. + ;; + *) + exit $result + ;; + esac done } diff --git a/test/packetimpact/testbench/BUILD b/test/packetimpact/testbench/BUILD index 199823419..838a10ffe 100644 --- a/test/packetimpact/testbench/BUILD +++ b/test/packetimpact/testbench/BUILD @@ -36,4 +36,5 @@ go_test( size = "small", srcs = ["layers_test.go"], library = ":testbench", + deps = ["//pkg/tcpip"], ) diff --git a/test/packetimpact/testbench/layers.go b/test/packetimpact/testbench/layers.go index 4d6625941..093a46e23 100644 --- a/test/packetimpact/testbench/layers.go +++ b/test/packetimpact/testbench/layers.go @@ -118,7 +118,7 @@ func stringLayer(l Layer) string { if v.IsNil() { continue } - ret = append(ret, fmt.Sprintf("%s:%v", t.Name, v)) + ret = append(ret, fmt.Sprintf("%s:%v", t.Name, reflect.Indirect(v))) } return fmt.Sprintf("&%s{%s}", t, strings.Join(ret, " ")) } diff --git a/test/packetimpact/testbench/layers_test.go b/test/packetimpact/testbench/layers_test.go index b39839625..8ffc26bf9 100644 --- a/test/packetimpact/testbench/layers_test.go +++ b/test/packetimpact/testbench/layers_test.go @@ -16,6 +16,8 @@ package testbench import "testing" +import "gvisor.dev/gvisor/pkg/tcpip" + func TestLayerMatch(t *testing.T) { var nilPayload *Payload noPayload := &Payload{} @@ -47,3 +49,96 @@ func TestLayerMatch(t *testing.T) { } } } + +func TestLayerStringFormat(t *testing.T) { + for _, tt := range []struct { + name string + l Layer + want string + }{ + { + name: "TCP", + l: &TCP{ + SrcPort: Uint16(34785), + DstPort: Uint16(47767), + SeqNum: Uint32(3452155723), + AckNum: Uint32(2596996163), + DataOffset: Uint8(5), + Flags: Uint8(20), + WindowSize: Uint16(64240), + Checksum: Uint16(0x2e2b), + }, + want: "&testbench.TCP{" + + "SrcPort:34785 " + + "DstPort:47767 " + + "SeqNum:3452155723 " + + "AckNum:2596996163 " + + "DataOffset:5 " + + "Flags:20 " + + "WindowSize:64240 " + + "Checksum:11819" + + "}", + }, + { + name: "UDP", + l: &UDP{ + SrcPort: Uint16(34785), + DstPort: Uint16(47767), + Length: Uint16(12), + }, + want: "&testbench.UDP{" + + "SrcPort:34785 " + + "DstPort:47767 " + + "Length:12" + + "}", + }, + { + name: "IPv4", + l: &IPv4{ + IHL: Uint8(5), + TOS: Uint8(0), + TotalLength: Uint16(44), + ID: Uint16(0), + Flags: Uint8(2), + FragmentOffset: Uint16(0), + TTL: Uint8(64), + Protocol: Uint8(6), + Checksum: Uint16(0x2e2b), + SrcAddr: Address(tcpip.Address([]byte{197, 34, 63, 10})), + DstAddr: Address(tcpip.Address([]byte{197, 34, 63, 20})), + }, + want: "&testbench.IPv4{" + + "IHL:5 " + + "TOS:0 " + + "TotalLength:44 " + + "ID:0 " + + "Flags:2 " + + "FragmentOffset:0 " + + "TTL:64 " + + "Protocol:6 " + + "Checksum:11819 " + + "SrcAddr:197.34.63.10 " + + "DstAddr:197.34.63.20" + + "}", + }, + { + name: "Ether", + l: &Ether{ + SrcAddr: LinkAddress(tcpip.LinkAddress([]byte{0x02, 0x42, 0xc5, 0x22, 0x3f, 0x0a})), + DstAddr: LinkAddress(tcpip.LinkAddress([]byte{0x02, 0x42, 0xc5, 0x22, 0x3f, 0x14})), + Type: NetworkProtocolNumber(4), + }, + want: "&testbench.Ether{" + + "SrcAddr:02:42:c5:22:3f:0a " + + "DstAddr:02:42:c5:22:3f:14 " + + "Type:4" + + "}", + }, + } { + t.Run(tt.name, func(t *testing.T) { + if got := tt.l.String(); got != tt.want { + t.Errorf("%s.String() = %s, want: %s", tt.name, got, tt.want) + } + }) + } +} diff --git a/test/syscalls/linux/BUILD b/test/syscalls/linux/BUILD index d0c431234..96ca39583 100644 --- a/test/syscalls/linux/BUILD +++ b/test/syscalls/linux/BUILD @@ -138,7 +138,6 @@ cc_library( hdrs = ["socket_netlink_route_util.h"], deps = [ ":socket_netlink_util", - "@com_google_absl//absl/types:optional", ], ) @@ -2026,6 +2025,8 @@ cc_binary( "//test/util:file_descriptor", "@com_google_absl//absl/strings", gtest, + ":ip_socket_test_util", + ":unix_domain_socket_test_util", "//test/util:temp_path", "//test/util:test_main", "//test/util:test_util", @@ -2802,13 +2803,13 @@ cc_binary( srcs = ["socket_netlink_route.cc"], linkstatic = 1, deps = [ + ":socket_netlink_route_util", ":socket_netlink_util", ":socket_test_util", "//test/util:capability_util", "//test/util:cleanup", "//test/util:file_descriptor", "@com_google_absl//absl/strings:str_format", - "@com_google_absl//absl/types:optional", gtest, "//test/util:test_main", "//test/util:test_util", diff --git a/test/syscalls/linux/aio.cc b/test/syscalls/linux/aio.cc index a33daff17..806d5729e 100644 --- a/test/syscalls/linux/aio.cc +++ b/test/syscalls/linux/aio.cc @@ -89,6 +89,7 @@ class AIOTest : public FileTest { FileTest::TearDown(); if (ctx_ != 0) { ASSERT_THAT(DestroyContext(), SyscallSucceeds()); + ctx_ = 0; } } @@ -188,14 +189,19 @@ TEST_F(AIOTest, BadWrite) { } TEST_F(AIOTest, ExitWithPendingIo) { - // Setup a context that is 5 entries deep. - ASSERT_THAT(SetupContext(5), SyscallSucceeds()); + // Setup a context that is 100 entries deep. + ASSERT_THAT(SetupContext(100), SyscallSucceeds()); struct iocb cb = CreateCallback(); struct iocb* cbs[] = {&cb}; // Submit a request but don't complete it to make it pending. - EXPECT_THAT(Submit(1, cbs), SyscallSucceeds()); + for (int i = 0; i < 100; ++i) { + EXPECT_THAT(Submit(1, cbs), SyscallSucceeds()); + } + + ASSERT_THAT(DestroyContext(), SyscallSucceeds()); + ctx_ = 0; } int Submitter(void* arg) { diff --git a/test/syscalls/linux/epoll.cc b/test/syscalls/linux/epoll.cc index a4f8f3cec..f57d38dc7 100644 --- a/test/syscalls/linux/epoll.cc +++ b/test/syscalls/linux/epoll.cc @@ -56,10 +56,6 @@ TEST(EpollTest, AllWritable) { struct epoll_event result[kFDsPerEpoll]; ASSERT_THAT(RetryEINTR(epoll_wait)(epollfd.get(), result, kFDsPerEpoll, -1), SyscallSucceedsWithValue(kFDsPerEpoll)); - // TODO(edahlgren): Why do some tests check epoll_event::data, and others - // don't? Does Linux actually guarantee that, in any of these test cases, - // epoll_wait will necessarily write out the epoll_events in the order that - // they were registered? for (int i = 0; i < kFDsPerEpoll; i++) { ASSERT_EQ(result[i].events, EPOLLOUT); } diff --git a/test/syscalls/linux/file_base.h b/test/syscalls/linux/file_base.h index 6f80bc97c..fb418e052 100644 --- a/test/syscalls/linux/file_base.h +++ b/test/syscalls/linux/file_base.h @@ -52,17 +52,6 @@ class FileTest : public ::testing::Test { test_file_fd_ = ASSERT_NO_ERRNO_AND_VALUE( Open(test_file_name_, O_CREAT | O_RDWR, S_IRUSR | S_IWUSR)); - // FIXME(edahlgren): enable when mknod syscall is supported. - // test_fifo_name_ = NewTempAbsPath(); - // ASSERT_THAT(mknod(test_fifo_name_.c_str()), S_IFIFO|0644, 0, - // SyscallSucceeds()); - // ASSERT_THAT(test_fifo_[1] = open(test_fifo_name_.c_str(), - // O_WRONLY), - // SyscallSucceeds()); - // ASSERT_THAT(test_fifo_[0] = open(test_fifo_name_.c_str(), - // O_RDONLY), - // SyscallSucceeds()); - ASSERT_THAT(pipe(test_pipe_), SyscallSucceeds()); ASSERT_THAT(fcntl(test_pipe_[0], F_SETFL, O_NONBLOCK), SyscallSucceeds()); } @@ -96,18 +85,12 @@ class FileTest : public ::testing::Test { CloseFile(); UnlinkFile(); ClosePipes(); - - // FIXME(edahlgren): enable when mknod syscall is supported. - // close(test_fifo_[0]); - // close(test_fifo_[1]); - // unlink(test_fifo_name_.c_str()); } + protected: std::string test_file_name_; - std::string test_fifo_name_; FileDescriptor test_file_fd_; - int test_fifo_[2]; int test_pipe_[2]; }; diff --git a/test/syscalls/linux/ip_socket_test_util.cc b/test/syscalls/linux/ip_socket_test_util.cc index bba022a41..d28dc0db6 100644 --- a/test/syscalls/linux/ip_socket_test_util.cc +++ b/test/syscalls/linux/ip_socket_test_util.cc @@ -177,17 +177,17 @@ SocketKind IPv6TCPUnboundSocket(int type) { PosixError IfAddrHelper::Load() { Release(); RETURN_ERROR_IF_SYSCALL_FAIL(getifaddrs(&ifaddr_)); - return PosixError(0); + return NoError(); } void IfAddrHelper::Release() { if (ifaddr_) { freeifaddrs(ifaddr_); + ifaddr_ = nullptr; } - ifaddr_ = nullptr; } -std::vector<std::string> IfAddrHelper::InterfaceList(int family) { +std::vector<std::string> IfAddrHelper::InterfaceList(int family) const { std::vector<std::string> names; for (auto ifa = ifaddr_; ifa != NULL; ifa = ifa->ifa_next) { if (ifa->ifa_addr == NULL || ifa->ifa_addr->sa_family != family) { @@ -198,7 +198,7 @@ std::vector<std::string> IfAddrHelper::InterfaceList(int family) { return names; } -sockaddr* IfAddrHelper::GetAddr(int family, std::string name) { +const sockaddr* IfAddrHelper::GetAddr(int family, std::string name) const { for (auto ifa = ifaddr_; ifa != NULL; ifa = ifa->ifa_next) { if (ifa->ifa_addr == NULL || ifa->ifa_addr->sa_family != family) { continue; @@ -210,7 +210,7 @@ sockaddr* IfAddrHelper::GetAddr(int family, std::string name) { return nullptr; } -PosixErrorOr<int> IfAddrHelper::GetIndex(std::string name) { +PosixErrorOr<int> IfAddrHelper::GetIndex(std::string name) const { return InterfaceIndex(name); } diff --git a/test/syscalls/linux/ip_socket_test_util.h b/test/syscalls/linux/ip_socket_test_util.h index 39fd6709d..9c3859fcd 100644 --- a/test/syscalls/linux/ip_socket_test_util.h +++ b/test/syscalls/linux/ip_socket_test_util.h @@ -110,10 +110,10 @@ class IfAddrHelper { PosixError Load(); void