diff options
Diffstat (limited to 'pkg/sentry/fsimpl')
31 files changed, 367 insertions, 130 deletions
diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go index e03a0c665..abd4f24e7 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go @@ -28,6 +28,9 @@ import ( "gvisor.dev/gvisor/pkg/sync" ) +// Name is the default filesystem name. +const Name = "devtmpfs" + // FilesystemType implements vfs.FilesystemType. type FilesystemType struct { initOnce sync.Once @@ -107,6 +110,7 @@ func (a *Accessor) wrapContext(ctx context.Context) *accessorContext { func (ac *accessorContext) Value(key interface{}) interface{} { switch key { case vfs.CtxMountNamespace: + ac.a.mntns.IncRef() return ac.a.mntns case vfs.CtxRoot: ac.a.root.IncRef() diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go index 73308a2b5..b6d52c015 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go @@ -29,7 +29,10 @@ func TestDevtmpfs(t *testing.T) { ctx := contexttest.Context(t) creds := auth.CredentialsFromContext(ctx) - vfsObj := vfs.New() + vfsObj := &vfs.VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + t.Fatalf("VFS init: %v", err) + } // Register tmpfs just so that we can have a root filesystem that isn't // devtmpfs. vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ diff --git a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go index 2015a8871..89caee3df 100644 --- a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go +++ b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go @@ -52,7 +52,10 @@ func setUp(b *testing.B, imagePath string) (context.Context, *vfs.VirtualFilesys creds := auth.CredentialsFromContext(ctx) // Create VFS. - vfsObj := vfs.New() + vfsObj := &vfs.VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + return nil, nil, nil, nil, err + } vfsObj.MustRegisterFilesystemType("extfs", ext.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) diff --git a/pkg/sentry/fsimpl/ext/directory.go b/pkg/sentry/fsimpl/ext/directory.go index ebb72b75e..bd6ede995 100644 --- a/pkg/sentry/fsimpl/ext/directory.go +++ b/pkg/sentry/fsimpl/ext/directory.go @@ -188,14 +188,14 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba childType = fs.ToInodeType(childInode.diskInode.Mode().FileType()) } - if !cb.Handle(vfs.Dirent{ + if err := cb.Handle(vfs.Dirent{ Name: child.diskDirent.FileName(), Type: fs.ToDirentType(childType), Ino: uint64(child.diskDirent.Inode()), NextOff: fd.off + 1, - }) { + }); err != nil { dir.childList.InsertBefore(child, fd.iter) - return nil + return err } fd.off++ } diff --git a/pkg/sentry/fsimpl/ext/ext_test.go b/pkg/sentry/fsimpl/ext/ext_test.go index 05f992826..29bb73765 100644 --- a/pkg/sentry/fsimpl/ext/ext_test.go +++ b/pkg/sentry/fsimpl/ext/ext_test.go @@ -65,7 +65,10 @@ func setUp(t *testing.T, imagePath string) (context.Context, *vfs.VirtualFilesys creds := auth.CredentialsFromContext(ctx) // Create VFS. - vfsObj := vfs.New() + vfsObj := &vfs.VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + t.Fatalf("VFS init: %v", err) + } vfsObj.MustRegisterFilesystemType("extfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) @@ -496,9 +499,9 @@ func newIterDirentCb() *iterDirentsCb { } // Handle implements vfs.IterDirentsCallback.Handle. -func (cb *iterDirentsCb) Handle(dirent vfs.Dirent) bool { +func (cb *iterDirentsCb) Handle(dirent vfs.Dirent) error { cb.dirents = append(cb.dirents, dirent) - return true + return nil } // TestIterDirents tests the FileDescriptionImpl.IterDirents functionality. diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index 6d4ebc2bf..5dbfc6250 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -65,8 +65,8 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba } for fd.off < int64(len(fd.dirents)) { - if !cb.Handle(fd.dirents[fd.off]) { - return nil + if err := cb.Handle(fd.dirents[fd.off]); err != nil { + return err } fd.off++ } diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 138adb9f7..5cfb0dc4c 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -400,6 +400,7 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b } vfsObj := rp.VirtualFilesystem() mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() parent.dirMu.Lock() defer parent.dirMu.Unlock() childVFSD := parent.vfsd.Child(name) @@ -934,7 +935,9 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if oldParent == newParent && oldName == newName { return nil } - if err := vfsObj.PrepareRenameDentry(vfs.MountNamespaceFromContext(ctx), &renamed.vfsd, replacedVFSD); err != nil { + mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() + if err := vfsObj.PrepareRenameDentry(mntns, &renamed.vfsd, replacedVFSD); err != nil { return err } if err := renamed.file.rename(ctx, newParent.file, newName); err != nil { diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index d0552bd99..d00850e25 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -52,6 +52,9 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// Name is the default filesystem name. +const Name = "9p" + // FilesystemType implements