diff options
author | Nicolas Lacasse <nlacasse@google.com> | 2019-01-14 20:33:29 -0800 |
---|---|---|
committer | Shentubot <shentubot@google.com> | 2019-01-14 20:34:28 -0800 |
commit | dc8450b5676d4c4ac9bcfa23cabd862e0060527d (patch) | |
tree | a4ef1ad59764f46f674b7003221ba8ae399b9e65 /pkg/sentry/fs | |
parent | 343ebe9789087b099ea7feae19879f5c24e59bf1 (diff) |
Remove fs.Handle, ramfs.Entry, and all the DeprecatedFileOperations.
More helper structs have been added to the fsutil package to make it easier to
implement fs.InodeOperations and fs.FileOperations.
PiperOrigin-RevId: 229305982
Change-Id: Ib6f8d3862f4216745116857913dbfa351530223b
Diffstat (limited to 'pkg/sentry/fs')
84 files changed, 2413 insertions, 3008 deletions
diff --git a/pkg/sentry/fs/BUILD b/pkg/sentry/fs/BUILD index 0fe2b14bf..6f368b0da 100644 --- a/pkg/sentry/fs/BUILD +++ b/pkg/sentry/fs/BUILD @@ -94,7 +94,8 @@ go_test( deps = [ ":fs", "//pkg/sentry/context", - "//pkg/sentry/fs/ramfs/test", + "//pkg/sentry/fs/fsutil", + "//pkg/sentry/fs/ramfs", "//pkg/sentry/fs/tmpfs", "//pkg/sentry/kernel/contexttest", "//pkg/sentry/usermem", diff --git a/pkg/sentry/fs/anon/anon.go b/pkg/sentry/fs/anon/anon.go index 743cf511f..a5e8c4f0d 100644 --- a/pkg/sentry/fs/anon/anon.go +++ b/pkg/sentry/fs/anon/anon.go @@ -28,16 +28,12 @@ import ( // with any real filesystem. Some types depend on completely pseudo // "anon" inodes (eventfds, epollfds, etc). func NewInode(ctx context.Context) *fs.Inode { - return fs.NewInode(fsutil.NewSimpleInodeOperations(fsutil.InodeSimpleAttributes{ - FSType: linux.ANON_INODE_FS_MAGIC, - UAttr: fs.WithCurrentTime(ctx, fs.UnstableAttr{ - Owner: fs.FileOwnerFromContext(ctx), - Perms: fs.FilePermissions{ - User: fs.PermMask{Read: true, Write: true}, - }, - Links: 1, - }), - }), fs.NewNonCachingMountSource(nil, fs.MountSourceFlags{}), fs.StableAttr{ + iops := &fsutil.SimpleFileInode{ + InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, fs.RootOwner, fs.FilePermissions{ + User: fs.PermMask{Read: true, Write: true}, + }, linux.ANON_INODE_FS_MAGIC), + } + return fs.NewInode(iops, fs.NewPseudoMountSource(), fs.StableAttr{ Type: fs.Anonymous, DeviceID: PseudoDevice.DeviceID(), InodeID: PseudoDevice.NextIno(), diff --git a/pkg/sentry/fs/ashmem/BUILD b/pkg/sentry/fs/ashmem/BUILD index 44ef82e64..2463111a8 100644 --- a/pkg/sentry/fs/ashmem/BUILD +++ b/pkg/sentry/fs/ashmem/BUILD @@ -28,6 +28,7 @@ go_library( "//pkg/sentry/usage", "//pkg/sentry/usermem", "//pkg/syserror", + "//pkg/waiter", ], ) diff --git a/pkg/sentry/fs/ashmem/area.go b/pkg/sentry/fs/ashmem/area.go index d7dd2c084..7c1b11464 100644 --- a/pkg/sentry/fs/ashmem/area.go +++ b/pkg/sentry/fs/ashmem/area.go @@ -28,6 +28,7 @@ import ( "gvisor.googlesource.com/gvisor/pkg/sentry/usage" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" "gvisor.googlesource.com/gvisor/pkg/syserror" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) const ( @@ -42,9 +43,10 @@ const ( // // +stateify savable type Area struct { - fsutil.NoFsync `state:"nosave"` - fsutil.DeprecatedFileOperations `state:"nosave"` - fsutil.NotDirReaddir `state:"nosave"` + waiter.AlwaysReady `state:"nosave"` + fsutil.FileNoFsync `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` ad *Device @@ -98,11 +100,6 @@ func (a *Area) Write(ctx context.Context, file *fs.File, src usermem.IOSequence, return 0, syserror.ENOSYS } -// Flush implements fs.FileOperations.Flush. -func (a *Area) Flush(ctx context.Context, file *fs.File) error { - return nil -} - // ConfigureMMap implements fs.FileOperations.ConfigureMMap. func (a *Area) ConfigureMMap(ctx context.Context, file *fs.File, opts *memmap.MMapOpts) error { a.mu.Lock() @@ -122,8 +119,7 @@ func (a *Area) ConfigureMMap(ctx context.Context, file *fs.File, opts *memmap.MM return syserror.ENOMEM } tmpfsInodeOps := tmpfs.NewInMemoryFile(ctx, usage.Tmpfs, fs.UnstableAttr{}, k) - // This is not backed by a real filesystem, so we pass in nil. - tmpfsInode := fs.NewInode(tmpfsInodeOps, fs.NewNonCachingMountSource(nil, fs.MountSourceFlags{}), fs.StableAttr{}) + tmpfsInode := fs.NewInode(tmpfsInodeOps, fs.NewPseudoMountSource(), fs.StableAttr{}) dirent := fs.NewDirent(tmpfsInode, namePrefix+"/"+a.name) tmpfsFile, err := tmpfsInode.GetFile(ctx, dirent, fs.FileFlags{Read: true, Write: true}) // Drop the extra reference on the Dirent. diff --git a/pkg/sentry/fs/ashmem/device.go b/pkg/sentry/fs/ashmem/device.go index 962da141b..5369d1b0d 100644 --- a/pkg/sentry/fs/ashmem/device.go +++ b/pkg/sentry/fs/ashmem/device.go @@ -16,49 +16,40 @@ package ashmem import ( - "sync" - + "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" - "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/time" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" - "gvisor.googlesource.com/gvisor/pkg/syserror" ) // Device implements fs.InodeOperations. // // +stateify savable type Device struct { - fsutil.DeprecatedFileOperations `state:"nosave"` + fsutil.InodeGenericChecker `state:"nosave"` fsutil.InodeNoExtendedAttributes `state:"nosave"` + fsutil.InodeNoopRelease `state:"nosave"` + fsutil.InodeNoopTruncate `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` fsutil.InodeNotDirectory `state:"nosave"` - fsutil.InodeNotRenameable `state:"nosave"` + fsutil.InodeNotMappable `state:"nosave"` fsutil.InodeNotSocket `state:"nosave"` fsutil.InodeNotSymlink `state:"nosave"` - fsutil.NoFsync `state:"nosave"` - fsutil.NoMappable `state:"nosave"` - fsutil.NoopWriteOut `state:"nosave"` - fsutil.NotDirReaddir `state:"nosave"` + fsutil.InodeVirtual `state:"nosave"` - mu sync.Mutex `state:"nosave"` - unstable fs.UnstableAttr + fsutil.InodeSimpleAttributes } +var _ fs.InodeOperations = (*Device)(nil) + // NewDevice creates and intializes a Device structure. func NewDevice(ctx context.Context, owner fs.FileOwner, fp fs.FilePermissions) *Device { return &Device{ - unstable: fs.WithCurrentTime(ctx, fs.UnstableAttr{ - Owner: owner, - Perms: fp, - Links: 1, - }), + InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, owner, fp, linux.ANON_INODE_FS_MAGIC), } } -// Release implements fs.InodeOperations.Release. -func (ad *Device) Release(context.Context) {} - // GetFile implements fs.InodeOperations.GetFile. func (ad *Device) GetFile(ctx context.Context, d *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { return fs.NewFile(ctx, d, flags, &Area{ @@ -67,105 +58,3 @@ func (ad *Device) GetFile(ctx context.Context, d *fs.Dirent, flags fs.FileFlags) perms: usermem.AnyAccess, }), nil } - -// UnstableAttr implements fs.InodeOperations.UnstableAttr. -func (ad *Device) UnstableAttr(ctx context.Context, inode *fs.Inode) (fs.UnstableAttr, error) { - ad.mu.Lock() - defer ad.mu.Unlock() - return ad.unstable, nil -} - -// Check implements fs.InodeOperations.Check. -func (ad *Device) Check(ctx context.Context, inode *fs.Inode, p fs.PermMask) bool { - return fs.ContextCanAccessFile(ctx, inode, p) -} - -// SetPermissions implements fs.InodeOperations.SetPermissions. -func (ad *Device) SetPermissions(ctx context.Context, inode *fs.Inode, fp fs.FilePermissions) bool { - ad.mu.Lock() - defer ad.mu.Unlock() - ad.unstable.Perms = fp - ad.unstable.StatusChangeTime = time.NowFromContext(ctx) - return true -} - -// SetOwner implements fs.InodeOperations.SetOwner. -func (ad *Device) SetOwner(ctx context.Context, inode *fs.Inode, owner fs.FileOwner) error { - ad.mu.Lock() - defer ad.mu.Unlock() - if owner.UID.Ok() { - ad.unstable.Owner.UID = owner.UID - } - if owner.GID.Ok() { - ad.unstable.Owner.GID = owner.GID - } - return nil -} - -// SetTimestamps implements fs.InodeOperations.SetTimestamps. -func (ad *Device) SetTimestamps(ctx context.Context, inode *fs.Inode, ts fs.TimeSpec) error { - if ts.ATimeOmit && ts.MTimeOmit { - return nil - } - - ad.mu.Lock() - defer ad.mu.Unlock() - - now := time.NowFromContext(ctx) - if !ts.ATimeOmit { - if ts.ATimeSetSystemTime { - ad.unstable.AccessTime = now - } else { - ad.unstable.AccessTime = ts.ATime - } - } - if !ts.MTimeOmit { - if ts.MTimeSetSystemTime { - ad.unstable.ModificationTime = now - } else { - ad.unstable.ModificationTime = ts.MTime - } - } - ad.unstable.StatusChangeTime = now - return nil -} - -// Truncate implements fs.InodeOperations.WriteOut. -// -// Ignored by ashmem. -func (ad *Device) Truncate(ctx context.Context, inode *fs.Inode, size int64) error { - return nil -} - -// AddLink implements fs.InodeOperations.AddLink. -// -// Ashmem doesn't support links, no-op. -func (ad *Device) AddLink() {} - -// DropLink implements fs.InodeOperations.DropLink. -// -// Ashmem doesn't support links, no-op. -func (ad *Device) DropLink() {} - -// NotifyStatusChange implements fs.InodeOperations.NotifyStatusChange. -func (ad *Device) NotifyStatusChange(ctx context.Context) { - ad.mu.Lock() - defer ad.mu.Unlock() - now := time.NowFromContext(ctx) - ad.unstable.ModificationTime = now - ad.unstable.StatusChangeTime = now -} - -// IsVirtual implements fs.InodeOperations.IsVirtual. -// -// Ashmem is virtual. -func (ad *Device) IsVirtual() bool { - return true -} - -// StatFS implements fs.InodeOperations.StatFS. -// -// Ashmem doesn't support querying for filesystem info. -func (ad *Device) StatFS(context.Context) (fs.Info, error) { - return fs.Info{}, syserror.ENOSYS -} diff --git a/pkg/sentry/fs/attr.go b/pkg/sentry/fs/attr.go index 59e060e3c..3523b068a 100644 --- a/pkg/sentry/fs/attr.go +++ b/pkg/sentry/fs/attr.go @@ -180,6 +180,53 @@ type UnstableAttr struct { Links uint64 } +// SetOwner sets the owner and group if they are valid. +// +// This method is NOT thread-safe. Callers must prevent concurrent calls. +func (ua *UnstableAttr) SetOwner(ctx context.Context, owner FileOwner) { + if owner.UID.Ok() { + ua.Owner.UID = owner.UID + } + if owner.GID.Ok() { + ua.Owner.GID = owner.GID + } + ua.StatusChangeTime = ktime.NowFromContext(ctx) +} + +// SetPermissions sets the permissions. +// +// This method is NOT thread-safe. Callers must prevent concurrent calls. +func (ua *UnstableAttr) SetPermissions(ctx context.Context, p FilePermissions) { + ua.Perms = p + ua.StatusChangeTime = ktime.NowFromContext(ctx) +} + +// SetTimestamps sets the timestamps according to the TimeSpec. +// +// This method is NOT thread-safe. Callers must prevent concurrent calls. +func (ua *UnstableAttr) SetTimestamps(ctx context.Context, ts TimeSpec) { + if ts.ATimeOmit && ts.MTimeOmit { + return + } + + now := ktime.NowFromContext(ctx) + if !ts.ATimeOmit { + if ts.ATimeSetSystemTime { + ua.AccessTime = now + } else { + ua.AccessTime = ts.ATime + } + } + if !ts.MTimeOmit { + if ts.MTimeSetSystemTime { + ua.ModificationTime = now + } else { + ua.ModificationTime = ts.MTime + } + } + ua.StatusChangeTime = now +} + // WithCurrentTime returns u with AccessTime == ModificationTime == current time. func WithCurrentTime(ctx context.Context, u UnstableAttr) UnstableAttr { t := ktime.NowFromContext(ctx) diff --git a/pkg/sentry/fs/binder/BUILD b/pkg/sentry/fs/binder/BUILD index a077b91d2..27155819e 100644 --- a/pkg/sentry/fs/binder/BUILD +++ b/pkg/sentry/fs/binder/BUILD @@ -16,11 +16,11 @@ go_library( "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", "//pkg/sentry/kernel", - "//pkg/sentry/kernel/time", "//pkg/sentry/memmap", "//pkg/sentry/platform", "//pkg/sentry/usage", "//pkg/sentry/usermem", "//pkg/syserror", + "//pkg/waiter", ], ) diff --git a/pkg/sentry/fs/binder/binder.go b/pkg/sentry/fs/binder/binder.go index e642c7f22..19cd55e65 100644 --- a/pkg/sentry/fs/binder/binder.go +++ b/pkg/sentry/fs/binder/binder.go @@ -24,12 +24,12 @@ import ( "gvisor.googlesource.com/gvisor/pkg/sentry/fs" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" - "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/time" "gvisor.googlesource.com/gvisor/pkg/sentry/memmap" "gvisor.googlesource.com/gvisor/pkg/sentry/platform" "gvisor.googlesource.com/gvisor/pkg/sentry/usage" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" "gvisor.googlesource.com/gvisor/pkg/syserror" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) const ( @@ -43,34 +43,29 @@ const ( // // +stateify savable type Device struct { + fsutil.InodeGenericChecker `state:"nosave"` fsutil.InodeNoExtendedAttributes `state:"nosave"` + fsutil.InodeNoopRelease `state:"nosave"` + fsutil.InodeNoopTruncate `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` fsutil.InodeNotDirectory `state:"nosave"` - fsutil.InodeNotRenameable `state:"nosave"` + fsutil.InodeNotMappable `state:"nosave"` fsutil.InodeNotSocket `state:"nosave"` fsutil.InodeNotSymlink `state:"nosave"` - fsutil.NoMappable `state:"nosave"` - fsutil.NoopWriteOut `state:"nosave"` - fsutil.DeprecatedFileOperations `state:"nosave"` + fsutil.InodeVirtual `state:"nosave"` - // mu protects unstable. - mu sync.Mutex `state:"nosave"` - unstable fs.UnstableAttr + fsutil.InodeSimpleAttributes } +var _ fs.InodeOperations = (*Device)(nil) + // NewDevice creates and intializes a Device structure. func NewDevice(ctx context.Context, owner fs.FileOwner, fp fs.FilePermissions) *Device { return &Device{ - unstable: fs.WithCurrentTime(ctx, fs.UnstableAttr{ - Owner: owner, - Perms: fp, - Links: 1, - }), + InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, owner, fp, 0), } } -// Release implements fs.InodeOperations.Release. -func (bd *Device) Release(context.Context) {} - // GetFile implements fs.InodeOperations.GetFile. // // TODO: Add functionality to GetFile: Additional fields will be @@ -85,115 +80,13 @@ func (bd *Device) GetFile(ctx context.Context, d *fs.Dirent, flags fs.FileFlags) }), nil } -// UnstableAttr implements fs.InodeOperations.UnstableAttr. -func (bd *Device) UnstableAttr(ctx context.Context, inode *fs.Inode) (fs.UnstableAttr, error) { - bd.mu.Lock() - defer bd.mu.Unlock() - return bd.unstable, nil -} - -// Check implements fs.InodeOperations.Check. -func (bd *Device) Check(ctx context.Context, inode *fs.Inode, p fs.PermMask) bool { - return fs.ContextCanAccessFile(ctx, inode, p) -} - -// SetPermissions implements fs.InodeOperations.SetPermissions. -func (bd *Device) SetPermissions(ctx context.Context, inode *fs.Inode, fp fs.FilePermissions) bool { - bd.mu.Lock() - defer bd.mu.Unlock() - bd.unstable.Perms = fp - bd.unstable.StatusChangeTime = time.NowFromContext(ctx) - return true -} - -// SetOwner implements fs.InodeOperations.SetOwner. -func (bd *Device) SetOwner(ctx context.Context, inode *fs.Inode, owner fs.FileOwner) error { - bd.mu.Lock() - defer bd.mu.Unlock() - if owner.UID.Ok() { - bd.unstable.Owner.UID = owner.UID - } - if owner.GID.Ok() { - bd.unstable.Owner.GID = owner.GID - } - return nil -} - -// SetTimestamps implements fs.InodeOperations.SetTimestamps. -func (bd *Device) SetTimestamps(ctx context.Context, inode *fs.Inode, ts fs.TimeSpec) error { - if ts.ATimeOmit && ts.MTimeOmit { - return nil - } - - bd.mu.Lock() - defer bd.mu.Unlock() - - now := time.NowFromContext(ctx) - if !ts.ATimeOmit { - if ts.ATimeSetSystemTime { - bd.unstable.AccessTime = now - } else { - bd.unstable.AccessTime = ts.ATime - } - } - if !ts.MTimeOmit { - if ts.MTimeSetSystemTime { - bd.unstable.ModificationTime = now - } else { - bd.unstable.ModificationTime = ts.MTime - } - } - bd.unstable.StatusChangeTime = now - return nil -} - -// Truncate implements fs.InodeOperations.WriteOut. -// -// Ignored for a character device, such as Binder. -func (bd *Device) Truncate(ctx context.Context, inode *fs.Inode, size int64) error { - return nil -} - -// AddLink implements fs.InodeOperations.AddLink. -// -// Binder doesn't support links, no-op. -func (bd *Device) AddLink() {} - -// DropLink implements fs.InodeOperations.DropLink. -// -// Binder doesn't support links, no-op. -func (bd *Device) DropLink() {} - -// NotifyStatusChange implements fs.InodeOperations.NotifyStatusChange. -func (bd *Device) NotifyStatusChange(ctx context.Context) { - bd.mu.Lock() - defer bd.mu.Unlock() - now := time.NowFromContext(ctx) - bd.unstable.ModificationTime = now - bd.unstable.StatusChangeTime = now -} - -// IsVirtual implements fs.InodeOperations.IsVirtual. -// -// Binder is virtual. -func (bd *Device) IsVirtual() bool { - return true -} - -// StatFS implements fs.InodeOperations.StatFS. -// -// Binder doesn't support querying for filesystem info. -func (bd *Device) StatFS(context.Context) (fs.Info, error) { - return fs.Info{}, syserror.ENOSYS -} - // Proc implements fs.FileOperations and fs.IoctlGetter. // // +stateify savable type Proc struct { - fsutil.NoFsync `state:"nosave"` - fsutil.DeprecatedFileOperations `state:"nosave"` - fsutil.NotDirReaddir `state:"nosave"` + waiter.AlwaysReady `state:"nosave"` + fsutil.FileNoFsync `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` bd *Device task *kernel.Task diff --git a/pkg/sentry/fs/dev/BUILD b/pkg/sentry/fs/dev/BUILD index b17b5202c..b9cfae05f 100644 --- a/pkg/sentry/fs/dev/BUILD +++ b/pkg/sentry/fs/dev/BUILD @@ -32,5 +32,6 @@ go_library( "//pkg/sentry/safemem", "//pkg/sentry/usermem", "//pkg/syserror", + "//pkg/waiter", ], ) diff --git a/pkg/sentry/fs/dev/dev.go b/pkg/sentry/fs/dev/dev.go index 3e127bf04..f8e8099f7 100644 --- a/pkg/sentry/fs/dev/dev.go +++ b/pkg/sentry/fs/dev/dev.go @@ -16,6 +16,8 @@ package dev import ( + "math" + "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ashmem" @@ -26,13 +28,6 @@ import ( "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" ) -// Dev is the root node. -// -// +stateify savable -type Dev struct { - ramfs.Dir -} - func newCharacterDevice(iops fs.InodeOperations, msrc *fs.MountSource) *fs.Inode { return fs.NewInode(iops, msrc, fs.StableAttr{ DeviceID: devDevice.DeviceID(), @@ -43,8 +38,7 @@ func newCharacterDevice(iops fs.InodeOperations, msrc *fs.MountSource) *fs.Inode } func newDirectory(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - iops := &ramfs.Dir{} - iops.InitDir(ctx, map[string]*fs.Inode{}, fs.RootOwner, fs.FilePermsFromMode(0555)) + iops := ramfs.NewDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) return fs.NewInode(iops, msrc, fs.StableAttr{ DeviceID: devDevice.DeviceID(), InodeID: devDevice.NextIno(), @@ -54,8 +48,7 @@ func newDirectory(ctx context.Context, msrc *fs.MountSource) *fs.Inode { } func newSymlink(ctx context.Context, target string, msrc *fs.MountSource) *fs.Inode { - iops := &ramfs.Symlink{} - iops.InitSymlink(ctx, fs.RootOwner, target) + iops := ramfs.NewSymlink(ctx, fs.RootOwner, target) return fs.NewInode(iops, msrc, fs.StableAttr{ DeviceID: devDevice.DeviceID(), InodeID: devDevice.NextIno(), @@ -66,8 +59,6 @@ func newSymlink(ctx context.Context, target string, msrc *fs.MountSource) *fs.In // New returns the root node of a device filesystem. func New(ctx context.Context, msrc *fs.MountSource, binderEnabled bool, ashmemEnabled bool) *fs.Inode { - d := &Dev{} - contents := map[string]*fs.Inode{ "fd": newSymlink(ctx, "/proc/self/fd", msrc), "stdin": newSymlink(ctx, "/proc/self/fd/0", msrc), @@ -114,11 +105,19 @@ func New(ctx context.Context, msrc *fs.MountSource, binderEnabled bool, ashmemEn contents["ashmem"] = newCharacterDevice(ashmem, msrc) } - d.InitDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) - return fs.NewInode(d, msrc, fs.StableAttr{ + iops := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return fs.NewInode(iops, msrc, fs.StableAttr{ DeviceID: devDevice.DeviceID(), InodeID: devDevice.NextIno(), BlockSize: usermem.PageSize, Type: fs.Directory, }) } + +// readZeros implements fs.FileOperations.Read with infinite null bytes. +type readZeros struct{} + +// Read implements fs.FileOperations.Read. +func (readZeros) Read(ctx context.Context, file *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { + return dst.ZeroOut(ctx, math.MaxInt64) +} diff --git a/pkg/sentry/fs/dev/fs.go b/pkg/sentry/fs/dev/fs.go index d96f4f423..abfe689f0 100644 --- a/pkg/sentry/fs/dev/fs.go +++ b/pkg/sentry/fs/dev/fs.go @@ -33,6 +33,8 @@ const ashmemEnabledKey = "ashmem_enabled" // +stateify savable type filesystem struct{} +var _ fs.Filesystem = (*filesystem)(nil) + func init() { fs.RegisterFilesystem(&filesystem{}) } diff --git a/pkg/sentry/fs/dev/full.go b/pkg/sentry/fs/dev/full.go index eeda646ab..cbdd40161 100644 --- a/pkg/sentry/fs/dev/full.go +++ b/pkg/sentry/fs/dev/full.go @@ -15,41 +15,64 @@ package dev import ( - "math" - "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" "gvisor.googlesource.com/gvisor/pkg/syserror" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) // fullDevice is used to implement /dev/full. // // +stateify savable type fullDevice struct { - ramfs.Entry + fsutil.InodeGenericChecker `state:"nosave"` + fsutil.InodeNoExtendedAttributes `state:"nosave"` + fsutil.InodeNoopRelease `state:"nosave"` + fsutil.InodeNoopTruncate `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` + fsutil.InodeNotDirectory `state:"nosave"` + fsutil.InodeNotMappable `state:"nosave"` + fsutil.InodeNotSocket `state:"nosave"` + fsutil.InodeNotSymlink `state:"nosave"` + fsutil.InodeVirtual `state:"nosave"` + + fsutil.InodeSimpleAttributes } +var _ fs.InodeOperations = (*fullDevice)(nil) + func newFullDevice(ctx context.Context, owner fs.FileOwner, mode linux.FileMode) *fullDevice { - f := &fullDevice{} - f.InitEntry(ctx, owner, fs.FilePermsFromMode(mode)) + f := &fullDevice{ + InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, owner, fs.FilePermsFromMode(mode), linux.TMPFS_MAGIC), + } return f } -// DeprecatedPwritev implements fs.InodeOperations.DeprecatedPwritev by -// returining ENOSPC. -func (f *fullDevice) DeprecatedPwritev(_ context.Context, _ usermem.IOSequence, _ int64) (int64, error) { - return 0, syserror.ENOSPC +// GetFile implements fs.InodeOperations.GetFile. +func (f *fullDevice) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + flags.Pread = true + return fs.NewFile(ctx, dirent, flags, &fullFileOperations{}), nil } -// DeprecatedPreadv implements fs.InodeOperations.DeprecatedPreadv. -func (f *fullDevice) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, _ int64) (int64, error) { - return dst.ZeroOut(ctx, math.MaxInt64) +// +stateify savable +type fullFileOperations struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + readZeros `state:"nosave"` } -// Truncate should be simply ignored for character devices on linux. -func (f *fullDevice) Truncate(context.Context, *fs.Inode, int64) error { - return nil +var _ fs.FileOperations = (*fullFileOperations)(nil) + +// Write implements FileOperations.Write. +func (fullFileOperations) Write(context.Context, *fs.File, usermem.IOSequence, int64) (int64, error) { + return 0, syserror.ENOSPC } diff --git a/pkg/sentry/fs/dev/null.go b/pkg/sentry/fs/dev/null.go index 68090f353..73fd09058 100644 --- a/pkg/sentry/fs/dev/null.go +++ b/pkg/sentry/fs/dev/null.go @@ -15,78 +15,104 @@ package dev import ( - "io" - "math" - "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/memmap" "gvisor.googlesource.com/gvisor/pkg/sentry/mm" "gvisor.googlesource.com/gvisor/pkg/sentry/platform" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) // +stateify savable type nullDevice struct { - ramfs.Entry + fsutil.InodeGenericChecker `state:"nosave"` + fsutil.InodeNoExtendedAttributes `state:"nosave"` + fsutil.InodeNoopRelease `state:"nosave"` + fsutil.InodeNoopTruncate `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` + fsutil.InodeNotDirectory `state:"nosave"` + fsutil.InodeNotMappable `state:"nosave"` + fsutil.InodeNotSocket `state:"nosave"` + fsutil.InodeNotSymlink `state:"nosave"` + fsutil.InodeVirtual `state:"nosave"` + + fsutil.InodeSimpleAttributes } +var _ fs.InodeOperations = (*nullDevice)(nil) + func newNullDevice(ctx context.Context, owner fs.FileOwner, mode linux.FileMode) *nullDevice { - n := &nullDevice{} - n.InitEntry(ctx, owner, fs.FilePermsFromMode(mode)) + n := &nullDevice{ + InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, owner, fs.FilePermsFromMode(mode), linux.TMPFS_MAGIC), + } return n } -// DeprecatedPreadv reads data from the device. -func (n *nullDevice) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { - return 0, io.EOF -} +// GetFile implements fs.FileOperations.GetFile. +func (n *nullDevice) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + flags.Pread = true + flags.Pwrite = true -// DeprecatedPwritev discards writes. -func (n *nullDevice) DeprecatedPwritev(_ context.Context, src usermem.IOSequence, offset int64) (int64, error) { - return src.NumBytes(), nil + return fs.NewFile(ctx, dirent, flags, &nullFileOperations{}), nil } -// Truncate should be simply ignored for character devices on linux. -func (n *nullDevice) Truncate(context.Context, *fs.Inode, int64) error { - return nil +// +stateify savable +type nullFileOperations struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRead `state:"nosave"` + fsutil.FileNoopWrite `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` } +var _ fs.FileOperations = (*nullFileOperations)(nil) + // +stateify savable type zeroDevice struct { nullDevice } +var _ fs.InodeOperations = (*zeroDevice)(nil) + func newZeroDevice(ctx context.Context, owner fs.FileOwner, mode linux.FileMode) *zeroDevice { - zd := &zeroDevice{} - zd.InitEntry(ctx, owner, fs.FilePermsFromMode(mode)) + zd := &zeroDevice{ + nullDevice: nullDevice{ + InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, owner, fs.FilePermsFromMode(mode), linux.TMPFS_MAGIC), + }, + } return zd } -// DeprecatedPreadv implements fs.InodeOperations.DeprecatedPreadv. -func (zd *zeroDevice) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { - return dst.ZeroOut(ctx, math.MaxInt64) -} - -// GetFile overrides ramfs.Entry.GetFile and returns a zeroFile instead. +// GetFile implements fs.FileOperations.GetFile. func (zd *zeroDevice) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { - // Allow pread(2) and pwrite(2) on this file. flags.Pread = true flags.Pwrite = true - return fs.NewFile(ctx, dirent, flags, &zeroFileOperations{ - FileOperations: &fsutil.Handle{HandleOperations: dirent.Inode.HandleOps()}, - }), nil + return fs.NewFile(ctx, dirent, flags, &zeroFileOperations{}), nil } // +stateify savable type zeroFileOperations struct { - fs.FileOperations + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNoopWrite `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + readZeros `state:"nosave"` } +var _ fs.FileOperations = (*zeroFileOperations)(nil) + // ConfigureMMap implements fs.FileOperations.ConfigureMMap. func (*zeroFileOperations) ConfigureMMap(ctx context.Context, file *fs.File, opts *memmap.MMapOpts) error { m, err := mm.NewSharedAnonMappable(opts.Length, platform.FromContext(ctx)) diff --git a/pkg/sentry/fs/dev/random.go b/pkg/sentry/fs/dev/random.go index 33e4913e4..837b7793a 100644 --- a/pkg/sentry/fs/dev/random.go +++ b/pkg/sentry/fs/dev/random.go @@ -19,37 +19,58 @@ import ( "gvisor.googlesource.com/gvisor/pkg/rand" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/safemem" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) // +stateify savable type randomDevice struct { - ramfs.Entry + fsutil.InodeGenericChecker `state:"nosave"` + fsutil.InodeNoExtendedAttributes `state:"nosave"` + fsutil.InodeNoopRelease `state:"nosave"` + fsutil.InodeNoopTruncate `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` + fsutil.InodeNotDirectory `state:"nosave"` + fsutil.InodeNotMappable `state:"nosave"` + fsutil.InodeNotSocket `state:"nosave"` + fsutil.InodeNotSymlink `state:"nosave"` + fsutil.InodeVirtual `state:"nosave"` + + fsutil.InodeSimpleAttributes } +var _ fs.InodeOperations = (*randomDevice)(nil) + func newRandomDevice(ctx context.Context, owner fs.FileOwner, mode linux.FileMode) *randomDevice { - r := &randomDevice{} - r.InitEntry(ctx, owner, fs.FilePermsFromMode(mode)) + r := &randomDevice{ + InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, owner, fs.FilePermsFromMode(mode), linux.TMPFS_MAGIC), + } return r } -// DeprecatedPreadv reads random data. -func (*randomDevice) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { - return dst.CopyOutFrom(ctx, safemem.FromIOReader{rand.Reader}) +// GetFile implements fs.InodeOperations.GetFile. +func (randomDevice) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &randomFileOperations{}), nil } -// DeprecatedPwritev implements fs.HandleOperations.DeprecatedPwritev. -func (*randomDevice) DeprecatedPwritev(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { - // On Linux, "Writing to /dev/random or /dev/urandom will update the - // entropy pool with the data written, but this will not result in a higher - // entropy count" - random(4). We don't need to support this, but we do - // need to support the write, so just make it a no-op a la /dev/null. - return src.NumBytes(), nil +// +stateify savable +type randomFileOperations struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNoopWrite `state:"nosave"` } -// Truncate should be simply ignored for character devices on linux. -func (r *randomDevice) Truncate(context.Context, *fs.Inode, int64) error { - return nil +var _ fs.FileOperations = (*randomFileOperations)(nil) + +// Read implements fs.FileOperations.Read. +func (randomFileOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, _ int64) (int64, error) { + return dst.CopyOutFrom(ctx, safemem.FromIOReader{rand.Reader}) } diff --git a/pkg/sentry/fs/dirent.go b/pkg/sentry/fs/dirent.go index c4918a11b..d6a19dc81 100644 --- a/pkg/sentry/fs/dirent.go +++ b/pkg/sentry/fs/dirent.go @@ -837,8 +837,8 @@ func (d *Dirent) CreateFifo(ctx context.Context, root *Dirent, name string, perm }) } -// getDotAttrs returns the DentAttrs corresponding to "." and ".." directories. -func (d *Dirent) getDotAttrs(root *Dirent) (DentAttr, DentAttr) { +// GetDotAttrs returns the DentAttrs corresponding to "." and ".." directories. +func (d *Dirent) GetDotAttrs(root *Dirent) (DentAttr, DentAttr) { // Get '.'. sattr := d.Inode.StableAttr dot := DentAttr{ @@ -870,7 +870,7 @@ func (d *Dirent) readdirFrozen(root *Dirent, offset int64, dirCtx *DirCtx) (int6 // Collect attrs for "." and "..". attrs := make(map[string]DentAttr) names := []string{".", ".."} - attrs["."], attrs[".."] = d.getDotAttrs(root) + attrs["."], attrs[".."] = d.GetDotAttrs(root) // Get info from all children. d.mu.Lock() @@ -965,7 +965,7 @@ func direntReaddir(ctx context.Context, d *Dirent, it DirIterator, root *Dirent, } // Collect attrs for "." and "..". - dot, dotdot := d.getDotAttrs(root) + dot, dotdot := d.GetDotAttrs(root) // Emit "." and ".." if the offset is low enough. if offset == 0 { diff --git a/pkg/sentry/fs/fdpipe/pipe.go b/pkg/sentry/fs/fdpipe/pipe.go index e3b830747..b4d11cb45 100644 --- a/pkg/sentry/fs/fdpipe/pipe.go +++ b/pkg/sentry/fs/fdpipe/pipe.go @@ -37,13 +37,13 @@ import ( // // +stateify savable type pipeOperations struct { - fsutil.PipeSeek `state:"nosave"` - fsutil.NotDirReaddir `state:"nosave"` - fsutil.NoFsync `state:"nosave"` - fsutil.NoopFlush `state:"nosave"` - fsutil.NoMMap `state:"nosave"` - fsutil.NoIoctl `state:"nosave"` - waiter.Queue `state:"nosave"` + fsutil.FilePipeSeek `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoFsync `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + waiter.Queue `state:"nosave"` // flags are the flags used to open the pipe. flags fs.FileFlags `state:".