Release(); - std::vector<std::string> InterfaceList(int family); + std::vector<std::string> InterfaceList(int family) const; - struct sockaddr* GetAddr(int family, std::string name); - PosixErrorOr<int> GetIndex(std::string name); + const sockaddr* GetAddr(int family, std::string name) const; + PosixErrorOr<int> GetIndex(std::string name) const; private: struct ifaddrs* ifaddr_; diff --git a/test/syscalls/linux/proc.cc b/test/syscalls/linux/proc.cc index 5a70f6c3b..79a625ebc 100644 --- a/test/syscalls/linux/proc.cc +++ b/test/syscalls/linux/proc.cc @@ -994,7 +994,7 @@ constexpr uint64_t kMappingSize = 100 << 20; // Tolerance on RSS comparisons to account for background thread mappings, // reclaimed pages, newly faulted pages, etc. -constexpr uint64_t kRSSTolerance = 5 << 20; +constexpr uint64_t kRSSTolerance = 10 << 20; // Capture RSS before and after an anonymous mapping with passed prot. void MapPopulateRSS(int prot, uint64_t* before, uint64_t* after) { @@ -1326,8 +1326,6 @@ TEST(ProcPidSymlink, SubprocessRunning) { SyscallSucceedsWithValue(sizeof(buf))); } -// FIXME(gvisor.dev/issue/164): Inconsistent behavior between gVisor and linux -// on proc files. TEST(ProcPidSymlink, SubprocessZombied) { ASSERT_NO_ERRNO(SetCapability(CAP_DAC_OVERRIDE, false)); ASSERT_NO_ERRNO(SetCapability(CAP_DAC_READ_SEARCH, false)); @@ -1337,7 +1335,7 @@ TEST(ProcPidSymlink, SubprocessZombied) { int want = EACCES; if (!IsRunningOnGvisor()) { auto version = ASSERT_NO_ERRNO_AND_VALUE(GetKernelVersion()); - if (version.major == 4 && version.minor > 3) { + if (version.major > 4 || (version.major == 4 && version.minor > 3)) { want = ENOENT; } } @@ -1350,30 +1348,25 @@ TEST(ProcPidSymlink, SubprocessZombied) { SyscallFailsWithErrno(want)); } - // FIXME(gvisor.dev/issue/164): Inconsistent behavior between gVisor and linux - // on proc files. + // FIXME(gvisor.dev/issue/164): Inconsistent behavior between linux on proc + // files. // // ~4.3: Syscall fails with EACCES. - // 4.17 & gVisor: Syscall succeeds and returns 1. + // 4.17: Syscall succeeds and returns 1. // - // EXPECT_THAT(ReadlinkWhileZombied("ns/pid", buf, sizeof(buf)), - // SyscallFailsWithErrno(EACCES)); + if (!IsRunningOnGvisor()) { + return; + } - // FIXME(gvisor.dev/issue/164): Inconsistent behavior between gVisor and linux - // on proc files. - // - // ~4.3: Syscall fails with EACCES. - // 4.17 & gVisor: Syscall succeeds and returns 1. - // - // EXPECT_THAT(ReadlinkWhileZombied("ns/user", buf, sizeof(buf)), - // SyscallFailsWithErrno(EACCES)); + EXPECT_THAT(ReadlinkWhileZombied("ns/pid", buf, sizeof(buf)), + SyscallFailsWithErrno(want)); + + EXPECT_THAT(ReadlinkWhileZombied("ns/user", buf, sizeof(buf)), + SyscallFailsWithErrno(want)); } // Test whether /proc/PID/ symlinks can be read for an exited process. TEST(ProcPidSymlink, SubprocessExited) { - // FIXME(gvisor.dev/issue/164): These all succeed on gVisor. - SKIP_IF(IsRunningOnGvisor()); - char buf[1]; EXPECT_THAT(ReadlinkWhileExited("exe", buf, sizeof(buf)), diff --git a/test/syscalls/linux/pwrite64.cc b/test/syscalls/linux/pwrite64.cc index b48fe540d..c2f72e010 100644 --- a/test/syscalls/linux/pwrite64.cc +++ b/test/syscalls/linux/pwrite64.cc @@ -27,14 +27,7 @@ namespace testing { namespace { -// This test is currently very rudimentary. -// -// TODO(edahlgren): -// * bad buffer states (EFAULT). -// * bad fds (wrong permission, wrong type of file, EBADF). -// * check offset is not incremented. -// * check for EOF. -// * writing to pipes, symlinks, special files. +// TODO(gvisor.dev/issue/2370): This test is currently very rudimentary. class Pwrite64 : public ::testing::Test { void SetUp() override { name_ = NewTempAbsPath(); diff --git a/test/syscalls/linux/sendfile_socket.cc b/test/syscalls/linux/sendfile_socket.cc index e94672679..c101fe9d2 100644 --- a/test/syscalls/linux/sendfile_socket.cc +++ b/test/syscalls/linux/sendfile_socket.cc @@ -23,6 +23,7 @@ #include "gtest/gtest.h" #include "absl/strings/string_view.h" +#include "test/syscalls/linux/ip_socket_test_util.