vfs.FilesystemType. type FilesystemType struct{} diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index eda781155..5650512e0 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -116,8 +116,8 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent Ino: stat.Ino, NextOff: 1, } - if !cb.Handle(dirent) { - return nil + if err := cb.Handle(dirent); err != nil { + return err } fd.off++ } @@ -132,8 +132,8 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent Ino: stat.Ino, NextOff: 2, } - if !cb.Handle(dirent) { - return nil + if err := cb.Handle(dirent); err != nil { + return err } fd.off++ } @@ -153,8 +153,8 @@ func (fd *GenericDirectoryFD) IterDirents(ctx context.Context, cb vfs.IterDirent Ino: stat.Ino, NextOff: fd.off + 1, } - if !cb.Handle(dirent) { - return nil + if err := cb.Handle(dirent); err != nil { + return err } fd.off++ } diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index ee98eb66a..292f58afd 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -544,6 +544,7 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() virtfs := rp.VirtualFilesystem() srcDirDentry := srcDirVFSD.Impl().(*Dentry) @@ -595,7 +596,10 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error parentDentry := vfsd.Parent().Impl().(*Dentry) parentDentry.dirMu.Lock() defer parentDentry.dirMu.Unlock() - if err := virtfs.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), vfsd); err != nil { + + mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() + if err := virtfs.PrepareDeleteDentry(mntns, vfsd); err != nil { return err } if err := parentDentry.inode.RmDir(ctx, rp.Component(), vfsd); err != nil { @@ -697,7 +701,9 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error parentDentry := vfsd.Parent().Impl().(*Dentry) parentDentry.dirMu.Lock() defer parentDentry.dirMu.Unlock() - if err := virtfs.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), vfsd); err != nil { + mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() + if err := virtfs.PrepareDeleteDentry(mntns, vfsd); err != nil { return err } if err := parentDentry.inode.Unlink(ctx, rp.Component(), vfsd); err != nil { diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index 96a16e654..0459fb305 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -45,7 +45,10 @@ type RootDentryFn func(*auth.Credentials, *filesystem) *kernfs.Dentry func newTestSystem(t *testing.T, rootFn RootDentryFn) *testutil.System { ctx := contexttest.Context(t) creds := auth.CredentialsFromContext(ctx) - v := vfs.New() + v := &vfs.VirtualFilesystem{} + if err := v.Init(); err != nil { + t.Fatalf("VFS init: %v", err) + } v.MustRegisterFilesystemType("testfs", &fsType{rootFn: rootFn}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index 12aac2e6a..a83245866 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -14,6 +14,7 @@ go_library( "tasks_net.go", "tasks_sys.go", ], + visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", "//pkg/context", diff --git a/pkg/sentry/fsimpl/proc/filesystem.go b/pkg/sentry/fsimpl/proc/filesystem.go index 11477b6a9..5c19d5522 100644 --- a/pkg/sentry/fsimpl/proc/filesystem.go +++ b/pkg/sentry/fsimpl/proc/filesystem.go @@ -26,15 +26,18 @@ import ( "gvisor.dev/gvisor/pkg/sentry/vfs" ) -// procFSType is the factory class for procfs. +// Name is the default filesystem name. +const Name = "proc" + +// FilesystemType is the factory class for procfs. // // +stateify savable -type procFSType struct{} +type FilesystemType struct{} -var _ vfs.FilesystemType = (*procFSType)(nil) +var _ vfs.FilesystemType = (*FilesystemType)(nil) // GetFilesystem implements vfs.FilesystemType. -func (ft *procFSType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { +func (ft *FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth.Credentials, source string, opts vfs.GetFilesystemOptions) (*vfs.Filesystem, *vfs.Dentry, error) { k := kernel.KernelFromContext(ctx) if k == nil { return nil, nil, fmt.Errorf("procfs requires a kernel") @@ -47,12 +50,13 @@ func (ft *procFSType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFile procfs := &kernfs.Filesystem{} procfs.VFSFilesystem().Init(vfsObj, procfs) - var data *InternalData + var cgroups map[string]string if opts.InternalData != nil { - data = opts.InternalData.(*InternalData) + data := opts.InternalData.