(fs.FileFlags)"` diff --git a/pkg/sentry/fs/file_operations.go b/pkg/sentry/fs/file_operations.go index 28e8e233d..81c6e2b5d 100644 --- a/pkg/sentry/fs/file_operations.go +++ b/pkg/sentry/fs/file_operations.go @@ -91,7 +91,7 @@ type FileOperations interface { Flush(ctx context.Context, file *File) error // ConfigureMMap mutates opts to implement mmap(2) for the file. Most - // implementations can either embed fsutil.NoMMap (if they don't support + // implementations can either embed fsutil.FileNoMMap (if they don't support // memory mapping) or call fsutil.GenericConfigureMMap with the appropriate // memmap.Mappable. ConfigureMMap(ctx context.Context, file *File, opts *memmap.MMapOpts) error diff --git a/pkg/sentry/fs/file_overlay_test.go b/pkg/sentry/fs/file_overlay_test.go index f121cbdda..a4ac58763 100644 --- a/pkg/sentry/fs/file_overlay_test.go +++ b/pkg/sentry/fs/file_overlay_test.go @@ -20,7 +20,8 @@ import ( "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - ramfstest "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs/test" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/contexttest" ) @@ -135,7 +136,7 @@ func TestReaddirRevalidation(t *testing.T) { // Get a handle to the dirent in the upper filesystem so that we can // modify it without going through the dirent. - upperDir := upper.InodeOperations.(*dir).InodeOperations.(*ramfstest.Dir) + upperDir := upper.InodeOperations.(*dir).InodeOperations.(*ramfs.Dir) // Check that overlay returns the files from both upper and lower. openDir, err := overlay.GetFile(ctx, fs.NewDirent(overlay, "stub"), fs.FileFlags{Read: true}) @@ -155,7 +156,7 @@ func TestReaddirRevalidation(t *testing.T) { if err := upperDir.Remove(ctx, upper, "a"); err != nil { t.Fatalf("error removing child: %v", err) } - upperDir.AddChild(ctx, "c", fs.NewInode(ramfstest.NewFile(ctx, fs.FilePermissions{}), + upperDir.AddChild(ctx, "c", fs.NewInode(fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermissions{}, 0), upper.MountSource, fs.StableAttr{Type: fs.RegularFile})) // Seek to beginning of the directory and do the readdir again. diff --git a/pkg/sentry/fs/filetest/filetest.go b/pkg/sentry/fs/filetest/filetest.go index 65ca196d9..40d84d9f2 100644 --- a/pkg/sentry/fs/filetest/filetest.go +++ b/pkg/sentry/fs/filetest/filetest.go @@ -31,14 +31,14 @@ import ( // TestFileOperations is an implementation of the File interface. It provides all // required methods. type TestFileOperations struct { - fsutil.NoopRelease `state:"nosave"` - fsutil.PipeSeek `state:"nosave"` - fsutil.NotDirReaddir `state:"nosave"` - fsutil.NoFsync `state:"nosave"` - fsutil.NoopFlush `state:"nosave"` - fsutil.NoMMap `state:"nosave"` - fsutil.NoIoctl `state:"nosave"` - waiter.AlwaysReady `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FilePipeSeek `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoFsync `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + waiter.AlwaysReady `state:"nosave"` } // NewTestFile creates and initializes a new test file. diff --git a/pkg/sentry/fs/fsutil/BUILD b/pkg/sentry/fs/fsutil/BUILD index 6834e1272..4965e1a5f 100644 --- a/pkg/sentry/fs/fsutil/BUILD +++ b/pkg/sentry/fs/fsutil/BUILD @@ -67,7 +67,6 @@ go_library( "frame_ref_set.go", "frame_ref_set_impl.go", "fsutil.go", - "handle.go", "host_file_mapper.go", "host_file_mapper_state.go", "host_file_mapper_unsafe.go", @@ -97,20 +96,6 @@ go_library( ) go_test( - name = "fsutil_x_test", - size = "small", - srcs = ["handle_test.go"], - deps = [ - ":fsutil", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", - "//pkg/sentry/fs", - "//pkg/sentry/fs/ramfs/test", - "//pkg/sentry/usermem", - ], -) - -go_test( name = "fsutil_test", size = "small", srcs = [ diff --git a/pkg/sentry/fs/fsutil/file.go b/pkg/sentry/fs/fsutil/file.go index 46db2e51c..0970f782b 100644 --- a/pkg/sentry/fs/fsutil/file.go +++ b/pkg/sentry/fs/fsutil/file.go @@ -24,12 +24,12 @@ import ( "gvisor.googlesource.com/gvisor/pkg/waiter" ) -// NoopRelease implements FileOperations.Release for files that have no +// FileNoopRelease implements fs.FileOperations.Release for files that have no // resources to release. -type NoopRelease struct{} +type FileNoopRelease struct{} // Release is a no-op. -func (NoopRelease) Release() {} +func (FileNoopRelease) Release() {} // SeekWithDirCursor is used to implement fs.FileOperations.Seek. If dirCursor // is not nil and the seek was on a directory, the cursor will be updated. @@ -127,71 +127,81 @@ func SeekWithDirCursor(ctx context.Context, file *fs.File, whence fs.SeekWhence, return current, syserror.EINVAL } -// GenericSeek implements FileOperations.Seek for files that use a generic -// seek implementation. -type GenericSeek struct{} +// FileGenericSeek implements fs.FileOperations.Seek for files that use a +// generic seek implementation. +type FileGenericSeek struct{} // Seek implements fs.FileOperations.Seek. -func (GenericSeek) Seek(ctx context.Context, file *fs.File, whence fs.SeekWhence, offset int64) (int64, error) { +func (FileGenericSeek) Seek(ctx context.Context, file *fs.File, whence fs.SeekWhence, offset int64) (int64, error) { return SeekWithDirCursor(ctx, file, whence, offset, nil) } -// ZeroSeek implements FileOperations.Seek for files that maintain a constant -// zero-value offset and require a no-op Seek. -type ZeroSeek struct{} +// FileZeroSeek implements fs.FileOperations.Seek for files that maintain a +// constant zero-value offset and require a no-op Seek. +type FileZeroSeek struct{} -// Seek implements FileOperations.Seek. -func (ZeroSeek) Seek(context.Context, *fs.File, fs.SeekWhence, int64) (int64, error) { +// Seek implements fs.FileOperations.Seek. +func (FileZeroSeek) Seek(context.Context, *fs.File, fs.SeekWhence, int64) (int64, error) { return 0, nil } -// PipeSeek implements FileOperations.Seek and can be used for files that behave -// like pipes (seeking is not supported). -type PipeSeek struct{} +// FileNoSeek implements fs.FileOperations.Seek to return EINVAL. +type FileNoSeek struct{} + +// Seek implements fs.FileOperations.Seek. +func (FileNoSeek) Seek(context.Context, *fs.File, fs.SeekWhence, int64) (int64, error) { + return 0, syserror.EINVAL +} -// Seek implements FileOperations.Seek. -func (PipeSeek) Seek(context.Context, *fs.File, fs.SeekWhence, int64) (int64, error) { +// FilePipeSeek implements fs.FileOperations.Seek and can be used for files +// that behave like pipes (seeking is not supported). +type FilePipeSeek struct{} + +// Seek implements fs.FileOperations.Seek. +func (FilePipeSeek) Seek(context.Context, *fs.File, fs.SeekWhence, int64) (int64, error) { return 0, syserror.ESPIPE } -// NotDirReaddir implements FileOperations.Readdir for non-directories. -type NotDirReaddir struct{} +// FileNotDirReaddir implements fs.FileOperations.Readdir for non-directories. +type FileNotDirReaddir struct{} -// Readdir implements FileOperations.NotDirReaddir. -func (NotDirReaddir) Readdir(context.Context, *fs.File, fs.DentrySerializer) (int64, error) { +// Readdir implements fs.FileOperations.FileNotDirReaddir. +func (FileNotDirReaddir) Readdir(context.Context, *fs.File, fs.DentrySerializer) (int64, error) { return 0, syserror.ENOTDIR } -// NoFsync implements FileOperations.Fsync for files that don't support syncing. -type NoFsync struct{} +// FileNoFsync implements fs.FileOperations.Fsync for files that don't support +// syncing. +type FileNoFsync struct{} -// Fsync implements FileOperations.Fsync. -func (NoFsync) Fsync(context.Context, *fs.File, int64, int64, fs.SyncType) error { +// Fsync implements fs.FileOperations.Fsync. +func (FileNoFsync) Fsync(context.Context, *fs.File, int64, int64, fs.SyncType) error { return syserror.EINVAL } -// NoopFsync implements FileOperations.Fsync for files that don't need to synced. -type NoopFsync struct{} +// FileNoopFsync implements fs.FileOperations.Fsync for files that don't need +// to synced. +type FileNoopFsync struct{} -// Fsync implements FileOperations.Fsync. -func (NoopFsync) Fsync(context.Context, *fs.File, int64, int64, fs.SyncType) error { +// Fsync implements fs.FileOperations.Fsync. +func (FileNoopFsync) Fsync(context.Context, *fs.File, int64, int64, fs.SyncType) error { return nil } -// NoopFlush implements FileOperations.Flush as a no-op. -type NoopFlush struct{} +// FileNoopFlush implements fs.FileOperations.Flush as a no-op. +type FileNoopFlush struct{} -// Flush implements FileOperations.Flush. -func (NoopFlush) Flush(context.Context, *fs.File) error { +// Flush implements fs.FileOperations.Flush. +func (FileNoopFlush) Flush(context.Context, *fs.File) error { return nil } -// NoMMap implements fs.FileOperations.Mappable for files that cannot +// FileNoMMap implements fs.FileOperations.Mappable for files that cannot // be memory mapped. -type NoMMap struct{} +type FileNoMMap struct{} // ConfigureMMap implements fs.FileOperations.ConfigureMMap. -func (NoMMap) ConfigureMMap(context.Context, *fs.File, *memmap.MMapOpts) error { +func (FileNoMMap) ConfigureMMap(context.Context, *fs.File, *memmap.MMapOpts) error { return syserror.ENODEV } @@ -204,26 +214,43 @@ func GenericConfigureMMap(file *fs.File, m memmap.Mappable, opts *memmap.MMapOpt return nil } -// NoIoctl implements fs.FileOperations.Ioctl for files that don't implement -// the ioctl syscall. -type NoIoctl struct{} +// FileNoIoctl implements fs.FileOperations.Ioctl for files that don't +// implement the ioctl syscall. +type FileNoIoctl struct{} // Ioctl implements fs.FileOperations.Ioctl. -func (NoIoctl) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { +func (FileNoIoctl) Ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { return 0, syserror.ENOTTY } -// DirFileOperations implements FileOperations for directories. +// DirFileOperations implements most of fs.FileOperations for directories, +// except for Readdir which the embedding type must implement. +type DirFileOperations struct { + waiter.AlwaysReady + FileGenericSeek + FileNoFsync + FileNoIoctl + FileNoMMap + FileNoopFlush + FileNoopRelease +} + +// Read implements fs.FileOperations.Read +func (*DirFileOperations) Read(context.Context, *fs.File, usermem.IOSequence, int64) (int64, error) { + return 0, syserror.EISDIR +} + +// Write implements fs.FileOperations.Write. +func (*DirFileOperations) Write(context.Context, *fs.File, usermem.IOSequence, int64) (int64, error) { + return 0, syserror.EISDIR +} + +// StaticDirFileOperations implements fs.FileOperations for directories with +// static children. // // +stateify savable -type DirFileOperations struct { - waiter.AlwaysReady `state:"nosave"` - NoopRelease `state:"nosave"` - GenericSeek `state:"nosave"` - NoFsync `state:"nosave"` - NoopFlush `state:"nosave"` - NoMMap `state:"nosave"` - NoIoctl `state:"nosave"` +type StaticDirFileOperations struct { + DirFileOperations // dentryMap is a SortedDentryMap used to implement Readdir. dentryMap *fs.SortedDentryMap @@ -233,37 +260,106 @@ type DirFileOperations struct { dirCursor string } -// NewDirFileOperations returns a new DirFileOperations that will iterate the -// given denty map. -func NewDirFileOperations(dentries *fs.SortedDentryMap) *DirFileOperations { - return &DirFileOperations{ +// NewStaticDirFileOperations returns a new StaticDirFileOperations that will +// iterate the given denty map. +func NewStaticDirFileOperations(dentries *fs.SortedDentryMap) *StaticDirFileOperations { + return &StaticDirFileOperations{ dentryMap: dentries, } } // IterateDir implements DirIterator.IterateDir. -func (dfo *DirFileOperations) IterateDir(ctx context.Context, dirCtx *fs.DirCtx, offset int) (int, error) { - n, err := fs.GenericReaddir(dirCtx, dfo.dentryMap) +func (sdfo *StaticDirFileOperations) IterateDir(ctx context.Context, dirCtx *fs.DirCtx, offset int) (int, error) { + n, err := fs.GenericReaddir(dirCtx, sdfo.dentryMap) return offset + n, err } -// Readdir implements FileOperations.Readdir. -func (dfo *DirFileOperations) Readdir(ctx context.Context, file *fs.File, serializer fs.DentrySerializer) (int64, error) { +// Readdir implements fs.FileOperations.Readdir. +func (sdfo *StaticDirFileOperations) Readdir(ctx context.Context, file *fs.File, serializer fs.DentrySerializer) (int64, error) { root := fs.RootFromContext(ctx) defer root.DecRef() dirCtx := &fs.DirCtx{ Serializer: serializer, - DirCursor: &dfo.dirCursor, + DirCursor: &sdfo.dirCursor, } - return fs.DirentReaddir(ctx, file.Dirent, dfo, root, dirCtx, file.Offset()) + return fs.DirentReaddir(ctx, file.Dirent, sdfo, root, dirCtx, file.Offset()) } -// Read implements FileOperations.Read -func (*DirFileOperations) Read(context.Context, *fs.File, usermem.IOSequence, int64) (int64, error) { - return 0, syserror.EISDIR +// NoReadWriteFile is a file that does not support reading or writing. +// +// +stateify savable +type NoReadWriteFile struct { + waiter.AlwaysReady `state:"nosave"` + FileGenericSeek `state:"nosave"` + FileNoIoctl `state:"nosave"` + FileNoMMap `state:"nosave"` + FileNoopFsync `state:"nosave"` + FileNoopFlush `state:"nosave"` + FileNoopRelease `state:"nosave"` + FileNoRead `state:"nosave"` + FileNoWrite `state:"nosave"` + FileNotDirReaddir `state:"nosave"` } -// Write implements FileOperations.Write. -func (*DirFileOperations) Write(context.Context, *fs.File, usermem.IOSequence, int64) (int64, error) { - return 0, syserror.EISDIR +var _ fs.FileOperations = (*NoReadWriteFile)(nil) + +// FileStaticContentReader is a helper to implement fs.FileOperations.Read with +// static content. +// +// +stateify savable +type FileStaticContentReader struct { + // content is immutable. + content []byte +} + +// NewFileStaticContentReader initializes a FileStaticContentReader with the +// given content. +func NewFileStaticContentReader(b []byte) FileStaticContentReader { + return FileStaticContentReader{ + content: b, + } +} + +// Read implements fs.FileOperations.Read. +func (scr *FileStaticContentReader) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { + if offset < 0 { + return 0, syserror.EINVAL + } + if offset >= int64(len(scr.content)) { + return 0, nil + } + n, err := dst.CopyOut(ctx, scr.content[offset:]) + return int64(n), err +} + +// FileNoopWrite implements fs.FileOperations.Write as a noop. +type FileNoopWrite struct{} + +// Write implements fs.FileOperations.Write. +func (FileNoopWrite) Write(_ context.Context, _ *fs.File, src usermem.IOSequence, _ int64) (int64, error) { + return src.NumBytes(), nil +} + +// FileNoRead implements fs.FileOperations.Read to return EINVAL. +type FileNoRead struct{} + +// Read implements fs.FileOperations.Read. +func (FileNoRead) Read(context.Context, *fs.File, usermem.IOSequence, int64) (int64, error) { + return 0, syserror.EINVAL +} + +// FileNoWrite implements fs.FileOperations.Write to return EINVAL. +type FileNoWrite struct{} + +// Write implements fs.FileOperations.Write. +func (FileNoWrite) Write(context.Context, *fs.File, usermem.IOSequence, int64) (int64, error) { + return 0, syserror.EINVAL +} + +// FileNoopRead implement fs.FileOperations.Read as a noop. +type FileNoopRead struct{} + +// Read implements fs.FileOperations.Read. +func (FileNoopRead) Read(context.Context, *fs.File, usermem.IOSequence, int64) (int64, error) { + return 0, nil } diff --git a/pkg/sentry/fs/fsutil/fsutil.go b/pkg/sentry/fs/fsutil/fsutil.go index 3d7f3732d..319c4841b 100644 --- a/pkg/sentry/fs/fsutil/fsutil.go +++ b/pkg/sentry/fs/fsutil/fsutil.go @@ -20,7 +20,5 @@ // - For fs.Inodes that require a page cache to be memory mapped, see // inode_cache.go. // -// - For fs.Files that implement fs.HandleOps, see handle.go. -// // - For anon fs.Inodes, see anon.go. package fsutil diff --git a/pkg/sentry/fs/fsutil/handle.go b/pkg/sentry/fs/fsutil/handle.go deleted file mode 100644 index 8920b72ee..000000000 --- a/pkg/sentry/fs/fsutil/handle.go +++ /dev/null @@ -1,128 +0,0 @@ -// Copyright 2018 Google LLC -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fsutil - -import ( - "gvisor.googlesource.com/gvisor/pkg/sentry/context" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - "gvisor.googlesource.com/gvisor/pkg/sentry/memmap" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" - "gvisor.googlesource.com/gvisor/pkg/syserror" - "gvisor.googlesource.com/gvisor/pkg/waiter" -) - -// Handle implements FileOperations. -// -// FIXME: Remove Handle entirely in favor of individual fs.File -// implementations using simple generic utilities. -// -// +stateify savable -type Handle struct { - NoopRelease `state:"nosave"` - NoIoctl `state:"nosave"` - HandleOperations fs.HandleOperations - - // dirCursor is the directory cursor. - dirCursor string -} - -// NewHandle returns a File backed by the Dirent and FileFlags. -func NewHandle(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags, hops fs.HandleOperations) *fs.File { - if !fs.IsPipe(dirent.Inode.StableAttr) && !fs.IsSocket(dirent.Inode.StableAttr) { - // Allow reading/writing at an arbitrary offset for non-pipes - // and non-sockets. - flags.Pread = true - flags.Pwrite = true - } - - return fs.NewFile(ctx, dirent, flags, &Handle{HandleOperations: hops}) -} - -// Readiness implements waiter.Waitable.Readiness. -func (h *Handle) Readiness(mask waiter.EventMask) waiter.EventMask { - return h.HandleOperations.Readiness(mask) -} - -// EventRegister implements waiter.Waitable.EventRegister. -func (h *Handle) EventRegister(e *waiter.Entry, mask waiter.EventMask) { - h.HandleOperations.EventRegister(e, mask) -} - -// EventUnregister implements waiter.Waitable.EventUnregister. -func (h *Handle) EventUnregister(e *waiter.Entry) { - h.HandleOperations.EventUnregister(e) -} - -// Readdir implements FileOperations.Readdir. -func (h *Handle) Readdir(ctx context.Context, file *fs.File, serializer fs.DentrySerializer) (int64, error) { - root := fs.RootFromContext(ctx) - defer root.DecRef() - dirCtx := &fs.DirCtx{ - Serializer: serializer, - DirCursor: &h.dirCursor, - } - n, err := fs.DirentReaddir(ctx, file.Dirent, h, root, dirCtx, file.Offset()) - return n, err -} - -// Seek implements FileOperations.Seek. -func (h *Handle) Seek(ctx context.Context, file *fs.File, whence fs.SeekWhence, offset int64) (int64, error) { - return SeekWithDirCursor(ctx, file, whence, offset, &h.dirCursor) -} - -// IterateDir implements DirIterator.IterateDir. -func (h *Handle) IterateDir(ctx context.Context, dirCtx *fs.DirCtx, offset int) (int, error) { - return h.HandleOperations.DeprecatedReaddir(ctx, dirCtx, offset) -} - -// Read implements FileOperations.Read. -func (h *Handle) Read(ctx context.Context, file *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { - return h.HandleOperations.DeprecatedPreadv(ctx, dst, offset) -} - -// Write implements FileOperations.Write. -func (h *Handle) Write(ctx context.Context, file *fs.File, src usermem.IOSequence, offset int64) (int64, error) { - return h.HandleOperations.DeprecatedPwritev(ctx, src, offset) -} - -// Fsync implements FileOperations.Fsync. -func (h *Handle) Fsync(ctx context.Context, file *fs.File, start int64, end int64, syncType fs.SyncType) error { - switch syncType { - case fs.SyncAll, fs.SyncData: - // Write out metadata. - if err := file.Dirent.Inode.WriteOut(ctx); err != nil { - return err - } - fallthrough - case fs.SyncBackingStorage: - // Use DeprecatedFsync to sync disks. - return h.HandleOperations.DeprecatedFsync() - } - panic("invalid sync type") -} - -// Flush implements FileOperations.Flush. -func (h *Handle) Flush(context.Context, *fs.File) error { - return h.HandleOperations.DeprecatedFlush() -} - -// ConfigureMMap implements FileOperations.ConfigureMMap. -func (h *Handle) ConfigureMMap(ctx context.Context, file *fs.File, opts *memmap.MMapOpts) error { - mappable := file.Dirent.Inode.Mappable() - if mappable == nil { - return syserror.ENODEV - } - return GenericConfigureMMap(file, mappable, opts) -} diff --git a/pkg/sentry/fs/fsutil/handle_test.go b/pkg/sentry/fs/fsutil/handle_test.go deleted file mode 100644 index 43e1a3bdf..000000000 --- a/pkg/sentry/fs/fsutil/handle_test.go +++ /dev/null @@ -1,227 +0,0 @@ -// Copyright 2018 Google LLC -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package fsutil_test - -import ( - "io" - "syscall" - "testing" - - "gvisor.googlesource.com/gvisor/pkg/sentry/context" - "gvisor.googlesource.com/gvisor/pkg/sentry/context/contexttest" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" - ramfstest "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs/test" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" -) - -type testInodeOperations struct { - fs.InodeOperations - fs.InodeType - FileSize int64 - writes uint - reads uint -} - -func (t *testInodeOperations) UnstableAttr(ctx context.Context, inode *fs.Inode) (fs.UnstableAttr, error) { - return fs.UnstableAttr{Size: t.FileSize}, nil -} - -// Check implements InodeOperations.Check. -func (t *testInodeOperations) Check(ctx context.Context, inode *fs.Inode, p fs.PermMask) bool { - return fs.ContextCanAccessFile(ctx, inode, p) -} - -func (t *testInodeOperations) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { - t.reads++ - return t.InodeOperations.DeprecatedPreadv(ctx, dst, offset) -} - -func (t *testInodeOperations) DeprecatedPwritev(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { - t.writes++ - return t.InodeOperations.DeprecatedPwritev(ctx, src, offset) -} - -// testHandle returns a handle for a test node. -// -// The size of the node is fixed at 20 bytes. -func testHandle(t *testing.T, flags fs.FileFlags, nt fs.InodeType) (*fs.File, *testInodeOperations) { - ctx := contexttest.Context(t) - m := fs.NewNonCachingMountSource(nil, fs.MountSourceFlags{}) - n := &testInodeOperations{ - InodeOperations: ramfstest.NewFile(ctx, fs.FilePermissions{User: fs.PermMask{Read: true, Write: true}}), - FileSize: 20, - } - d := fs.NewDirent(fs.NewInode(n, m, fs.StableAttr{Type: nt}), "test") - return fsutil.NewHandle(ctx, d, flags, d.Inode.HandleOps()), n -} - -func TestHandleOps(t *testing.T) { - h, n := testHandle(t, fs.FileFlags{Read: true, Write: true}, fs.RegularFile) - defer h.DecRef() - - // Make sure a write request works. - if n, err := h.Writev(contexttest.Context(t), usermem.BytesIOSequence([]byte("a"))); n != 1 || err != nil { - t.Fatalf("Writev: got (%d, %v), wanted (1, nil)", n, err) - } - if n.writes != 1 { - t.Errorf("found %d writes, expected 1", n.writes) - } - - // Make sure a read request works. - dst := make([]byte, 1) - if n, err := h.Preadv(contexttest.Context(t), usermem.BytesIOSequence(dst), 0); n != 1 || (err != nil && err != io.EOF) { - t.Errorf("Preadv: got (%d, %v), wanted (1, nil or EOF)", n, err) - } - if dst[0] != 'a' { - t.Errorf("Preadv: read %q, wanted 'a'", dst[0]) - } - if n.reads != 1 { - t.Errorf("found %d reads, expected 1", n.reads) - } -} - -type seekTest struct { - whence fs.SeekWhence - offset int64 - result int64 - err error -} - -type seekSuite struct { - nodeType fs.InodeType - cases []seekTest -} - -// FIXME: This is currently missing fs.SeekEnd tests due to the -// fact that NullInodeOperations returns an error on stat. -func TestHandleSeek(t *testing.T) { - ts := []seekSuite{ - { - nodeType: fs.RegularFile, - cases: []seekTest{ - {fs.SeekSet, 0, 0, nil}, - {fs.SeekSet, 10, 10, nil}, - {fs.SeekSet, -5, 10, syscall.EINVAL}, - {fs.SeekCurrent, -1, 9, nil}, - {fs.SeekCurrent, 2, 11, nil}, - {fs.SeekCurrent, -12, 11, syscall.EINVAL}, - {fs.SeekEnd, -1, 19, nil}, - {fs.SeekEnd, 0, 20, nil}, - {fs.SeekEnd, 2, 22, nil}, - }, - }, - { - nodeType: fs.Directory, - cases: []seekTest{ - {fs.SeekSet, 0, 0, nil}, - {fs.SeekSet, 10, 0, syscall.EINVAL}, - {fs.SeekSet, -5, 0, syscall.EINVAL}, - {fs.SeekCurrent, 0, 0, nil}, - {fs.SeekCurrent, 11, 0, syscall.EINVAL}, - {fs.SeekCurrent, -6, 0, syscall.EINVAL}, - {fs.SeekEnd, 0, 0, syscall.EINVAL}, - {fs.SeekEnd, -1, 0, syscall.EINVAL}, - {fs.SeekEnd, 2, 0, syscall.EINVAL}, - }, - }, - { - nodeType: fs.Symlink, - cases: []seekTest{ - {fs.SeekSet, 5, 0, syscall.EINVAL}, - {fs.SeekSet, -5, 0, syscall.EINVAL}, - {fs.SeekSet, 0, 0, syscall.EINVAL}, - {fs.SeekCurrent, 5, 0, syscall.EINVAL}, - {fs.SeekCurrent, -5, 0, syscall.EINVAL}, - {fs.SeekCurrent, 0, 0, syscall.EINVAL}, - {fs.SeekEnd, 5, 0, syscall.EINVAL}, - {fs.SeekEnd, -5, 0, syscall.EINVAL}, - {fs.SeekEnd, 0, 0, syscall.EINVAL}, - }, - }, - { - nodeType: fs.Pipe, - cases: []seekTest{ - {fs.SeekSet, 5, 0, syscall.ESPIPE}, - {fs.SeekSet, -5, 0, syscall.ESPIPE}, - {fs.SeekSet, 0, 0, syscall.ESPIPE}, - {fs.SeekCurrent, 5, 0, syscall.ESPIPE}, - {fs.SeekCurrent, -5, 0, syscall.ESPIPE}, - {fs.SeekCurrent, 0, 0, syscall.ESPIPE}, - {fs.SeekEnd, 5, 0, syscall.ESPIPE}, - {fs.SeekEnd, -5, 0, syscall.ESPIPE}, - {fs.SeekEnd, 0, 0, syscall.ESPIPE}, - }, - }, - { - nodeType: fs.Socket, - cases: []seekTest{ - {fs.SeekSet, 5, 0, syscall.ESPIPE}, - {fs.SeekSet, -5, 0, syscall.ESPIPE}, - {fs.SeekSet, 0, 0, syscall.ESPIPE}, - {fs.SeekCurrent, 5, 0, syscall.ESPIPE}, - {fs.SeekCurrent, -5, 0, syscall.ESPIPE}, - {fs.SeekCurrent, 0, 0, syscall.ESPIPE}, - {fs.SeekEnd, 5, 0, syscall.ESPIPE}, - {fs.SeekEnd, -5, 0, syscall.ESPIPE}, - {fs.SeekEnd, 0, 0, syscall.ESPIPE}, - }, - }, - { - nodeType: fs.CharacterDevice, - cases: []seekTest{ - {fs.SeekSet, 5, 0, nil}, - {fs.SeekSet, -5, 0, nil}, - {fs.SeekSet, 0, 0, nil}, - {fs.SeekCurrent, 5, 0, nil}, - {fs.SeekCurrent, -5, 0, nil}, - {fs.SeekCurrent, 0, 0, nil}, - {fs.SeekEnd, 5, 0, nil}, - {fs.SeekEnd, -5, 0, nil}, - {fs.SeekEnd, 0, 0, nil}, - }, - }, - { - nodeType: fs.BlockDevice, - cases: []seekTest{ - {fs.SeekSet, 0, 0, nil}, - {fs.SeekSet, 10, 10, nil}, - {fs.SeekSet, -5, 10, syscall.EINVAL}, - {fs.SeekCurrent, -1, 9, nil}, - {fs.SeekCurrent, 2, 11, nil}, - {fs.SeekCurrent, -12, 11, syscall.EINVAL}, - {fs.SeekEnd, -1, 19, nil}, - {fs.SeekEnd, 0, 20, nil}, - {fs.SeekEnd, 2, 22, nil}, - }, - }, - } - - for _, s := range ts { - h, _ := testHandle(t, fs.FileFlags{Read: true, Write: true}, s.nodeType) - defer h.DecRef() - - for _, c := range s.cases { - // Try the given seek. - offset, err := h.Seek(contexttest.Context(t), c.whence, c.offset) - if err != c.err { - t.Errorf("seek(%s, %d) on %s had unexpected error: expected %v, got %v", c.whence, c.offset, s.nodeType, c.err, err) - } - if err == nil && offset != c.result { - t.Errorf("seek(%s, %d) on %s had bad result: expected %v, got %v", c.whence, c.offset, s.nodeType, c.result, offset) - } - } - } -} diff --git a/pkg/sentry/fs/fsutil/inode.go b/pkg/sentry/fs/fsutil/inode.go index d4db1c2de..f1f5ec1de 100644 --- a/pkg/sentry/fs/fsutil/inode.go +++ b/pkg/sentry/fs/fsutil/inode.go @@ -15,213 +15,270 @@ package fsutil import ( + "sync" + "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" ktime "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/time" "gvisor.googlesource.com/gvisor/pkg/sentry/memmap" "gvisor.googlesource.com/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" "gvisor.googlesource.com/gvisor/pkg/syserror" "gvisor.googlesource.com/gvisor/pkg/waiter" ) -// NewSimpleInodeOperations constructs fs.InodeOperations from InodeSimpleAttributes. -func NewSimpleInodeOperations(i InodeSimpleAttributes) fs.InodeOperations { - return &simpleInodeOperations{InodeSimpleAttributes: i} +// SimpleFileInode is a simple implementation of InodeOperations. +// +// +stateify savable +type SimpleFileInode struct { + InodeGenericChecker `state:"nosave"` + InodeNoExtendedAttributes `state:"nosave"` + InodeNoopRelease `state:"nosave"` + InodeNoopWriteOut `state:"nosave"` + InodeNotDirectory `state:"nosave"` + InodeNotMappable `state:"nosave"` + InodeNotOpenable `state:"nosave"` + InodeNotSocket `state:"nosave"` + InodeNotSymlink `state:"nosave"` + InodeNotTruncatable `state:"nosave"` + InodeNotVirtual `state:"nosave"` + + InodeSimpleAttributes +} + +// NewSimpleFileInode returns a new SimpleFileInode. +func NewSimpleFileInode(ctx context.Context, owner fs.FileOwner, perms fs.FilePermissions, typ uint64) *SimpleFileInode { + return &SimpleFileInode{ + InodeSimpleAttributes: NewInodeSimpleAttributes(ctx, owner, perms, typ), + } } -// simpleInodeOperations is a simple implementation of Inode. +// NoReadWriteFileInode is an implementation of InodeOperations that supports +// opening files that are not readable or writeable. // // +stateify savable -type simpleInodeOperations struct { - DeprecatedFileOperations `state:"nosave"` +type NoReadWriteFileInode struct { + InodeGenericChecker `state:"nosave"` + InodeNoExtendedAttributes `state:"nosave"` + InodeNoopRelease `state:"nosave"` + InodeNoopWriteOut `state:"nosave"` InodeNotDirectory `state:"nosave"` + InodeNotMappable `state:"nosave"` InodeNotSocket `state:"nosave"` - InodeNotRenameable `state:"nosave"` - InodeNotOpenable `state:"nosave"` - InodeNotVirtual `state:"nosave"` InodeNotSymlink `state:"nosave"` - InodeNoExtendedAttributes `state:"nosave"` - NoMappable `state:"nosave"` - NoopWriteOut `state:"nosave"` + InodeNotTruncatable `state:"nosave"` + InodeNotVirtual `state:"nosave"` InodeSimpleAttributes } -// InodeSimpleAttributes implements a subset of the Inode interface. It provides -// read-only access to attributes. +// NewNoReadWriteFileInode returns a new NoReadWriteFileInode. +func NewNoReadWriteFileInode(ctx context.Context, owner fs.FileOwner, perms fs.FilePermissions, typ uint64) *NoReadWriteFileInode { + return &NoReadWriteFileInode{ + InodeSimpleAttributes: NewInodeSimpleAttributes(ctx, owner, perms, typ), + } +} + +// GetFile implements fs.InodeOperations.GetFile. +func (*NoReadWriteFileInode) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &NoReadWriteFile{}), nil +} + +// InodeSimpleAttributes implements methods for updating in-memory unstable +// attributes. // // +stateify savable type InodeSimpleAttributes struct { - // FSType is the filesystem type reported by StatFS. + // FSType is the immutable filesystem type that will be returned by + // StatFS. FSType uint64 - // UAttr are the unstable attributes of the Inode. - UAttr fs.UnstableAttr + // mu protects unstable. + mu sync.RWMutex `state:"nosave"` + Unstable fs.UnstableAttr } -// Release implements fs.InodeOperations.Release. -func (i *InodeSimpleAttributes) Release(context.Context) {} - -// StatFS implements fs.InodeOperations.StatFS. -func (i *InodeSimpleAttributes) StatFS(context.Context) (fs.Info, error) { - return fs.Info{Type: i.FSType}, nil +// NewInodeSimpleAttributes returns a new InodeSimpleAttributes. +func NewInodeSimpleAttributes(ctx context.Context, owner fs.FileOwner, perms fs.FilePermissions, typ uint64) InodeSimpleAttributes { + return InodeSimpleAttributes{ + FSType: typ, + Unstable: fs.WithCurrentTime(ctx, fs.UnstableAttr{ + Owner: owner, + Perms: perms, + }), + } } // UnstableAttr implements fs.InodeOperations.UnstableAttr. -func (i *InodeSimpleAttributes) UnstableAttr(context.Context, *fs.Inode) (fs.UnstableAttr, error) { - return i.UAttr, nil +func (i *InodeSimpleAttributes) UnstableAttr(ctx context.Context, _ *fs.Inode) (fs.UnstableAttr, error) { + i.mu.RLock() + u := i.Unstable + i.mu.RUnlock() + return u, nil } -// Check implements fs.InodeOperations.Check. -func (i *InodeSimpleAttributes) Check(ctx context.Context, inode *fs.Inode, p fs.PermMask) bool { - return fs.ContextCanAccessFile(ctx, inode, p) +// SetPermissions implements fs.InodeOperations.SetPermissions. +func (i *InodeSimpleAttributes) SetPermissions(ctx context.Context, _ *fs.Inode, p fs.FilePermissions) bool { + i.mu.Lock() + i.Unstable.SetPermissions(ctx, p) + i.mu.Unlock() + return true +} + +// SetOwner implements fs.InodeOperations.SetOwner. +func (i *InodeSimpleAttributes) SetOwner(ctx context.Context, _ *fs.Inode, owner fs.FileOwner) error { + i.mu.Lock() + i.Unstable.SetOwner(ctx, owner) + i.mu.Unlock() + return nil +} + +// SetTimestamps implements fs.InodeOperations.SetTimestamps. +func (i *InodeSimpleAttributes) SetTimestamps(ctx context.Context, _ *fs.Inode, ts fs.TimeSpec) error { + i.mu.Lock() + i.Unstable.SetTimestamps(ctx, ts) + i.mu.Unlock() + return nil } // AddLink implements fs.InodeOperations.AddLink. -func (*InodeSimpleAttributes) AddLink() {} +func (i *InodeSimpleAttributes) AddLink() { + i.mu.Lock() + i.Unstable.Links++ + i.mu.Unlock() +} // DropLink implements fs.InodeOperations.DropLink. -func (*InodeSimpleAttributes) DropLink() {} - -// NotifyStatusChange implements fs.fs.InodeOperations. -func (i *InodeSimpleAttributes) NotifyStatusChange(ctx context.Context) { - i.UAttr.StatusChangeTime = ktime.NowFromContext(ctx) +func (i *InodeSimpleAttributes) DropLink() { + i.mu.Lock() + i.Unstable.Links-- + i.mu.Unlock() } -// SetPermissions implements fs.InodeOperations.SetPermissions. -func (*InodeSimpleAttributes) SetPermissions(context.Context, *fs.Inode, fs.FilePermissions) bool { - return false +// StatFS implements fs.InodeOperations.StatFS. +func (i *InodeSimpleAttributes) StatFS(context.Context) (fs.Info, error) { + if i.FSType == 0 { + return fs.Info{}, syserror.ENOSYS + } + return fs.Info{Type: i.FSType}, nil } -// SetOwner implements fs.InodeOperations.SetOwner. -func (*InodeSimpleAttributes) SetOwner(context.Context, *fs.Inode, fs.FileOwner) error { - return syserror.EINVAL +// NotifyAccess updates the access time. +func (i *InodeSimpleAttributes) NotifyAccess(ctx context.Context) { + i.mu.Lock() + i.Unstable.AccessTime = ktime.NowFromContext(ctx) + i.mu.Unlock() } -// SetTimestamps implements fs.InodeOperations.SetTimestamps. -func (*InodeSimpleAttributes) SetTimestamps(context.Context, *fs.Inode, fs.TimeSpec) error { - return syserror.EINVAL +// NotifyModification updates the modification time. +func (i *InodeSimpleAttributes) NotifyModification(ctx context.Context) { + i.mu.Lock() + i.Unstable.ModificationTime = ktime.NowFromContext(ctx) + i.mu.Unlock() } -// Truncate implements fs.InodeOperations.Truncate. -func (*InodeSimpleAttributes) Truncate(context.Context, *fs.Inode, int64) error { - return syserror.EINVAL +// NotifyStatusChange updates the status change time. +func (i *InodeSimpleAttributes) NotifyStatusChange(ctx context.Context) { + i.mu.Lock() + i.Unstable.StatusChangeTime = ktime.NowFromContext(ctx) + i.mu.Unlock() } -// InMemoryAttributes implements utilities for updating in-memory unstable -// attributes and extended attributes. It is not thread-safe. -// -// Users need not initialize Xattrs to non-nil (it will be initialized -// when the first extended attribute is set. +// InodeSimpleExtendedAttributes implements +// fs.InodeOperations.