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/util/file_descriptor.h" #include "test/util/temp_path.h" @@ -35,61 +36,39 @@ namespace { class SendFileTest : public ::testing::TestWithParam<int> { protected: - PosixErrorOr<std::tuple<int, int>> Sockets() { + PosixErrorOr<std::unique_ptr<SocketPair>> Sockets(int type) { // Bind a server socket. int family = GetParam(); - struct sockaddr server_addr = {}; switch (family) { case AF_INET: { - struct sockaddr_in* server_addr_in = - reinterpret_cast<struct sockaddr_in*>(&server_addr); - server_addr_in->sin_family = family; - server_addr_in->sin_addr.s_addr = INADDR_ANY; - break; + if (type == SOCK_STREAM) { + return SocketPairKind{ + "TCP", AF_INET, type, 0, + TCPAcceptBindSocketPairCreator(AF_INET, type, 0, false)} + .Create(); + } else { + return SocketPairKind{ + "UDP", AF_INET, type, 0, + UDPBidirectionalBindSocketPairCreator(AF_INET, type, 0, false)} + .Create(); + } } case AF_UNIX: { - struct sockaddr_un* server_addr_un = - reinterpret_cast<struct sockaddr_un*>(&server_addr); - server_addr_un->sun_family = family; - server_addr_un->sun_path[0] = '\0'; - break; + if (type == SOCK_STREAM) { + return SocketPairKind{ + "UNIX", AF_UNIX, type, 0, + FilesystemAcceptBindSocketPairCreator(AF_UNIX, type, 0)} + .Create(); + } else { + return SocketPairKind{ + "UNIX", AF_UNIX, type, 0, + FilesystemBidirectionalBindSocketPairCreator(AF_UNIX, type, 0)} + .Create(); + } } default: return PosixError(EINVAL); } - int server = socket(family, SOCK_STREAM, 0); - if (bind(server, &server_addr, sizeof(server_addr)) < 0) { - return PosixError(errno); - } - if (listen(server, 1) < 0) { - close(server); - return PosixError(errno); - } - - // Fetch the address; both are anonymous. - socklen_t length = sizeof(server_addr); - if (getsockname(server, &server_addr, &length) < 0) { - close(server); - return PosixError(errno); - } - - // Connect the client. - int client = socket(family, SOCK_STREAM, 0); - if (connect(client, &server_addr, length) < 0) { - close(server); - close(client); - return PosixError(errno); - } - - // Accept on the server. - int server_client = accept(server, nullptr, 0); - if (server_client < 0) { - close(server); - close(client); - return PosixError(errno); - } - close(server); - return std::make_tuple(client, server_client); } }; @@ -106,9 +85,7 @@ TEST_P(SendFileTest, SendMultiple) { const TempPath out_file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); // Create sockets. - std::tuple<int, int> fds = ASSERT_NO_ERRNO_AND_VALUE(Sockets()); - const FileDescriptor server(std::get<0>(fds)); - FileDescriptor client(std::get<1>(fds)); // non-const, reset is used. + auto socks = ASSERT_NO_ERRNO_AND_VALUE(Sockets(SOCK_STREAM)); // Thread that reads data from socket and dumps to a file. ScopedThread th([&] { @@ -118,7 +95,7 @@ TEST_P(SendFileTest, SendMultiple) { // Read until socket is closed. char buf[10240]; for (int cnt = 0;; cnt++) { - int r = RetryEINTR(read)(server.get(), buf, sizeof(buf)); + int r = RetryEINTR(read)(socks->first_fd(), buf, sizeof(buf)); // We cannot afford to save on every read() call. if (cnt % 1000 == 0) { ASSERT_THAT(r, SyscallSucceeds()); @@ -152,7 +129,7 @@ TEST_P(SendFileTest, SendMultiple) { << ", remain=" << remain << std::endl; // Send data and verify that sendfile returns the correct value. - int res = sendfile(client.get(), inf.get(), nullptr, remain); + int res = sendfile(socks->second_fd(), inf.get(), nullptr, remain); // We cannot afford to save on every sendfile() call. if (cnt % 120 == 0) { MaybeSave(); @@ -169,7 +146,7 @@ TEST_P(SendFileTest, SendMultiple) { } // Close socket to stop thread. - client.reset(); + close(socks->release_second_fd()); th.Join(); // Verify that the output file has the correct data. @@ -183,9 +160,7 @@ TEST_P(SendFileTest, SendMultiple) { TEST_P(SendFileTest, Shutdown) { // Create a socket. - std::tuple<int, int> fds = ASSERT_NO_ERRNO_AND_VALUE(Sockets()); - const FileDescriptor client(std::get<0>(fds)); - FileDescriptor server(std::get<1>(fds)); // non-const, reset below. + auto socks = ASSERT_NO_ERRNO_AND_VALUE(Sockets(SOCK_STREAM)); // If this is a TCP socket, then turn off linger. if (GetParam() == AF_INET) { @@ -193,7 +168,7 @@ TEST_P(SendFileTest, Shutdown) { sl.l_onoff = 1; sl.l_linger = 0; ASSERT_THAT( - setsockopt(server.get(), SOL_SOCKET, SO_LINGER, &sl, sizeof(sl)), + setsockopt(socks->first_fd(), SOL_SOCKET, SO_LINGER, &sl, sizeof(sl)), SyscallSucceeds()); } @@ -212,12 +187,12 @@ TEST_P(SendFileTest, Shutdown) { ScopedThread t([&]() { size_t done = 0; while (done < data.size()) { - int n = RetryEINTR(read)(server.get(), data.data(), data.size()); + int n = RetryEINTR(read)(socks->first_fd(), data.data(), data.size()); ASSERT_THAT(n, SyscallSucceeds()); done += n; } // Close the server side socket. - server.reset(); + close(socks->release_first_fd()); }); // Continuously stream from the file to the socket. Note we do not assert @@ -225,7 +200,7 @@ TEST_P(SendFileTest, Shutdown) { // data is written. Eventually, we should get a connection reset error. while (1) { off_t offset = 0; // Always read from the start. - int n = sendfile(client.get(), inf.get(), &offset, data.size()); + int n = sendfile(socks->second_fd(), inf.get(), &offset, data.size()); EXPECT_THAT(n, AnyOf(SyscallFailsWithErrno(ECONNRESET), SyscallFailsWithErrno(EPIPE), SyscallSucceeds())); if (n <= 0) { @@ -234,6 +209,20 @@ TEST_P(SendFileTest, Shutdown) { } } +TEST_P(SendFileTest, SendpageFromEmptyFileToUDP) { + auto socks = ASSERT_NO_ERRNO_AND_VALUE(Sockets(SOCK_DGRAM)); + + TempPath file = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateFile()); + const FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(file.path(), O_RDWR)); + + // The value to the count argument has to be so that it is impossible to + // allocate a buffer of this size. In Linux, sendfile transfer at most + // 0x7ffff000 (MAX_RW_COUNT) bytes. + EXPECT_THAT(sendfile(socks->first_fd(), fd.get(), 0x0, 0x8000000000004), + SyscallSucceedsWithValue(0)); +} + INSTANTIATE_TEST_SUITE_P(AddressFamily, SendFileTest, ::testing::Values(AF_UNIX, AF_INET)); diff --git a/test/syscalls/linux/socket_ipv4_udp_unbound_external_networking.cc b/test/syscalls/linux/socket_ipv4_udp_unbound_external_networking.cc index 40e673625..d690d9564 100644 --- a/test/syscalls/linux/socket_ipv4_udp_unbound_external_networking.cc +++ b/test/syscalls/linux/socket_ipv4_udp_unbound_external_networking.cc @@ -45,37 +45,31 @@ void IPv4UDPUnboundExternalNetworkingSocketTest::SetUp() { got_if_infos_ = false; // Get interface list. - std::vector<std::string> if_names; ASSERT_NO_ERRNO(if_helper_.Load()); - if_names = if_helper_.InterfaceList(AF_INET); + std::vector<std::string> if_names = if_helper_.InterfaceList(AF_INET); if (if_names.size() != 2) { return; } // Figure out which interface is where. - int lo = 0, eth = 1; - if (if_names[lo] != "lo") { - lo = 1; - eth = 0; - } - - if (if_names[lo] != "lo") { - return; - } - - lo_if_idx_ = ASSERT_NO_ERRNO_AND_VALUE(if_helper_.GetIndex(if_names[lo])); - lo_if_addr_ = if_helper_.GetAddr(AF_INET, if_names[lo]); - if (lo_if_addr_ == nullptr) { + std::string lo = if_names[0]; + std::string eth = if_names[1]; + if (lo != "lo") std::swap(lo, eth); + if (lo != "lo") return; + + lo_if_idx_ = ASSERT_NO_ERRNO_AND_VALUE(if_helper_.GetIndex(lo)); + auto lo_if_addr = if_helper_.GetAddr(AF_INET, lo); + if (lo_if_addr == nullptr) { return; } - lo_if_sin_addr_ = reinterpret_cast<sockaddr_in*>(lo_if_addr_)->sin_addr; + lo_if_addr_ = *reinterpret_cast<const sockaddr_in*>(lo_if_addr); - eth_if_idx_ = ASSERT_NO_ERRNO_AND_VALUE(if_helper_.GetIndex(if_names[eth])); - eth_if_addr_ = if_helper_.GetAddr(AF_INET, if_names[eth]); - if (eth_if_addr_ == nullptr) { + eth_if_idx_ = ASSERT_NO_ERRNO_AND_VALUE(if_helper_.GetIndex(eth)); + auto eth_if_addr = if_helper_.GetAddr(AF_INET, eth); + if (eth_if_addr == nullptr) { return; } - eth_if_sin_addr_ = reinterpret_cast<sockaddr_in*>(eth_if_addr_)->sin_addr; + eth_if_addr_ = *reinterpret_cast<const sockaddr_in*>(eth_if_addr); got_if_infos_ = true; } @@ -242,7 +236,7 @@ TEST_P(IPv4UDPUnboundExternalNetworkingSocketTest, // Bind the non-receiving socket to the unicast ethernet address. auto norecv_addr = rcv1_addr; reinterpret_cast<sockaddr_in*>(&norecv_addr.addr)->sin_addr = - eth_if_sin_addr_; + eth_if_addr_.sin_addr; ASSERT_THAT(bind(norcv->get(), reinterpret_cast<sockaddr*>(&norecv_addr.addr), norecv_addr.addr_len), SyscallSucceedsWithValue(0)); @@ -1028,7 +1022,7 @@ TEST_P(IPv4UDPUnboundExternalNetworkingSocketTest, auto sender = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); ip_mreqn iface = {}; iface.imr_ifindex = lo_if_idx_; - iface.imr_address = eth_if_sin_addr_; + iface.imr_address = eth_if_addr_.sin_addr; ASSERT_THAT(setsockopt(sender->get(), IPPROTO_IP, IP_MULTICAST_IF, &iface, sizeof(iface)), SyscallSucceeds()); @@ -1058,7 +1052,7 @@ TEST_P(IPv4UDPUnboundExternalNetworkingSocketTest, SKIP_IF(IsRunningOnGvisor()); // Verify the received source address. - EXPECT_EQ(eth_if_sin_addr_.s_addr, src_addr_in->sin_addr.s_addr); + EXPECT_EQ(eth_if_addr_.sin_addr.s_addr, src_addr_in->sin_addr.s_addr); } // Check that when we are bound to one interface we can set IP_MULTICAST_IF to @@ -1075,7 +1069,8 @@ TEST_P(IPv4UDPUnboundExternalNetworkingSocketTest, // Create sender and bind to eth interface. auto sender = ASSERT_NO_ERRNO_AND_VALUE(NewSocket()); - ASSERT_THAT(bind(sender->get(), eth_if_addr_, sizeof(sockaddr_in)), + ASSERT_THAT(bind(sender->get(), reinterpret_cast<sockaddr*>(ð_if_addr_), + sizeof(eth_if_addr_)), SyscallSucceeds()); // Run through all possible combinations of index and address for @@ -1085,9 +1080,9 @@ TEST_P(IPv4UDPUnboundExternalNetworkingSocketTest, struct in_addr imr_address; } test_data[] = { {lo_if_idx_, {}}, - {0, lo_if_sin_addr_}, - {lo_if_idx_, lo_if_sin_addr_}, - {lo_if_idx_, eth_if_sin_addr_}, + {0, lo_if_addr_.sin_addr}, + {lo_if_idx_, lo_if_addr_.sin_addr}, + {lo_if_idx_, eth_if_addr_.sin_addr}, }; for (auto t : test_data) { ip_mreqn iface = {}; diff --git a/test/syscalls/linux/socket_ipv4_udp_unbound_external_networking.h b/test/syscalls/linux/socket_ipv4_udp_unbound_external_networking.h index bec2e96ee..10b90b1e0 100644 --- a/test/syscalls/linux/socket_ipv4_udp_unbound_external_networking.h +++ b/test/syscalls/linux/socket_ipv4_udp_unbound_external_networking.h @@ -36,10 +36,8 @@ class IPv4UDPUnboundExternalNetworkingSocketTest : public SimpleSocketTest { // Interface infos. int lo_if_idx_; int eth_if_idx_; - sockaddr* lo_if_addr_; - sockaddr* eth_if_addr_; - in_addr lo_if_sin_addr_; - in_addr eth_if_sin_addr_; + sockaddr_in lo_if_addr_; + sockaddr_in eth_if_addr_; }; } // namespace testing diff --git a/test/syscalls/linux/socket_netlink_route.cc b/test/syscalls/linux/socket_netlink_route.cc index 2efb96bc3..fbe61c5a0 100644 --- a/test/syscalls/linux/socket_netlink_route.cc +++ b/test/syscalls/linux/socket_netlink_route.cc @@ -26,7 +26,7 @@ #include "gtest/gtest.h" #include "absl/strings/str_format.h" -#include "absl/types/optional.h" +#include "test/syscalls/linux/socket_netlink_route_util.h" #include "test/syscalls/linux/socket_netlink_util.h" #include "test/syscalls/linux/socket_test_util.h" #include "test/util/capability_util.h" @@ -118,24 +118,6 @@ void CheckGetLinkResponse(const struct nlmsghdr* hdr, int seq, int port) { // TODO(mpratt): Check ifinfomsg contents and following attrs. } -PosixError DumpLinks( - const FileDescriptor& fd, uint32_t seq, - const std::function<void(const struct nlmsghdr* hdr)>& fn) { - struct request { - struct nlmsghdr hdr; - struct ifinfomsg ifm; - }; - - struct request req = {}; - req.hdr.nlmsg_len = sizeof(req); - req.hdr.nlmsg_type = RTM_GETLINK; - req.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP; - req.hdr.nlmsg_seq = seq; - req.ifm.ifi_family = AF_UNSPEC; - - return NetlinkRequestResponse(fd, &req, sizeof(req), fn, false); -} - TEST(NetlinkRouteTest, GetLinkDump) { FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); @@ -161,37 +143,6 @@ TEST(NetlinkRouteTest, GetLinkDump) { EXPECT_TRUE(loopbackFound); } -struct Link { - int index; - std::string name; -}; - -PosixErrorOr<absl::optional<Link>> FindLoopbackLink() { - ASSIGN_OR_RETURN_ERRNO(FileDescriptor fd, NetlinkBoundSocket(NETLINK_ROUTE)); - - absl::optional<Link> link; - RETURN_IF_ERRNO(DumpLinks(fd, kSeq, [&](const struct nlmsghdr* hdr) { - if (hdr->nlmsg_type != RTM_NEWLINK || - hdr->nlmsg_len < NLMSG_SPACE(sizeof(struct ifinfomsg))) { - return; - } - const struct ifinfomsg* msg = - reinterpret_cast<const struct ifinfomsg*>(NLMSG_DATA(hdr)); - if (msg->ifi_type == ARPHRD_LOOPBACK) { - const auto* rta = FindRtAttr(hdr, msg, IFLA_IFNAME); - if (rta == nullptr) { - // Ignore links that do not have a name. - return; - } - - link = Link(); - link->index = msg->ifi_index; - link->name = std::string(reinterpret_cast<const char*>(RTA_DATA(rta))); - } - })); - return link; -} - // CheckLinkMsg checks a netlink message against an expected link. void CheckLinkMsg(const struct nlmsghdr* hdr, const Link& link) { ASSERT_THAT(hdr->nlmsg_type, Eq(RTM_NEWLINK)); @@ -209,9 +160,7 @@ void CheckLinkMsg(const struct nlmsghdr* hdr, const Link& link) { } TEST(NetlinkRouteTest, GetLinkByIndex) { - absl::optional<Link> loopback_link = - ASSERT_NO_ERRNO_AND_VALUE(FindLoopbackLink()); - ASSERT_TRUE(loopback_link.has_value()); + Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink()); FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); @@ -227,13 +176,13 @@ TEST(NetlinkRouteTest, GetLinkByIndex) { req.hdr.nlmsg_flags = NLM_F_REQUEST; req.hdr.nlmsg_seq = kSeq; req.ifm.ifi_family = AF_UNSPEC; - req.ifm.ifi_index = loopback_link->index; + req.ifm.ifi_index = loopback_link.index; bool found = false; ASSERT_NO_ERRNO(NetlinkRequestResponse( fd, &req, sizeof(req), [&](const struct nlmsghdr* hdr) { - CheckLinkMsg(hdr, *loopback_link); + CheckLinkMsg(hdr, loopback_link); found = true; }, false)); @@ -241,9 +190,7 @@ TEST(NetlinkRouteTest, GetLinkByIndex) { } TEST(NetlinkRouteTest, GetLinkByName) { - absl::optional<Link> loopback_link = - ASSERT_NO_ERRNO_AND_VALUE(FindLoopbackLink()); - ASSERT_TRUE(loopback_link.has_value()); + Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink()); FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); @@ -262,8 +209,8 @@ TEST(NetlinkRouteTest, GetLinkByName) { req.hdr.nlmsg_seq = kSeq; req.ifm.ifi_family = AF_UNSPEC; req.rtattr.rta_type = IFLA_IFNAME; - req.rtattr.rta_len = RTA_LENGTH(loopback_link->name.size() + 1); - strncpy(req.ifname, loopback_link->name.c_str(), sizeof(req.ifname)); + req.rtattr.rta_len = RTA_LENGTH(loopback_link.name.size() + 1); + strncpy(req.ifname, loopback_link.name.c_str(), sizeof(req.ifname)); req.hdr.nlmsg_len = NLMSG_LENGTH(sizeof(req.ifm)) + NLMSG_ALIGN(req.rtattr.rta_len); @@ -271,7 +218,7 @@ TEST(NetlinkRouteTest, GetLinkByName) { ASSERT_NO_ERRNO(NetlinkRequestResponse( fd, &req, sizeof(req), [&](const struct nlmsghdr* hdr) { - CheckLinkMsg(hdr, *loopback_link); + CheckLinkMsg(hdr, loopback_link); found = true; }, false)); @@ -523,9 +470,7 @@ TEST(NetlinkRouteTest, LookupAll) { TEST(NetlinkRouteTest, AddAddr) { SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN))); - absl::optional<Link> loopback_link = - ASSERT_NO_ERRNO_AND_VALUE(FindLoopbackLink()); - ASSERT_TRUE(loopback_link.has_value()); + Link loopback_link = ASSERT_NO_ERRNO_AND_VALUE(LoopbackLink()); FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(NetlinkBoundSocket(NETLINK_ROUTE)); @@ -545,7 +490,7 @@ TEST(NetlinkRouteTest, AddAddr) { req.ifa.ifa_prefixlen = 24; req.ifa.ifa_flags = 0; req.ifa.ifa_scope = 0; - req.ifa.ifa_index = loopback_link->index; + req.ifa.ifa_index = loopback_link.index; req.rtattr.rta_type = IFA_LOCAL; req.rtattr.rta_len = RTA_LENGTH(sizeof(req.addr)); inet_pton(AF_INET, "10.0.0.1", &req.addr); diff --git a/test/syscalls/linux/socket_netlink_route_util.cc b/test/syscalls/linux/socket_netlink_route_util.cc index 53eb3b6b2..bde1dbb4d 100644 --- a/test/syscalls/linux/socket_netlink_route_util.cc +++ b/test/syscalls/linux/socket_netlink_route_util.cc @@ -18,7 +18,6 @@ #include <linux/netlink.h> #include <linux/rtnetlink.h> -#include "absl/types/optional.h" #include "test/syscalls/linux/socket_netlink_util.h" namespace gvisor { @@ -73,14 +72,14 @@ PosixErrorOr<std::vector<Link>> DumpLinks() { return links; } -PosixErrorOr<absl::optional<Link>> FindLoopbackLink() { +PosixErrorOr<Link> LoopbackLink() { ASSIGN_OR_RETURN_ERRNO(auto links, DumpLinks()); for (const auto& link : links) { if (link.type == ARPHRD_LOOPBACK) { - return absl::optional<Link>(link); + return link; } } - return absl::optional<Link>(); + return PosixError(ENOENT, "loopback link not found"); } PosixError LinkAddLocalAddr(int index, int family, int prefixlen, diff --git a/test/syscalls/linux/socket_netlink_route_util.h b/test/syscalls/linux/socket_netlink_route_util.h index 2c018e487..149c4a7f6 100644 --- a/test/syscalls/linux/socket_netlink_route_util.h +++ b/test/syscalls/linux/socket_netlink_route_util.h @@ -20,7 +20,6 @@ #include <vector> -#include "absl/types/optional.h" #include "test/syscalls/linux/socket_netlink_util.h" namespace gvisor { @@ -37,7 +36,8 @@ PosixError DumpLinks(const FileDescriptor& fd, uint32_t seq, PosixErrorOr<std::vector<Link>> DumpLinks(); -PosixErrorOr<absl::optional<Link>> FindLoopbackLink(); +// Returns the loopback link on the system. ENOENT if not found. +PosixErrorOr<Link> LoopbackLink(); // LinkAddLocalAddr sets IFA_LOCAL attribute on the interface. PosixError LinkAddLocalAddr(int index, int family, int prefixlen, diff --git a/test/syscalls/linux/tuntap.cc b/test/syscalls/linux/tuntap.cc index 53ad2dda3..6195b11e1 100644 --- a/test/syscalls/linux/tuntap.cc +++ b/test/syscalls/linux/tuntap.cc @@ -56,14 +56,14 @@ PosixErrorOr<std::set<std::string>> DumpLinkNames() { return names; } -PosixErrorOr<absl::optional<Link>> GetLinkByName(const std::string& name) { +PosixErrorOr<Link> GetLinkByName(const std::string& name) { ASSIGN_OR_RETURN_ERRNO(auto links, DumpLinks()); for (const auto& link : links) { if (link.name == name) { - return absl::optional<Link>(link); + return link; } } - return absl::optional<Link>(); + return PosixError(ENOENT, "interface not found"); } struct pihdr { @@ -242,7 +242,7 @@ TEST_F(TuntapTest, InvalidReadWrite) { TEST_F(TuntapTest, WriteToDownDevice) { SKIP_IF(!ASSERT_NO_ERRNO_AND_VALUE(HaveCapability(CAP_NET_ADMIN))); - // FIXME: gVisor always creates enabled/up'd interfaces. + // FIXME(b/110961832): gVisor always creates enabled/up'd interfaces. SKIP_IF(IsRunningOnGvisor()); FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(kDevNetTun, O_RDWR)); @@ -268,23 +268,21 @@ PosixErrorOr<FileDescriptor> OpenAndAttachTap( return PosixError(errno); } - ASSIGN_OR_RETURN_ERRNO(absl::optional<Link> link, GetLinkByName(dev_name)); - if (!link.has_value()) { - return PosixError(ENOENT, "no link"); - } + ASSIGN_OR_RETURN_ERRNO(auto link, GetLinkByName(dev_name)); // Interface setup. struct in_addr addr; inet_pton(AF_INET, dev_ipv4_addr.c_str(), &addr); - EXPECT_NO_ERRNO(LinkAddLocalAddr(link->index, AF_INET, /*prefixlen=*/24, - &addr, sizeof(addr))); + EXPECT_NO_ERRNO(LinkAddLocalAddr(link.index, AF_INET, /*prefixlen=*/24, &addr, + sizeof(addr))); if (!IsRunningOnGvisor()) { - // FIXME: gVisor doesn't support setting MAC address on interfaces yet. - RETURN_IF_ERRNO(LinkSetMacAddr(link->index, kMacA, sizeof(kMacA))); + // FIXME(b/110961832): gVisor doesn't support setting MAC address on + // interfaces yet. + RETURN_IF_ERRNO(LinkSetMacAddr(link.index, kMacA, sizeof(kMacA))); - // FIXME: gVisor always creates enabled/up'd interfaces. - RETURN_IF_ERRNO(LinkChangeFlags(link->index, IFF_UP, IFF_UP)); + // FIXME(b/110961832): gVisor always creates enabled/up'd interfaces. + RETURN_IF_ERRNO(LinkChangeFlags(link.index, IFF_UP, IFF_UP)); } return fd; diff --git a/test/syscalls/linux/utimes.cc b/test/syscalls/linux/utimes.cc index 3a927a430..22e6d1a85 100644 --- a/test/syscalls/linux/utimes.cc +++ b/test/syscalls/linux/utimes.cc @@ -34,17 +34,10 @@ namespace testing { namespace { -// TODO(b/36516566): utimes(nullptr) does not pick the "now" time in the -// application's time domain, so when asserting that times are within a window, -// we expand the window to allow for differences between the time domains. -constexpr absl::Duration kClockSlack = absl::Milliseconds(100); - // TimeBoxed runs fn, setting before and after to (coarse realtime) times // guaranteed* to come before and after fn started and completed, respectively. // // fn may be called more than once if the clock is adjusted. -// -// * See the comment on kClockSlack. gVisor breaks this guarantee. void TimeBoxed(absl::Time* before, absl::Time* after, std::function<void()> const& fn) { do { @@ -69,12 +62,6 @@ void TimeBoxed(absl::Time* before, absl::Time* after, // which could lead to test failures, but that is very unlikely to happen. continue; } - - if (IsRunningOnGvisor()) { - // See comment on kClockSlack. - *before -= kClockSlack; - *after += kClockSlack; - } } while (*after < *before); } @@ -235,10 +222,7 @@ void TestUtimensat(int dirFd, std::string const& path) { EXPECT_GE(mtime3, before); EXPECT_LE(mtime3, after); - if (!IsRunningOnGvisor()) { - // FIXME(b/36516566): Gofers set atime and mtime to different "now" times. - EXPECT_EQ(atime3, mtime3); - } + EXPECT_EQ(atime3, mtime3); } TEST(UtimensatTest, OnAbsPath) { diff --git a/test/syscalls/linux/write.cc b/test/syscalls/linux/write.cc index 9b219cfd6..39b5b2f56 100644 --- a/test/syscalls/linux/write.cc +++ b/test/syscalls/linux/write.cc @@ -31,14 +31,8 @@ namespace gvisor { namespace testing { namespace { -// This test is currently very rudimentary. -// -// TODO(edahlgren): -// * bad buffer states (EFAULT). -// * bad fds (wrong permission, wrong type of file, EBADF). -// * check offset is incremented. -// * check for EOF. -// * writing to pipes, symlinks, special files. + +// TODO(gvisor.dev/issue/2370): This test is currently very rudimentary. class WriteTest : public ::testing::Test { public: ssize_t WriteBytes(int fd, int bytes) { diff --git a/tools/bazeldefs/platforms.bzl b/tools/bazeldefs/platforms.bzl index 92b0b5fc0..132040c20 100644 --- a/tools/bazeldefs/platforms.bzl +++ b/tools/bazeldefs/platforms.bzl @@ -2,15 +2,10 @@ # Platform to associated tags. platforms = { - "ptrace": [ - # TODO(b/120560048): Make the tests run without this tag. - "no-sandbox", - ], + "ptrace": [], "kvm": [ "manual", "local", - # TODO(b/120560048): Make the tests run without this tag. - "no-sandbox", ], } diff --git a/tools/go_generics/defs.bzl b/tools/go_generics/defs.bzl index c5be52ecd..8c9995fd4 100644 --- a/tools/go_generics/defs.bzl +++ b/tools/go_generics/defs.bzl @@ -105,7 +105,6 @@ def _go_template_instance_impl(ctx): executable = ctx.executable._tool, ) - # TODO: How can we get the dependencies out? return struct( files = depset([output]), ) |