(*InternalData) + cgroups = data.Cgroups } - _, dentry := newTasksInode(procfs, k, pidns, data.Cgroups) + _, dentry := newTasksInode(procfs, k, pidns, cgroups) return procfs.VFSFilesystem(), dentry.VFSDentry(), nil } diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go index 102af0e93..f3f4e49b4 100644 --- a/pkg/sentry/fsimpl/proc/subtasks.go +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -105,8 +105,8 @@ func (i *subtasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallb Ino: i.inoGen.NextIno(), NextOff: offset + 1, } - if !cb.Handle(dirent) { - return offset, nil + if err := cb.Handle(dirent); err != nil { + return offset, err } offset++ } diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index ebe21630c..10c08fa90 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -73,9 +73,9 @@ func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNames "meminfo": newDentry(root, inoGen.NextIno(), 0444, &meminfoData{}), "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/mounts"), "net": newNetDir(root, inoGen, k), - "stat": newDentry(root, inoGen.NextIno(), 0444, &statData{}), + "stat": newDentry(root, inoGen.NextIno(), 0444, &statData{k: k}), "uptime": newDentry(root, inoGen.NextIno(), 0444, &uptimeData{}), - "version": newDentry(root, inoGen.NextIno(), 0444, &versionData{}), + "version": newDentry(root, inoGen.NextIno(), 0444, &versionData{k: k}), } inode := &tasksInode{ @@ -151,8 +151,8 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback Ino: i.inoGen.NextIno(), NextOff: offset + 1, } - if !cb.Handle(dirent) { - return offset, nil + if err := cb.Handle(dirent); err != nil { + return offset, err } offset++ } @@ -163,8 +163,8 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback Ino: i.inoGen.NextIno(), NextOff: offset + 1, } - if !cb.Handle(dirent) { - return offset, nil + if err := cb.Handle(dirent); err != nil { + return offset, err } offset++ } @@ -196,8 +196,8 @@ func (i *tasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback Ino: i.inoGen.NextIno(), NextOff: FIRST_PROCESS_ENTRY + 2 + int64(tid) + 1, } - if !cb.Handle(dirent) { - return offset, nil + if err := cb.Handle(dirent); err != nil { + return offset, err } offset++ } diff --git a/pkg/sentry/fsimpl/proc/tasks_net.go b/pkg/sentry/fsimpl/proc/tasks_net.go index 608fec017..d4e1812d8 100644 --- a/pkg/sentry/fsimpl/proc/tasks_net.go +++ b/pkg/sentry/fsimpl/proc/tasks_net.go @@ -39,7 +39,10 @@ import ( func newNetDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry { var contents map[string]*kernfs.Dentry - if stack := k.NetworkStack(); stack != nil { + // TODO(gvisor.dev/issue/1833): Support for using the network stack in the + // network namespace of the calling process. We should make this per-process, + // a.k.a. /proc/PID/net, and make /proc/net a symlink to /proc/self/net. + if stack := k.RootNetworkNamespace().Stack(); stack != nil { const ( arp = "IP address HW type Flags HW address Mask Device\n" netlink = "sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode\n" diff --git a/pkg/sentry/fsimpl/proc/tasks_sys.go b/pkg/sentry/fsimpl/proc/tasks_sys.go index c7ce74883..3d5dc463c 100644 --- a/pkg/sentry/fsimpl/proc/tasks_sys.go +++ b/pkg/sentry/fsimpl/proc/tasks_sys.go @@ -50,7 +50,9 @@ func newSysDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *k func newSysNetDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry { var contents map[string]*kernfs.Dentry - if stack := k.NetworkStack(); stack != nil { + // TODO(gvisor.dev/issue/1833): Support for using the network stack in the + // network namespace of the calling process. + if stack := k.RootNetworkNamespace().Stack(); stack != nil { contents = map[string]*kernfs.Dentry{ "ipv4": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ "tcp_sack": newDentry(root, inoGen.NextIno(), 0644, &tcpSackData{stack: stack}), diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index 6fc3524db..c5d531fe0 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -90,8 +90,7 @@ func setup(t *testing.T) *testutil.System { ctx := k.SupervisorContext() creds := auth.CredentialsFromContext(ctx) - vfsObj := vfs.New() - vfsObj.MustRegisterFilesystemType("procfs", &procFSType{}, &vfs.RegisterFilesystemTypeOptions{ + k.VFS().MustRegisterFilesystemType(Name, &FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) fsOpts := vfs.GetFilesystemOptions{ @@ -102,11 +101,11 @@ func setup(t *testing.T) *testutil.System { }, }, } - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "procfs", &fsOpts) + mntns, err := k.VFS().NewMountNamespace(ctx, creds, "", Name, &fsOpts) if err != nil { t.Fatalf("NewMountNamespace(): %v", err) } - return testutil.NewSystem(ctx, t, vfsObj, mntns) + return testutil.NewSystem(ctx, t, k.VFS(), mntns) } func TestTasksEmpty(t *testing.T) { @@ -131,7 +130,7 @@ func TestTasks(t *testing.T) { var tasks []*kernel.Task for i := 0; i < 5; i++ { tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc) + task, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc, s.MntNs, s.Root, s.Root) if err != nil { t.Fatalf("CreateTask(): %v", err) } @@ -213,7 +212,7 @@ func TestTasksOffset(t *testing.T) { k := kernel.KernelFromContext(s.Ctx) for i := 0; i < 3; i++ { tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - if _, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc); err != nil { + if _, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc, s.MntNs, s.Root, s.Root); err != nil { t.Fatalf("CreateTask(): %v", err) } } @@ -337,7 +336,7 @@ func TestTask(t *testing.T) { k := kernel.KernelFromContext(s.Ctx) tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - _, err := testutil.CreateTask(s.Ctx, "name", tc) + _, err := testutil.CreateTask(s.Ctx, "name", tc, s.MntNs, s.Root, s.Root) if