{Get,Set,List}xattr. // // +stateify savable -type InMemoryAttributes struct { - Unstable fs.UnstableAttr - Xattrs map[string][]byte +type InodeSimpleExtendedAttributes struct { + // mu protects xattrs. + mu sync.RWMutex `state:"nosave"` + xattrs map[string][]byte } -// SetPermissions updates the permissions to p. -func (i *InMemoryAttributes) SetPermissions(ctx context.Context, p fs.FilePermissions) bool { - i.Unstable.Perms = p - i.Unstable.StatusChangeTime = ktime.NowFromContext(ctx) - return true +// Getxattr implements fs.InodeOperations.Getxattr. +func (i *InodeSimpleExtendedAttributes) Getxattr(_ *fs.Inode, name string) ([]byte, error) { + i.mu.RLock() + value, ok := i.xattrs[name] + i.mu.RUnlock() + if !ok { + return nil, syserror.ENOATTR + } + return value, nil } -// SetOwner updates the file owner to owner. -func (i *InMemoryAttributes) SetOwner(ctx context.Context, owner fs.FileOwner) error { - if owner.UID.Ok() { - i.Unstable.Owner.UID = owner.UID - } - if owner.GID.Ok() { - i.Unstable.Owner.GID = owner.GID +// Setxattr implements fs.InodeOperations.Setxattr. +func (i *InodeSimpleExtendedAttributes) Setxattr(_ *fs.Inode, name string, value []byte) error { + i.mu.Lock() + if i.xattrs == nil { + i.xattrs = make(map[string][]byte) } + i.xattrs[name] = value + i.mu.Unlock() return nil } -// SetTimestamps sets the timestamps to ts. -func (i *InMemoryAttributes) SetTimestamps(ctx context.Context, ts fs.TimeSpec) error { - if ts.ATimeOmit && ts.MTimeOmit { - return nil - } - - now := ktime.NowFromContext(ctx) - if !ts.ATimeOmit { - if ts.ATimeSetSystemTime { - i.Unstable.AccessTime = now - } else { - i.Unstable.AccessTime = ts.ATime - } - } - if !ts.MTimeOmit { - if ts.MTimeSetSystemTime { - i.Unstable.ModificationTime = now - } else { - i.Unstable.ModificationTime = ts.MTime - } +// Listxattr implements fs.InodeOperations.Listxattr. +func (i *InodeSimpleExtendedAttributes) Listxattr(_ *fs.Inode) (map[string]struct{}, error) { + i.mu.RLock() + names := make(map[string]struct{}, len(i.xattrs)) + for name := range i.xattrs { + names[name] = struct{}{} } - i.Unstable.StatusChangeTime = now - return nil + i.mu.RUnlock() + return names, nil } -// TouchAccessTime updates access time to the current time. -func (i *InMemoryAttributes) TouchAccessTime(ctx context.Context) { - i.Unstable.AccessTime = ktime.NowFromContext(ctx) -} +// staticFile is a file with static contents. It is returned by +// InodeStaticFileGetter.GetFile. +// +// +stateify savable +type staticFile struct { + waiter.AlwaysReady `state:"nosave"` + FileGenericSeek `state:"nosave"` + FileNoIoctl `state:"nosave"` + FileNoMMap `state:"nosave"` + FileNoopFsync `state:"nosave"` + FileNoopFlush `state:"nosave"` + FileNoopRelease `state:"nosave"` + FileNoopWrite `state:"nosave"` + FileNotDirReaddir `state:"nosave"` -// TouchModificationTime updates modification and status change -// time to the current time. -func (i *InMemoryAttributes) TouchModificationTime(ctx context.Context) { - now := ktime.NowFromContext(ctx) - i.Unstable.ModificationTime = now - i.Unstable.StatusChangeTime = now + FileStaticContentReader } -// TouchStatusChangeTime updates status change time to the current time. -func (i *InMemoryAttributes) TouchStatusChangeTime(ctx context.Context) { - i.Unstable.StatusChangeTime = ktime.NowFromContext(ctx) -} +// InodeNoStatFS implement StatFS by retuning ENOSYS. +type InodeNoStatFS struct{} -// Getxattr returns the extended attribute at name or ENOATTR if -// it isn't set. -func (i *InMemoryAttributes) Getxattr(name string) ([]byte, error) { - if value, ok := i.Xattrs[name]; ok { - return value, nil - } - return nil, syserror.ENOATTR +// StatFS implements fs.InodeOperations.StatFS. +func (InodeNoStatFS) StatFS(context.Context) (fs.Info, error) { + return fs.Info{}, syserror.ENOSYS } -// Setxattr sets the extended attribute at name to value. -func (i *InMemoryAttributes) Setxattr(name string, value []byte) error { - if i.Xattrs == nil { - i.Xattrs = make(map[string][]byte) - } - i.Xattrs[name] = value - return nil +// InodeStaticFileGetter implements GetFile for a file with static contents. +// +// +stateify savable +type InodeStaticFileGetter struct { + Contents []byte } -// Listxattr returns the set of all currently set extended attributes. -func (i *InMemoryAttributes) Listxattr() (map[string]struct{}, error) { - names := make(map[string]struct{}, len(i.Xattrs)) - for name := range i.Xattrs { - names[name] = struct{}{} - } - return names, nil +// GetFile implements fs.InodeOperations.GetFile. +func (i *InodeStaticFileGetter) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &staticFile{ + FileStaticContentReader: NewFileStaticContentReader(i.Contents), + }), nil } -// NoMappable returns a nil memmap.Mappable. -type NoMappable struct{} +// InodeNotMappable returns a nil memmap.Mappable. +type InodeNotMappable struct{} // Mappable implements fs.InodeOperations.Mappable. -func (NoMappable) Mappable(*fs.Inode) memmap.Mappable { +func (InodeNotMappable) Mappable(*fs.Inode) memmap.Mappable { return nil } -// NoopWriteOut is a no-op implementation of Inode.WriteOut. -type NoopWriteOut struct{} +// InodeNoopWriteOut is a no-op implementation of fs.InodeOperations.WriteOut. +type InodeNoopWriteOut struct{} // WriteOut is a no-op. -func (NoopWriteOut) WriteOut(context.Context, *fs.Inode) error { +func (InodeNoopWriteOut) WriteOut(context.Context, *fs.Inode) error { return nil } @@ -273,6 +330,11 @@ func (InodeNotDirectory) RemoveDirectory(context.Context, *fs.Inode, string) err return syserror.ENOTDIR } +// Rename implements fs.FileOperations.Rename. +func (InodeNotDirectory) Rename(context.Context, *fs.Inode, string, *fs.Inode, string) error { + return syserror.EINVAL +} + // InodeNotSocket can be used by Inodes that are not sockets. type InodeNotSocket struct{} @@ -281,7 +343,31 @@ func (InodeNotSocket) BoundEndpoint(*fs.Inode, string) transport.BoundEndpoint { return nil } -// InodeNotRenameable can be used by Inodes that cannot be renamed. +// InodeNotTruncatable can be used by Inodes that cannot be truncated. +type InodeNotTruncatable struct{} + +// Truncate implements fs.InodeOperations.Truncate. +func (InodeNotTruncatable) Truncate(context.Context, *fs.Inode, int64) error { + return syserror.EINVAL +} + +// InodeIsDirTruncate implements fs.InodeOperations.Truncate for directories. +type InodeIsDirTruncate struct{} + +// Truncate implements fs.InodeOperations.Truncate. +func (InodeIsDirTruncate) Truncate(context.Context, *fs.Inode, int64) error { + return syserror.EISDIR +} + +// InodeNoopTruncate implements fs.InodeOperations.Truncate as a noop. +type InodeNoopTruncate struct{} + +// Truncate implements fs.InodeOperations.Truncate. +func (InodeNoopTruncate) Truncate(context.Context, *fs.Inode, int64) error { + return nil +} + +// InodeNotRenameable can be used by Inodes that cannot be truncated. type InodeNotRenameable struct{} // Rename implements fs.InodeOperations.Rename. @@ -305,6 +391,14 @@ func (InodeNotVirtual) IsVirtual() bool { return false } +// InodeVirtual can be used by Inodes that are virtual. +type InodeVirtual struct{} + +// IsVirtual implements fs.InodeOperations.IsVirtual. +func (InodeVirtual) IsVirtual() bool { + return true +} + // InodeNotSymlink can be used by Inodes that are not symlinks. type InodeNotSymlink struct{} @@ -337,50 +431,17 @@ func (InodeNoExtendedAttributes) Listxattr(*fs.Inode) (map[string]struct{}, erro return nil, syserror.EOPNOTSUPP } -// DeprecatedFileOperations panics if any deprecated Inode method is called. -type DeprecatedFileOperations struct{} +// InodeNoopRelease implements fs.InodeOperations.Release as a noop. +type InodeNoopRelease struct{} -// Readiness implements fs.InodeOperations.Waitable.Readiness. -func (DeprecatedFileOperations) Readiness(waiter.EventMask) waiter.EventMask { - panic("not implemented") -} - -// EventRegister implements fs.InodeOperations.Waitable.EventRegister. -func (DeprecatedFileOperations) EventRegister(*waiter.Entry, waiter.EventMask) { - panic("not implemented") -} - -// EventUnregister implements fs.InodeOperations.Waitable.EventUnregister. -func (DeprecatedFileOperations) EventUnregister(*waiter.Entry) { - panic("not implemented") -} - -// DeprecatedPreadv implements fs.InodeOperations.DeprecatedPreadv. -func (DeprecatedFileOperations) DeprecatedPreadv(context.Context, usermem.IOSequence, int64) (int64, error) { - panic("not implemented") -} - -// DeprecatedPwritev implements fs.InodeOperations.DeprecatedPwritev. -func (DeprecatedFileOperations) DeprecatedPwritev(context.Context, usermem.IOSequence, int64) (int64, error) { - panic("not implemented") -} - -// DeprecatedReaddir implements fs.InodeOperations.DeprecatedReaddir. -func (DeprecatedFileOperations) DeprecatedReaddir(context.Context, *fs.DirCtx, int) (int, error) { - panic("not implemented") -} - -// DeprecatedFsync implements fs.InodeOperations.DeprecatedFsync. -func (DeprecatedFileOperations) DeprecatedFsync() error { - panic("not implemented") -} +// Release implements fs.InodeOperations.Release. +func (InodeNoopRelease) Release(context.Context) {} -// DeprecatedFlush implements fs.InodeOperations.DeprecatedFlush. -func (DeprecatedFileOperations) DeprecatedFlush() error { - panic("not implemented") -} +// InodeGenericChecker implements fs.InodeOperations.Check with a generic +// implementation. +type InodeGenericChecker struct{} -// DeprecatedMappable implements fs.InodeOperations.DeprecatedMappable. -func (DeprecatedFileOperations) DeprecatedMappable(context.Context, *fs.Inode) (memmap.Mappable, bool) { - panic("not implemented") +// Check implements fs.InodeOperations.Check. +func (InodeGenericChecker) Check(ctx context.Context, inode *fs.Inode, p fs.PermMask) bool { + return fs.ContextCanAccessFile(ctx, inode, p) } diff --git a/pkg/sentry/fs/fsutil/inode_cached_test.go b/pkg/sentry/fs/fsutil/inode_cached_test.go index ce5201a40..9c9391511 100644 --- a/pkg/sentry/fs/fsutil/inode_cached_test.go +++ b/pkg/sentry/fs/fsutil/inode_cached_test.go @@ -261,15 +261,11 @@ func (noopMappingSpace) Invalidate(ar usermem.AddrRange, opts memmap.InvalidateO } func anonInode(ctx context.Context) *fs.Inode { - return fs.NewInode(NewSimpleInodeOperations(InodeSimpleAttributes{ - UAttr: fs.WithCurrentTime(ctx, fs.UnstableAttr{ - Owner: fs.FileOwnerFromContext(ctx), - Perms: fs.FilePermissions{ - User: fs.PermMask{Read: true, Write: true}, - }, - Links: 1, - }), - }), fs.NewNonCachingMountSource(nil, fs.MountSourceFlags{}), fs.StableAttr{ + return fs.NewInode(&SimpleFileInode{ + InodeSimpleAttributes: NewInodeSimpleAttributes(ctx, fs.FileOwnerFromContext(ctx), fs.FilePermissions{ + User: fs.PermMask{Read: true, Write: true}, + }, 0), + }, fs.NewPseudoMountSource(), fs.StableAttr{ Type: fs.Anonymous, BlockSize: usermem.PageSize, }) diff --git a/pkg/sentry/fs/gofer/file.go b/pkg/sentry/fs/gofer/file.go index 6d961813d..3578b07a0 100644 --- a/pkg/sentry/fs/gofer/file.go +++ b/pkg/sentry/fs/gofer/file.go @@ -37,7 +37,7 @@ var openedWX = metric.MustCreateNewUint64Metric("/gofer/opened_write_execute_fil // // +stateify savable type fileOperations struct { - fsutil.NoIoctl `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` waiter.AlwaysReady `state:"nosave"` // inodeOperations is the inodeOperations backing the file. It is protected diff --git a/pkg/sentry/fs/gofer/fs.go b/pkg/sentry/fs/gofer/fs.go index ed30cb1f1..2dc000c6f 100644 --- a/pkg/sentry/fs/gofer/fs.go +++ b/pkg/sentry/fs/gofer/fs.go @@ -75,11 +75,11 @@ var ( // ErrNoTransport is returned when there is no 'trans' option. ErrNoTransport = errors.New("missing required option: 'trans='") - // ErrNoReadFD is returned when there is no 'rfdno' option. - ErrNoReadFD = errors.New("missing required option: 'rfdno='") + // ErrFileNoReadFD is returned when there is no 'rfdno' option. + ErrFileNoReadFD = errors.New("missing required option: 'rfdno='") - // ErrNoWriteFD is returned when there is no 'wfdno' option. - ErrNoWriteFD = errors.New("missing required option: 'wfdno='") + // ErrFileNoWriteFD is returned when there is no 'wfdno' option. + ErrFileNoWriteFD = errors.New("missing required option: 'wfdno='") ) // filesystem is a 9p client. @@ -87,6 +87,8 @@ var ( // +stateify savable type filesystem struct{} +var _ fs.Filesystem = (*filesystem)(nil) + func init() { fs.RegisterFilesystem(&filesystem{}) } @@ -160,14 +162,14 @@ func options(data string) (opts, error) { // Check for the required 'rfdno=' option. srfd, ok := options[readFDKey] if !ok { - return o, ErrNoReadFD + return o, ErrFileNoReadFD } delete(options, readFDKey) // Check for the required 'wfdno=' option. swfd, ok := options[writeFDKey] if !ok { - return o, ErrNoWriteFD + return o, ErrFileNoWriteFD } delete(options, writeFDKey) diff --git a/pkg/sentry/fs/gofer/inode.go b/pkg/sentry/fs/gofer/inode.go index 7c6e5b025..f0dc99fd0 100644 --- a/pkg/sentry/fs/gofer/inode.go +++ b/pkg/sentry/fs/gofer/inode.go @@ -40,7 +40,6 @@ import ( type inodeOperations struct { fsutil.InodeNotVirtual `state:"nosave"` fsutil.InodeNoExtendedAttributes `state:"nosave"` - fsutil.DeprecatedFileOperations `state:"nosave"` // fileState implements fs.CachedFileObject. It exists // to break a circular load dependency between inodeOperations diff --git a/pkg/sentry/fs/host/file.go b/pkg/sentry/fs/host/file.go index bc6ee7aa4..4e84d1d6c 100644 --- a/pkg/sentry/fs/host/file.go +++ b/pkg/sentry/fs/host/file.go @@ -36,8 +36,8 @@ import ( // // +stateify savable type fileOperations struct { - fsutil.NoIoctl `state:"nosave"` - fsutil.NoopRelease `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` // iops are the Inode operations for this file. iops *inodeOperations `state:"wait"` diff --git a/pkg/sentry/fs/host/fs.go b/pkg/sentry/fs/host/fs.go index 54cbb94f9..d2ba38449 100644 --- a/pkg/sentry/fs/host/fs.go +++ b/pkg/sentry/fs/host/fs.go @@ -58,6 +58,8 @@ type Filesystem struct { paths []string } +var _ fs.Filesystem = (*Filesystem)(nil) + // Name is the identifier of this file system. func (*Filesystem) Name() string { return FilesystemName diff --git a/pkg/sentry/fs/host/inode.go b/pkg/sentry/fs/host/inode.go index 08754bd6b..6ff6c3254 100644 --- a/pkg/sentry/fs/host/inode.go +++ b/pkg/sentry/fs/host/inode.go @@ -39,7 +39,6 @@ import ( type inodeOperations struct { fsutil.InodeNotVirtual `state:"nosave"` fsutil.InodeNoExtendedAttributes `state:"nosave"` - fsutil.DeprecatedFileOperations `state:"nosave"` // fileState implements fs.CachedFileObject. It exists // to break a circular load dependency between inodeOperations diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go index fa3beb111..d32f52d55 100644 --- a/pkg/sentry/fs/inode.go +++ b/pkg/sentry/fs/inode.go @@ -356,11 +356,10 @@ func (i *Inode) AddLink() { if i.overlay != nil { // FIXME: Remove this from InodeOperations altogether. // - // This interface (including DropLink and NotifyStatusChange) - // is only used by ramfs to update metadata of children. These - // filesystems should _never_ have overlay Inodes cached as - // children. So explicitly disallow this scenario and avoid plumbing - // Dirents through to do copy up. + // 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 + // scenario and avoid plumbing Dirents through to do copy up. panic("overlay Inodes cached in ramfs directories are not supported") } i.InodeOperations.AddLink() @@ -375,15 +374,6 @@ func (i *Inode) DropLink() { i.InodeOperations.DropLink() } -// NotifyStatusChange calls i.InodeOperations.NotifyStatusChange. -func (i *Inode) NotifyStatusChange(ctx context.Context) { - if i.overlay != nil { - // Same as AddLink. - panic("overlay Inodes cached in ramfs directories are not supported") - } - i.InodeOperations.NotifyStatusChange(ctx) -} - // IsVirtual calls i.InodeOperations.IsVirtual. func (i *Inode) IsVirtual() bool { if i.overlay != nil { @@ -401,17 +391,6 @@ func (i *Inode) StatFS(ctx context.Context) (Info, error) { return i.InodeOperations.StatFS(ctx) } -// HandleOps extracts HandleOperations from i. -func (i *Inode) HandleOps() HandleOperations { - if i.overlay != nil { - return overlayHandleOps(i.overlay) - } - if h, ok := i.InodeOperations.(HandleOperations); ok { - return h - } - return nil -} - // CheckOwnership checks whether `ctx` owns this Inode or may act as its owner. // Compare Linux's fs/inode.c:inode_owner_or_capable(). func (i *Inode) CheckOwnership(ctx context.Context) bool { diff --git a/pkg/sentry/fs/inode_operations.go b/pkg/sentry/fs/inode_operations.go index 77973ce79..db40b5256 100644 --- a/pkg/sentry/fs/inode_operations.go +++ b/pkg/sentry/fs/inode_operations.go @@ -21,8 +21,6 @@ import ( ktime "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/time" "gvisor.googlesource.com/gvisor/pkg/sentry/memmap" "gvisor.googlesource.com/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" - "gvisor.googlesource.com/gvisor/pkg/waiter" ) var ( @@ -303,83 +301,5 @@ type InodeOperations interface { // StatFS returns a filesystem Info implementation or an error. If // the filesystem does not support this operation (maybe in the future // it will), then ENOSYS should be returned. - // - // Move to MountSourceOperations. StatFS(context.Context) (Info, error) - - HandleOperations -} - -// HandleOperations are extended InodeOperations that are only implemented -// for file systems that use fs/handle.go:Handle to generate open Files. -// -// Handle is deprecated; these methods are deprecated as well. -// -// Filesystems are encouraged to implement the File interface directly -// instead of using Handle. To indicate that the below methods should never -// be called, embed DeprecatedFileOperations to satisfy this interface. -type HandleOperations interface { - waiter.Waitable - - // DeprecatedPreadv is deprecated in favor of filesystems - // implementing File.Preadv directly. - // - // DeprecatedPreadv reads up to dst.NumBytes() bytes into dst, starting at - // the given offset, and returns the number of bytes read. - // - // Preadv may return a partial read result before EOF is reached. - // - // If a symlink, Preadv reads the target value of the symlink. - // - // Preadv should not check for readable permissions. - DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) - - // DeprecatedPwritev is deprecated in favor of filesystems - // implementing File.Pwritev directly. - // - // DeprecatedPwritev writes up to src.NumBytes() bytes from src to the - // Inode, starting at the given offset and returns the number of bytes - // written. - // - // Pwritev should not check that the Inode has writable permissions. - DeprecatedPwritev(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) - - // DeprecatedReaddir is deprecated in favor of filesystems - // implementing File.Readdir directly. - // - // DeprecatedReaddir emits directory entries by calling dirCtx.EmitDir, - // beginning with the entry at offset. - // - // Entries for "." and ".." must *not* be included. - // - // If the offset returned is the same as the argument offset, then - // nothing has been serialized. This is equivalent to reaching EOF. - // In this case serializer.Written() should return 0. - // - // The order of entries to emit must be consistent between Readdir - // calls, and must start with the given offset. - // - // The caller must ensure that this operation is permitted. - DeprecatedReaddir(ctx context.Context, dirCtx *DirCtx, offset int) (int, error) - - // DeprecatedFsync is deprecated in favor of filesystems implementing - // File.Fsync directly. - // - // DeprecatedFsync syncs a file. - DeprecatedFsync() error - - // DeprecatedMappable is deprecated in favor of filesystems implementing - // File.Mappable directly. - // - // DeprecatedMappable returns a Mappable if the Inode can be mapped. - DeprecatedMappable(ctx context.Context, inode *Inode) (memmap.Mappable, bool) - - // DeprecatedFlush is deprecated in favor of filesystems implementing - // File.Flush directly. - // - // DeprecatedFlush flushes a file. - // - // Implementations may choose to free up memory or complete pending I/O - // but also may implement Flush as a no-op. - DeprecatedFlush() error } diff --git a/pkg/sentry/fs/inode_overlay.go b/pkg/sentry/fs/inode_overlay.go index 512a0da28..77a2623ef 100644 --- a/pkg/sentry/fs/inode_overlay.go +++ b/pkg/sentry/fs/inode_overlay.go @@ -596,19 +596,6 @@ func overlayStatFS(ctx context.Context, o *overlayEntry) (Info, error) { return i, nil } -func overlayHandleOps(o *overlayEntry) HandleOperations { - // Hot path. Avoid defers. - var hops HandleOperations - o.copyMu.RLock() - if o.upper != nil { - hops = o.upper.HandleOps() - } else { - hops = o.lower.HandleOps() - } - o.copyMu.RUnlock() - return hops -} - // NewTestOverlayDir returns an overlay Inode for tests. // // If `revalidate` is true, then the upper filesystem will require diff --git a/pkg/sentry/fs/inode_overlay_test.go b/pkg/sentry/fs/inode_overlay_test.go index 9e922d008..bc91be226 100644 --- a/pkg/sentry/fs/inode_overlay_test.go +++ b/pkg/sentry/fs/inode_overlay_test.go @@ -19,7 +19,8 @@ import ( "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - ramfstest "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs/test" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/contexttest" "gvisor.googlesource.com/gvisor/pkg/syserror" ) @@ -376,7 +377,8 @@ type dir struct { // List of negative child names. negative []string - // Whether DeprecatedReaddir has been called on this dir. + // ReaddirCalled records whether Readdir was called on a file + // corresponding to this inode. ReaddirCalled bool } @@ -390,10 +392,19 @@ func (d *dir) Getxattr(inode *fs.Inode, name string) ([]byte, error) { return nil, syserror.ENOATTR } -// DeprecatedReaddir implements InodeOperations.DeprecatedReaddir. -func (d *dir) DeprecatedReaddir(ctx context.Context, dirctx *fs.DirCtx, offset int) (int, error) { - d.ReaddirCalled = true - return d.InodeOperations.DeprecatedReaddir(ctx, dirctx, offset) +// GetFile implements InodeOperations.GetFile. +func (d *dir) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + file, err := d.InodeOperations.GetFile(ctx, dirent, flags) + if err != nil { + return nil, err + } + defer file.DecRef() + // Wrap the file's FileOperations in a dirFile. + fops := &dirFile{ + FileOperations: file.FileOperations, + inode: d, + } + return fs.NewFile(ctx, dirent, flags, fops), nil } type dirContent struct { @@ -401,12 +412,45 @@ type dirContent struct { dir bool } +type dirFile struct { + fs.FileOperations + inode *dir +} + +type inode struct { + fsutil.InodeGenericChecker `state:"nosave"` + fsutil.InodeNoExtendedAttributes `state:"nosave"` + fsutil.InodeNoopRelease `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` + fsutil.InodeNotDirectory `state:"nosave"` + fsutil.InodeNotMappable `state:"nosave"` + fsutil.InodeNotSocket `state:"nosave"` + fsutil.InodeNotSymlink `state:"nosave"` + fsutil.InodeNotTruncatable `state:"nosave"` + fsutil.InodeNotVirtual `state:"nosave"` + + fsutil.InodeSimpleAttributes + fsutil.InodeStaticFileGetter +} + +// Readdir implements fs.FileOperations.Readdir. It sets the ReaddirCalled +// field on the inode. +func (f *dirFile) Readdir(ctx context.Context, file *fs.File, ser fs.DentrySerializer) (int64, error) { + f.inode.ReaddirCalled = true + return f.FileOperations.Readdir(ctx, file, ser) +} + func newTestRamfsInode(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - return fs.NewInode(ramfstest.NewFile(ctx, fs.FilePermissions{}), msrc, fs.StableAttr{Type: fs.RegularFile}) + inode := fs.NewInode(&inode{ + InodeStaticFileGetter: fsutil.InodeStaticFileGetter{ + Contents: []byte("foobar"), + }, + }, msrc, fs.StableAttr{Type: fs.RegularFile}) + return inode } func newTestRamfsDir(ctx context.Context, contains []dirContent, negative []string) *fs.Inode { - msrc := fs.NewCachingMountSource(nil, fs.MountSourceFlags{}) + msrc := fs.NewPseudoMountSource() contents := make(map[string]*fs.Inode) for _, c := range contains { if c.dir { @@ -415,7 +459,7 @@ func newTestRamfsDir(ctx context.Context, contains []dirContent, negative []stri contents[c.name] = newTestRamfsInode(ctx, msrc) } } - dops := ramfstest.NewDir(ctx, contents, fs.FilePermissions{ + dops := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermissions{ User: fs.PermMask{Read: true, Execute: true}, }) return fs.NewInode(&dir{ diff --git a/pkg/sentry/fs/mock.go b/pkg/sentry/fs/mock.go index 6bfcda6bb..abfdc6a25 100644 --- a/pkg/sentry/fs/mock.go +++ b/pkg/sentry/fs/mock.go @@ -16,7 +16,6 @@ package fs import ( "gvisor.googlesource.com/gvisor/pkg/sentry/context" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" "gvisor.googlesource.com/gvisor/pkg/syserror" ) @@ -151,16 +150,6 @@ func (n *MockInodeOperations) Truncate(ctx context.Context, inode *Inode, size i return nil } -// DeprecatedPwritev implements fs.InodeOperations.DeprecatedPwritev. -func (n *MockInodeOperations) DeprecatedPwritev(context.Context, usermem.IOSequence, int64) (int64, error) { - return 0, nil -} - -// DeprecatedReaddir implements fs.InodeOperations.DeprecatedReaddir. -func (n *MockInodeOperations) DeprecatedReaddir(context.Context, *DirCtx, int) (int, error) { - return 0, nil -} - // Remove implements fs.InodeOperations.Remove. func (n *MockInodeOperations) Remove(context.Context, *Inode, string) error { return nil diff --git a/pkg/sentry/fs/mount.go b/pkg/sentry/fs/mount.go index 24e28ddb2..dd6e64b4c 100644 --- a/pkg/sentry/fs/mount.go +++ b/pkg/sentry/fs/mount.go @@ -43,8 +43,6 @@ type DirentOperations interface { // MountSourceOperations contains filesystem specific operations. type MountSourceOperations interface { // TODO: Add: - // - // StatFS() (Info, error) // BlockSize() int64 // FS() Filesystem @@ -249,7 +247,7 @@ func (msrc *MountSource) FlushDirentRefs() { } // NewCachingMountSource returns a generic mount that will cache dirents -// aggressively. Filesystem may be nil if there is no backing filesystem. +// aggressively. func NewCachingMountSource(filesystem Filesystem, flags MountSourceFlags) *MountSource { return NewMountSource(&SimpleMountSourceOperations{ keep: true, @@ -258,7 +256,6 @@ func NewCachingMountSource(filesystem Filesystem, flags MountSourceFlags) *Mount } // NewNonCachingMountSource returns a generic mount that will never cache dirents. -// Filesystem may be nil if there is no backing filesystem. func NewNonCachingMountSource(filesystem Filesystem, flags MountSourceFlags) *MountSource { return NewMountSource(&SimpleMountSourceOperations{ keep: false, @@ -275,6 +272,15 @@ func NewRevalidatingMountSource(filesystem Filesystem, flags MountSourceFlags) * }, filesystem, flags) } +// NewPseudoMountSource returns a "pseudo" mount source that is not backed by +// an actual filesystem. It is always non-caching. +func NewPseudoMountSource() *MountSource { + return NewMountSource(&SimpleMountSourceOperations{ + keep: false, + revalidate: false, + }, nil, MountSourceFlags{}) +} + // SimpleMountSourceOperations implements MountSourceOperations. // // +stateify savable diff --git a/pkg/sentry/fs/mounts_test.go b/pkg/sentry/fs/mounts_test.go index 7d682d99b..54000614f 100644 --- a/pkg/sentry/fs/mounts_test.go +++ b/pkg/sentry/fs/mounts_test.go @@ -19,7 +19,8 @@ import ( "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - ramfstest "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs/test" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/contexttest" ) @@ -29,15 +30,15 @@ import ( // |-bar (file) func createMountNamespace(ctx context.Context) (*fs.MountNamespace, error) { perms := fs.FilePermsFromMode(0777) - m := fs.NewNonCachingMountSource(nil, fs.MountSourceFlags{}) + m := fs.NewPseudoMountSource() - barFile := ramfstest.NewFile(ctx, perms) - fooDir := ramfstest.NewDir(ctx, map[string]*fs.Inode{ + barFile := fsutil.NewSimpleFileInode(ctx, fs.RootOwner, perms, 0) + fooDir := ramfs.NewDir(ctx, map[string]*fs.Inode{ "bar": fs.NewInode(barFile, m, fs.StableAttr{Type: fs.RegularFile}), - }, perms) - rootDir := ramfstest.NewDir(ctx, map[string]*fs.Inode{ + }, fs.RootOwner, perms) + rootDir := ramfs.NewDir(ctx, map[string]*fs.Inode{ "foo": fs.NewInode(fooDir, m, fs.StableAttr{Type: fs.Directory}), - }, perms) + }, fs.RootOwner, perms) return fs.NewMountNamespace(ctx, fs.NewInode(rootDir, m, fs.StableAttr{Type: fs.Directory})) } diff --git a/pkg/sentry/fs/proc/BUILD b/pkg/sentry/fs/proc/BUILD index aff3c3c01..74954f213 100644 --- a/pkg/sentry/fs/proc/BUILD +++ b/pkg/sentry/fs/proc/BUILD @@ -8,9 +8,9 @@ go_library( "cpuinfo.go", "exec_args.go", "fds.go", - "file.go", "filesystems.go", "fs.go", + "inode.go", "loadavg.go", "meminfo.go", "mounts.go", @@ -32,6 +32,7 @@ go_library( "//pkg/abi/linux", "//pkg/sentry/context", "//pkg/sentry/fs", + "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/proc/device", "//pkg/sentry/fs/proc/seqfile", "//pkg/sentry/fs/ramfs", @@ -45,6 +46,7 @@ go_library( "//pkg/sentry/usage", "//pkg/sentry/usermem", "//pkg/syserror", + "//pkg/waiter", ], ) diff --git a/pkg/sentry/fs/proc/cpuinfo.go b/pkg/sentry/fs/proc/cpuinfo.go index f8be06dc3..f756c45bf 100644 --- a/pkg/sentry/fs/proc/cpuinfo.go +++ b/pkg/sentry/fs/proc/cpuinfo.go @@ -15,52 +15,21 @@ package proc import ( - "io" - "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" ) -// cpuinfo is a file describing the CPU capabilities. -// -// Presently cpuinfo never changes, so it doesn't need to be a SeqFile. -// -// +stateify savable -type cpuinfo struct { - ramfs.Entry - - // k is the system kernel. - k *kernel.Kernel -} - -// DeprecatedPreadv implements fs.InodeOperations.DeprecatedPreadv. -func (c *cpuinfo) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { - features := c.k.FeatureSet() +func newCPUInfo(ctx context.Context, msrc *fs.MountSource) *fs.Inode { + k := kernel.KernelFromContext(ctx) + features := k.FeatureSet() if features == nil { // Kernel is always initialized with a FeatureSet. panic("cpuinfo read with nil FeatureSet") } - contents := make([]byte, 0, 1024) - for i, max := uint(0), c.k.ApplicationCores(); i < max; i++ { + for i, max := uint(0), k.ApplicationCores(); i < max; i++ { contents = append(contents, []byte(features.CPUInfo(i))...) } - if offset >= int64(len(contents)) { - return 0, io.EOF - } - - n, err := dst.CopyOut(ctx, contents[offset:]) - return int64(n), err -} - -func (p *proc) newCPUInfo(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - f := &cpuinfo{ - k: p.k, - } - f.