err != nil { t.Fatalf("CreateTask(): %v", err) } @@ -352,7 +351,7 @@ func TestProcSelf(t *testing.T) { k := kernel.KernelFromContext(s.Ctx) tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := testutil.CreateTask(s.Ctx, "name", tc) + task, err := testutil.CreateTask(s.Ctx, "name", tc, s.MntNs, s.Root, s.Root) if err != nil { t.Fatalf("CreateTask(): %v", err) } @@ -433,7 +432,7 @@ func TestTree(t *testing.T) { var tasks []*kernel.Task for i := 0; i < 5; i++ { tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits()) - task, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc) + task, err := testutil.CreateTask(s.Ctx, fmt.Sprintf("name-%d", i), tc, s.MntNs, s.Root, s.Root) if err != nil { t.Fatalf("CreateTask(): %v", err) } diff --git a/pkg/sentry/fsimpl/sys/BUILD b/pkg/sentry/fsimpl/sys/BUILD index 66c0d8bc8..a741e2bb6 100644 --- a/pkg/sentry/fsimpl/sys/BUILD +++ b/pkg/sentry/fsimpl/sys/BUILD @@ -7,6 +7,7 @@ go_library( srcs = [ "sys.go", ], + visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", "//pkg/context", diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index d693fceae..c36c4fa11 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -28,6 +28,9 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) +// Name is the default filesystem name. +const Name = "sysfs" + // FilesystemType implements vfs.FilesystemType. type FilesystemType struct{} diff --git a/pkg/sentry/fsimpl/sys/sys_test.go b/pkg/sentry/fsimpl/sys/sys_test.go index 8b1cf0bd0..4b3602d47 100644 --- a/pkg/sentry/fsimpl/sys/sys_test.go +++ b/pkg/sentry/fsimpl/sys/sys_test.go @@ -34,16 +34,15 @@ func newTestSystem(t *testing.T) *testutil.System { } ctx := k.SupervisorContext() creds := auth.CredentialsFromContext(ctx) - v := vfs.New() - v.MustRegisterFilesystemType("sysfs", sys.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + k.VFS().MustRegisterFilesystemType(sys.Name, sys.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) - mns, err := v.NewMountNamespace(ctx, creds, "", "sysfs", &vfs.GetFilesystemOptions{}) + mns, err := k.VFS().NewMountNamespace(ctx, creds, "", sys.Name, &vfs.GetFilesystemOptions{}) if err != nil { t.Fatalf("Failed to create new mount namespace: %v", err) } - return testutil.NewSystem(ctx, t, v, mns) + return testutil.NewSystem(ctx, t, k.VFS(), mns) } func TestReadCPUFile(t *testing.T) { diff --git a/pkg/sentry/fsimpl/testutil/BUILD b/pkg/sentry/fsimpl/testutil/BUILD index efd5974c4..e4f36f4ae 100644 --- a/pkg/sentry/fsimpl/testutil/BUILD +++ b/pkg/sentry/fsimpl/testutil/BUILD @@ -16,7 +16,7 @@ go_library( "//pkg/cpuid", "//pkg/fspath", "//pkg/memutil", - "//pkg/sentry/fs", + "//pkg/sentry/fsimpl/tmpfs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/sched", diff --git a/pkg/sentry/fsimpl/testutil/kernel.go b/pkg/sentry/fsimpl/testutil/kernel.go index 89f8c4915..488478e29 100644 --- a/pkg/sentry/fsimpl/testutil/kernel.go +++ b/pkg/sentry/fsimpl/testutil/kernel.go @@ -24,7 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/memutil" - "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/sched" @@ -33,6 +33,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/time" + "gvisor.dev/gvisor/pkg/sentry/vfs" // Platforms are plugable. _ "gvisor.dev/gvisor/pkg/sentry/platform/kvm" @@ -99,36 +100,41 @@ func Boot() (*kernel.Kernel, error) { return nil, fmt.Errorf("initializing kernel: %v", err) } - ctx := k.SupervisorContext() + kernel.VFS2Enabled = true - // Create mount namespace without root as it's the minimum required to create - // the global thread group. - mntns, err := fs.NewMountNamespace(ctx, nil) - if err != nil { - return nil, err + if err := k.VFS().Init(); err != nil { + return nil, fmt.Errorf("VFS init: %v", err) } + k.VFS().MustRegisterFilesystemType(tmpfs.Name, &tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserMount: true, + AllowUserList: true, + }) + ls, err := limits.NewLinuxLimitSet() if err != nil { return nil, err } - tg := k.NewThreadGroup(mntns, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, ls) + tg := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, ls) k.TestOnly_SetGlobalInit(tg) return k, nil } // CreateTask creates a new bare bones task for tests. -func CreateTask(ctx context.Context, name string, tc *kernel.ThreadGroup) (*kernel.Task, error) { +func CreateTask(ctx context.Context, name string, tc *kernel.ThreadGroup, mntns *vfs.MountNamespace, root, cwd vfs.VirtualDentry) (*kernel.Task, error) { k := kernel.KernelFromContext(ctx) config := &kernel.TaskConfig{ Kernel: k, ThreadGroup: tc, TaskContext: &kernel.TaskContext{Name: name}, Credentials: auth.CredentialsFromContext(ctx), + NetworkNamespace: k.RootNetworkNamespace(), AllowedCPUMask: sched.NewFullCPUSet(k.ApplicationCores()), UTSNamespace: kernel.UTSNamespaceFromContext(ctx), IPCNamespace: kernel.IPCNamespaceFromContext(ctx), AbstractSocketNamespace: kernel.NewAbstractSocketNamespace(), + MountNamespaceVFS2: mntns, + FSContext: kernel.NewFSContextVFS2(root, cwd, 0022), } return k.TaskSet().NewTask(config) } diff --git a/pkg/sentry/fsimpl/testutil/testutil.go b/pkg/sentry/fsimpl/testutil/testutil.go index 69fd84ddd..e16808c63 100644 --- a/pkg/sentry/fsimpl/testutil/testutil.go +++ b/pkg/sentry/fsimpl/testutil/testutil.go @@ -41,12 +41,12 @@ type System struct { Creds *auth.Credentials VFS *vfs.VirtualFilesystem Root vfs.VirtualDentry - mns *vfs.MountNamespace + MntNs *vfs.MountNamespace } // NewSystem constructs a System. // -// Precondition: Caller must hold a reference on mns, whose ownership +// Precondition: Caller must hold a reference on MntNs, whose ownership // is transferred to the new System. func NewSystem(ctx context.Context, t *testing.T, v *vfs.VirtualFilesystem, mns *vfs.MountNamespace) *System { s := &System{ @@ -54,7 +54,7 @@ func NewSystem(ctx context.Context, t *testing.T, v *vfs.VirtualFilesystem, mns Ctx: ctx, Creds: auth.CredentialsFromContext(ctx), VFS: v, - mns: mns, + MntNs: mns, Root: mns.Root(), } return s @@ -75,7 +75,7 @@ func (s *System) WithSubtest(t *testing.T) *System { Ctx: s.Ctx, Creds: s.Creds, VFS: s.VFS, - mns: s.mns, + MntNs: s.MntNs, Root: s.Root, } } @@ -90,7 +90,7 @@ func (s *System) WithTemporaryContext(ctx context.Context) *System { Ctx: ctx, Creds: s.Creds, VFS: s.VFS, - mns: s.mns, + MntNs: s.MntNs, Root: s.Root, } } @@ -98,7 +98,7 @@ func (s *System) WithTemporaryContext(ctx context.Context) *System { // Destroy release resources associated with a test system. func (s *System) Destroy() { s.Root.DecRef() - s.mns.DecRef() // Reference on mns passed to NewSystem. + s.MntNs.DecRef() // Reference on MntNs passed to NewSystem. } // ReadToEnd reads the contents of fd until EOF to a string. @@ -226,7 +226,7 @@ func (d *DirentCollector) SkipDotsChecks(value bool) { } // Handle implements vfs.IterDirentsCallback.Handle. -func (d *DirentCollector) Handle(dirent vfs.Dirent) bool { +func (d *DirentCollector) Handle(dirent vfs.Dirent) error { d.mu.Lock() if d.dirents == nil { d.dirents = make(map[string]*vfs.Dirent) @@ -234,7 +234,7 @@ func (d *DirentCollector) Handle(dirent vfs.Dirent) bool { d.order = append(d.order, &dirent) d.dirents[dirent.Name] = &dirent d.mu.Unlock() - return true + return nil } // Count returns the number of dirents currently in the collector. diff --git a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go index 9fce5e4b4..383133e44 100644 --- a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go +++ b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go @@ -175,7 +175,10 @@ func BenchmarkVFS2MemfsStat(b *testing.B) { creds := auth.CredentialsFromContext(ctx) // Create VFS. - vfsObj := vfs.New() + vfsObj := vfs.VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + b.Fatalf("VFS init: %v", err) + } vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) @@ -366,7 +369,10 @@ func BenchmarkVFS2MemfsMountStat(b *testing.B) { creds := auth.CredentialsFromContext(ctx) // Create VFS. - vfsObj := vfs.New() + vfsObj := vfs.VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + b.Fatalf("VFS init: %v", err) + } vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go index dc0d27cf9..b4380af38 100644 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ b/pkg/sentry/fsimpl/tmpfs/directory.go @@ -74,25 +74,25 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba defer fs.mu.Unlock() if fd.off == 0 { - if !cb.Handle(vfs.Dirent{ + if err := cb.Handle(vfs.Dirent{ Name: ".", Type: linux.DT_DIR, Ino: vfsd.Impl().(*dentry).inode.ino, NextOff: 1, - }) { - return nil + }); err != nil { + return err } fd.off++ } if fd.off == 1 { parentInode := vfsd.ParentOrSelf().Impl().(*dentry).inode - if !cb.Handle(vfs.Dirent{ + if err := cb.Handle(vfs.Dirent{ Name: "..", Type: parentInode.direntType(), Ino: parentInode.ino, NextOff: 2, - }) { - return nil + }); err != nil { + return err } fd.off++ } @@ -111,14 +111,14 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba for child != nil { // Skip other directoryFD iterators. if child.inode != nil { - if !cb.Handle(vfs.Dirent{ + if err := cb.Handle(vfs.Dirent{ Name: child.vfsd.Name(), Type: child.inode.direntType(), Ino: child.inode.ino, NextOff: fd.off + 1, - }) { + }); err != nil { dir.childList.InsertBefore(child, fd.iter) - return nil + return err } fd.off++ } diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 8785452b6..e1b551422 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -16,7 +16,6 @@ package tmpfs import ( "fmt" - "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" @@ -347,10 +346,9 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open return nil, err } if opts.Flags&linux.O_TRUNC != 0 { - impl.mu.Lock() - impl.data.Truncate(0, impl.memFile) - atomic.StoreUint64(&impl.size, 0) - impl.mu.Unlock() + if _, err := impl.truncate(0); err != nil { + return nil, err + } } return &fd.vfsfd, nil case *directory: @@ -486,7 +484,9 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa vfsObj := rp.VirtualFilesystem() oldParentDir := oldParent.inode.impl.(*directory) newParentDir := newParent.inode.impl.(*directory) - if err := vfsObj.PrepareRenameDentry(vfs.MountNamespaceFromContext(ctx), renamedVFSD, replacedVFSD); err != nil { + mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() + if err := vfsObj.PrepareRenameDentry(mntns, renamedVFSD, replacedVFSD); err != nil { return err } if replaced != nil { @@ -543,7 +543,9 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error } defer mnt.EndWrite() vfsObj := rp.VirtualFilesystem() - if err := vfsObj.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), childVFSD); err != nil { + mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() + if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { return err } parent.inode.impl.(*directory).childList.Remove(child) @@ -631,7 +633,9 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error } defer mnt.EndWrite() vfsObj := rp.VirtualFilesystem() - if err := vfsObj.PrepareDeleteDentry(vfs.MountNamespaceFromContext(ctx), childVFSD); err != nil { + mntns := vfs.MountNamespaceFromContext(ctx) + defer mntns.DecRef() + if err := vfsObj.PrepareDeleteDentry(mntns, childVFSD); err != nil { return err } parent.inode.impl.(*directory).childList.Remove(child) diff --git a/pkg/sentry/fsimpl/tmpfs/pipe_test.go b/pkg/sentry/fsimpl/tmpfs/pipe_test.go index 5ee7f2a72..1614f2c39 100644 --- a/pkg/sentry/fsimpl/tmpfs/pipe_test.go +++ b/pkg/sentry/fsimpl/tmpfs/pipe_test.go @@ -151,7 +151,10 @@ func setup(t *testing.T) (context.Context, *auth.Credentials, *vfs.VirtualFilesy creds := auth.CredentialsFromContext(ctx) // Create VFS. - vfsObj := vfs.New() + vfsObj := &vfs.VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + t.Fatalf("VFS init: %v", err) + } vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index dab346a41..711442424 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -15,6 +15,7 @@ package tmpfs import ( + "fmt" "io" "math" "sync/atomic" @@ -22,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/safemem" + "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -34,25 +36,53 @@ import ( "gvisor.dev/gvisor/pkg/usermem" ) +// regularFile is a regular (=S_IFREG) tmpfs file. type regularFile struct { inode inode // memFile is a platform.File used to allocate pages to this regularFile. memFile *pgalloc.MemoryFile - // mu protects the fields below. - mu sync.RWMutex + // mapsMu protects mappings. + mapsMu sync.Mutex `state:"nosave"` + + // mappings tracks mappings of the file into memmap.MappingSpaces. + // + // Protected by mapsMu. + mappings memmap.MappingSet + + // writableMappingPages tracks how many pages of virtual memory are mapped + // as potentially writable from this file. If a page has multiple mappings, + // each mapping is counted separately. + // + // This counter is susceptible to overflow as we can potentially count + // mappings from many VMAs. We count pages rather than bytes to slightly + // mitigate this. + // + // Protected by mapsMu. + writableMappingPages uint64 + + // dataMu protects the fields below. + dataMu sync.RWMutex // data maps offsets into the file to offsets into memFile that store // the file's data. + // + // Protected by dataMu. data fsutil.FileRangeSet - // size is the size of data, but accessed using atomic memory - // operations to avoid locking in inode.stat(). - size uint64 - // seals represents file seals on this inode. + // + // Protected by dataMu. seals uint32 + + // size is the size of data. + // + // Protected by both dataMu and inode.mu; reading it requires holding + // either mutex, while writing requires holding both AND using atomics. + // Readers that do not require consistency (like Stat) may read the + // value atomically without holding either lock. + size uint64 } func (fs *filesystem) newRegularFile(creds *auth.Credentials, mode linux.FileMode) *inode { @@ -66,39 +96,170 @@ func (fs *filesystem) newRegularFile(creds *auth.Credentials, mode linux.FileMod // truncate grows or shrinks the file to the given size. It returns true if the // file size was updated. -func (rf *regularFile) truncate(size uint64) (bool, error) { - rf.mu.Lock() - defer rf.mu.Unlock() +func (rf *regularFile) truncate(newSize uint64) (bool, error) { + rf.inode.mu.Lock() + defer rf.inode.mu.Unlock() + return rf.truncateLocked(newSize) +} - if size == rf.size { +// Preconditions: rf.inode.mu must be held. +func (rf *regularFile) truncateLocked(newSize uint64) (bool, error) { + oldSize := rf.size + if newSize == oldSize { // Nothing to do. return false, nil } - if size > rf.size { - // Growing the file. + // Need to hold inode.mu and dataMu while modifying size. + rf.dataMu.Lock() + if newSize > oldSize { + // Can we grow the file? if rf.seals&linux.F_SEAL_GROW != 0 { - // Seal does not allow growth. + rf.dataMu.Unlock() return false, syserror.EPERM } - rf.size = size + // We only need to update the file size. + atomic.StoreUint64(&rf.size, newSize) + rf.dataMu.Unlock() return true, nil } - // Shrinking the file + // We are shrinking the file. First check if this is allowed. if rf.seals&linux.F_SEAL_SHRINK != 0 { - // Seal does not allow shrink. + rf.dataMu.Unlock() return false, syserror.EPERM } - // TODO(gvisor.dev/issues/1197): Invalidate mappings once we have - // mappings. + // Update the file size. + atomic.StoreUint64(&rf.size, newSize) + rf.dataMu.Unlock() + + // Invalidate past translations of truncated pages. + oldpgend := fs.OffsetPageEnd(int64(oldSize)) + newpgend := fs.OffsetPageEnd(int64(newSize)) + if newpgend < oldpgend { + rf.mapsMu.Lock() + rf.mappings.Invalidate(memmap.MappableRange{newpgend, oldpgend}, memmap.InvalidateOpts{ + // Compare Linux's mm/shmem.c:shmem_setattr() => + // mm/memory.c:unmap_mapping_range(evencows=1). + InvalidatePrivate: true, + }) + rf.mapsMu.Unlock() + } - rf.data.Truncate(size, rf.memFile) - rf.size = size + // We are now guaranteed that there are no translations of truncated pages, + // and can remove them. + rf.dataMu.Lock() + rf.data.Truncate(newSize, rf.memFile) + rf.dataMu.Unlock() return true, nil } +// AddMapping implements memmap.Mappable.AddMapping. +func (rf *regularFile) AddMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) error { + rf.mapsMu.Lock() + defer rf.mapsMu.Unlock() + rf.dataMu.RLock() + defer rf.dataMu.RUnlock() + + // Reject writable mapping if F_SEAL_WRITE is set. + if rf.seals&linux.F_SEAL_WRITE != 0 && writable { + return syserror.EPERM + } + + rf.mappings.AddMapping(ms, ar, offset, writable) + if writable { + pagesBefore := rf.writableMappingPages + + // ar is guaranteed to be page aligned per memmap.Mappable. + rf.writableMappingPages += uint64(ar.Length() / usermem.PageSize) + + if rf.writableMappingPages < pagesBefore { + panic(fmt.Sprintf("Overflow while mapping potentially writable pages pointing to a tmpfs file. Before %v, after %v", pagesBefore, rf.writableMappingPages)) + } + } + + return nil +} + +// RemoveMapping implements memmap.Mappable.RemoveMapping. +func (rf *regularFile) RemoveMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, writable bool) { + rf.mapsMu.Lock() + defer rf.mapsMu.Unlock() + + rf.mappings.RemoveMapping(ms, ar, offset, writable) + + if writable { + pagesBefore := rf.writableMappingPages + + // ar is guaranteed to be page aligned per memmap.Mappable. + rf.writableMappingPages -= uint64(ar.Length() / usermem.PageSize) + + if rf.writableMappingPages > pagesBefore { + panic(fmt.Sprintf("Underflow while unmapping potentially writable pages pointing to a tmpfs file. Before %v, after %v", pagesBefore, rf.writableMappingPages)) + } + } +} + +// CopyMapping implements memmap.Mappable.CopyMapping. +func (rf *regularFile) CopyMapping(ctx context.Context, ms memmap.MappingSpace, srcAR, dstAR usermem.AddrRange, offset uint64, writable bool) error { + return rf.AddMapping(ctx, ms, dstAR, offset, writable) +} + +// Translate implements memmap.Mappable.Translate. +func (rf *regularFile) Translate(ctx context.Context, required, optional memmap.MappableRange, at usermem.AccessType) ([]memmap.Translation, error) { + rf.dataMu.Lock() + defer rf.dataMu.Unlock() + + // Constrain translations to f.attr.Size (rounded up) to prevent + // translation to pages that may be concurrently truncated. + pgend := fs.OffsetPageEnd(int64(rf.size)) + var beyondEOF bool + if required.End > pgend { + if required.Start >= pgend { + return nil, &memmap.BusError{io.EOF} + } + beyondEOF = true + required.End = pgend + } + if optional.End > pgend { + optional.End = pgend + } + + cerr := rf.data.Fill(ctx, required, optional, rf.memFile, usage.Tmpfs, func(_ context.Context, dsts safemem.BlockSeq, _ uint64) (uint64, error) { + // Newly-allocated pages are zeroed, so we don't need to do anything. + return dsts.NumBytes(), nil + }) + + var ts []memmap.Translation + var translatedEnd uint64 + for seg := rf.data.FindSegment(required.Start); seg.Ok() && seg.Start() < required.End; seg, _ = seg.NextNonEmpty() { + segMR := seg.Range().Intersect(optional) + ts = append(ts, memmap.Translation{ + Source: segMR, + File: rf.memFile, + Offset: seg.FileRangeOf(segMR).Start, + Perms: usermem.AnyAccess, + }) + translatedEnd = segMR.End + } + + // Don't return the error returned by f.data.Fill if it occurred outside of + // required. + if translatedEnd < required.End && cerr != nil { + return ts, &memmap.BusError{cerr} + } + if beyondEOF { + return ts, &memmap.BusError{io.EOF} + } + return ts, nil +} + +// InvalidateUnsavable implements memmap.Mappable.InvalidateUnsavable. +func (*regularFile) InvalidateUnsavable(context.Context) error { + return nil +} + type regularFileFD struct { fileDescription @@ -152,8 +313,10 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off // Overflow. return 0, syserror.EFBIG } + f.inode.mu.Lock() rw := getRegularFileReadWriter(f, offset) n, err := src.CopyInTo(ctx, rw) + f.inode.mu.Unlock() putRegularFileReadWriter(rw) return n, err } @@ -215,6 +378,12 @@ func (fd *regularFileFD) UnlockPOSIX(ctx context.Context, uid lock.UniqueID, rng return nil } +// ConfigureMMap implements vfs.FileDescriptionImpl.ConfigureMMap. +func (fd *regularFileFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOpts) error { + file := fd.inode().impl.