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(0444)) - - return newFile(f, msrc, fs.SpecialFile, nil) + return newStaticProcInode(ctx, msrc, contents) } diff --git a/pkg/sentry/fs/proc/exec_args.go b/pkg/sentry/fs/proc/exec_args.go index b4896053f..ddda67f54 100644 --- a/pkg/sentry/fs/proc/exec_args.go +++ b/pkg/sentry/fs/proc/exec_args.go @@ -18,12 +18,14 @@ import ( "fmt" "io" + "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" "gvisor.googlesource.com/gvisor/pkg/syserror" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) // execArgType enumerates the types of exec arguments that are exposed through @@ -35,12 +37,12 @@ const ( environExecArg ) -// execArgFile is a file containing the exec args (either cmdline or environ) +// execArgInode is a inode containing the exec args (either cmdline or environ) // for a given task. // // +stateify savable -type execArgFile struct { - ramfs.Entry +type execArgInode struct { + fsutil.SimpleFileInode // arg is the type of exec argument this file contains. arg execArgType @@ -49,21 +51,52 @@ type execArgFile struct { t *kernel.Task } +var _ fs.InodeOperations = (*execArgInode)(nil) + // newExecArgFile creates a file containing the exec args of the given type. -func newExecArgFile(t *kernel.Task, msrc *fs.MountSource, arg execArgType) *fs.Inode { +func newExecArgInode(t *kernel.Task, msrc *fs.MountSource, arg execArgType) *fs.Inode { if arg != cmdlineExecArg && arg != environExecArg { panic(fmt.Sprintf("unknown exec arg type %v", arg)) } - f := &execArgFile{ - arg: arg, - t: t, + f := &execArgInode{ + SimpleFileInode: *fsutil.NewSimpleFileInode(t, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + arg: arg, + t: t, } - f.InitEntry(t, fs.RootOwner, fs.FilePermsFromMode(0444)) - return newFile(f, msrc, fs.SpecialFile, t) + return newProcInode(f, msrc, fs.SpecialFile, t) +} + +// GetFile implements fs.InodeOperations.GetFile. +func (i *execArgInode) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &execArgFile{ + arg: i.arg, + t: i.t, + }), nil } -// DeprecatedPreadv reads the exec arg from the process's address space.. -func (f *execArgFile) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { +// +stateify savable +type execArgFile struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopWrite `state:"nosave"` + + // arg is the type of exec argument this file contains. + arg execArgType + + // t is the Task to read the exec arg line from. + t *kernel.Task +} + +var _ fs.FileOperations = (*execArgFile)(nil) + +// Read reads the exec arg from the process's address space.. +func (f *execArgFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { if offset < 0 { return 0, syserror.EINVAL } diff --git a/pkg/sentry/fs/proc/fds.go b/pkg/sentry/fs/proc/fds.go index 5acbce75e..b8a0a5eff 100644 --- a/pkg/sentry/fs/proc/fds.go +++ b/pkg/sentry/fs/proc/fds.go @@ -21,11 +21,11 @@ import ( "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/device" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/kdefs" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" "gvisor.googlesource.com/gvisor/pkg/syserror" ) @@ -55,7 +55,7 @@ func walkDescriptors(t *kernel.Task, p string, toInode func(*fs.File, kernel.FDF // readDescriptors reads fds in the task starting at offset, and calls the // toDentAttr callback for each to get a DentAttr, which it then emits. This is // a helper for implementing fs.InodeOperations.Readdir. -func readDescriptors(t *kernel.Task, c *fs.DirCtx, offset int, toDentAttr func(int) fs.DentAttr) (int, error) { +func readDescriptors(t *kernel.Task, c *fs.DirCtx, offset int64, toDentAttr func(int) fs.DentAttr) (int64, error) { var fds kernel.FDs t.WithMuLocked(func(t *kernel.Task) { if fdm := t.FDMap(); fdm != nil { @@ -69,7 +69,7 @@ func readDescriptors(t *kernel.Task, c *fs.DirCtx, offset int, toDentAttr func(i } // Find the fd to start at. - idx := sort.SearchInts(fdInts, offset) + idx := sort.SearchInts(fdInts, int(offset)) if idx == len(fdInts) { return offset, nil } @@ -80,28 +80,32 @@ func readDescriptors(t *kernel.Task, c *fs.DirCtx, offset int, toDentAttr func(i name := strconv.FormatUint(uint64(fd), 10) if err := c.DirEmit(name, toDentAttr(fd)); err != nil { // Returned offset is the next fd to serialize. - return fd, err + return int64(fd), err } } // We serialized them all. Next offset should be higher than last // serialized fd. - return fd + 1, nil + return int64(fd + 1), nil } -// fd is a single file in /proc/TID/fd/. +// fd implements fs.InodeOperations for a file in /proc/TID/fd/. type fd struct { ramfs.Symlink *fs.File } +var _ fs.InodeOperations = (*fd)(nil) + // newFd returns a new fd based on an existing file. // // This inherits one reference to the file. func newFd(t *kernel.Task, f *fs.File, msrc *fs.MountSource) *fs.Inode { - fd := &fd{File: f} - // RootOwner by default, is overridden in UnstableAttr() - fd.InitSymlink(t, fs.RootOwner, "") - return newFile(fd, msrc, fs.Symlink, t) + fd := &fd{ + // RootOwner overridden by taskOwnedInodeOps.UnstableAttrs(). + Symlink: *ramfs.NewSymlink(t, fs.RootOwner, ""), + File: f, + } + return newProcInode(fd, msrc, fs.Symlink, t) } // GetFile returns the fs.File backing this fd. The dirent and flags @@ -142,7 +146,7 @@ func (f *fd) Close() error { return nil } -// fdDir implements /proc/TID/fd. +// fdDir is an InodeOperations for /proc/TID/fd. // // +stateify savable type fdDir struct { @@ -154,11 +158,15 @@ type fdDir struct { t *kernel.Task } +var _ fs.InodeOperations = (*fdDir)(nil) + // newFdDir creates a new fdDir. func newFdDir(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - f := &fdDir{t: t} - f.InitDir(t, nil, fs.RootOwner, fs.FilePermissions{User: fs.PermMask{Read: true, Execute: true}}) - return newFile(f, msrc, fs.SpecialDirectory, t) + f := &fdDir{ + Dir: *ramfs.NewDir(t, nil, fs.RootOwner, fs.FilePermissions{User: fs.PermMask{Read: true, Execute: true}}), + t: t, + } + return newProcInode(f, msrc, fs.SpecialDirectory, t) } // Check implements InodeOperations.Check. @@ -191,49 +199,55 @@ func (f *fdDir) Lookup(ctx context.Context, dir *fs.Inode, p string) (*fs.Dirent return fs.NewDirent(n, p), nil } -// DeprecatedReaddir lists fds in /proc/TID/fd. -func (f *fdDir) DeprecatedReaddir(ctx context.Context, dirCtx *fs.DirCtx, offset int) (int, error) { - return readDescriptors(f.t, dirCtx, offset, func(fd int) fs.DentAttr { - return fs.GenericDentAttr(fs.Symlink, device.ProcDevice) +// GetFile implements fs.FileOperations.GetFile. +func (f *fdDir) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + fops := &fdDirFile{ + isInfoFile: false, + t: f.t, + } + return fs.NewFile(ctx, dirent, flags, fops), nil +} + +// +stateify savable +type fdDirFile struct { + fsutil.DirFileOperations `state:"nosave"` + + isInfoFile bool + + t *kernel.Task +} + +var _ fs.FileOperations = (*fdDirFile)(nil) + +// Readdir implements fs.FileOperations.Readdir. +func (f *fdDirFile) Readdir(ctx context.Context, file *fs.File, ser fs.DentrySerializer) (int64, error) { + dirCtx := &fs.DirCtx{ + Serializer: ser, + } + typ := fs.RegularFile + if f.isInfoFile { + typ = fs.Symlink + } + return readDescriptors(f.t, dirCtx, file.Offset(), func(fd int) fs.DentAttr { + return fs.GenericDentAttr(typ, device.ProcDevice) }) } -// fdInfo is a single file in /proc/TID/fdinfo/. +// fdInfoInode is a single file in /proc/TID/fdinfo/. // // +stateify savable -type fdInfo struct { - ramfs.File +type fdInfoInode struct { + staticFileInodeOps file *fs.File flags fs.FileFlags fdFlags kernel.FDFlags } -// newFdInfo returns a new fdInfo based on an existing file. -func newFdInfo(t *kernel.Task, file *fs.File, fdFlags kernel.FDFlags, msrc *fs.MountSource) *fs.Inode { - fdi := &fdInfo{file: file, flags: file.Flags(), fdFlags: fdFlags} - fdi.InitFile(t, fs.RootOwner, fs.FilePermissions{User: fs.PermMask{Read: true}}) - // TODO: Get pos, locks, and other data. For now we only - // have flags. - // See https://www.kernel.org/doc/Documentation/filesystems/proc.txt - - flags := file.Flags().ToLinux() | fdFlags.ToLinuxFileFlags() - fdi.Append([]byte(fmt.Sprintf("flags:\t0%o\n", flags))) - return newFile(fdi, msrc, fs.SpecialFile, t) -} - -// DeprecatedPwritev implements fs.HandleOperations.DeprecatedPwritev. -func (*fdInfo) DeprecatedPwritev(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { - return 0, ramfs.ErrInvalidOp -} - -// Truncate implements fs.InodeOperations.Truncate. -func (*fdInfo) Truncate(ctx context.Context, inode *fs.Inode, size int64) error { - return ramfs.ErrInvalidOp -} +var _ fs.InodeOperations = (*fdInfoInode)(nil) -func (f *fdInfo) Release(ctx context.Context) { - f.File.Release(ctx) +// Release implements fs.InodeOperations.Release. +func (f *fdInfoInode) Release(ctx context.Context) { f.file.DecRef() } @@ -249,25 +263,37 @@ type fdInfoDir struct { // newFdInfoDir creates a new fdInfoDir. func newFdInfoDir(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - fdid := &fdInfoDir{t: t} - fdid.InitDir(t, nil, fs.RootOwner, fs.FilePermsFromMode(0500)) - return newFile(fdid, msrc, fs.SpecialDirectory, t) + fdid := &fdInfoDir{ + Dir: *ramfs.NewDir(t, nil, fs.RootOwner, fs.FilePermsFromMode(0500)), + t: t, + } + return newProcInode(fdid, msrc, fs.SpecialDirectory, t) } // Lookup loads an fd in /proc/TID/fdinfo into a Dirent. func (fdid *fdInfoDir) Lookup(ctx context.Context, dir *fs.Inode, p string) (*fs.Dirent, error) { - n, err := walkDescriptors(fdid.t, p, func(file *fs.File, fdFlags kernel.FDFlags) *fs.Inode { - return newFdInfo(fdid.t, file, fdFlags, dir.MountSource) + inode, err := walkDescriptors(fdid.t, p, func(file *fs.File, fdFlags kernel.FDFlags) *fs.Inode { + // TODO: Using a static inode here means that the + // data can be out-of-date if, for instance, the flags on the + // FD change before we read this file. We should switch to + // generating the data on Read(). Also, we should include pos, + // locks, and other data. For now we only have flags. + // See https://www.kernel.org/doc/Documentation/filesystems/proc.txt + flags := file.Flags().ToLinux() | fdFlags.ToLinuxFileFlags() + contents := []byte(fmt.Sprintf("flags:\t0%o\n", flags)) + return newStaticProcInode(ctx, dir.MountSource, contents) }) if err != nil { return nil, err } - return fs.NewDirent(n, p), nil + return fs.NewDirent(inode, p), nil } -// DeprecatedReaddir lists fds in /proc/TID/fdinfo. -func (fdid *fdInfoDir) DeprecatedReaddir(ctx context.Context, dirCtx *fs.DirCtx, offset int) (int, error) { - return readDescriptors(fdid.t, dirCtx, offset, func(fd int) fs.DentAttr { - return fs.GenericDentAttr(fs.RegularFile, device.ProcDevice) - }) +// GetFile implements fs.FileOperations.GetFile. +func (fdid *fdInfoDir) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + fops := &fdDirFile{ + isInfoFile: true, + t: fdid.t, + } + return fs.NewFile(ctx, dirent, flags, fops), nil } diff --git a/pkg/sentry/fs/proc/file.go b/pkg/sentry/fs/proc/file.go deleted file mode 100644 index f659e590a..000000000 --- a/pkg/sentry/fs/proc/file.go +++ /dev/null @@ -1,58 +0,0 @@ -// Copyright 2018 Google LLC -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "gvisor.googlesource.com/gvisor/pkg/sentry/context" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/device" - "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" -) - -// +stateify savable -type file struct { - fs.InodeOperations - - // nodeType is the file type of this file. - nodeType fs.InodeType - - // t is the associated kernel task that owns this file. - t *kernel.Task -} - -func newFile(node fs.InodeOperations, msrc *fs.MountSource, nodeType fs.InodeType, t *kernel.Task) *fs.Inode { - iops := &file{node, nodeType, t} - sattr := fs.StableAttr{ - DeviceID: device.ProcDevice.DeviceID(), - InodeID: device.ProcDevice.NextIno(), - BlockSize: usermem.PageSize, - Type: nodeType, - } - return fs.NewInode(iops, msrc, sattr) -} - -// UnstableAttr returns all attributes of this file. -func (f *file) UnstableAttr(ctx context.Context, inode *fs.Inode) (fs.UnstableAttr, error) { - uattr, err := f.InodeOperations.UnstableAttr(ctx, inode) - if err != nil { - return fs.UnstableAttr{}, err - } - if f.t != nil { - creds := f.t.Credentials() - uattr.Owner = fs.FileOwner{creds.EffectiveKUID, creds.EffectiveKGID} - } - return uattr, nil -} diff --git a/pkg/sentry/fs/proc/inode.go b/pkg/sentry/fs/proc/inode.go new file mode 100644 index 000000000..3c36af5ea --- /dev/null +++ b/pkg/sentry/fs/proc/inode.go @@ -0,0 +1,96 @@ +// Copyright 2018 Google LLC +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package proc + +import ( + "gvisor.googlesource.com/gvisor/pkg/abi/linux" + "gvisor.googlesource.com/gvisor/pkg/sentry/context" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/device" + "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" + "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" +) + +// taskOwnedInodeOps wraps an fs.InodeOperations and overrides the UnstableAttr +// method to return the task as the owner. +// +// +stateify savable +type taskOwnedInodeOps struct { + fs.InodeOperations + + // t is the task that owns this file. + t *kernel.Task +} + +// UnstableAttr implement fs.InodeOperations.UnstableAttr. +func (i *taskOwnedInodeOps) UnstableAttr(ctx context.Context, inode *fs.Inode) (fs.UnstableAttr, error) { + uattr, err := i.InodeOperations.UnstableAttr(ctx, inode) + if err != nil { + return fs.UnstableAttr{}, err + } + // Set the task owner as the file owner. + creds := i.t.Credentials() + uattr.Owner = fs.FileOwner{creds.EffectiveKUID, creds.EffectiveKGID} + return uattr, nil +} + +// staticFileInodeOps is an InodeOperations implementation that can be used to +// return file contents which are constant. This file is not writable and will +// always have mode 0444. +// +// +stateify savable +type staticFileInodeOps struct { + fsutil.InodeGenericChecker `state:"nosave"` + fsutil.InodeNoExtendedAttributes `state:"nosave"` + fsutil.InodeNoopRelease `state:"nosave"` + fsutil.InodeNoopTruncate `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` + fsutil.InodeNotDirectory `state:"nosave"` + fsutil.InodeNotMappable `state:"nosave"` + fsutil.InodeNotSocket `state:"nosave"` + fsutil.InodeNotSymlink `state:"nosave"` + fsutil.InodeVirtual `state:"nosave"` + + fsutil.InodeSimpleAttributes + fsutil.InodeStaticFileGetter +} + +var _ fs.InodeOperations = (*staticFileInodeOps)(nil) + +// newStaticFileInode returns a procfs InodeOperations with static contents. +func newStaticProcInode(ctx context.Context, msrc *fs.MountSource, contents []byte) *fs.Inode { + iops := &staticFileInodeOps{ + InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + InodeStaticFileGetter: fsutil.InodeStaticFileGetter{ + Contents: contents, + }, + } + return newProcInode(iops, msrc, fs.SpecialFile, nil) +} + +// newProcInode creates a new inode from the given inode operations. +func newProcInode(iops fs.InodeOperations, msrc *fs.MountSource, typ fs.InodeType, t *kernel.Task) *fs.Inode { + sattr := fs.StableAttr{ + DeviceID: device.ProcDevice.DeviceID(), + InodeID: device.ProcDevice.NextIno(), + BlockSize: usermem.PageSize, + Type: typ, + } + if t != nil { + iops = &taskOwnedInodeOps{iops, t} + } + return fs.NewInode(iops, msrc, sattr) +} diff --git a/pkg/sentry/fs/proc/net.go b/pkg/sentry/fs/proc/net.go index 2806d6035..3ed85a538 100644 --- a/pkg/sentry/fs/proc/net.go +++ b/pkg/sentry/fs/proc/net.go @@ -28,35 +28,36 @@ import ( // newNet creates a new proc net entry. func (p *proc) newNetDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) + var contents map[string]*fs.Inode if s := p.k.NetworkStack(); s != nil && s.SupportsIPv6() { - d.AddChild(ctx, "dev", seqfile.NewSeqFileInode(ctx, &netDev{s: s}, msrc)) - d.AddChild(ctx, "if_inet6", seqfile.NewSeqFileInode(ctx, &ifinet6{s: s}, msrc)) - - // The following files are simple stubs until they are implemented in - // netstack, if the file contains a header the stub is just the header - // otherwise it is an empty file. - d.AddChild(ctx, "arp", p.newStubProcFSFile(ctx, msrc, []byte("IP address HW type Flags HW address Mask Device"))) - d.AddChild(ctx, "ipv6_route", p.newStubProcFSFile(ctx, msrc, []byte(""))) - d.AddChild(ctx, "netlink", p.newStubProcFSFile(ctx, msrc, []byte("sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode"))) - d.AddChild(ctx, "netstat", p.newStubProcFSFile(ctx, msrc, []byte("TcpExt: SyncookiesSent SyncookiesRecv SyncookiesFailed EmbryonicRsts PruneCalled RcvPruned OfoPruned OutOfWindowIcmps LockDroppedIcmps ArpFilter TW TWRecycled TWKilled PAWSPassive PAWSActive PAWSEstab DelayedACKs DelayedACKLocked DelayedACKLost ListenOverflows ListenDrops TCPPrequeued TCPDirectCopyFromBacklog TCPDirectCopyFromPrequeue TCPPrequeueDropped TCPHPHits TCPHPHitsToUser TCPPureAcks TCPHPAcks TCPRenoRecovery TCPSackRecovery TCPSACKReneging TCPFACKReorder TCPSACKReorder TCPRenoReorder TCPTSReorder TCPFullUndo TCPPartialUndo TCPDSACKUndo TCPLossUndo TCPLostRetransmit TCPRenoFailures TCPSackFailures TCPLossFailures TCPFastRetrans TCPForwardRetrans TCPSlowStartRetrans TCPTimeouts TCPLossProbes TCPLossProbeRecovery TCPRenoRecoveryFail TCPSackRecoveryFail TCPSchedulerFailed TCPRcvCollapsed TCPDSACKOldSent TCPDSACKOfoSent TCPDSACKRecv TCPDSACKOfoRecv TCPAbortOnData TCPAbortOnClose TCPAbortOnMemory TCPAbortOnTimeout TCPAbortOnLinger TCPAbortFailed TCPMemoryPressures TCPSACKDiscard TCPDSACKIgnoredOld TCPDSACKIgnoredNoUndo TCPSpuriousRTOs TCPMD5NotFound TCPMD5Unexpected TCPMD5Failure TCPSackShifted TCPSackMerged TCPSackShiftFallback TCPBacklogDrop TCPMinTTLDrop TCPDeferAcceptDrop IPReversePathFilter TCPTimeWaitOverflow TCPReqQFullDoCookies TCPReqQFullDrop TCPRetransFail TCPRcvCoalesce TCPOFOQueue TCPOFODrop TCPOFOMerge TCPChallengeACK TCPSYNChallenge TCPFastOpenActive TCPFastOpenActiveFail TCPFastOpenPassive TCPFastOpenPassiveFail TCPFastOpenListenOverflow TCPFastOpenCookieReqd TCPSpuriousRtxHostQueues BusyPollRxPackets TCPAutoCorking TCPFromZeroWindowAdv TCPToZeroWindowAdv TCPWantZeroWindowAdv TCPSynRetrans TCPOrigDataSent TCPHystartTrainDetect TCPHystartTrainCwnd TCPHystartDelayDetect TCPHystartDelayCwnd TCPACKSkippedSynRecv TCPACKSkippedPAWS TCPACKSkippedSeq TCPACKSkippedFinWait2 TCPACKSkippedTimeWait TCPACKSkippedChallenge TCPWinProbe TCPKeepAlive TCPMTUPFail TCPMTUPSuccess"))) - d.AddChild(ctx, "packet", p.newStubProcFSFile(ctx, msrc, []byte("sk RefCnt Type Proto Iface R Rmem User Inode"))) - d.AddChild(ctx, "protocols", p.newStubProcFSFile(ctx, msrc, []byte("protocol size sockets memory press maxhdr slab module cl co di ac io in de sh ss gs se re sp bi br ha uh gp em"))) - - // Linux sets these values to: nsec per usec, psched tick in ns, 1000000, - // high res timer ticks per sec (ClockGetres returns 1ns resolution). - psched := fmt.Sprintf("%08x %08x %08x %08x\n", uint64(time.Microsecond/time.Nanosecond), 64, 1000000, uint64(time.Second/time.Nanosecond)) - d.AddChild(ctx, "psched", p.newStubProcFSFile(ctx, msrc, []byte(psched))) - - d.AddChild(ctx, "ptype", p.newStubProcFSFile(ctx, msrc, []byte("Type Device Function"))) - d.AddChild(ctx, "route", p.newStubProcFSFile(ctx, msrc, []byte("Iface Destination Gateway Flags RefCnt Use Metric Mask MTU Window IRTT"))) - d.AddChild(ctx, "tcp", p.newStubProcFSFile(ctx, msrc, []byte(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode"))) - d.AddChild(ctx, "tcp6", p.newStubProcFSFile(ctx, msrc, []byte(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode"))) - d.AddChild(ctx, "udp", p.newStubProcFSFile(ctx, msrc, []byte(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode ref pointer drops"))) - d.AddChild(ctx, "udp6", p.newStubProcFSFile(ctx, msrc, []byte(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode"))) + contents = map[string]*fs.Inode{ + "dev": seqfile.NewSeqFileInode(ctx, &netDev{s: s}, msrc), + "if_inet6": seqfile.NewSeqFileInode(ctx, &ifinet6{s: s}, msrc), + + // The following files are simple stubs until they are + // implemented in netstack, if the file contains a + // header the stub is just the header otherwise it is + // an empty file. + "arp": newStaticProcInode(ctx, msrc, []byte("IP address HW type Flags HW address Mask Device")), + "ipv6_route": newStaticProcInode(ctx, msrc, []byte("")), + "netlink": newStaticProcInode(ctx, msrc, []byte("sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode")), + "netstat": newStaticProcInode(ctx, msrc, []byte("TcpExt: SyncookiesSent SyncookiesRecv SyncookiesFailed EmbryonicRsts PruneCalled RcvPruned OfoPruned OutOfWindowIcmps LockDroppedIcmps ArpFilter TW TWRecycled TWKilled PAWSPassive PAWSActive PAWSEstab DelayedACKs DelayedACKLocked DelayedACKLost ListenOverflows ListenDrops TCPPrequeued TCPDirectCopyFromBacklog TCPDirectCopyFromPrequeue TCPPrequeueDropped TCPHPHits TCPHPHitsToUser TCPPureAcks TCPHPAcks TCPRenoRecovery TCPSackRecovery TCPSACKReneging TCPFACKReorder TCPSACKReorder TCPRenoReorder TCPTSReorder TCPFullUndo TCPPartialUndo TCPDSACKUndo TCPLossUndo TCPLostRetransmit TCPRenoFailures TCPSackFailures TCPLossFailures TCPFastRetrans TCPForwardRetrans TCPSlowStartRetrans TCPTimeouts TCPLossProbes TCPLossProbeRecovery TCPRenoRecoveryFail TCPSackRecoveryFail TCPSchedulerFailed TCPRcvCollapsed TCPDSACKOldSent TCPDSACKOfoSent TCPDSACKRecv TCPDSACKOfoRecv TCPAbortOnData TCPAbortOnClose TCPAbortOnMemory TCPAbortOnTimeout TCPAbortOnLinger TCPAbortFailed TCPMemoryPressures TCPSACKDiscard TCPDSACKIgnoredOld TCPDSACKIgnoredNoUndo TCPSpuriousRTOs TCPMD5NotFound TCPMD5Unexpected TCPMD5Failure TCPSackShifted TCPSackMerged TCPSackShiftFallback TCPBacklogDrop TCPMinTTLDrop TCPDeferAcceptDrop IPReversePathFilter TCPTimeWaitOverflow TCPReqQFullDoCookies TCPReqQFullDrop TCPRetransFail TCPRcvCoalesce TCPOFOQueue TCPOFODrop TCPOFOMerge TCPChallengeACK TCPSYNChallenge TCPFastOpenActive TCPFastOpenActiveFail TCPFastOpenPassive TCPFastOpenPassiveFail TCPFastOpenListenOverflow TCPFastOpenCookieReqd TCPSpuriousRtxHostQueues BusyPollRxPackets TCPAutoCorking TCPFromZeroWindowAdv TCPToZeroWindowAdv TCPWantZeroWindowAdv TCPSynRetrans TCPOrigDataSent TCPHystartTrainDetect TCPHystartTrainCwnd TCPHystartDelayDetect TCPHystartDelayCwnd TCPACKSkippedSynRecv TCPACKSkippedPAWS TCPACKSkippedSeq TCPACKSkippedFinWait2 TCPACKSkippedTimeWait TCPACKSkippedChallenge TCPWinProbe TCPKeepAlive TCPMTUPFail TCPMTUPSuccess")), + "packet": newStaticProcInode(ctx, msrc, []byte("sk RefCnt Type Proto Iface R Rmem User Inode")), + "protocols": newStaticProcInode(ctx, msrc, []byte("protocol size sockets memory press maxhdr slab module cl co di ac io in de sh ss gs se re sp bi br ha uh gp em")), + // Linux sets psched values to: nsec per usec, psched + // tick in ns, 1000000, high res timer ticks per sec + // (ClockGetres returns 1ns resolution). + "psched": newStaticProcInode(ctx, msrc, []byte(fmt.Sprintf("%08x %08x %08x %08x\n", uint64(time.Microsecond/time.Nanosecond), 64, 1000000, uint64(time.Second/time.Nanosecond)))), + "ptype": newStaticProcInode(ctx, msrc, []byte("Type Device Function")), + "route": newStaticProcInode(ctx, msrc, []byte("Iface Destination Gateway Flags RefCnt Use Metric Mask MTU Window IRTT")), + "tcp": newStaticProcInode(ctx, msrc, []byte(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode")), + "tcp6": newStaticProcInode(ctx, msrc, []byte(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode")), + "udp": newStaticProcInode(ctx, msrc, []byte(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode ref pointer drops")), + "udp6": newStaticProcInode(ctx, msrc, []byte(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode")), + } } - return newFile(d, msrc, fs.SpecialDirectory, nil) + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } // ifinet6 implements seqfile.SeqSource for /proc/net/if_inet6. diff --git a/pkg/sentry/fs/proc/proc.go b/pkg/sentry/fs/proc/proc.go index 70e549c31..d1c699418 100644 --- a/pkg/sentry/fs/proc/proc.go +++ b/pkg/sentry/fs/proc/proc.go @@ -17,18 +17,17 @@ package proc import ( "fmt" - "io" "sort" "strconv" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/device" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" "gvisor.googlesource.com/gvisor/pkg/sentry/socket/rpcinet" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" "gvisor.googlesource.com/gvisor/pkg/syserror" ) @@ -46,32 +45,6 @@ type proc struct { pidns *kernel.PIDNamespace } -// stubProcFSFile is a file type that can be used to return file contents -// which are constant. This file is not writable and will always have mode -// 0444. -// -// +stateify savable -type stubProcFSFile struct { - ramfs.Entry - - // contents are the immutable file contents that will always be returned. - contents []byte -} - -// DeprecatedPreadv implements fs.InodeOperations.DeprecatedPreadv. -func (s *stubProcFSFile) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { - if offset < 0 { - return 0, syserror.EINVAL - } - - if offset >= int64(len(s.contents)) { - return 0, io.EOF - } - - n, err := dst.CopyOut(ctx, s.contents[offset:]) - return int64(n), err -} - // New returns the root node of a partial simple procfs. func New(ctx context.Context, msrc *fs.MountSource) (*fs.Inode, error) { k := kernel.KernelFromContext(ctx) @@ -83,29 +56,39 @@ func New(ctx context.Context, msrc *fs.MountSource) (*fs.Inode, error) { return nil, fmt.Errorf("procfs requires a PID namespace") } - p := &proc{k: k, pidns: pidns} - p.InitDir(ctx, map[string]*fs.Inode{ + // Note that these are just the static members. There are dynamic + // members populated in Readdir and Lookup below. + contents := map[string]*fs.Inode{ + "cpuinfo": newCPUInfo(ctx, msrc), "filesystems": seqfile.NewSeqFileInode(ctx, &filesystemsData{}, msrc), "loadavg": seqfile.NewSeqFileInode(ctx, &loadavgData{}, msrc), "meminfo": seqfile.NewSeqFileInode(ctx, &meminfoData{k}, msrc), - "mounts": newMountsSymlink(ctx, msrc), + "mounts": newProcInode(ramfs.NewSymlink(ctx, fs.RootOwner, "self/mounts"), msrc, fs.Symlink, nil), + "self": newSelf(ctx, pidns, msrc), "stat": seqfile.NewSeqFileInode(ctx, &statData{k}, msrc), + "thread-self": newThreadSelf(ctx, pidns, msrc), + "uptime": newUptime(ctx, msrc), "version": seqfile.NewSeqFileInode(ctx, &versionData{k}, msrc), - }, fs.RootOwner, fs.FilePermsFromMode(0555)) + } + + // Construct the proc InodeOperations. + p := &proc{ + Dir: *ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)), + k: k, + pidns: pidns, + } + + // Add more contents that need proc to be initialized. + p.AddChild(ctx, "sys", p.newSysDir(ctx, msrc)) - p.AddChild(ctx, "cpuinfo", p.newCPUInfo(ctx, msrc)) // If we're using rpcinet we will let it manage /proc/net. if _, ok := p.k.NetworkStack().(*rpcinet.Stack); ok { - p.AddChild(ctx, "net", newRPCInetProcNet(ctx, msrc)) + contents["net"] = newRPCInetProcNet(ctx, msrc) } else { - p.AddChild(ctx, "net", p.newNetDir(ctx, msrc)) + contents["net"] = p.newNetDir(ctx, msrc) } - p.AddChild(ctx, "self", p.newSelf(ctx, msrc)) - p.AddChild(ctx, "sys", p.newSysDir(ctx, msrc)) - p.AddChild(ctx, "thread-self", p.newThreadSelf(ctx, msrc)) - p.AddChild(ctx, "uptime", p.newUptime(ctx, msrc)) - return newFile(p, msrc, fs.SpecialDirectory, nil), nil + return newProcInode(p, msrc, fs.SpecialDirectory, nil), nil } // self is a magical link. @@ -118,26 +101,21 @@ type self struct { } // newSelf returns a new "self" node. -func (p *proc) newSelf(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - s := &self{pidns: p.pidns} - s.InitSymlink(ctx, fs.RootOwner, "") - return newFile(s, msrc, fs.Symlink, nil) +func newSelf(ctx context.Context, pidns *kernel.PIDNamespace, msrc *fs.MountSource) *fs.Inode { + s := &self{ + Symlink: *ramfs.NewSymlink(ctx, fs.RootOwner, ""), + pidns: pidns, + } + return newProcInode(s, msrc, fs.Symlink, nil) } // newThreadSelf returns a new "threadSelf" node. -func (p *proc) newThreadSelf(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - s := &threadSelf{pidns: p.pidns} - s.InitSymlink(ctx, fs.RootOwner, "") - return newFile(s, msrc, fs.Symlink, nil) -} - -// newStubProcFsFile returns a procfs file with constant contents. -func (p *proc) newStubProcFSFile(ctx context.Context, msrc *fs.MountSource, c []byte) *fs.Inode { - u := &stubProcFSFile{ - contents: c, +func newThreadSelf(ctx context.Context, pidns *kernel.PIDNamespace, msrc *fs.MountSource) *fs.Inode { + s := &threadSelf{ + Symlink: *ramfs.NewSymlink(ctx, fs.RootOwner, ""), + pidns: pidns, } - u.