(*regularFile) + return vfs.GenericConfigureMMap(&fd.vfsfd, file, opts) +} + // regularFileReadWriter implements safemem.Reader and Safemem.Writer. type regularFileReadWriter struct { file *regularFile @@ -244,14 +413,15 @@ func putRegularFileReadWriter(rw *regularFileReadWriter) { // ReadToBlocks implements safemem.Reader.ReadToBlocks. func (rw *regularFileReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { - rw.file.mu.RLock() + rw.file.dataMu.RLock() + defer rw.file.dataMu.RUnlock() + size := rw.file.size // Compute the range to read (limited by file size and overflow-checked). - if rw.off >= rw.file.size { - rw.file.mu.RUnlock() + if rw.off >= size { return 0, io.EOF } - end := rw.file.size + end := size if rend := rw.off + dsts.NumBytes(); rend > rw.off && rend < end { end = rend } @@ -265,7 +435,6 @@ func (rw *regularFileReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, er // Get internal mappings. ims, err := rw.file.memFile.MapInternal(seg.FileRangeOf(seg.Range().Intersect(mr)), usermem.Read) if err != nil { - rw.file.mu.RUnlock() return done, err } @@ -275,7 +444,6 @@ func (rw *regularFileReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, er rw.off += uint64(n) dsts = dsts.DropFirst64(n) if err != nil { - rw.file.mu.RUnlock() return done, err } @@ -291,7 +459,6 @@ func (rw *regularFileReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, er rw.off += uint64(n) dsts = dsts.DropFirst64(n) if err != nil { - rw.file.mu.RUnlock() return done, err } @@ -299,13 +466,16 @@ func (rw *regularFileReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, er seg, gap = gap.NextSegment(), fsutil.FileRangeGapIterator{} } } - rw.file.mu.RUnlock() return done, nil } // WriteFromBlocks implements safemem.Writer.WriteFromBlocks. +// +// Preconditions: inode.mu must be held. func (rw *regularFileReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { - rw.file.mu.Lock() + // Hold dataMu so we can modify size. + rw.file.dataMu.Lock() + defer rw.file.dataMu.Unlock() // Compute the range to write (overflow-checked). end := rw.off + srcs.NumBytes() @@ -316,7 +486,6 @@ func (rw *regularFileReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, // Check if seals prevent either file growth or all writes. switch { case rw.file.seals&linux.F_SEAL_WRITE != 0: // Write sealed - rw.file.mu.Unlock() return 0, syserror.EPERM case end > rw.file.size && rw.file.seals&linux.F_SEAL_GROW != 0: // Grow sealed // When growth is sealed, Linux effectively allows writes which would @@ -338,7 +507,6 @@ func (rw *regularFileReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, } if end <= rw.off { // Truncation would result in no data being written. - rw.file.mu.Unlock() return 0, syserror.EPERM } } @@ -395,9 +563,8 @@ exitLoop: // If the write ends beyond the file's previous size, it causes the // file to grow. if rw.off > rw.file.size { - atomic.StoreUint64(&rw.file.size, rw.off) + rw.file.size = rw.off } - rw.file.mu.Unlock() return done, retErr } diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go index e9f71e334..0399725cf 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go @@ -40,7 +40,11 @@ var nextFileID int64 func newTmpfsRoot(ctx context.Context) (*vfs.VirtualFilesystem, vfs.VirtualDentry, func(), error) { creds := auth.CredentialsFromContext(ctx) - vfsObj := vfs.New() + vfsObj := &vfs.VirtualFilesystem{} + if err := vfsObj.Init(); err != nil { + return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("VFS init: %v", err) + } + vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 2108d0f4d..521206305 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -18,9 +18,10 @@ // Lock order: // // filesystem.mu -// regularFileFD.offMu -// regularFile.mu // inode.mu +// regularFileFD.offMu +// regularFile.mapsMu +// regularFile.dataMu package tmpfs import ( @@ -40,6 +41,9 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) +// Name is the default filesystem name. +const Name = "tmpfs" + // FilesystemType implements vfs.FilesystemType. type FilesystemType struct{} @@ -223,12 +227,15 @@ func (i *inode) tryIncRef() bool { func (i *inode) decRef() { if refs := atomic.AddInt64(&i.refs, -1); refs == 0 { - // This is unnecessary; it's mostly to simulate what tmpfs would do. if regFile, ok := i.impl.(*regularFile); ok { - regFile.mu.Lock() + // Hold inode.mu and regFile.dataMu while mutating + // size. + i.mu.Lock() + regFile.dataMu.Lock() regFile.data.DropAll(regFile.memFile) atomic.StoreUint64(®File.size, 0) - regFile.mu.Unlock() + regFile.dataMu.Unlock() + i.mu.Unlock() } } else if refs < 0 { panic("tmpfs.inode.decRef() called without holding a reference") @@ -317,7 +324,7 @@ func (i *inode) setStat(stat linux.Statx) error { if mask&linux.STATX_SIZE != 0 { switch impl := i.impl.(type) { case *regularFile: - updated, err := impl.truncate(stat.Size) + updated, err := impl.truncateLocked(stat.Size) if err != nil { return err } |