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(0444)) - return newFile(u, msrc, fs.SpecialFile, nil) + return newProcInode(s, msrc, fs.Symlink, nil) } // Readlink implements fs.InodeOperations.Readlink. @@ -145,13 +123,13 @@ func (s *self) Readlink(ctx context.Context, inode *fs.Inode) (string, error) { if t := kernel.TaskFromContext(ctx); t != nil { tgid := s.pidns.IDOfThreadGroup(t.ThreadGroup()) if tgid == 0 { - return "", ramfs.ErrNotFound + return "", syserror.ENOENT } return strconv.FormatUint(uint64(tgid), 10), nil } // Who is reading this link? - return "", ramfs.ErrInvalidOp + return "", syserror.EINVAL } // threadSelf is more magical than "self" link. @@ -169,13 +147,13 @@ func (s *threadSelf) Readlink(ctx context.Context, inode *fs.Inode) (string, err tgid := s.pidns.IDOfThreadGroup(t.ThreadGroup()) tid := s.pidns.IDOfTask(t) if tid == 0 || tgid == 0 { - return "", ramfs.ErrNotFound + return "", syserror.ENOENT } return fmt.Sprintf("%d/task/%d", tgid, tid), nil } // Who is reading this link? - return "", ramfs.ErrInvalidOp + return "", syserror.EINVAL } // Lookup loads an Inode at name into a Dirent. @@ -204,25 +182,44 @@ func (p *proc) Lookup(ctx context.Context, dir *fs.Inode, name string) (*fs.Dire return fs.NewDirent(td, name), nil } -// Readdir synthesizes proc contents. -func (p *proc) DeprecatedReaddir(ctx context.Context, dirCtx *fs.DirCtx, offset int) (int, error) { - // Serialize normal contents. - _, err := p.Dir.DeprecatedReaddir(ctx, dirCtx, offset) - if err != nil { - return offset, err +// GetFile implements fs.InodeOperations. +func (p *proc) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &rootProcFile{iops: p}), nil +} + +// rootProcFile implements fs.FileOperations for the proc directory. +// +// +stateify savable +type rootProcFile struct { + fsutil.DirFileOperations `state:"nosave"` + + iops *proc +} + +var _ fs.FileOperations = (*rootProcFile)(nil) + +// Readdir implements fs.FileOperations.Readdir. +func (rpf *rootProcFile) Readdir(ctx context.Context, file *fs.File, ser fs.DentrySerializer) (int64, error) { + offset := file.Offset() + dirCtx := &fs.DirCtx{ + Serializer: ser, } - m := make(map[string]fs.DentAttr) - var names []string + // Get normal directory contents from ramfs dir. + names, m := rpf.iops.Dir.Children() - // Add special files. - m["sys"] = fs.GenericDentAttr(fs.SpecialFile, device.ProcDevice) - names = append(names, "sys") + // Add dot and dotdot. + root := fs.RootFromContext(ctx) + defer root.DecRef() + dot, dotdot := file.Dirent.GetDotAttrs(root) + names = append(names, ".", "..") + m["."] = dot + m[".."] = dotdot // Collect tasks. // Per linux we only include it in directory listings if it's the leader. // But for whatever crazy reason, you can still walk to the given node. - for _, tg := range p.pidns.ThreadGroups() { + for _, tg := range rpf.iops.pidns.ThreadGroups() { if leader := tg.Leader(); leader != nil { name := strconv.FormatUint(uint64(tg.ID()), 10) m[name] = fs.GenericDentAttr(fs.SpecialDirectory, device.ProcDevice) @@ -230,7 +227,7 @@ func (p *proc) DeprecatedReaddir(ctx context.Context, dirCtx *fs.DirCtx, offset } } - if offset >= len(m) { + if offset >= int64(len(m)) { return offset, nil } sort.Strings(names) @@ -241,12 +238,5 @@ func (p *proc) DeprecatedReaddir(ctx context.Context, dirCtx *fs.DirCtx, offset } offset++ } - return offset, err -} - -// newMountsSymlink returns a symlink to "self/mounts". -func newMountsSymlink(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - s := &ramfs.Symlink{} - s.InitSymlink(ctx, fs.RootOwner, "self/mounts") - return newFile(s, msrc, fs.Symlink, nil) + return offset, nil } diff --git a/pkg/sentry/fs/proc/rpcinet_proc.go b/pkg/sentry/fs/proc/rpcinet_proc.go index d025069df..65faa21f2 100644 --- a/pkg/sentry/fs/proc/rpcinet_proc.go +++ b/pkg/sentry/fs/proc/rpcinet_proc.go @@ -20,32 +20,72 @@ import ( "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" "gvisor.googlesource.com/gvisor/pkg/sentry/socket/rpcinet" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" + "gvisor.googlesource.com/gvisor/pkg/syserror" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) -// rpcinetFile implments fs.InodeOperations as RPCs. -type rpcinetFile struct { - ramfs.Entry +// rpcInetInode implments fs.InodeOperations. +type rpcInetInode struct { + fsutil.SimpleFileInode - // filepath is the full path of this rpcinetFile. + // filepath is the full path of this rpcInetInode. filepath string k *kernel.Kernel } -// DeprecatedPreadv implements fs.InodeOperations.DeprecatedPreadv. -// This method can panic if an rpcinetFile was created without an rpcinet +func newRPCInetInode(ctx context.Context, msrc *fs.MountSource, filepath string, mode linux.FileMode) *fs.Inode { + f := &rpcInetInode{ + SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(mode), linux.PROC_SUPER_MAGIC), + filepath: filepath, + k: kernel.KernelFromContext(ctx), + } + return newProcInode(f, msrc, fs.SpecialFile, nil) +} + +// GetFile implements fs.InodeOperations.GetFile. +func (i *rpcInetInode) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + flags.Pread = true + flags.Pwrite = true + fops := &rpcInetFile{ + inode: i, + } + return fs.NewFile(ctx, dirent, flags, fops), nil +} + +// rpcInetFile implements fs.FileOperations as RPCs. +type rpcInetFile struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + + inode *rpcInetInode +} + +// Read implements fs.FileOperations.Read. +// +// This method can panic if an rpcInetInode was created without an rpcinet // stack. -func (r rpcinetFile) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { - s, ok := r.k.NetworkStack().(*rpcinet.Stack) +func (f *rpcInetFile) Read(ctx context.Context, file *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { + if offset < 0 { + return 0, syserror.EINVAL + } + s, ok := f.inode.k.NetworkStack().(*rpcinet.Stack) if !ok { panic("Network stack is not a rpcinet.") } - contents, se := s.RPCReadFile(r.filepath) + contents, se := s.RPCReadFile(f.inode.filepath) if se != nil || offset >= int64(len(contents)) { return 0, io.EOF } @@ -54,16 +94,12 @@ func (r rpcinetFile) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequenc return int64(n), err } -// Truncate implements fs.InodeOperations.Truncate. -func (r rpcinetFile) Truncate(context.Context, *fs.Inode, int64) error { - return nil -} - -// DeprecatedPwritev implements fs.InodeOperations.DeprecatedPwritev. -// This method can panic if an rpcinetFile was created without an rpcinet +// Write implements fs.FileOperations.Write. +// +// This method can panic if an rpcInetInode was created without an rpcInet // stack. -func (r rpcinetFile) DeprecatedPwritev(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { - s, ok := r.k.NetworkStack().(*rpcinet.Stack) +func (f *rpcInetFile) Write(ctx context.Context, file *fs.File, src usermem.IOSequence, offset int64) (int64, error) { + s, ok := f.inode.k.NetworkStack().(*rpcinet.Stack) if !ok { panic("Network stack is not a rpcinet.") } @@ -78,116 +114,102 @@ func (r rpcinetFile) DeprecatedPwritev(ctx context.Context, src usermem.IOSequen return int64(n), err } - written, se := s.RPCWriteFile(r.filepath, b) + written, se := s.RPCWriteFile(f.inode.filepath, b) return int64(written), se.ToError() } -func newRPCProcFSFile(ctx context.Context, msrc *fs.MountSource, filepath string, mode linux.FileMode) *fs.Inode { - f := &rpcinetFile{ - filepath: filepath, - k: kernel.KernelFromContext(ctx), - } - f.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(mode)) - - fi := newFile(f, msrc, fs.SpecialFile, nil) - return fi -} - // newRPCInetProcNet will build an inode for /proc/net. func newRPCInetProcNet(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - - // Add all the files we want to forward for /proc/net. - d.AddChild(ctx, "arp", newRPCProcFSFile(ctx, msrc, "/proc/net/arp", 0444)) - d.AddChild(ctx, "dev", newRPCProcFSFile(ctx, msrc, "/proc/net/dev", 0444)) - d.AddChild(ctx, "if_inet6", newRPCProcFSFile(ctx, msrc, "/proc/net/if_inet6", 0444)) - d.AddChild(ctx, "ipv6_route", newRPCProcFSFile(ctx, msrc, "/proc/net/ipv6_route", 0444)) - d.AddChild(ctx, "netlink", newRPCProcFSFile(ctx, msrc, "/proc/net/netlink", 0444)) - d.AddChild(ctx, "netstat", newRPCProcFSFile(ctx, msrc, "/proc/net/netstat", 0444)) - d.AddChild(ctx, "packet", newRPCProcFSFile(ctx, msrc, "/proc/net/packet", 0444)) - d.AddChild(ctx, "protocols", newRPCProcFSFile(ctx, msrc, "/proc/net/protocols", 0444)) - d.AddChild(ctx, "psched", newRPCProcFSFile(ctx, msrc, "/proc/net/psched", 0444)) - d.AddChild(ctx, "ptype", newRPCProcFSFile(ctx, msrc, "/proc/net/ptype", 0444)) - d.AddChild(ctx, "route", newRPCProcFSFile(ctx, msrc, "/proc/net/route", 0444)) - d.AddChild(ctx, "tcp", newRPCProcFSFile(ctx, msrc, "/proc/net/tcp", 0444)) - d.AddChild(ctx, "tcp6", newRPCProcFSFile(ctx, msrc, "/proc/net/tcp6", 0444)) - d.AddChild(ctx, "udp", newRPCProcFSFile(ctx, msrc, "/proc/net/udp", 0444)) - d.AddChild(ctx, "udp6", newRPCProcFSFile(ctx, msrc, "/proc/net/udp6", 0444)) - - return newFile(d, msrc, fs.SpecialDirectory, nil) + contents := map[string]*fs.Inode{ + "arp": newRPCInetInode(ctx, msrc, "/proc/net/arp", 0444), + "dev": newRPCInetInode(ctx, msrc, "/proc/net/dev", 0444), + "if_inet6": newRPCInetInode(ctx, msrc, "/proc/net/if_inet6", 0444), + "ipv6_route": newRPCInetInode(ctx, msrc, "/proc/net/ipv6_route", 0444), + "netlink": newRPCInetInode(ctx, msrc, "/proc/net/netlink", 0444), + "netstat": newRPCInetInode(ctx, msrc, "/proc/net/netstat", 0444), + "packet": newRPCInetInode(ctx, msrc, "/proc/net/packet", 0444), + "protocols": newRPCInetInode(ctx, msrc, "/proc/net/protocols", 0444), + "psched": newRPCInetInode(ctx, msrc, "/proc/net/psched", 0444), + "ptype": newRPCInetInode(ctx, msrc, "/proc/net/ptype", 0444), + "route": newRPCInetInode(ctx, msrc, "/proc/net/route", 0444), + "tcp": newRPCInetInode(ctx, msrc, "/proc/net/tcp", 0444), + "tcp6": newRPCInetInode(ctx, msrc, "/proc/net/tcp6", 0444), + "udp": newRPCInetInode(ctx, msrc, "/proc/net/udp", 0444), + "udp6": newRPCInetInode(ctx, msrc, "/proc/net/udp6", 0444), + } + + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } // newRPCInetProcSysNet will build an inode for /proc/sys/net. func newRPCInetProcSysNet(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - d.AddChild(ctx, "ipv4", newRPCInetSysNetIPv4Dir(ctx, msrc)) - d.AddChild(ctx, "core", newRPCInetSysNetCore(ctx, msrc)) + contents := map[string]*fs.Inode{ + "ipv4": newRPCInetSysNetIPv4Dir(ctx, msrc), + "core": newRPCInetSysNetCore(ctx, msrc), + } - return newFile(d, msrc, fs.SpecialDirectory, nil) + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } // newRPCInetSysNetCore builds the /proc/sys/net/core directory. func newRPCInetSysNetCore(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - - // Add all the files we want to forward over RPC for /proc/sys/net/core - d.AddChild(ctx, "default_qdisc", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/default_qdisc", 0444)) - d.AddChild(ctx, "message_burst", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/message_burst", 0444)) - d.AddChild(ctx, "message_cost", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/message_cost", 0444)) - d.AddChild(ctx, "optmem_max", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/optmem_max", 0444)) - d.AddChild(ctx, "rmem_default", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/rmem_default", 0444)) - d.AddChild(ctx, "rmem_max", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/rmem_max", 0444)) - d.AddChild(ctx, "somaxconn", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/somaxconn", 0444)) - d.AddChild(ctx, "wmem_default", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/wmem_default", 0444)) - d.AddChild(ctx, "wmem_max", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/core/wmem_max", 0444)) - - return newFile(d, msrc, fs.SpecialDirectory, nil) + contents := map[string]*fs.Inode{ + "default_qdisc": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/default_qdisc", 0444), + "message_burst": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/message_burst", 0444), + "message_cost": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/message_cost", 0444), + "optmem_max": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/optmem_max", 0444), + "rmem_default": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/rmem_default", 0444), + "rmem_max": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/rmem_max", 0444), + "somaxconn": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/somaxconn", 0444), + "wmem_default": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/wmem_default", 0444), + "wmem_max": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/wmem_max", 0444), + } + + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } // newRPCInetSysNetIPv4Dir builds the /proc/sys/net/ipv4 directory. func newRPCInetSysNetIPv4Dir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - - // Add all the files we want to forward over RPC for /proc/sys/net/ipv4. - d.AddChild(ctx, "ip_local_port_range", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/ip_local_port_range", 0444)) - d.AddChild(ctx, "ip_local_reserved_ports", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/ip_local_reserved_ports", 0444)) - d.AddChild(ctx, "ipfrag_time", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/ipfrag_time", 0444)) - d.AddChild(ctx, "ip_nonlocal_bind", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/ip_nonlocal_bind", 0444)) - d.AddChild(ctx, "ip_no_pmtu_disc", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/ip_no_pmtu_disc", 0444)) - - d.AddChild(ctx, "tcp_allowed_congestion_control", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_allowed_congestion_control", 0444)) - d.AddChild(ctx, "tcp_available_congestion_control", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_available_congestion_control", 0444)) - d.AddChild(ctx, "tcp_base_mss", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_base_mss", 0444)) - d.AddChild(ctx, "tcp_congestion_control", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_congestion_control", 0644)) - d.AddChild(ctx, "tcp_dsack", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_dsack", 0644)) - d.AddChild(ctx, "tcp_early_retrans", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_early_retrans", 0644)) - d.AddChild(ctx, "tcp_fack", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_fack", 0644)) - d.AddChild(ctx, "tcp_fastopen", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_fastopen", 0644)) - d.AddChild(ctx, "tcp_fastopen_key", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_fastopen_key", 0444)) - d.AddChild(ctx, "tcp_fin_timeout", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_fin_timeout", 0644)) - d.AddChild(ctx, "tcp_invalid_ratelimit", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_invalid_ratelimit", 0444)) - d.AddChild(ctx, "tcp_keepalive_intvl", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_intvl", 0644)) - d.AddChild(ctx, "tcp_keepalive_probes", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_probes", 0644)) - d.AddChild(ctx, "tcp_keepalive_time", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_time", 0644)) - d.AddChild(ctx, "tcp_mem", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_mem", 0444)) - d.AddChild(ctx, "tcp_mtu_probing", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_mtu_probing", 0644)) - d.AddChild(ctx, "tcp_no_metrics_save", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_no_metrics_save", 0444)) - d.AddChild(ctx, "tcp_probe_interval", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_probe_interval", 0444)) - d.AddChild(ctx, "tcp_probe_threshold", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_probe_threshold", 0444)) - d.AddChild(ctx, "tcp_retries1", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_retries1", 0644)) - d.AddChild(ctx, "tcp_retries2", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_retries2", 0644)) - d.AddChild(ctx, "tcp_rfc1337", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_rfc1337", 0444)) - d.AddChild(ctx, "tcp_rmem", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_rmem", 0444)) - d.AddChild(ctx, "tcp_sack", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_sack", 0644)) - d.AddChild(ctx, "tcp_slow_start_after_idle", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_slow_start_after_idle", 0644)) - d.AddChild(ctx, "tcp_synack_retries", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_synack_retries", 0644)) - d.AddChild(ctx, "tcp_syn_retries", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_syn_retries", 0644)) - d.AddChild(ctx, "tcp_timestamps", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_timestamps", 0644)) - d.AddChild(ctx, "tcp_wmem", newRPCProcFSFile(ctx, msrc, "/proc/sys/net/ipv4/tcp_wmem", 0444)) - - return newFile(d, msrc, fs.SpecialDirectory, nil) + contents := map[string]*fs.Inode{ + "ip_local_port_range": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ip_local_port_range", 0444), + "ip_local_reserved_ports": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ip_local_reserved_ports", 0444), + "ipfrag_time": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ipfrag_time", 0444), + "ip_nonlocal_bind": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ip_nonlocal_bind", 0444), + "ip_no_pmtu_disc": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ip_no_pmtu_disc", 0444), + "tcp_allowed_congestion_control": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_allowed_congestion_control", 0444), + "tcp_available_congestion_control": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_available_congestion_control", 0444), + "tcp_base_mss": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_base_mss", 0444), + "tcp_congestion_control": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_congestion_control", 0644), + "tcp_dsack": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_dsack", 0644), + "tcp_early_retrans": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_early_retrans", 0644), + "tcp_fack": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_fack", 0644), + "tcp_fastopen": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_fastopen", 0644), + "tcp_fastopen_key": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_fastopen_key", 0444), + "tcp_fin_timeout": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_fin_timeout", 0644), + "tcp_invalid_ratelimit": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_invalid_ratelimit", 0444), + "tcp_keepalive_intvl": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_intvl", 0644), + "tcp_keepalive_probes": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_probes", 0644), + "tcp_keepalive_time": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_time", 0644), + "tcp_mem": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_mem", 0444), + "tcp_mtu_probing": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_mtu_probing", 0644), + "tcp_no_metrics_save": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_no_metrics_save", 0444), + "tcp_probe_interval": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_probe_interval", 0444), + "tcp_probe_threshold": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_probe_threshold", 0444), + "tcp_retries1": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_retries1", 0644), + "tcp_retries2": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_retries2", 0644), + "tcp_rfc1337": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_rfc1337", 0444), + "tcp_rmem": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_rmem", 0444), + "tcp_sack": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_sack", 0644), + "tcp_slow_start_after_idle": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_slow_start_after_idle", 0644), + "tcp_synack_retries": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_synack_retries", 0644), + "tcp_syn_retries": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_syn_retries", 0644), + "tcp_timestamps": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_timestamps", 0644), + "tcp_wmem": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_wmem", 0444), + } + + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } diff --git a/pkg/sentry/fs/proc/seqfile/BUILD b/pkg/sentry/fs/proc/seqfile/BUILD index 53c475652..b4ba64e10 100644 --- a/pkg/sentry/fs/proc/seqfile/BUILD +++ b/pkg/sentry/fs/proc/seqfile/BUILD @@ -8,12 +8,15 @@ go_library( importpath = "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/seqfile", visibility = ["//pkg/sentry:internal"], deps = [ + "//pkg/abi/linux", "//pkg/sentry/context", "//pkg/sentry/fs", + "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/proc/device", - "//pkg/sentry/fs/ramfs", "//pkg/sentry/kernel/time", "//pkg/sentry/usermem", + "//pkg/syserror", + "//pkg/waiter", ], ) @@ -26,7 +29,7 @@ go_test( "//pkg/sentry/context", "//pkg/sentry/context/contexttest", "//pkg/sentry/fs", - "//pkg/sentry/fs/ramfs/test", + "//pkg/sentry/fs/ramfs", "//pkg/sentry/usermem", ], ) diff --git a/pkg/sentry/fs/proc/seqfile/seqfile.go b/pkg/sentry/fs/proc/seqfile/seqfile.go index 0499ba65b..16fc6789e 100644 --- a/pkg/sentry/fs/proc/seqfile/seqfile.go +++ b/pkg/sentry/fs/proc/seqfile/seqfile.go @@ -18,12 +18,15 @@ import ( "io" "sync" + "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/device" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" ktime "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/time" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" + "gvisor.googlesource.com/gvisor/pkg/syserror" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) // SeqHandle is a helper handle to seek in the file. @@ -87,7 +90,18 @@ func (s *SeqGenerationCounter) IsCurrent(generation int64) bool { // // +stateify savable type SeqFile struct { - ramfs.Entry + fsutil.InodeGenericChecker `state:"nosave"` + fsutil.InodeNoopRelease `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` + fsutil.InodeNotDirectory `state:"nosave"` + fsutil.InodeNotMappable `state:"nosave"` + fsutil.InodeNotSocket `state:"nosave"` + fsutil.InodeNotSymlink `state:"nosave"` + fsutil.InodeNotTruncatable `state:"nosave"` + fsutil.InodeVirtual `state:"nosave"` + + fsutil.InodeSimpleExtendedAttributes + fsutil.InodeSimpleAttributes // mu protects the fields below. mu sync.Mutex `state:"nosave"` @@ -99,11 +113,14 @@ type SeqFile struct { lastRead int64 } +var _ fs.InodeOperations = (*SeqFile)(nil) + // NewSeqFile returns a seqfile suitable for use by external consumers. func NewSeqFile(ctx context.Context, source SeqSource) *SeqFile { - s := &SeqFile{SeqSource: source} - s.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(0444)) - return s + return &SeqFile{ + InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + SeqSource: source, + } } // NewSeqFileInode returns an Inode with SeqFile InodeOperations. @@ -120,11 +137,19 @@ func NewSeqFileInode(ctx context.Context, source SeqSource, msrc *fs.MountSource // UnstableAttr returns unstable attributes of the SeqFile. func (s *SeqFile) UnstableAttr(ctx context.Context, inode *fs.Inode) (fs.UnstableAttr, error) { - uattr, _ := s.Entry.UnstableAttr(ctx, inode) + uattr, err := s.InodeSimpleAttributes.UnstableAttr(ctx, inode) + if err != nil { + return fs.UnstableAttr{}, err + } uattr.ModificationTime = ktime.NowFromContext(ctx) return uattr, nil } +// GetFile implements fs.InodeOperations.GetFile. +func (s *SeqFile) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &seqFileOperations{seqFile: s}), nil +} + // findIndexAndOffset finds the unit that corresponds to a certain offset. // Returns the unit and the offset within the unit. If there are not enough // units len(data) and leftover offset is returned. @@ -139,36 +164,74 @@ func findIndexAndOffset(data []SeqData, offset int64) (int, int64) { return len(data), offset } -// DeprecatedPreadv reads from the file at the given offset. -func (s *SeqFile) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { - s.mu.Lock() - defer s.mu.Unlock() +// updateSourceLocked requires that s.mu is held. +func (s *SeqFile) updateSourceLocked(ctx context.Context, record int) { + var h SeqHandle + if record == 0 { + h = nil + } else { + h = s.source[record-1].Handle + } + // Save what we have previously read. + s.source = s.source[:record] + var newSource []SeqData + newSource, s.generation = s.SeqSource.ReadSeqFileData(ctx, h) + s.source = append(s.source, newSource...) +} + +// seqFileOperations implements fs.FileOperations. +// +// +stateify savable +type seqFileOperations struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + + seqFile *SeqFile +} + +var _ fs.FileOperations = (*seqFileOperations)(nil) + +// Write implements fs.FileOperations.Write. +func (*seqFileOperations) Write(context.Context, *fs.File, usermem.IOSequence, int64) (int64, error) { + return 0, syserror.EACCES +} + +// Read implements fs.FileOperations.Read. +func (sfo *seqFileOperations) Read(ctx context.Context, file *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { + sfo.seqFile.mu.Lock() + defer sfo.seqFile.mu.Unlock() - s.Entry.NotifyAccess(ctx) - defer func() { s.lastRead = offset }() + sfo.seqFile.NotifyAccess(ctx) + defer func() { sfo.seqFile.lastRead = offset }() updated := false // Try to find where we should start reading this file. - i, recordOffset := findIndexAndOffset(s.source, offset) - if i == len(s.source) { + i, recordOffset := findIndexAndOffset(sfo.seqFile.source, offset) + if i == len(sfo.seqFile.source) { // Ok, we're at EOF. Let's first check to see if there might be // more data available to us. If there is more data, add it to // the end and try reading again. - if !s.SeqSource.NeedsUpdate(s.generation) { + if !sfo.seqFile.SeqSource.NeedsUpdate(sfo.seqFile.generation) { return 0, io.EOF } - oldLen := len(s.source) - s.updateSourceLocked(ctx, len(s.source)) + oldLen := len(sfo.seqFile.source) + sfo.seqFile.updateSourceLocked(ctx, len(sfo.seqFile.source)) updated = true // We know that we had consumed everything up until this point // so we search in the new slice instead of starting over. - i, recordOffset = findIndexAndOffset(s.source[oldLen:], recordOffset) + i, recordOffset = findIndexAndOffset(sfo.seqFile.source[oldLen:], recordOffset) i += oldLen // i is at most the length of the slice which is - // len(s.source) - oldLen. So at most i will be equal to - // len(s.source). - if i == len(s.source) { + // len(sfo.seqFile.source) - oldLen. So at most i will be equal to + // len(sfo.seqFile.source). + if i == len(sfo.seqFile.source) { return 0, io.EOF } } @@ -178,7 +241,7 @@ func (s *SeqFile) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, // before continuing on to the next. We don't refresh our data source // before this record is completed. if recordOffset != 0 { - n, err := dst.CopyOut(ctx, s.source[i].Buf[recordOffset:]) + n, err := dst.CopyOut(ctx, sfo.seqFile.source[i].Buf[recordOffset:]) done += int64(n) dst = dst.DropFirst(n) if dst.NumBytes() == 0 || err != nil { @@ -190,15 +253,15 @@ func (s *SeqFile) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, // Next/New unit, update the source file if necessary. Make an extra // check to see if we've seeked backwards and if so always update our // data source. - if !updated && (s.SeqSource.NeedsUpdate(s.generation) || s.lastRead > offset) { - s.updateSourceLocked(ctx, i) + if !updated && (sfo.seqFile.SeqSource.NeedsUpdate(sfo.seqFile.generation) || sfo.seqFile.lastRead > offset) { + sfo.seqFile.updateSourceLocked(ctx, i) // recordOffset is 0 here and we won't update records behind the // current one so recordOffset is still 0 even though source // just got updated. Just read the next record. } // Finish by reading all the available data. - for _, buf := range s.source[i:] { + for _, buf := range sfo.seqFile.source[i:] { n, err := dst.CopyOut(ctx, buf.Buf) done += int64(n) dst = dst.DropFirst(n) @@ -214,23 +277,3 @@ func (s *SeqFile) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, } return done, nil } - -// updateSourceLocked requires that s.mu is held. -func (s *SeqFile) updateSourceLocked(ctx context.Context, record int) { - var h SeqHandle - if record == 0 { - h = nil - } else { - h = s.source[record-1].Handle - } - // Save what we have previously read. - s.source = s.source[:record] - var newSource []SeqData - newSource, s.generation = s.SeqSource.ReadSeqFileData(ctx, h) - s.source = append(s.source, newSource...) -} - -// DeprecatedPwritev is always denied. -func (*SeqFile) DeprecatedPwritev(context.Context, usermem.IOSequence, int64) (int64, error) { - return 0, ramfs.ErrDenied -} diff --git a/pkg/sentry/fs/proc/seqfile/seqfile_test.go b/pkg/sentry/fs/proc/seqfile/seqfile_test.go index f9a2ca38e..35403ab7f 100644 --- a/pkg/sentry/fs/proc/seqfile/seqfile_test.go +++ b/pkg/sentry/fs/proc/seqfile/seqfile_test.go @@ -23,7 +23,7 @@ import ( "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/context/contexttest" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - ramfstest "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs/test" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" ) @@ -91,10 +91,15 @@ type testTable struct { expectedError error } -func runTableTests(ctx context.Context, table []testTable, n fs.InodeOperations) error { +func runTableTests(ctx context.Context, table []testTable, dirent *fs.Dirent) error { for _, tt := range table { + file, err := dirent.Inode.InodeOperations.GetFile(ctx, dirent, fs.FileFlags{Read: true}) + if err != nil { + return fmt.Errorf("GetFile returned error: %v", err) + } + data := make([]byte, tt.readBufferSize) - resultLen, err := n.DeprecatedPreadv(ctx, usermem.BytesIOSequence(data), tt.offset) + resultLen, err := file.Preadv(ctx, usermem.BytesIOSequence(data), tt.offset) if err != tt.expectedError { return fmt.Errorf("t.Preadv(len: %v, offset: %v) (error) => %v expected %v", tt.readBufferSize, tt.offset, err, tt.expectedError) } @@ -115,12 +120,12 @@ func TestSeqFile(t *testing.T) { testSource.Init() // Create a file that can be R/W. - m := fs.NewNonCachingMountSource(nil, fs.MountSourceFlags{}) + m := fs.NewPseudoMountSource() ctx := contexttest.Context(t) contents := map[string]*fs.Inode{ "foo": NewSeqFileInode(ctx, testSource, m), } - root := ramfstest.NewDir(ctx, contents, fs.FilePermsFromMode(0777)) + root := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0777)) // How about opening it? inode := fs.NewInode(root, m, fs.StableAttr{Type: fs.Directory}) @@ -129,9 +134,13 @@ func TestSeqFile(t *testing.T) { t.Fatalf("failed to walk to foo for n2: %v", err) } n2 := dirent2.Inode.InodeOperations + file2, err := n2.GetFile(ctx, dirent2, fs.FileFlags{Read: true, Write: true}) + if err != nil { + t.Fatalf("GetFile returned error: %v", err) + } // Writing? - if _, err := n2.DeprecatedPwritev(nil, usermem.BytesIOSequence([]byte("test")), 0); err == nil { + if _, err := file2.Writev(ctx, usermem.BytesIOSequence([]byte("test"))); err == nil { t.Fatalf("managed to write to n2: %v", err) } @@ -141,7 +150,6 @@ func TestSeqFile(t *testing.T) { t.Fatalf("failed to walk to foo: %v", err) } n3 := dirent3.Inode.InodeOperations - if n2 != n3 { t.Error("got n2 != n3, want same") } @@ -170,13 +178,13 @@ func TestSeqFile(t *testing.T) { // Read the last 3 bytes. {97, 10, testSource.actual[9].Buf[7:], nil}, } - if err := runTableTests(ctx, table, n2); err != nil { + if err := runTableTests(ctx, table, dirent2); err != nil { t.Errorf("runTableTest failed with testSource.update = %v : %v", testSource.update, err) } // Disable updates and do it again. testSource.update = false - if err := runTableTests(ctx, table, n2); err != nil { + if err := runTableTests(ctx, table, dirent2); err != nil { t.Errorf("runTableTest failed with testSource.update = %v: %v", testSource.update, err) } } @@ -188,25 +196,24 @@ func TestSeqFileFileUpdated(t *testing.T) { testSource.update = true // Create a file that can be R/W. - m := fs.NewNonCachingMountSource(nil, fs.MountSourceFlags{}) + m := fs.NewPseudoMountSource() ctx := contexttest.Context(t) contents := map[string]*fs.Inode{ "foo": NewSeqFileInode(ctx, testSource, m), } - root := ramfstest.NewDir(ctx, contents, fs.FilePermsFromMode(0777)) + root := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0777)) // How about opening it? inode := fs.NewInode(root, m, fs.StableAttr{Type: fs.Directory}) dirent2, err := root.Lookup(ctx, inode, "foo") if err != nil { - t.Fatalf("failed to walk to foo for n2: %v", err) + t.Fatalf("failed to walk to foo for dirent2: %v", err) } - n2 := dirent2.Inode.InodeOperations table := []testTable{ {0, 16, flatten(testSource.actual[0].Buf, testSource.actual[1].Buf[:6]), nil}, } - if err := runTableTests(ctx, table, n2); err != nil { + if err := runTableTests(ctx, table, dirent2); err != nil { t.Errorf("runTableTest failed: %v", err) } // Delete the first entry. @@ -224,7 +231,7 @@ func TestSeqFileFileUpdated(t *testing.T) { // Read the following two lines. {30, 20, flatten(testSource.actual[3].Buf, testSource.actual[4].Buf), nil}, } - if err := runTableTests(ctx, table, n2); err != nil { + if err := runTableTests(ctx, table, dirent2); err != nil { t.Errorf("runTableTest failed after removing first entry: %v", err) } @@ -238,7 +245,7 @@ func TestSeqFileFileUpdated(t *testing.T) { table = []testTable{ {50, 20, flatten(testSource.actual[4].Buf, testSource.actual[5].Buf), nil}, } - if err := runTableTests(ctx, table, n2); err != nil { + if err := runTableTests(ctx, table, dirent2); err != nil { t.Errorf("runTableTest failed after adding middle entry: %v", err) } // This will be used in a later test. @@ -249,7 +256,7 @@ func TestSeqFileFileUpdated(t *testing.T) { table = []testTable{ {20, 20, []byte{}, io.EOF}, } - if err := runTableTests(ctx, table, n2); err != nil { + if err := runTableTests(ctx, table, dirent2); err != nil { t.Errorf("runTableTest failed after removing all entries: %v", err) } // Restore some of the data. @@ -257,7 +264,7 @@ func TestSeqFileFileUpdated(t *testing.T) { table = []testTable{ {6, 20, testSource.actual[0].Buf[6:], nil}, } - if err := runTableTests(ctx, table, n2); err != nil { + if err := runTableTests(ctx, table, dirent2); err != nil { t.Errorf("runTableTest failed after adding first entry back: %v", err) } @@ -266,7 +273,7 @@ func TestSeqFileFileUpdated(t *testing.T) { table = []testTable{ {30, 20, flatten(testSource.actual[3].Buf, testSource.actual[4].Buf), nil}, } - if err := runTableTests(ctx, table, n2); err != nil { + if err := runTableTests(ctx, table, dirent2); err != nil { t.Errorf("runTableTest failed after extending testSource: %v", err) } } diff --git a/pkg/sentry/fs/proc/sys.go b/pkg/sentry/fs/proc/sys.go index 54562508d..ee6b9f262 100644 --- a/pkg/sentry/fs/proc/sys.go +++ b/pkg/sentry/fs/proc/sys.go @@ -22,39 +22,15 @@ import ( "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" "gvisor.googlesource.com/gvisor/pkg/sentry/socket/rpcinet" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) -// hostname is a file containing the system hostname. -// -// +stateify savable -type hostname struct { - ramfs.Entry -} - -// DeprecatedPreadv implements fs.InodeOperations.DeprecatedPreadv. -func (hostname) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { - utsns := kernel.UTSNamespaceFromContext(ctx) - contents := []byte(utsns.HostName() + "\n") - - if offset >= int64(len(contents)) { - return 0, io.EOF - } - - n, err := dst.CopyOut(ctx, contents[offset:]) - return int64(n), err -} - -func (p *proc) newHostname(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - h := &hostname{} - h.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(0444)) - return newFile(h, msrc, fs.SpecialFile, nil) -} - // mmapMinAddrData backs /proc/sys/vm/mmap_min_addr. // // +stateify savable @@ -101,36 +77,84 @@ func (*overcommitMemory) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandl } func (p *proc) newKernelDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - d.AddChild(ctx, "hostname", p.newHostname(ctx, msrc)) - - d.AddChild(ctx, "shmmax", p.newStubProcFSFile(ctx, msrc, []byte(strconv.FormatUint(linux.SHMMAX, 10)))) - d.AddChild(ctx, "shmall", p.newStubProcFSFile(ctx, msrc, []byte(strconv.FormatUint(linux.SHMALL, 10)))) - d.AddChild(ctx, "shmmni", p.newStubProcFSFile(ctx, msrc, []byte(strconv.FormatUint(linux.SHMMNI, 10)))) - return newFile(d, msrc, fs.SpecialDirectory, nil) + h := hostname{ + SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + } + + children := map[string]*fs.Inode{ + "hostname": newProcInode(&h, msrc, fs.SpecialFile, nil), + "shmall": newStaticProcInode(ctx, msrc, []byte(strconv.FormatUint(linux.SHMALL, 10))), + "shmmax": newStaticProcInode(ctx, msrc, []byte(strconv.FormatUint(linux.SHMMAX, 10))), + "shmmni": newStaticProcInode(ctx, msrc, []byte(strconv.FormatUint(linux.SHMMNI, 10))), + } + + d := ramfs.NewDir(ctx, children, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } func (p *proc) newVMDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - d.AddChild(ctx, "mmap_min_addr", seqfile.NewSeqFileInode(ctx, &mmapMinAddrData{p.k}, msrc)) - d.AddChild(ctx, "overcommit_memory", seqfile.NewSeqFileInode(ctx, &overcommitMemory{}, msrc)) - return newFile(d, msrc, fs.SpecialDirectory, nil) + children := map[string]*fs.Inode{ + "mmap_min_addr": seqfile.NewSeqFileInode(ctx, &mmapMinAddrData{p.k}, msrc), + "overcommit_memory": seqfile.NewSeqFileInode(ctx, &overcommitMemory{}, msrc), + } + d := ramfs.NewDir(ctx, children, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } func (p *proc) newSysDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - d.AddChild(ctx, "kernel", p.newKernelDir(ctx, msrc)) - d.AddChild(ctx, "vm", p.newVMDir(ctx, msrc)) + children := map[string]*fs.Inode{ + "kernel": p.newKernelDir(ctx, msrc), + "vm": p.newVMDir(ctx, msrc), + } // If we're using rpcinet we will let it manage /proc/sys/net. if _, ok := p.k.NetworkStack().(*rpcinet.Stack); ok { - d.AddChild(ctx, "net", newRPCInetProcSysNet(ctx, msrc)) + children["net"] = newRPCInetProcSysNet(ctx, msrc) } else { - d.AddChild(ctx, "net", p.newSysNetDir(ctx, msrc)) + children["net"] = p.newSysNetDir(ctx, msrc) } - return newFile(d, msrc, fs.SpecialDirectory, nil) + d := ramfs.NewDir(ctx, children, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) +} + +// hostname is the inode for a file containing the system hostname. +// +// +stateify savable +type hostname struct { + fsutil.SimpleFileInode +} + +// GetFile implements fs.InodeOperations.GetFile. +func (h *hostname) GetFile(ctx context.Context, d *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, d, flags, &hostnameFile{}), nil +} + +var _ fs.InodeOperations = (*hostname)(nil) + +// +stateify savable +type hostnameFile struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoSeek `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoWrite `state:"nosave"` } + +// Read implements fs.FileOperations.Read. +func (hf *hostnameFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { + utsns := kernel.UTSNamespaceFromContext(ctx) + contents := []byte(utsns.HostName() + "\n") + if offset >= int64(len(contents)) { + return 0, io.EOF + } + n, err := dst.CopyOut(ctx, contents[offset:]) + return int64(n), err + +} + +var _ fs.FileOperations = (*hostnameFile)(nil) diff --git a/pkg/sentry/fs/proc/sys_net.go b/pkg/sentry/fs/proc/sys_net.go index b50d43d70..42e9bc47f 100644 --- a/pkg/sentry/fs/proc/sys_net.go +++ b/pkg/sentry/fs/proc/sys_net.go @@ -17,13 +17,17 @@ package proc import ( "fmt" "io" + "sync" + "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/device" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/inet" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) type tcpMemDir int @@ -33,21 +37,37 @@ const ( tcpWMem ) +// tcpMemInode is used to read/write the size of netstack tcp buffers. +// +// TODO: If we have multiple proc mounts, concurrent writes can +// leave netstack and the proc files in an inconsistent state. Since we set the +// buffer size from these proc files on restore, we may also race and end up in +// an inconsistent state on restore. +// // +stateify savable -type tcpMem struct { - ramfs.Entry - s inet.Stack `state:"wait"` +type tcpMemInode struct { + fsutil.SimpleFileInode + dir tcpMemDir + s inet.Stack `state:"wait"` + + // size stores the tcp buffer size during save, and sets the buffer + // size in netstack in restore. We must save/restore this here, since + // netstack itself is stateless. size inet.TCPBufferSize - dir tcpMemDir -} -func newTCPMem(s inet.Stack, size inet.TCPBufferSize, dir tcpMemDir) *tcpMem { - return &tcpMem{s: s, size: size, dir: dir} + // mu protects against concurrent reads/writes to files based on this + // inode. + mu sync.Mutex `state:"nosave"` } -func newTCPMemInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack, size inet.TCPBufferSize, dir tcpMemDir) *fs.Inode { - tm := newTCPMem(s, size, dir) - tm.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(0644)) +var _ fs.InodeOperations = (*tcpMemInode)(nil) + +func newTCPMemInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack, dir tcpMemDir) *fs.Inode { + tm := &tcpMemInode{ + SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + s: s, + dir: dir, + } sattr := fs.StableAttr{ DeviceID: device.ProcDevice.DeviceID(), InodeID: device.ProcDevice.NextIno(), @@ -57,62 +77,105 @@ func newTCPMemInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack, siz return fs.NewInode(tm, msrc, sattr) } -// DeprecatedPreadv implements fs.InodeOperations.DeprecatedPreadv. -func (m *tcpMem) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { +// GetFile implements fs.InodeOperations.GetFile. +func (m *tcpMemInode) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + flags.Pread = true + return fs.NewFile(ctx, dirent, flags, &tcpMemFile{tcpMemInode: m}), nil +} + +// +stateify savable +type tcpMemFile struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + + tcpMemInode *tcpMemInode +} + +var _ fs.FileOperations = (*tcpMemFile)(nil) + +// Read implements fs.FileOperations.Read. +func (f *tcpMemFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { if offset != 0 { return 0, io.EOF } - s := fmt.Sprintf("%d\t%d\t%d\n", m.size.Min, m.size.Default, m.size.Max) + f.tcpMemInode.mu.Lock() + defer f.tcpMemInode.mu.Unlock() + + size, err := readSize(f.tcpMemInode.dir, f.tcpMemInode.s) + if err != nil { + return 0, err + } + s := fmt.Sprintf("%d\t%d\t%d\n", size.Min, size.Default, size.Max) n, err := dst.CopyOut(ctx, []byte(s)) return int64(n), err } -// Truncate implements fs.InodeOperations.Truncate. -func (*tcpMem) Truncate(context.Context, *fs.Inode, int64) error { - return nil -} - -// DeprecatedPwritev implements fs.InodeOperations.DeprecatedPwritev. -func (m *tcpMem) DeprecatedPwritev(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { +// Write implements fs.FileOperations.Write. +func (f *tcpMemFile) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, offset int64) (int64, error) { if src.NumBytes() == 0 { return 0, nil } - src = src.TakeFirst(usermem.PageSize - 1) + f.tcpMemInode.mu.Lock() + defer f.tcpMemInode.mu.Unlock() - buf := []int32{int32(m.size.Min), int32(m.size.Default), int32(m.size.Max)} + src = src.TakeFirst(usermem.PageSize - 1) + size, err := readSize(f.tcpMemInode.dir, f.tcpMemInode.s) + if err != nil { + return 0, err + } + buf := []int32{int32(size.Min), int32(size.Default), int32(size.Max)} n, cperr := usermem.CopyInt32StringsInVec(ctx, src.IO, src.Addrs, buf, src.Opts) - m.size = inet.TCPBufferSize{ + newSize := inet.TCPBufferSize{ Min: int(buf[0]), Default: int(buf[1]), Max: int(buf[2]), } - if err := m.writeSize(); err != nil { + if err := writeSize(f.tcpMemInode.dir, f.tcpMemInode.s, newSize); err != nil { return n, err } return n, cperr } -func (m *tcpMem) writeSize() error { - switch m.dir { +func readSize(dirType tcpMemDir, s inet.Stack) (inet.TCPBufferSize, error) { + switch dirType { + case tcpRMem: + return s.TCPReceiveBufferSize() + case tcpWMem: + return s.TCPSendBufferSize() + default: + panic(fmt.Sprintf("unknown tcpMemFile type: %v", dirType)) + } +} + +func writeSize(dirType tcpMemDir, s inet.Stack, size inet.TCPBufferSize) error { + switch dirType { case tcpRMem: - return m.s.SetTCPReceiveBufferSize(m.size) + return s.SetTCPReceiveBufferSize(size) case tcpWMem: - return m.s.SetTCPSendBufferSize(m.size) + return s.SetTCPSendBufferSize(size) default: - panic(fmt.Sprintf("unknown tcpMem.dir: %v", m.dir)) + panic(fmt.Sprintf("unknown tcpMemFile type: %v", dirType)) } } // +stateify savable type tcpSack struct { - ramfs.Entry - s inet.Stack `state:"wait"` + stack inet.Stack `state:"wait"` enabled *bool + fsutil.SimpleFileInode } func newTCPSackInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *fs.Inode { - ts := &tcpSack{s: s} - ts.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(0644)) + ts := &tcpSack{ + SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + stack: s, + } sattr := fs.StableAttr{ DeviceID: device.ProcDevice.DeviceID(), InodeID: device.ProcDevice.NextIno(), @@ -122,21 +185,48 @@ func newTCPSackInode(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *f return fs.NewInode(ts, msrc, sattr) } -func (s *tcpSack) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { +// GetFile implements fs.InodeOperations.GetFile. +func (s *tcpSack) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + flags.Pread = true + flags.Pwrite = true + return fs.NewFile(ctx, dirent, flags, &tcpSackFile{ + tcpSack: s, + stack: s.stack, + }), nil +} + +// +stateify savable +type tcpSackFile struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + + tcpSack *tcpSack + + stack inet.Stack `state:"wait"` +} + +// Read implements fs.FileOperations.Read. +func (f *tcpSackFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { if offset != 0 { return 0, io.EOF } - if s.enabled == nil { - sack, err := s.s.TCPSACKEnabled() + if f.tcpSack.enabled == nil { + sack, err := f.stack.TCPSACKEnabled() if err != nil { return 0, err } - s.enabled = &sack + f.tcpSack.enabled = &sack } val := "0\n" - if *s.enabled { + if *f.tcpSack.enabled { // Technically, this is not quite compatible with Linux. Linux // stores these as an integer, so if you write "2" into // tcp_sack, you should get 2 back. Tough luck. @@ -146,13 +236,8 @@ func (s *tcpSack) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, return int64(n), err } -// Truncate implements fs.InodeOperations.Truncate. -func (*tcpSack) Truncate(context.Context, *fs.Inode, int64) error { - return nil -} - -// DeprecatedPwritev implements fs.InodeOperations.DeprecatedPwritev. -func (s *tcpSack) DeprecatedPwritev(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { +// Write implements fs.FileOperations.Write. +func (f *tcpSackFile) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, offset int64) (int64, error) { if src.NumBytes() == 0 { return 0, nil } @@ -163,100 +248,104 @@ func (s *tcpSack) DeprecatedPwritev(ctx context.Context, src usermem.IOSequence, if err != nil { return n, err } - if s.enabled == nil { - s.enabled = new(bool) + if f.tcpSack.enabled == nil { + f.tcpSack.enabled = new(bool) } - *s.enabled = v != 0 - return n, s.s.SetTCPSACKEnabled(*s.enabled) + *f.tcpSack.enabled = v != 0 + return n, f.tcpSack.stack.SetTCPSACKEnabled(*f.tcpSack.enabled) } func (p *proc) newSysNetCore(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - // The following files are simple stubs until they are implemented in // netstack, most of these files are configuration related. We use the // value closest to the actual netstack behavior or any empty file, // all of these files will have mode 0444 (read-only for all users). - d.AddChild(ctx, "default_qdisc", p.newStubProcFSFile(ctx, msrc, []byte("pfifo_fast"))) - d.AddChild(ctx, "message_burst", p.newStubProcFSFile(ctx, msrc, []byte("10"))) - d.AddChild(ctx, "message_cost", p.newStubProcFSFile(ctx, msrc, []byte("5"))) - d.AddChild(ctx, "optmem_max", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "rmem_default", p.newStubProcFSFile(ctx, msrc, []byte("212992"))) - d.AddChild(ctx, "rmem_max", p.newStubProcFSFile(ctx, msrc, []byte("212992"))) - d.AddChild(ctx, "somaxconn", p.newStubProcFSFile(ctx, msrc, []byte("128"))) - d.AddChild(ctx, "wmem_default", p.newStubProcFSFile(ctx, msrc, []byte("212992"))) - d.AddChild(ctx, "wmem_max", p.newStubProcFSFile(ctx, msrc, []byte("212992"))) - - return newFile(d, msrc, fs.SpecialDirectory, nil) + contents := map[string]*fs.Inode{ + "default_qdisc": newStaticProcInode(ctx, msrc, []byte("pfifo_fast")), + "message_burst": newStaticProcInode(ctx, msrc, []byte("10")), + "message_cost": newStaticProcInode(ctx, msrc, []byte("5")), + "optmem_max": newStaticProcInode(ctx, msrc, []byte("0")), + "rmem_default": newStaticProcInode(ctx, msrc, []byte("212992")), + "rmem_max": newStaticProcInode(ctx, msrc, []byte("212992")), + "somaxconn": newStaticProcInode(ctx, msrc, []byte("128")), + "wmem_default": newStaticProcInode(ctx, msrc, []byte("212992")), + "wmem_max": newStaticProcInode(ctx, msrc, []byte("212992")), + } + + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } func (p *proc) newSysNetIPv4Dir(ctx context.Context, msrc *fs.MountSource, s inet.Stack) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) + contents := map[string]*fs.Inode{ + // Add tcp_sack. + "tcp_sack": newTCPSackInode(ctx, msrc, s), + + // The following files are simple stubs until they are + // implemented in netstack, most of these files are + // configuration related. We use the value closest to the + // actual netstack behavior or any empty file, all of these + // files will have mode 0444 (read-only for all users). + "ip_local_port_range": newStaticProcInode(ctx, msrc, []byte("16000 65535")), + "ip_local_reserved_ports": newStaticProcInode(ctx, msrc, []byte("")), + "ipfrag_time": newStaticProcInode(ctx, msrc, []byte("30")), + "ip_nonlocal_bind": newStaticProcInode(ctx, msrc, []byte("0")), + "ip_no_pmtu_disc": newStaticProcInode(ctx, msrc, []byte("1")), + + // tcp_allowed_congestion_control tell the user what they are + // able to do as an unprivledged process so we leave it empty. + "tcp_allowed_congestion_control": newStaticProcInode(ctx, msrc, []byte("")), + "tcp_available_congestion_control": newStaticProcInode(ctx, msrc, []byte("reno")), + "tcp_congestion_control": newStaticProcInode(ctx, msrc, []byte("reno")), + + // Many of the following stub files are features netstack + // doesn't support. The unsupported features return "0" to + // indicate they are disabled. + "tcp_base_mss": newStaticProcInode(ctx, msrc, []byte("1280")), + "tcp_dsack": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_early_retrans": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_fack": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_fastopen": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_fastopen_key": newStaticProcInode(ctx, msrc, []byte("")), + "tcp_invalid_ratelimit": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_keepalive_intvl": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_keepalive_probes": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_keepalive_time": newStaticProcInode(ctx, msrc, []byte("7200")), + "tcp_mtu_probing": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_no_metrics_save": newStaticProcInode(ctx, msrc, []byte("1")), + "tcp_probe_interval": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_probe_threshold": newStaticProcInode(ctx, msrc, []byte("0")), + "tcp_retries1": newStaticProcInode(ctx, msrc, []byte("3")), + "tcp_retries2": newStaticProcInode(ctx, msrc, []byte("15")), + "tcp_rfc1337": newStaticProcInode(ctx, msrc, []byte("1")), + "tcp_slow_start_after_idle": newStaticProcInode(ctx, msrc, []byte("1")), + "tcp_synack_retries": newStaticProcInode(ctx, msrc, []byte("5")), + "tcp_syn_retries": newStaticProcInode(ctx, msrc, []byte("3")), + "tcp_timestamps": newStaticProcInode(ctx, msrc, []byte("1")), + } // Add tcp_rmem. - if rs, err := s.TCPReceiveBufferSize(); err == nil { - d.AddChild(ctx, "tcp_rmem", newTCPMemInode(ctx, msrc, s, rs, tcpRMem)) + if _, err := s.TCPReceiveBufferSize(); err == nil { + contents["tcp_rmem"] = newTCPMemInode(ctx, msrc, s, tcpRMem) } // Add tcp_wmem. - if ss, err := s.TCPSendBufferSize(); err == nil { - d.AddChild(ctx, "tcp_wmem", newTCPMemInode(ctx, msrc, s, ss, tcpWMem)) + if _, err := s.TCPSendBufferSize(); err == nil { + contents["tcp_wmem"] = newTCPMemInode(ctx, msrc, s, tcpWMem) } - // Add tcp_sack. - d.AddChild(ctx, "tcp_sack", newTCPSackInode(ctx, msrc, s)) - - // The following files are simple stubs until they are implemented in - // netstack, most of these files are configuration related. We use the - // value closest to the actual netstack behavior or any empty file, - // all of these files will have mode 0444 (read-only for all users). - d.AddChild(ctx, "ip_local_port_range", p.newStubProcFSFile(ctx, msrc, []byte("16000 65535"))) - d.AddChild(ctx, "ip_local_reserved_ports", p.newStubProcFSFile(ctx, msrc, []byte(""))) - d.AddChild(ctx, "ipfrag_time", p.newStubProcFSFile(ctx, msrc, []byte("30"))) - d.AddChild(ctx, "ip_nonlocal_bind", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "ip_no_pmtu_disc", p.newStubProcFSFile(ctx, msrc, []byte("1"))) - - // tcp_allowed_congestion_control tell the user what they are able to do as an - // unprivledged process so we leave it empty. - d.AddChild(ctx, "tcp_allowed_congestion_control", p.newStubProcFSFile(ctx, msrc, []byte(""))) - d.AddChild(ctx, "tcp_available_congestion_control", p.newStubProcFSFile(ctx, msrc, []byte("reno"))) - d.AddChild(ctx, "tcp_congestion_control", p.newStubProcFSFile(ctx, msrc, []byte("reno"))) - - // Many of the following stub files are features netstack doesn't support - // and are therefore "0" for disabled. - d.AddChild(ctx, "tcp_base_mss", p.newStubProcFSFile(ctx, msrc, []byte("1280"))) - d.AddChild(ctx, "tcp_dsack", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_early_retrans", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_fack", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_fastopen", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_fastopen_key", p.newStubProcFSFile(ctx, msrc, []byte(""))) - d.AddChild(ctx, "tcp_invalid_ratelimit", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_keepalive_intvl", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_keepalive_probes", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_keepalive_time", p.newStubProcFSFile(ctx, msrc, []byte("7200"))) - d.AddChild(ctx, "tcp_mtu_probing", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_no_metrics_save", p.newStubProcFSFile(ctx, msrc, []byte("1"))) - d.AddChild(ctx, "tcp_probe_interval", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_probe_threshold", p.newStubProcFSFile(ctx, msrc, []byte("0"))) - d.AddChild(ctx, "tcp_retries1", p.newStubProcFSFile(ctx, msrc, []byte("3"))) - d.AddChild(ctx, "tcp_retries2", p.newStubProcFSFile(ctx, msrc, []byte("15"))) - d.AddChild(ctx, "tcp_rfc1337", p.newStubProcFSFile(ctx, msrc, []byte("1"))) - d.AddChild(ctx, "tcp_slow_start_after_idle", p.newStubProcFSFile(ctx, msrc, []byte("1"))) - d.AddChild(ctx, "tcp_synack_retries", p.newStubProcFSFile(ctx, msrc, []byte("5"))) - d.AddChild(ctx, "tcp_syn_retries", p.newStubProcFSFile(ctx, msrc, []byte("3"))) - d.AddChild(ctx, "tcp_timestamps", p.newStubProcFSFile(ctx, msrc, []byte("1"))) - - return newFile(d, msrc, fs.SpecialDirectory, nil) + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } func (p *proc) newSysNetDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(ctx, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) + var contents map[string]*fs.Inode if s := p.k.NetworkStack(); s != nil { - d.AddChild(ctx, "ipv4", p.newSysNetIPv4Dir(ctx, msrc, s)) - d.AddChild(ctx, "core", p.newSysNetCore(ctx, msrc, s)) + contents = map[string]*fs.Inode{ + "ipv4": p.newSysNetIPv4Dir(ctx, msrc, s), + "core": p.newSysNetCore(ctx, msrc, s), + } } - return newFile(d, msrc, fs.SpecialDirectory, nil) + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + return newProcInode(d, msrc, fs.SpecialDirectory, nil) } diff --git a/pkg/sentry/fs/proc/sys_net_state.go b/pkg/sentry/fs/proc/sys_net_state.go index 7f46776c0..5f481a1cf 100644 --- a/pkg/sentry/fs/proc/sys_net_state.go +++ b/pkg/sentry/fs/proc/sys_net_state.go @@ -16,17 +16,26 @@ package proc import "fmt" +// beforeSave is invoked by stateify. +func (t *tcpMemInode) beforeSave() { + size, err := readSize(t.dir, t.s) + if err != nil { + panic(fmt.Sprintf("failed to read TCP send / receive buffer sizes: %v", err)) + } + t.size = size +} + // afterLoad is invoked by stateify. -func (m *tcpMem) afterLoad() { - if err := m.writeSize(); err != nil { - panic(fmt.Sprintf("failed to write previous TCP send / receive buffer sizes [%v]: %v", m.size, err)) +func (t *tcpMemInode) afterLoad() { + if err := writeSize(t.dir, t.s, t.size); err != nil { + panic(fmt.Sprintf("failed to write previous TCP send / receive buffer sizes [%v]: %v", t.size, err)) } } // afterLoad is invoked by stateify. func (s *tcpSack) afterLoad() { if s.enabled != nil { - if err := s.s.SetTCPSACKEnabled(*s.enabled); err != nil { + if err := s.stack.SetTCPSACKEnabled(*s.enabled); err != nil { panic(fmt.Sprintf("failed to set previous TCP sack configuration [%v]: %v", *s.enabled, err)) } } diff --git a/pkg/sentry/fs/proc/sys_net_test.go b/pkg/sentry/fs/proc/sys_net_test.go index 0ce9d30f1..ea0d94fce 100644 --- a/pkg/sentry/fs/proc/sys_net_test.go +++ b/pkg/sentry/fs/proc/sys_net_test.go @@ -26,13 +26,14 @@ func TestQuerySendBufferSize(t *testing.T) { ctx := context.Background() s := inet.NewTestStack() s.TCPSendBufSize = inet.TCPBufferSize{100, 200, 300} - tm := newTCPMem(s, s.TCPSendBufSize, tcpWMem) + tmi := &tcpMemInode{s: s, dir: tcpWMem} + tmf := &tcpMemFile{tcpMemInode: tmi} buf := make([]byte, 100) dst := usermem.BytesIOSequence(buf) - n, err := tm.DeprecatedPreadv(ctx, dst, 0) + n, err := tmf.Read(ctx, nil, dst, 0) if err != nil { - t.Fatalf("DeprecatedPreadv failed: %v", err) + t.Fatalf("Read failed: %v", err) } if got, want := string(buf[:n]), "100\t200\t300\n"; got != want { @@ -44,13 +45,14 @@ func TestQueryRecvBufferSize(t *testing.T) { ctx := context.Background() s := inet.NewTestStack() s.TCPRecvBufSize = inet.TCPBufferSize{100, 200, 300} - tm := newTCPMem(s, s.TCPRecvBufSize, tcpRMem) + tmi := &tcpMemInode{s: s, dir: tcpRMem} + tmf := &tcpMemFile{tcpMemInode: tmi} buf := make([]byte, 100) dst := usermem.BytesIOSequence(buf) - n, err := tm.DeprecatedPreadv(ctx, dst, 0) + n, err := tmf.Read(ctx, nil, dst, 0) if err != nil { - t.Fatalf("DeprecatedPreadv failed: %v", err) + t.Fatalf("Read failed: %v", err) } if got, want := string(buf[:n]), "100\t200\t300\n"; got != want { @@ -85,12 +87,13 @@ func TestConfigureSendBufferSize(t *testing.T) { s := inet.NewTestStack() for _, c := range cases { s.TCPSendBufSize = c.initial - tm := newTCPMem(s, c.initial, tcpWMem) + tmi := &tcpMemInode{s: s, dir: tcpWMem} + tmf := &tcpMemFile{tcpMemInode: tmi} // Write the values. src := usermem.BytesIOSequence([]byte(c.str)) - if n, err := tm.DeprecatedPwritev(ctx, src, 0); n != int64(len(c.str)) || err != nil { - t.Errorf("DeprecatedPwritev, case = %q: got (%d, %v), wanted (%d, nil)", c.str, n, err, len(c.str)) + if n, err := tmf.Write(ctx, nil, src, 0); n != int64(len(c.str)) || err != nil { + t.Errorf("Write, case = %q: got (%d, %v), wanted (%d, nil)", c.str, n, err, len(c.str)) } // Read the values from the stack and check them. @@ -105,12 +108,13 @@ func TestConfigureRecvBufferSize(t *testing.T) { s := inet.NewTestStack() for _, c := range cases { s.TCPRecvBufSize = c.initial - tm := newTCPMem(s, c.initial, tcpRMem) + tmi := &tcpMemInode{s: s, dir: tcpRMem} + tmf := &tcpMemFile{tcpMemInode: tmi} // Write the values. src := usermem.BytesIOSequence([]byte(c.str)) - if n, err := tm.DeprecatedPwritev(ctx, src, 0); n != int64(len(c.str)) || err != nil { - t.Errorf("DeprecatedPwritev, case = %q: got (%d, %v), wanted (%d, nil)", c.str, n, err, len(c.str)) + if n, err := tmf.Write(ctx, nil, src, 0); n != int64(len(c.str)) || err != nil { + t.Errorf("Write, case = %q: got (%d, %v), wanted (%d, nil)", c.str, n, err, len(c.str)) } // Read the values from the stack and check them. diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index 91bda8a95..41981a973 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -24,6 +24,7 @@ import ( "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/device" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" @@ -32,6 +33,7 @@ import ( "gvisor.googlesource.com/gvisor/pkg/sentry/usage" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" "gvisor.googlesource.com/gvisor/pkg/syserror" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) // getTaskMM returns t's MemoryManager. If getTaskMM succeeds, the MemoryManager's @@ -57,19 +59,19 @@ func getTaskMM(t *kernel.Task) (*mm.MemoryManager, error) { type taskDir struct { ramfs.Dir - // t is the associated kernel task that owns this file. - t *kernel.Task + t *kernel.Task + pidns *kernel.PIDNamespace } +var _ fs.InodeOperations = (*taskDir)(nil) + // newTaskDir creates a new proc task entry. func newTaskDir(t *kernel.Task, msrc *fs.MountSource, pidns *kernel.PIDNamespace, showSubtasks bool) *fs.Inode { - d := &taskDir{t: t} - // TODO: Set EUID/EGID based on dumpability. - d.InitDir(t, map[string]*fs.Inode{ + contents := map[string]*fs.Inode{ "auxv": newAuxvec(t, msrc), - "cmdline": newExecArgFile(t, msrc, cmdlineExecArg), + "cmdline": newExecArgInode(t, msrc, cmdlineExecArg), "comm": newComm(t, msrc), - "environ": newExecArgFile(t, msrc, environExecArg), + "environ": newExecArgInode(t, msrc, environExecArg), "exe": newExe(t, msrc), "fd": newFdDir(t, msrc), "fdinfo": newFdInfoDir(t, msrc), @@ -87,11 +89,18 @@ func newTaskDir(t *kernel.Task, msrc *fs.MountSource, pidns *kernel.PIDNamespace "statm": newStatm(t, msrc), "status": newStatus(t, msrc, pidns), "uid_map": newUIDMap(t, msrc), - }, fs.RootOwner, fs.FilePermsFromMode(0555)) + } if showSubtasks { - d.AddChild(t, "task", newSubtasks(t, msrc, pidns)) + contents["task"] = newSubtasks(t, msrc, pidns) } - return newFile(d, msrc, fs.SpecialDirectory, t) + + // TODO: Set EUID/EGID based on dumpability. + d := &taskDir{ + Dir: *ramfs.NewDir(t, contents, fs.RootOwner, fs.FilePermsFromMode(0555)), + t: t, + pidns: pidns, + } + return newProcInode(d, msrc, fs.SpecialDirectory, t) } // subtasks represents a /proc/TID/task directory. @@ -100,15 +109,19 @@ func newTaskDir(t *kernel.Task, msrc *fs.MountSource, pidns *kernel.PIDNamespace type subtasks struct { ramfs.Dir - t *kernel.Task - + t *kernel.Task pidns *kernel.PIDNamespace } +var _ fs.InodeOperations = (*subtasks)(nil) + func newSubtasks(t *kernel.Task, msrc *fs.MountSource, pidns *kernel.PIDNamespace) *fs.Inode { - s := &subtasks{t: t, pidns: pidns} - s.InitDir(t, nil, fs.RootOwner, fs.FilePermsFromMode(0555)) - return newFile(s, msrc, fs.SpecialDirectory, t) + s := &subtasks{ + Dir: *ramfs.NewDir(t, nil, fs.RootOwner, fs.FilePermsFromMode(0555)), + t: t, + pidns: pidns, + } + return newProcInode(s, msrc, fs.SpecialDirectory, t) } // UnstableAttr returns unstable attributes of the subtasks. @@ -123,35 +136,52 @@ func (s *subtasks) UnstableAttr(ctx context.Context, inode *fs.Inode) (fs.Unstab return uattr, nil } -// Lookup loads an Inode in a task's subtask directory into a Dirent. -func (s *subtasks) Lookup(ctx context.Context, dir *fs.Inode, p string) (*fs.Dirent, error) { - tid, err := strconv.ParseUint(p, 10, 32) - if err != nil { - return nil, syserror.ENOENT - } +// GetFile implements fs.InodeOperations.GetFile. +func (s *subtasks) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &subtasksFile{t: s.t, pidns: s.pidns}), nil +} - task := s.pidns.TaskWithID(kernel.ThreadID(tid)) - if task == nil { - return nil, syserror.ENOENT - } - if task.ThreadGroup() != s.t.ThreadGroup() { - return nil, syserror.ENOENT - } +// +stateify savable +type subtasksFile struct { + fsutil.DirFileOperations `state:"nosave"` - td := newTaskDir(task, dir.MountSource, s.pidns, false) - return fs.NewDirent(td, p), nil + t *kernel.Task + pidns *kernel.PIDNamespace } -// DeprecatedReaddir lists a task's subtask directory. -func (s *subtasks) DeprecatedReaddir(ctx context.Context, dirCtx *fs.DirCtx, offset int) (int, error) { - tasks := s.t.ThreadGroup().MemberIDs(s.pidns) +// Readdir implements fs.FileOperations.Readdir. +func (f *subtasksFile) Readdir(ctx context.Context, file *fs.File, ser fs.DentrySerializer) (int64, error) { + dirCtx := fs.DirCtx{ + Serializer: ser, + } + + // Note that unlike most Readdir implementations, the offset here is + // not an index into the subtasks, but rather the TID of the next + // subtask to emit. + offset := file.Offset() + + if offset == 0 { + // Serialize "." and "..". + root := fs.RootFromContext(ctx) + defer root.DecRef() + dot, dotdot := file.Dirent.GetDotAttrs(root) + if err := dirCtx.DirEmit(".", dot); err != nil { + return offset, err + } + if err := dirCtx.DirEmit("..", dotdot); err != nil { + return offset, err + } + } + + // Serialize tasks. + tasks := f.t.ThreadGroup().MemberIDs(f.pidns) taskInts := make([]int, 0, len(tasks)) for _, tid := range tasks { taskInts = append(taskInts, int(tid)) } // Find the task to start at. - idx := sort.SearchInts(taskInts, offset) + idx := sort.SearchInts(taskInts, int(offset)) if idx == len(taskInts) { return offset, nil } @@ -163,12 +193,33 @@ func (s *subtasks) DeprecatedReaddir(ctx context.Context, dirCtx *fs.DirCtx, off attr := fs.GenericDentAttr(fs.SpecialDirectory, device.ProcDevice) if err := dirCtx.DirEmit(name, attr); err != nil { // Returned offset is next tid to serialize. - return tid, err + return int64(tid), err } } // We serialized them all. Next offset should be higher than last // serialized tid. - return tid + 1, nil + return int64(tid) + 1, nil +} + +var _ fs.FileOperations = (*subtasksFile)(nil) + +// Lookup loads an Inode in a task's subtask directory into a Dirent. +func (s *subtasks) Lookup(ctx context.Context, dir *fs.Inode, p string) (*fs.Dirent, error) { + tid, err := strconv.ParseUint(p, 10, 32) + if err != nil { + return nil, syserror.ENOENT + } + + task := s.pidns.TaskWithID(kernel.ThreadID(tid)) + if task == nil { + return nil, syserror.ENOENT + } + if task.ThreadGroup() != s.t.ThreadGroup() { + return nil, syserror.ENOENT + } + + td := newTaskDir(task, dir.MountSource, s.pidns, false) + return fs.NewDirent(td, p), nil } // exe is an fs.InodeOperations symlink for the /proc/PID/exe file. @@ -181,9 +232,11 @@ type exe struct { } func newExe(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - exeSymlink := &exe{t: t} - exeSymlink.InitSymlink(t, fs.RootOwner, "") - return newFile(exeSymlink, msrc, fs.Symlink, t) + exeSymlink := &exe{ + Symlink: *ramfs.NewSymlink(t, fs.RootOwner, ""), + t: t, + } + return newProcInode(exeSymlink, msrc, fs.Symlink, t) } func (e *exe) executable() (d *fs.Dirent, err error) { @@ -231,55 +284,48 @@ func (e *exe) Readlink(ctx context.Context, inode *fs.Inode) (string, error) { return n, nil } -// namespaceFile represents a file in the namespacefs, such as the files in -// /proc/<pid>/ns. +// namespaceSymlink represents a symlink in the namespacefs, such as the files +// in /proc/<pid>/ns. // // +stateify savable -type namespaceFile struct { +type namespaceSymlink struct { ramfs.Symlink t *kernel.Task } -func newNamespaceFile(t *kernel.Task, msrc *fs.MountSource, name string) *fs.Inode { - n := &namespaceFile{t: t} - n.InitSymlink(t, fs.RootOwner, "") - +func newNamespaceSymlink(t *kernel.Task, msrc *fs.MountSource, name string) *fs.Inode { // TODO: 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. - n.Target = fmt.Sprintf("%s:[%d]", name, device.ProcDevice.NextIno()) - - return newFile(n, msrc, fs.Symlink, t) + target := fmt.Sprintf("%s:[%d]", name, device.ProcDevice.NextIno()) + n := &namespaceSymlink{ + Symlink: *ramfs.NewSymlink(t, fs.RootOwner, target), + t: t, + } + return newProcInode(n, msrc, fs.Symlink, t) } // Getlink implements fs.InodeOperations.Getlink. -func (n *namespaceFile) Getlink(ctx context.Context, inode *fs.Inode) (*fs.Dirent, error) { +func (n *namespaceSymlink) Getlink(ctx context.Context, inode *fs.Inode) (*fs.Dirent, error) { if !kernel.ContextCanTrace(ctx, n.t, false) { return nil, syserror.EACCES } // Create a new regular file to fake the namespace file. - node := &ramfs.Entry{} - node.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(0777)) - sattr := fs.StableAttr{ - DeviceID: device.ProcDevice.DeviceID(), - InodeID: device.ProcDevice.NextIno(), - BlockSize: usermem.PageSize, - Type: fs.RegularFile, - } - return fs.NewDirent(fs.NewInode(node, inode.MountSource, sattr), n.Symlink.Target), nil + iops := fsutil.NewNoReadWriteFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0777), linux.PROC_SUPER_MAGIC) + return fs.NewDirent(newProcInode(iops, inode.MountSource, fs.RegularFile, nil), n.Symlink.Target), nil } func newNamespaceDir(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - d := &ramfs.Dir{} - d.InitDir(t, map[string]*fs.Inode{ - "net": newNamespaceFile(t, msrc, "net"), - "pid": newNamespaceFile(t, msrc, "pid"), - "user": newNamespaceFile(t, msrc, "user"), - }, fs.RootOwner, fs.FilePermsFromMode(0511)) - return newFile(d, msrc, fs.SpecialDirectory, t) + contents := map[string]*fs.Inode{ + "net": newNamespaceSymlink(t, msrc, "net"), + "pid": newNamespaceSymlink(t, msrc, "pid"), + "user": newNamespaceSymlink(t, msrc, "user"), + } + d := ramfs.NewDir(t, contents, fs.RootOwner, fs.FilePermsFromMode(0511)) + return newProcInode(d, msrc, fs.SpecialDirectory, t) } // mapsData implements seqfile.SeqSource for /proc/[pid]/maps. @@ -290,7 +336,7 @@ type mapsData struct { } func newMaps(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - return newFile(seqfile.NewSeqFile(t, &mapsData{t}), msrc, fs.SpecialFile, t) + return newProcInode(seqfile.NewSeqFile(t, &mapsData{t}), msrc, fs.SpecialFile, t) } func (md *mapsData) mm() *mm.MemoryManager { @@ -330,7 +376,7 @@ type smapsData struct { } func newSmaps(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - return newFile(seqfile.NewSeqFile(t, &smapsData{t}), msrc, fs.SpecialFile, t) + return newProcInode(seqfile.NewSeqFile(t, &smapsData{t}), msrc, fs.SpecialFile, t) } func (sd *smapsData) mm() *mm.MemoryManager { @@ -376,7 +422,7 @@ type taskStatData struct { } func newTaskStat(t *kernel.Task, msrc *fs.MountSource, showSubtasks bool, pidns *kernel.PIDNamespace) *fs.Inode { - return newFile(seqfile.NewSeqFile(t, &taskStatData{t, showSubtasks /* tgstats */, pidns}), msrc, fs.SpecialFile, t) + return newProcInode(seqfile.NewSeqFile(t, &taskStatData{t, showSubtasks /* tgstats */, pidns}), msrc, fs.SpecialFile, t) } // NeedsUpdate returns whether the generation is old or not. @@ -450,7 +496,7 @@ type statmData struct { } func newStatm(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - return newFile(seqfile.NewSeqFile(t, &statmData{t}), msrc, fs.SpecialFile, t) + return newProcInode(seqfile.NewSeqFile(t, &statmData{t}), msrc, fs.SpecialFile, t) } // NeedsUpdate implements seqfile.SeqSource.NeedsUpdate. @@ -487,7 +533,7 @@ type statusData struct { } func newStatus(t *kernel.Task, msrc *fs.MountSource, pidns *kernel.PIDNamespace) *fs.Inode { - return newFile(seqfile.NewSeqFile(t, &statusData{t, pidns}), msrc, fs.SpecialFile, t) + return newProcInode(seqfile.NewSeqFile(t, &statusData{t, pidns}), msrc, fs.SpecialFile, t) } // NeedsUpdate implements seqfile.SeqSource.NeedsUpdate. @@ -552,7 +598,7 @@ type ioData struct { } func newIO(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - return newFile(seqfile.NewSeqFile(t, &ioData{t.ThreadGroup()}), msrc, fs.SpecialFile, t) + return newProcInode(seqfile.NewSeqFile(t, &ioData{t.ThreadGroup()}), msrc, fs.SpecialFile, t) } // NeedsUpdate returns whether the generation is old or not. @@ -590,25 +636,49 @@ func (i *ioData) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se // // +stateify savable type comm struct { - ramfs.Entry + fsutil.SimpleFileInode t *kernel.Task } // newComm returns a new comm file. func newComm(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - c := &comm{t: t} - c.InitEntry(t, fs.RootOwner, fs.FilePermsFromMode(0444)) - return newFile(c, msrc, fs.SpecialFile, t) + c := &comm{ + SimpleFileInode: *fsutil.NewSimpleFileInode(t, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + t: t, + } + return newProcInode(c, msrc, fs.SpecialFile, t) +} + +// GetFile implements fs.InodeOperations.GetFile. +func (c *comm) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &commFile{t: c.t}), nil +} + +// +stateify savable +type commFile struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoWrite `state:"nosave"` + + t *kernel.Task } -// DeprecatedPreadv reads the current command name. -func (c *comm) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { +var _ fs.FileOperations = (*commFile)(nil) + +// Read implements fs.FileOperations.Read. +func (f *commFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { if offset < 0 { return 0, syserror.EINVAL } - buf := []byte(c.t.Name() + "\n") + buf := []byte(f.t.Name() + "\n") if offset >= int64(len(buf)) { return 0, io.EOF } @@ -621,25 +691,47 @@ func (c *comm) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, off // // +stateify savable type auxvec struct { - ramfs.Entry + fsutil.SimpleFileInode t *kernel.Task } // newAuxvec returns a new auxvec file. func newAuxvec(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { - a := &auxvec{t: t} - a.InitEntry(t, fs.RootOwner, fs.FilePermsFromMode(0400)) - return newFile(a, msrc, fs.SpecialFile, t) + a := &auxvec{ + SimpleFileInode: *fsutil.NewSimpleFileInode(t, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + t: t, + } + return newProcInode(a, msrc, fs.SpecialFile, t) +} + +// GetFile implements fs.InodeOperations.GetFile. +func (a *auxvec) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &auxvecFile{t: a.t}), nil +} + +// +stateify savable +type auxvecFile struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoWrite `state:"nosave"` + + t *kernel.Task } -// DeprecatedPreadv reads the current auxiliary vector. -func (a *auxvec) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { +// Read implements fs.FileOperations.Read. +func (f *auxvecFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { if offset < 0 { return 0, syserror.EINVAL } - m, err := getTaskMM(a.t) + m, err := getTaskMM(f.t) if err != nil { return 0, err } diff --git a/pkg/sentry/fs/proc/uid_gid_map.go b/pkg/sentry/fs/proc/uid_gid_map.go index f70399686..815c40b7f 100644 --- a/pkg/sentry/fs/proc/uid_gid_map.go +++ b/pkg/sentry/fs/proc/uid_gid_map.go @@ -90,12 +90,13 @@ func newGIDMap(t *kernel.Task, msrc *fs.MountSource) *fs.Inode { } func newIDMap(t *kernel.Task, msrc *fs.MountSource, gids bool) *fs.Inode { - imsf := &idMapSeqFile{seqfile.SeqFile{SeqSource: &idMapSeqSource{ - t: t, - gids: gids, - }}} - imsf.InitEntry(t, fs.RootOwner, fs.FilePermsFromMode(0644)) - return newFile(imsf, msrc, fs.SpecialFile, t) + imsf := &idMapSeqFile{ + *seqfile.NewSeqFile(t, &idMapSeqSource{ + t: t, + gids: gids, + }), + } + return newProcInode(imsf, msrc, fs.SpecialFile, t) } func (imsf *idMapSeqFile) source() *idMapSeqSource { @@ -106,8 +107,8 @@ func (imsf *idMapSeqFile) source() *idMapSeqSource { // Linux 3.18, the limit is five lines." - user_namespaces(7) const maxIDMapLines = 5 -// DeprecatedPwritev implements fs.InodeOperations.DeprecatedPwritev. -func (imsf *idMapSeqFile) DeprecatedPwritev(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { +// Write implements fs.FileOperations.Write. +func (imsf *idMapSeqFile) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, offset int64) (int64, error) { // "In addition, the number of bytes written to the file must be less than // the system page size, and the write must be performed at the start of // the file ..." - user_namespaces(7) diff --git a/pkg/sentry/fs/proc/uptime.go b/pkg/sentry/fs/proc/uptime.go index 80c7ce0b4..40d0fd1fd 100644 --- a/pkg/sentry/fs/proc/uptime.go +++ b/pkg/sentry/fs/proc/uptime.go @@ -18,42 +18,64 @@ import ( "fmt" "io" + "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" ktime "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/time" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" "gvisor.googlesource.com/gvisor/pkg/syserror" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) // uptime is a file containing the system uptime. // // +stateify savable type uptime struct { - ramfs.Entry + fsutil.SimpleFileInode // The "start time" of the sandbox. startTime ktime.Time } // newUptime returns a new uptime file. -func (p *proc) newUptime(ctx context.Context, msrc *fs.MountSource) *fs.Inode { +func newUptime(ctx context.Context, msrc *fs.MountSource) *fs.Inode { u := &uptime{ - startTime: ktime.NowFromContext(ctx), + SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(0444), linux.PROC_SUPER_MAGIC), + startTime: ktime.NowFromContext(ctx), } - u.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(0444)) - return newFile(u, msrc, fs.SpecialFile, nil) + return newProcInode(u, msrc, fs.SpecialFile, nil) } -// DeprecatedPreadv reads the current uptime. -func (u *uptime) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { +// GetFile implements fs.InodeOperations.GetFile. +func (u *uptime) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &uptimeFile{startTime: u.startTime}), nil +} + +// +stateify savable +type uptimeFile struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoWrite `state:"nosave"` + + startTime ktime.Time +} + +// Read implements fs.FileOperations.Read. +func (f *uptimeFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { if offset < 0 { return 0, syserror.EINVAL } now := ktime.NowFromContext(ctx) // Pretend that we've spent zero time sleeping (second number). - s := []byte(fmt.Sprintf("%.2f 0.00\n", now.Sub(u.startTime).Seconds())) + s := []byte(fmt.Sprintf("%.2f 0.00\n", now.Sub(f.startTime).Seconds())) if offset >= int64(len(s)) { return 0, io.EOF } diff --git a/pkg/sentry/fs/ramfs/BUILD b/pkg/sentry/fs/ramfs/BUILD index a93ad6240..a476c9cce 100644 --- a/pkg/sentry/fs/ramfs/BUILD +++ b/pkg/sentry/fs/ramfs/BUILD @@ -6,8 +6,6 @@ go_library( name = "ramfs", srcs = [ "dir.go", - "file.go", - "ramfs.go", "socket.go", "symlink.go", "tree.go", @@ -15,14 +13,12 @@ go_library( importpath = "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs", visibility = ["//pkg/sentry:internal"], deps = [ - "//pkg/secio", + "//pkg/abi/linux", "//pkg/sentry/context", "//pkg/sentry/fs", "//pkg/sentry/fs/anon", "//pkg/sentry/fs/fsutil", "//pkg/sentry/kernel/time", - "//pkg/sentry/memmap", - "//pkg/sentry/safemem", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usermem", "//pkg/syserror", diff --git a/pkg/sentry/fs/ramfs/dir.go b/pkg/sentry/fs/ramfs/dir.go index 0a911b155..729f37694 100644 --- a/pkg/sentry/fs/ramfs/dir.go +++ b/pkg/sentry/fs/ramfs/dir.go @@ -18,10 +18,12 @@ import ( "sync" "syscall" + "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" + ktime "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/time" "gvisor.googlesource.com/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" "gvisor.googlesource.com/gvisor/pkg/syserror" ) @@ -47,7 +49,17 @@ type CreateOps struct { // // +stateify savable type Dir struct { - Entry + fsutil.InodeGenericChecker `state:"nosave"` + fsutil.InodeIsDirTruncate `state:"nosave"` + fsutil.InodeNoopRelease `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` + fsutil.InodeNotMappable `state:"nosave"` + fsutil.InodeNotSocket `state:"nosave"` + fsutil.InodeNotSymlink `state:"nosave"` + fsutil.InodeVirtual `state:"nosave"` + + fsutil.InodeSimpleAttributes + fsutil.InodeSimpleExtendedAttributes // CreateOps may be provided. // @@ -64,17 +76,23 @@ type Dir struct { children map[string]*fs.Inode // dentryMap is a sortedDentryMap containing entries for all children. - // Its entries ar kept up-to-date with d.children. + // Its entries are kept up-to-date with d.children. dentryMap *fs.SortedDentryMap } -// InitDir initializes a directory. -func (d *Dir) InitDir(ctx context.Context, contents map[string]*fs.Inode, owner fs.FileOwner, perms fs.FilePermissions) { - d.InitEntry(ctx, owner, perms) +var _ fs.InodeOperations = (*Dir)(nil) + +// NewDir returns a new Dir with the given contents and attributes. +func NewDir(ctx context.Context, contents map[string]*fs.Inode, owner fs.FileOwner, perms fs.FilePermissions) *Dir { + d := &Dir{ + InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, owner, perms, linux.RAMFS_MAGIC), + } + if contents == nil { contents = make(map[string]*fs.Inode) } d.children = contents + // Build the entries map ourselves, rather than calling addChildLocked, // because it will be faster. entries := make(map[string]fs.DentAttr, len(contents)) @@ -88,6 +106,8 @@ func (d *Dir) InitDir(ctx context.Context, contents map[string]*fs.Inode, owner // Directories have an extra link, corresponding to '.'. d.AddLink() + + return d } // addChildLocked add the child inode, inheriting its reference. @@ -124,17 +144,24 @@ func (d *Dir) FindChild(name string) (*fs.Inode, bool) { return child, ok } +// Children returns the names and DentAttrs of all children. It can be used to +// implement Readdir for types that embed ramfs.Dir. +func (d *Dir) Children() ([]string, map[string]fs.DentAttr) { + d.mu.Lock() + defer d.mu.Unlock() + return d.dentryMap.GetAll() +} + // removeChildLocked attempts to remove an entry from this directory. -// This Entry's mutex must be held. It returns the removed Inode. func (d *Dir) removeChildLocked(ctx context.Context, name string) (*fs.Inode, error) { inode, ok := d.children[name] if !ok { - return nil, ErrNotFound + return nil, syserror.EACCES } delete(d.children, name) d.dentryMap.Remove(name) - d.Entry.NotifyModification(ctx) + d.NotifyModification(ctx) // If the child was a subdirectory, then we must decrement this dir's // link count which was the child's ".." directory entry. @@ -143,7 +170,7 @@ func (d *Dir) removeChildLocked(ctx context.Context, name string) (*fs.Inode, er } // Update ctime. - inode.NotifyStatusChange(ctx) + inode.InodeOperations.NotifyStatusChange(ctx) // Given we're now removing this inode to the directory we must also // decrease its link count. Similarly it is increased in addChildLocked. @@ -152,8 +179,8 @@ func (d *Dir) removeChildLocked(ctx context.Context, name string) (*fs.Inode, er return inode, nil } -// RemoveEntry attempts to remove an entry from this directory. -func (d *Dir) RemoveEntry(ctx context.Context, name string) error { +// Remove removes the named non-directory. +func (d *Dir) Remove(ctx context.Context, _ *fs.Inode, name string) error { d.mu.Lock() defer d.mu.Unlock() inode, err := d.removeChildLocked(ctx, name) @@ -166,27 +193,23 @@ func (d *Dir) RemoveEntry(ctx context.Context, name string) error { return nil } -// Remove removes the named non-directory. -func (d *Dir) Remove(ctx context.Context, dir *fs.Inode, name string) error { - return d.RemoveEntry(ctx, name) -} - // RemoveDirectory removes the named directory. -func (d *Dir) RemoveDirectory(ctx context.Context, dir *fs.Inode, name string) error { +func (d *Dir) RemoveDirectory(ctx context.Context, _ *fs.Inode, name string) error { d.mu.Lock() defer d.mu.Unlock() - n, err := d.walkLocked(ctx, name) + // Get the child and make sure it is not empty. + childInode, err := d.walkLocked(ctx, name) if err != nil { return err } - dirCtx := &fs.DirCtx{} - if _, err := n.HandleOps().DeprecatedReaddir(ctx, dirCtx, 0); err != nil { + if ok, err := hasChildren(ctx, childInode); err != nil { return err + } else if ok { + return syserror.ENOTEMPTY } - if len(dirCtx.DentAttrs()) > 0 { - return ErrNotEmpty - } + + // Child was empty. Proceed with removal. inode, err := d.removeChildLocked(ctx, name) if err != nil { return err @@ -195,11 +218,11 @@ func (d *Dir) RemoveDirectory(ctx context.Context, dir *fs.Inode, name string) e // Remove our reference on the inode. inode.DecRef() - return err + return nil } // Lookup loads an inode at p into a Dirent. -func (d *Dir) Lookup(ctx context.Context, dir *fs.Inode, p string) (*fs.Dirent, error) { +func (d *Dir) Lookup(ctx context.Context, _ *fs.Inode, p string) (*fs.Dirent, error) { d.mu.Lock() defer d.mu.Unlock() @@ -214,9 +237,9 @@ func (d *Dir) Lookup(ctx context.Context, dir *fs.Inode, p string) (*fs.Dirent, return fs.NewDirent(inode, p), nil } -// walkLocked must be called with this Entry's mutex held. +// walkLocked must be called with d.mu held. func (d *Dir) walkLocked(ctx context.Context, p string) (*fs.Inode, error) { - d.Entry.NotifyAccess(ctx) + d.NotifyAccess(ctx) // Lookup a child node. if inode, ok := d.children[p]; ok { @@ -244,7 +267,7 @@ func (d *Dir) createInodeOperationsCommon(ctx context.Context, name string, make } d.addChildLocked(name, inode) - d.Entry.NotifyModification(ctx) + d.NotifyModification(ctx) return inode, nil } @@ -252,7 +275,7 @@ func (d *Dir) createInodeOperationsCommon(ctx context.Context, name string, make // Create creates a new Inode with the given name and returns its File. func (d *Dir) Create(ctx context.Context, dir *fs.Inode, name string, flags fs.FileFlags, perms fs.FilePermissions) (*fs.File, error) { if d.CreateOps == nil || d.CreateOps.NewFile == nil { - return nil, ErrDenied + return nil, syserror.EACCES } inode, err := d.createInodeOperationsCommon(ctx, name, func() (*fs.Inode, error) { @@ -274,7 +297,7 @@ func (d *Dir) Create(ctx context.Context, dir *fs.Inode, name string, flags fs.F // CreateLink returns a new link. func (d *Dir) CreateLink(ctx context.Context, dir *fs.Inode, oldname, newname string) error { if d.CreateOps == nil || d.CreateOps.NewSymlink == nil { - return ErrDenied + return syserror.EACCES } _, err := d.createInodeOperationsCommon(ctx, newname, func() (*fs.Inode, error) { return d.NewSymlink(ctx, dir, oldname) @@ -292,10 +315,10 @@ func (d *Dir) CreateHardLink(ctx context.Context, dir *fs.Inode, target *fs.Inod // The link count will be incremented in addChildLocked. d.addChildLocked(name, target) - d.Entry.NotifyModification(ctx) + d.NotifyModification(ctx) // Update ctime. - target.NotifyStatusChange(ctx) + target.InodeOperations.NotifyStatusChange(ctx) return nil } @@ -303,7 +326,7 @@ func (d *Dir) CreateHardLink(ctx context.Context, dir *fs.Inode, target *fs.Inod // CreateDirectory returns a new subdirectory. func (d *Dir) CreateDirectory(ctx context.Context, dir *fs.Inode, name string, perms fs.FilePermissions) error { if d.CreateOps == nil || d.CreateOps.NewDir == nil { - return ErrDenied + return syserror.EACCES } _, err := d.createInodeOperationsCommon(ctx, name, func() (*fs.Inode, error) { return d.NewDir(ctx, dir, perms) @@ -316,7 +339,7 @@ func (d *Dir) CreateDirectory(ctx context.Context, dir *fs.Inode, name string, p // Bind implements fs.InodeOperations.Bind. func (d *Dir) Bind(ctx context.Context, dir *fs.Inode, name string, ep transport.BoundEndpoint, perms fs.FilePermissions) (*fs.Dirent, error) { if d.CreateOps == nil || d.CreateOps.NewBoundEndpoint == nil { - return nil, ErrDenied + return nil, syserror.EACCES } inode, err := d.createInodeOperationsCommon(ctx, name, func() (*fs.Inode, error) { return d.NewBoundEndpoint(ctx, dir, ep, perms) @@ -335,7 +358,7 @@ func (d *Dir) Bind(ctx context.Context, dir *fs.Inode, name string, ep transport // CreateFifo implements fs.InodeOperations.CreateFifo. func (d *Dir) CreateFifo(ctx context.Context, dir *fs.Inode, name string, perms fs.FilePermissions) error { if d.CreateOps == nil || d.CreateOps.NewFifo == nil { - return ErrDenied + return syserror.EACCES } _, err := d.createInodeOperationsCommon(ctx, name, func() (*fs.Inode, error) { return d.NewFifo(ctx, dir, perms) @@ -343,29 +366,125 @@ func (d *Dir) CreateFifo(ctx context.Context, dir *fs.Inode, name string, perms return err } -func (d *Dir) readdirLocked(ctx context.Context, dirCtx *fs.DirCtx, offset int) (int, error) { - // Serialize the entries in dentryMap. - n, err := fs.GenericReaddir(dirCtx, d.dentryMap) +// GetFile implements fs.InodeOperations.GetFile. +func (d *Dir) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + flags.Pread = true + return fs.NewFile(ctx, dirent, flags, &dirFileOperations{dir: d}), nil +} - // Touch the access time. - d.Entry.NotifyAccess(ctx) +// Rename implements fs.InodeOperations.Rename. +func (*Dir) Rename(ctx context.Context, oldParent *fs.Inode, oldName string, newParent *fs.Inode, newName string) error { + return Rename(ctx, oldParent.InodeOperations, oldName, newParent.InodeOperations, newName) +} +// dirFileOperations implements fs.FileOperations for a ramfs directory. +// +// +stateify savable +type dirFileOperations struct { + fsutil.DirFileOperations `state:"nosave"` + + // dirCursor contains the name of the last directory entry that was + // serialized. + dirCursor string + + // dir is the ramfs dir that this file corresponds to. + dir *Dir +} + +var _ fs.FileOperations = (*dirFileOperations)(nil) + +// Seek implements fs.FileOperations.Seek. +func (dfo *dirFileOperations) Seek(ctx context.Context, file *fs.File, whence fs.SeekWhence, offset int64) (int64, error) { + return fsutil.SeekWithDirCursor(ctx, file, whence, offset, &dfo.dirCursor) +} + +// IterateDir implements DirIterator.IterateDir. +func (dfo *dirFileOperations) IterateDir(ctx context.Context, dirCtx *fs.DirCtx, offset int) (int, error) { + dfo.dir.mu.Lock() + defer dfo.dir.mu.Unlock() + + n, err := fs.GenericReaddir(dirCtx, dfo.dir.dentryMap) return offset + n, err } -// DeprecatedReaddir emits the entries contained in this directory. -func (d *Dir) DeprecatedReaddir(ctx context.Context, dirCtx *fs.DirCtx, offset int) (int, error) { - d.mu.Lock() - defer d.mu.Unlock() - return d.readdirLocked(ctx, dirCtx, offset) +// Readdir implements FileOperations.Readdir. +func (dfo *dirFileOperations) Readdir(ctx context.Context, file *fs.File, serializer fs.DentrySerializer) (int64, error) { + root := fs.RootFromContext(ctx) + defer root.DecRef() + dirCtx := &fs.DirCtx{ + Serializer: serializer, + DirCursor: &dfo.dirCursor, + } + dfo.dir.mu.Lock() + dfo.dir.InodeSimpleAttributes.Unstable.AccessTime = ktime.NowFromContext(ctx) + dfo.dir.mu.Unlock() + return fs.DirentReaddir(ctx, file.Dirent, dfo, root, dirCtx, file.Offset()) } -// DeprecatedPreadv always returns ErrIsDirectory -func (*Dir) DeprecatedPreadv(context.Context, usermem.IOSequence, int64) (int64, error) { - return 0, ErrIsDirectory +// hasChildren is a helper method that determines whether an arbitrary inode +// (not necessarily ramfs) has any children. +func hasChildren(ctx context.Context, inode *fs.Inode) (bool, error) { + // Take an extra ref on inode which will be given to the dirent and + // dropped when that dirent is destroyed. + inode.IncRef() + d := fs.NewTransientDirent(inode) + defer d.DecRef() + + file, err := inode.GetFile(ctx, d, fs.FileFlags{Read: true}) + if err != nil { + return false, err + } + defer file.DecRef() + + ser := &fs.CollectEntriesSerializer{} + if err := file.Readdir(ctx, ser); err != nil { + return false, err + } + // We will always write "." and "..", so ignore those two. + if ser.Written() > 2 { + return true, nil + } + return false, nil } -// DeprecatedPwritev always returns ErrIsDirectory -func (*Dir) DeprecatedPwritev(context.Context, usermem.IOSequence, int64) (int64, error) { - return 0, ErrIsDirectory +// Rename renames from a *ramfs.Dir to another *ramfs.Dir. +func Rename(ctx context.Context, oldParent fs.InodeOperations, oldName string, newParent fs.InodeOperations, newName string) error { + op, ok := oldParent.(*Dir) + if !ok { + return syserror.EXDEV + } + np, ok := newParent.(*Dir) + if !ok { + return syserror.EXDEV + } + + np.mu.Lock() + defer np.mu.Unlock() + + // Check whether the ramfs entry to be replaced is a non-empty directory. + if replaced, ok := np.children[newName]; ok { + if fs.IsDir(replaced.StableAttr) { + if ok, err := hasChildren(ctx, replaced); err != nil { + return err + } else if ok { + return syserror.ENOTEMPTY + } + } + } + + // Be careful, we may have already grabbed this mutex above. + if op != np { + op.mu.Lock() + defer op.mu.Unlock() + } + + // Do the swap. + n := op.children[oldName] + op.removeChildLocked(ctx, oldName) + np.addChildLocked(newName, n) + + // Update ctime. + n.InodeOperations.NotifyStatusChange(ctx) + + return nil } diff --git a/pkg/sentry/fs/ramfs/file.go b/pkg/sentry/fs/ramfs/file.go deleted file mode 100644 index b7fc98ffc..000000000 --- a/pkg/sentry/fs/ramfs/file.go +++ /dev/null @@ -1,150 +0,0 @@ -// Copyright 2018 Google LLC -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ramfs - -import ( - "io" - "sync" - - "gvisor.googlesource.com/gvisor/pkg/secio" - "gvisor.googlesource.com/gvisor/pkg/sentry/context" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - "gvisor.googlesource.com/gvisor/pkg/sentry/safemem" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" - "gvisor.googlesource.com/gvisor/pkg/syserror" -) - -// File represents a unique file. It uses a simple byte slice as storage, and -// thus should only be used for small files. -// -// A File is not mappable. -// -// +stateify savable -type File struct { - Entry - - // mu protects the fields below. - mu sync.Mutex `state:"nosave"` - - // data tracks backing data for the file. - data []byte -} - -// InitFile initializes a file. -func (f *File) InitFile(ctx context.Context, owner fs.FileOwner, perms fs.FilePermissions) { - f.InitEntry(ctx, owner, perms) -} - -// UnstableAttr returns unstable attributes of this ramfs file. -func (f *File) UnstableAttr(ctx context.Context, inode *fs.Inode) (fs.UnstableAttr, error) { - f.mu.Lock() - defer f.mu.Unlock() - - uattr, _ := f.Entry.UnstableAttr(ctx, inode) - uattr.Size = int64(len(f.data)) - uattr.Usage = f.usageLocked() - - return uattr, nil -} - -// usageLocked returns the disk usage. Caller must hold f.mu. -func (f *File) usageLocked() int64 { - return int64(len(f.data)) -} - -// Append appends the given data. This is for internal use. -func (f *File) Append(data []byte) { - f.mu.Lock() - defer f.mu.Unlock() - f.data = append(f.data, data...) -} - -// Truncate truncates this node. -func (f *File) Truncate(ctx context.Context, inode *fs.Inode, l int64) error { - f.mu.Lock() - defer f.mu.Unlock() - if l < int64(len(f.data)) { - // Remove excess bytes. - f.data = f.data[:l] - return nil - } else if l > int64(len(f.data)) { - // Create a new slice with size l, and copy f.data into it. - d := make([]byte, l) - copy(d, f.data) - f.data = d - } - f.Entry.NotifyModification(ctx) - return nil -} - -// ReadAt implements io.ReaderAt. -func (f *File) ReadAt(data []byte, offset int64) (int, error) { - if offset < 0 { - return 0, ErrInvalidOp - } - if offset >= int64(len(f.data)) { - return 0, io.EOF - } - n := copy(data, f.data[offset:]) - // Did we read past the end? - if offset+int64(len(data)) >= int64(len(f.data)) { - return n, io.EOF - } - return n, nil -} - -// DeprecatedPreadv reads into a collection of slices from a given offset. -func (f *File) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { - f.mu.Lock() - defer f.mu.Unlock() - if offset >= int64(len(f.data)) { - return 0, io.EOF - } - n, err := dst.CopyOut(ctx, f.data[offset:]) - if n > 0 { - f.Entry.NotifyAccess(ctx) - } - return int64(n), err -} - -// WriteAt implements io.WriterAt. -func (f *File) WriteAt(data []byte, offset int64) (int, error) { - if offset < 0 { - return 0, ErrInvalidOp - } - newLen := offset + int64(len(data)) - if newLen < 0 { - // Overflow. - return 0, syserror.EINVAL - } - if newLen > int64(len(f.data)) { - // Copy f.data into new slice with expanded length. - d := make([]byte, newLen) - copy(d, f.data) - f.data = d - } - return copy(f.data[offset:], data), nil -} - -// DeprecatedPwritev writes from a collection of slices at a given offset. -func (f *File) DeprecatedPwritev(ctx context.Context, src usermem.IOSequence, offset int64) (int64, error) { - f.mu.Lock() - defer f.mu.Unlock() - n, err := src.CopyInTo(ctx, safemem.FromIOWriter{secio.NewOffsetWriter(f, offset)}) - if n > 0 { - f.Entry.NotifyModification(ctx) - } - return n, err -} diff --git a/pkg/sentry/fs/ramfs/ramfs.go b/pkg/sentry/fs/ramfs/ramfs.go deleted file mode 100644 index d77688a34..000000000 --- a/pkg/sentry/fs/ramfs/ramfs.go +++ /dev/null @@ -1,441 +0,0 @@ -// Copyright 2018 Google LLC -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package ramfs implements an in-memory file system that can be associated with -// any device. -package ramfs - -import ( - "errors" - "sync" - "syscall" - - "gvisor.googlesource.com/gvisor/pkg/sentry/context" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" - ktime "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/time" - "gvisor.googlesource.com/gvisor/pkg/sentry/memmap" - "gvisor.googlesource.com/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" - "gvisor.googlesource.com/gvisor/pkg/syserror" - "gvisor.googlesource.com/gvisor/pkg/waiter" -) - -var ( - // ErrInvalidOp indicates the operation is not valid. - ErrInvalidOp = errors.New("invalid operation") - - // ErrDenied indicates the operation was denied. - ErrDenied = errors.New("operation denied") - - // ErrNotFound indicates that a node was not found on a walk. - ErrNotFound = errors.New("node not found") - - // ErrCrossDevice indicates a cross-device link or rename. - ErrCrossDevice = errors.New("can't link across filesystems") - - // ErrIsDirectory indicates that the operation failed because - // the node is a directory. - ErrIsDirectory = errors.New("is a directory") - - // ErrNotDirectory indicates that the operation failed because - // the node is a not directory. - ErrNotDirectory = errors.New("not a directory") - - // ErrNotEmpty indicates that the operation failed because the - // directory is not empty. - ErrNotEmpty = errors.New("directory not empty") -) - -// Entry represents common internal state for file and directory nodes. -// This may be used by other packages to easily create ramfs files. -// -// +stateify savable -type Entry struct { - waiter.AlwaysReady `state:"nosave"` - fsutil.NoMappable `state:"nosave"` - fsutil.NoopWriteOut `state:"nosave"` - fsutil.InodeNotSocket `state:"nosave"` - - // mu protects the fields below. - mu sync.Mutex `state:"nosave"` - - // unstable is unstable attributes. - unstable fs.UnstableAttr - - // xattrs are the extended attributes of the Entry. - xattrs map[string][]byte -} - -// InitEntry initializes an entry. -func (e *Entry) InitEntry(ctx context.Context, owner fs.FileOwner, p fs.FilePermissions) { - e.InitEntryWithAttr(ctx, fs.WithCurrentTime(ctx, fs.UnstableAttr{ - Owner: owner, - Perms: p, - // Always start unlinked. - Links: 0, - })) -} - -// InitEntryWithAttr initializes an entry with a complete set of attributes. -func (e *Entry) InitEntryWithAttr(ctx context.Context, uattr fs.UnstableAttr) { - e.unstable = uattr - e.xattrs = make(map[string][]byte) -} - -// UnstableAttr implements fs.InodeOperations.UnstableAttr. -func (e *Entry) UnstableAttr(ctx context.Context, inode *fs.Inode) (fs.UnstableAttr, error) { - e.mu.Lock() - attr := e.unstable - e.mu.Unlock() - return attr, nil -} - -// Check implements fs.InodeOperations.Check. -func (*Entry) Check(ctx context.Context, inode *fs.Inode, p fs.PermMask) bool { - return fs.ContextCanAccessFile(ctx, inode, p) -} - -// Getxattr implements fs.InodeOperations.Getxattr. -func (e *Entry) Getxattr(inode *fs.Inode, name string) ([]byte, error) { - // Hot path. Avoid defers. - e.mu.Lock() - value, ok := e.xattrs[name] - e.mu.Unlock() - if ok { - return value, nil - } - return nil, syserror.ENOATTR -} - -// Setxattr implements fs.InodeOperations.Setxattr. -func (e *Entry) Setxattr(inode *fs.Inode, name string, value []byte) error { - e.mu.Lock() - e.xattrs[name] = value - e.mu.Unlock() - return nil -} - -// Listxattr implements fs.InodeOperations.Listxattr. -func (e *Entry) Listxattr(inode *fs.Inode) (map[string]struct{}, error) { - e.mu.Lock() - names := make(map[string]struct{}, len(e.xattrs)) - for name := range e.xattrs { - names[name] = struct{}{} - } - e.mu.Unlock() - return names, nil -} - -// GetFile returns a fs.File backed by the dirent argument and flags. -func (*Entry) GetFile(ctx context.Context, d *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { - return fsutil.NewHandle(ctx, d, flags, d.Inode.HandleOps()), nil -} - -// SetPermissions always sets the permissions. -func (e *Entry) SetPermissions(ctx context.Context, inode *fs.Inode, p fs.FilePermissions) bool { - e.mu.Lock() - e.unstable.Perms = p - e.unstable.StatusChangeTime = ktime.NowFromContext(ctx) - e.mu.Unlock() - return true -} - -// SetOwner always sets ownership. -func (e *Entry) SetOwner(ctx context.Context, inode *fs.Inode, owner fs.FileOwner) error { - e.mu.Lock() - if owner.UID.Ok() { - e.unstable.Owner.UID = owner.UID - } - if owner.GID.Ok() { - e.unstable.Owner.GID = owner.GID - } - e.mu.Unlock() - return nil -} - -// SetTimestamps sets the timestamps. -func (e *Entry) SetTimestamps(ctx context.Context, inode *fs.Inode, ts fs.TimeSpec) error { - if ts.ATimeOmit && ts.MTimeOmit { - return nil - } - - e.mu.Lock() - now := ktime.NowFromContext(ctx) - if !ts.ATimeOmit { - if ts.ATimeSetSystemTime { - e.unstable.AccessTime = now - } else { - e.unstable.AccessTime = ts.ATime - } - } - if !ts.MTimeOmit { - if ts.MTimeSetSystemTime { - e.unstable.ModificationTime = now - } else { - e.unstable.ModificationTime = ts.MTime - } - } - e.unstable.StatusChangeTime = now - e.mu.Unlock() - return nil -} - -// NotifyStatusChange updates the status change time (ctime). -func (e *Entry) NotifyStatusChange(ctx context.Context) { - e.mu.Lock() - e.unstable.StatusChangeTime = ktime.NowFromContext(ctx) - e.mu.Unlock() -} - -// StatusChangeTime returns the last status change time for this node. -func (e *Entry) StatusChangeTime() ktime.Time { - e.mu.Lock() - t := e.unstable.StatusChangeTime - e.mu.Unlock() - return t -} - -// NotifyModification updates the modification time and the status change time. -func (e *Entry) NotifyModification(ctx context.Context) { - e.mu.Lock() - now := ktime.NowFromContext(ctx) - e.unstable.ModificationTime = now - e.unstable.StatusChangeTime = now - e.mu.Unlock() -} - -// ModificationTime returns the last modification time for this node. -func (e *Entry) ModificationTime() ktime.Time { - e.mu.Lock() - t := e.unstable.ModificationTime - e.mu.Unlock() - return t -} - -// NotifyAccess updates the access time. -func (e *Entry) NotifyAccess(ctx context.Context) { - e.mu.Lock() - now := ktime.NowFromContext(ctx) - e.unstable.AccessTime = now - e.mu.Unlock() -} - -// AccessTime returns the last access time for this node. -func (e *Entry) AccessTime() ktime.Time { - e.mu.Lock() - t := e.unstable.AccessTime - e.mu.Unlock() - return t -} - -// Permissions returns permissions on this entry. -func (e *Entry) Permissions() fs.FilePermissions { - e.mu.Lock() - p := e.unstable.Perms - e.mu.Unlock() - return p -} - -// Lookup is not supported by default. -func (*Entry) Lookup(context.Context, *fs.Inode, string) (*fs.Dirent, error) { - return nil, ErrInvalidOp -} - -// Create is not supported by default. -func (*Entry) Create(context.Context, *fs.Inode, string, fs.FileFlags, fs.FilePermissions) (*fs.File, error) { - return nil, ErrInvalidOp -} - -// CreateLink is not supported by default. -func (*Entry) CreateLink(context.Context, *fs.Inode, string, string) error { - return ErrInvalidOp -} - -// CreateHardLink is not supported by default. -func (*Entry) CreateHardLink(context.Context, *fs.Inode, *fs.Inode, string) error { - return ErrInvalidOp -} - -// IsVirtual returns true. -func (*Entry) IsVirtual() bool { - return true -} - -// CreateDirectory is not supported by default. -func (*Entry) CreateDirectory(context.Context, *fs.Inode, string, fs.FilePermissions) error { - return ErrInvalidOp -} - -// Bind is not supported by default. -func (*Entry) Bind(context.Context, *fs.Inode, string, transport.BoundEndpoint, fs.FilePermissions) (*fs.Dirent, error) { - return nil, ErrInvalidOp -} - -// CreateFifo implements fs.InodeOperations.CreateFifo. CreateFifo is not supported by -// default. -func (*Entry) CreateFifo(context.Context, *fs.Inode, string, fs.FilePermissions) error { - return ErrInvalidOp -} - -// Remove is not supported by default. -func (*Entry) Remove(context.Context, *fs.Inode, string) error { - return ErrInvalidOp -} - -// RemoveDirectory is not supported by default. -func (*Entry) RemoveDirectory(context.Context, *fs.Inode, string) error { - return ErrInvalidOp -} - -// StatFS always returns ENOSYS. -func (*Entry) StatFS(context.Context) (fs.Info, error) { - return fs.Info{}, syscall.ENOSYS -} - -// Rename implements fs.InodeOperations.Rename. -func (e *Entry) Rename(ctx context.Context, oldParent *fs.Inode, oldName string, newParent *fs.Inode, newName string) error { - return Rename(ctx, oldParent.InodeOperations, oldName, newParent.InodeOperations, newName) -} - -// Rename renames from a *ramfs.Dir to another *ramfs.Dir. -func Rename(ctx context.Context, oldParent fs.InodeOperations, oldName string, newParent fs.InodeOperations, newName string) error { - op, ok := oldParent.(*Dir) - if !ok { - return ErrCrossDevice - } - np, ok := newParent.(*Dir) - if !ok { - return ErrCrossDevice - } - - np.mu.Lock() - defer np.mu.Unlock() - - // Check whether the ramfs entry to be replaced is a non-empty directory. - if replaced, ok := np.children[newName]; ok { - if fs.IsDir(replaced.StableAttr) { - // FIXME: simplify by pinning children of ramfs-backed directories - // in the Dirent tree: this allows us to generalize ramfs operations without - // relying on an implementation of Readdir (which may do anything, like require - // that the file be open ... which would be reasonable). - dirCtx := &fs.DirCtx{} - _, err := replaced.HandleOps().DeprecatedReaddir(ctx, dirCtx, 0) - if err != nil { - return err - } - attrs := dirCtx.DentAttrs() - - // ramfs-backed directories should not contain "." and "..", but we do this - // just in case. - delete(attrs, ".") - delete(attrs, "..") - - // If the directory to be replaced is not empty, reject the rename. - if len(attrs) != 0 { - return ErrNotEmpty - } - } - } - - // Be careful, we may have already grabbed this mutex above. - if op != np { - op.mu.Lock() - defer op.mu.Unlock() - } - - // Do the swap. - n := op.children[oldName] - op.removeChildLocked(ctx, oldName) - np.addChildLocked(newName, n) - - // Update ctime. - n.NotifyStatusChange(ctx) - - return nil -} - -// Truncate is not supported by default. -func (*Entry) Truncate(context.Context, *fs.Inode, int64) error { - return ErrInvalidOp -} - -// Readlink always returns ENOLINK. -func (*Entry) Readlink(context.Context, *fs.Inode) (string, error) { - return "", syscall.ENOLINK -} - -// Getlink always returns ENOLINK. -func (*Entry) Getlink(context.Context, *fs.Inode) (*fs.Dirent, error) { - return nil, syscall.ENOLINK -} - -// Release is a no-op. -func (e *Entry) Release(context.Context) {} - -// AddLink implements InodeOperationss.AddLink. -func (e *Entry) AddLink() { - e.mu.Lock() - e.unstable.Links++ - e.mu.Unlock() -} - -// DropLink implements InodeOperationss.DropLink. -func (e *Entry) DropLink() { - e.mu.Lock() - e.unstable.Links-- - e.mu.Unlock() -} - -// DeprecatedReaddir is not supported by default. -func (*Entry) DeprecatedReaddir(context.Context, *fs.DirCtx, int) (int, error) { - return 0, ErrNotDirectory -} - -// DeprecatedPreadv always returns ErrInvalidOp. -func (*Entry) DeprecatedPreadv(context.Context, usermem.IOSequence, int64) (int64, error) { - return 0, ErrInvalidOp -} - -// DeprecatedPwritev always returns ErrInvalidOp. -func (*Entry) DeprecatedPwritev(context.Context, usermem.IOSequence, int64) (int64, error) { - return 0, ErrInvalidOp -} - -// DeprecatedFsync is a noop. -func (*Entry) DeprecatedFsync() error { - // Ignore, this is in memory. - return nil -} - -// DeprecatedFlush always returns nil. -func (*Entry) DeprecatedFlush() error { - return nil -} - -// DeprecatedMappable implements fs.InodeOperations.DeprecatedMappable. -func (*Entry) DeprecatedMappable(context.Context, *fs.Inode) (memmap.Mappable, bool) { - return nil, false -} - -func init() { - // Register ramfs errors. - syserror.AddErrorTranslation(ErrInvalidOp, syscall.EINVAL) - syserror.AddErrorTranslation(ErrDenied, syscall.EACCES) - syserror.AddErrorTranslation(ErrNotFound, syscall.ENOENT) - syserror.AddErrorTranslation(ErrCrossDevice, syscall.EXDEV) - syserror.AddErrorTranslation(ErrIsDirectory, syscall.EISDIR) - syserror.AddErrorTranslation(ErrNotDirectory, syscall.ENOTDIR) - syserror.AddErrorTranslation(ErrNotEmpty, syscall.ENOTEMPTY) -} diff --git a/pkg/sentry/fs/ramfs/socket.go b/pkg/sentry/fs/ramfs/socket.go index 8c81478c8..2c1295897 100644 --- a/pkg/sentry/fs/ramfs/socket.go +++ b/pkg/sentry/fs/ramfs/socket.go @@ -15,25 +15,42 @@ package ramfs import ( + "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) // Socket represents a socket. // // +stateify savable type Socket struct { - Entry + fsutil.InodeGenericChecker `state:"nosave"` + fsutil.InodeNoopRelease `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` + fsutil.InodeNotDirectory `state:"nosave"` + fsutil.InodeNotMappable `state:"nosave"` + fsutil.InodeNotSymlink `state:"nosave"` + fsutil.InodeNotTruncatable `state:"nosave"` + fsutil.InodeVirtual `state:"nosave"` + + fsutil.InodeSimpleAttributes + fsutil.InodeSimpleExtendedAttributes // ep is the bound endpoint. ep transport.BoundEndpoint } -// InitSocket initializes a socket. -func (s *Socket) InitSocket(ctx context.Context, ep transport.BoundEndpoint, owner fs.FileOwner, perms fs.FilePermissions) { - s.InitEntry(ctx, owner, perms) - s.ep = ep +var _ fs.InodeOperations = (*Socket)(nil) + +// NewSocket returns a new Socket. +func NewSocket(ctx context.Context, ep transport.BoundEndpoint, owner fs.FileOwner, perms fs.FilePermissions) *Socket { + return &Socket{ + InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, owner, perms, linux.SOCKFS_MAGIC), + ep: ep, + } } // BoundEndpoint returns the socket data. @@ -42,3 +59,24 @@ func (s *Socket) BoundEndpoint(*fs.Inode, string) transport.BoundEndpoint { // care about the path argument. return s.ep } + +// GetFile implements fs.FileOperations.GetFile. +func (s *Socket) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &socketFileOperations{}), nil +} + +// +stateify savable +type socketFileOperations struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNoRead `state:"nosave"` + fsutil.FileNoSeek `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoWrite `state:"nosave"` +} + +var _ fs.FileOperations = (*socketFileOperations)(nil) diff --git a/pkg/sentry/fs/ramfs/symlink.go b/pkg/sentry/fs/ramfs/symlink.go index a21fac2c7..47dae380b 100644 --- a/pkg/sentry/fs/ramfs/symlink.go +++ b/pkg/sentry/fs/ramfs/symlink.go @@ -15,44 +15,55 @@ package ramfs import ( - "sync" - + "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" + "gvisor.googlesource.com/gvisor/pkg/waiter" ) // Symlink represents a symlink. // // +stateify savable type Symlink struct { - Entry + fsutil.InodeGenericChecker `state:"nosave"` + fsutil.InodeNoopRelease `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` + fsutil.InodeNotDirectory `state:"nosave"` + fsutil.InodeNotMappable `state:"nosave"` + fsutil.InodeNotTruncatable `state:"nosave"` + fsutil.InodeNotSocket `state:"nosave"` + fsutil.InodeVirtual `state:"nosave"` - mu sync.Mutex `state:"nosave"` + fsutil.InodeSimpleAttributes + fsutil.InodeSimpleExtendedAttributes // Target is the symlink target. Target string } -// InitSymlink initializes a symlink, pointing to the given target. -// A symlink is assumed to always have permissions 0777. -func (s *Symlink) InitSymlink(ctx context.Context, owner fs.FileOwner, target string) { - s.InitEntry(ctx, owner, fs.FilePermsFromMode(0777)) - s.Target = target +var _ fs.InodeOperations = (*Symlink)(nil) + +// NewSymlink returns a new Symlink. +func NewSymlink(ctx context.Context, owner fs.FileOwner, target string) *Symlink { + // A symlink is assumed to always have permissions 0777. + return &Symlink{ + InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, owner, fs.FilePermsFromMode(0777), linux.RAMFS_MAGIC), + Target: target, + } } // UnstableAttr returns all attributes of this ramfs symlink. func (s *Symlink) UnstableAttr(ctx context.Context, inode *fs.Inode) (fs.UnstableAttr, error) { - uattr, _ := s.Entry.UnstableAttr(ctx, inode) + uattr, err := s.InodeSimpleAttributes.UnstableAttr(ctx, inode) + if err != nil { + return fs.UnstableAttr{}, err + } uattr.Size = int64(len(s.Target)) uattr.Usage = uattr.Size return uattr, nil } -// Check implements InodeOperations.Check. -func (s *Symlink) Check(ctx context.Context, inode *fs.Inode, p fs.PermMask) bool { - return fs.ContextCanAccessFile(ctx, inode, p) -} - // SetPermissions on a symlink is always rejected. func (s *Symlink) SetPermissions(context.Context, *fs.Inode, fs.FilePermissions) bool { return false @@ -60,10 +71,7 @@ func (s *Symlink) SetPermissions(context.Context, *fs.Inode, fs.FilePermissions) // Readlink reads the symlink value. func (s *Symlink) Readlink(ctx context.Context, _ *fs.Inode) (string, error) { - s.mu.Lock() - defer s.mu.Unlock() - - s.Entry.NotifyAccess(ctx) + s.NotifyAccess(ctx) return s.Target, nil } @@ -72,3 +80,24 @@ func (s *Symlink) Readlink(ctx context.Context, _ *fs.Inode) (string, error) { func (*Symlink) Getlink(context.Context, *fs.Inode) (*fs.Dirent, error) { return nil, fs.ErrResolveViaReadlink } + +// GetFile implements fs.FileOperations.GetFile. +func (s *Symlink) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return fs.NewFile(ctx, dirent, flags, &symlinkFileOperations{}), nil +} + +// +stateify savable +type symlinkFileOperations struct { + waiter.AlwaysReady `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileNoRead `state:"nosave"` + fsutil.FileNoSeek `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoWrite `state:"nosave"` +} + +var _ fs.FileOperations = (*symlinkFileOperations)(nil) diff --git a/pkg/sentry/fs/ramfs/test/BUILD b/pkg/sentry/fs/ramfs/test/BUILD deleted file mode 100644 index 187eac49d..000000000 --- a/pkg/sentry/fs/ramfs/test/BUILD +++ /dev/null @@ -1,16 +0,0 @@ -package(licenses = ["notice"]) # Apache 2.0 - -load("//tools/go_stateify:defs.bzl", "go_library") - -go_library( - name = "test", - testonly = 1, - srcs = ["test.go"], - importpath = "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs/test", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/sentry/context", - "//pkg/sentry/fs", - "//pkg/sentry/fs/ramfs", - ], -) diff --git a/pkg/sentry/fs/ramfs/test/test.go b/pkg/sentry/fs/ramfs/test/test.go deleted file mode 100644 index 11bff7729..000000000 --- a/pkg/sentry/fs/ramfs/test/test.go +++ /dev/null @@ -1,46 +0,0 @@ -// Copyright 2018 Google LLC -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package test provides a simple ramfs-based filesystem for use in testing. -package test - -import ( - "gvisor.googlesource.com/gvisor/pkg/sentry/context" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" -) - -// Dir is a simple ramfs.Dir that supports save/restore as-is. -type Dir struct { - ramfs.Dir -} - -// NewDir returns a simple ramfs directory with the passed contents. -func NewDir(ctx context.Context, contents map[string]*fs.Inode, perms fs.FilePermissions) *Dir { - d := &Dir{} - d.InitDir(ctx, contents, fs.RootOwner, perms) - return d -} - -// File is a simple ramfs.File that supports save/restore as-is. -type File struct { - ramfs.File -} - -// NewFile returns a simple ramfs File. -func NewFile(ctx context.Context, perms fs.FilePermissions) *File { - f := &File{} - f.InitFile(ctx, fs.RootOwner, perms) - return f -} diff --git a/pkg/sentry/fs/ramfs/tree.go b/pkg/sentry/fs/ramfs/tree.go index 29a70f698..f6d5ffdec 100644 --- a/pkg/sentry/fs/ramfs/tree.go +++ b/pkg/sentry/fs/ramfs/tree.go @@ -60,8 +60,7 @@ func makeSubdir(ctx context.Context, msrc *fs.MountSource, root *Dir, subdir str // emptyDir returns an empty *ramfs.Dir that is traversable but not writable. func emptyDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - dir := &Dir{} - dir.InitDir(ctx, make(map[string]*fs.Inode), fs.RootOwner, fs.FilePermsFromMode(0555)) + dir := NewDir(ctx, make(map[string]*fs.Inode), fs.RootOwner, fs.FilePermsFromMode(0555)) return fs.NewInode(dir, msrc, fs.StableAttr{ DeviceID: anon.PseudoDevice.DeviceID(), InodeID: anon.PseudoDevice.NextIno(), diff --git a/pkg/sentry/fs/ramfs/tree_test.go b/pkg/sentry/fs/ramfs/tree_test.go index 54df2143c..8bee9cfc1 100644 --- a/pkg/sentry/fs/ramfs/tree_test.go +++ b/pkg/sentry/fs/ramfs/tree_test.go @@ -22,7 +22,7 @@ import ( ) func TestMakeDirectoryTree(t *testing.T) { - mount := fs.NewNonCachingMountSource(nil, fs.MountSourceFlags{}) + mount := fs.NewPseudoMountSource() for _, test := range []struct { name string diff --git a/pkg/sentry/fs/sys/BUILD b/pkg/sentry/fs/sys/BUILD index 5ba23d5da..7de928e16 100644 --- a/pkg/sentry/fs/sys/BUILD +++ b/pkg/sentry/fs/sys/BUILD @@ -13,12 +13,13 @@ go_library( importpath = "gvisor.googlesource.com/gvisor/pkg/sentry/fs/sys", visibility = ["//pkg/sentry:internal"], deps = [ + "//pkg/abi/linux", "//pkg/sentry/context", "//pkg/sentry/device", "//pkg/sentry/fs", + "//pkg/sentry/fs/fsutil", "//pkg/sentry/fs/ramfs", "//pkg/sentry/kernel", "//pkg/sentry/usermem", - "//pkg/syserror", ], ) diff --git a/pkg/sentry/fs/sys/devices.go b/pkg/sentry/fs/sys/devices.go index e64aa0edc..8b728a4e4 100644 --- a/pkg/sentry/fs/sys/devices.go +++ b/pkg/sentry/fs/sys/devices.go @@ -16,43 +16,50 @@ package sys import ( "fmt" - "io" + "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" - "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" - "gvisor.googlesource.com/gvisor/pkg/syserror" ) // +stateify savable type cpunum struct { - ramfs.Entry + fsutil.InodeGenericChecker `state:"nosave"` + fsutil.InodeNoExtendedAttributes `state:"nosave"` + fsutil.InodeNoopRelease `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` + fsutil.InodeNotDirectory `state:"nosave"` + fsutil.InodeNotMappable `state:"nosave"` + fsutil.InodeNotSocket `state:"nosave"` + fsutil.InodeNotSymlink `state:"nosave"` + fsutil.InodeNotVirtual `state:"nosave"` + fsutil.InodeNotTruncatable `state:"nosave"` + + fsutil.InodeSimpleAttributes + fsutil.InodeStaticFileGetter + + // k is the system kernel. + k *kernel.Kernel } -func (c *cpunum) DeprecatedPreadv(ctx context.Context, dst usermem.IOSequence, offset int64) (int64, error) { - if offset < 0 { - return 0, syserror.EINVAL - } +var _ fs.InodeOperations = (*cpunum)(nil) +func newPossible(ctx context.Context, msrc *fs.MountSource) *fs.Inode { + var maxCore uint k := kernel.KernelFromContext(ctx) - if k == nil { - return 0, io.EOF + if k != nil { + maxCore = k.ApplicationCores() - 1 } + contents := []byte(fmt.Sprintf("0-%d\n", maxCore)) - str := []byte(fmt.Sprintf("0-%d\n", k.ApplicationCores()-1)) - if offset >= int64(len(str)) { - return 0, io.EOF + c := &cpunum{ + InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, fs.RootOwner, fs.FilePermsFromMode(0444), linux.SYSFS_MAGIC), + InodeStaticFileGetter: fsutil.InodeStaticFileGetter{ + Contents: contents, + }, } - - n, err := dst.CopyOut(ctx, str[offset:]) - return int64(n), err -} - -func newPossible(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - c := &cpunum{} - c.InitEntry(ctx, fs.RootOwner, fs.FilePermsFromMode(0444)) return newFile(c, msrc) } diff --git a/pkg/sentry/fs/sys/fs.go b/pkg/sentry/fs/sys/fs.go index 5ce33f87f..301fef038 100644 --- a/pkg/sentry/fs/sys/fs.go +++ b/pkg/sentry/fs/sys/fs.go @@ -24,6 +24,8 @@ import ( // +stateify savable type filesystem struct{} +var _ fs.Filesystem = (*filesystem)(nil) + func init() { fs.RegisterFilesystem(&filesystem{}) } diff --git a/pkg/sentry/fs/sys/sys.go b/pkg/sentry/fs/sys/sys.go index 7cc1942c7..c5b56fe69 100644 --- a/pkg/sentry/fs/sys/sys.go +++ b/pkg/sentry/fs/sys/sys.go @@ -22,13 +22,6 @@ import ( "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" ) -// sys is a root sys node. -// -// +stateify savable -type sys struct { - ramfs.Dir -} - func newFile(node fs.InodeOperations, msrc *fs.MountSource) *fs.Inode { sattr := fs.StableAttr{ DeviceID: sysfsDevice.DeviceID(), @@ -40,8 +33,7 @@ func newFile(node fs.InodeOperations, msrc *fs.MountSource) *fs.Inode { } func newDir(ctx context.Context, msrc *fs.MountSource, contents map[string]*fs.Inode) *fs.Inode { - d := &sys{} - d.InitDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) + d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) return fs.NewInode(d, msrc, fs.StableAttr{ DeviceID: sysfsDevice.DeviceID(), InodeID: sysfsDevice.NextIno(), diff --git a/pkg/sentry/fs/timerfd/timerfd.go b/pkg/sentry/fs/timerfd/timerfd.go index 7423e816c..b26466b9d 100644 --- a/pkg/sentry/fs/timerfd/timerfd.go +++ b/pkg/sentry/fs/timerfd/timerfd.go @@ -33,12 +33,12 @@ import ( // // +stateify savable type TimerOperations struct { - fsutil.ZeroSeek `state:"nosave"` - fsutil.NotDirReaddir `state:"nosave"` - fsutil.NoFsync `state:"nosave"` - fsutil.NoopFlush `state:"nosave"` - fsutil.NoMMap `state:"nosave"` - fsutil.NoIoctl `state:"nosave"` + fsutil.FileZeroSeek `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoFsync `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` events waiter.Queue `state:"zerovalue"` timer *ktime.Timer diff --git a/pkg/sentry/fs/tmpfs/BUILD b/pkg/sentry/fs/tmpfs/BUILD index 9065cdd5d..14c7a9e62 100644 --- a/pkg/sentry/fs/tmpfs/BUILD +++ b/pkg/sentry/fs/tmpfs/BUILD @@ -23,11 +23,13 @@ go_library( "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/pipe", + "//pkg/sentry/kernel/time", "//pkg/sentry/memmap", "//pkg/sentry/safemem", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", "//pkg/sentry/usermem", + "//pkg/syserror", "//pkg/waiter", ], ) diff --git a/pkg/sentry/fs/tmpfs/file_regular.go b/pkg/sentry/fs/tmpfs/file_regular.go index 1f9d69909..2c1eb0fd2 100644 --- a/pkg/sentry/fs/tmpfs/file_regular.go +++ b/pkg/sentry/fs/tmpfs/file_regular.go @@ -28,13 +28,13 @@ import ( // // +stateify savable type regularFileOperations struct { - waiter.AlwaysReady `state:"nosave"` - fsutil.NoopRelease `state:"nosave"` - fsutil.GenericSeek `state:"nosave"` - fsutil.NotDirReaddir `state:"nosave"` - fsutil.NoopFsync `state:"nosave"` - fsutil.NoopFlush `state:"nosave"` - fsutil.NoIoctl `state:"nosave"` + waiter.AlwaysReady `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoopFsync `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` // iops is the InodeOperations of a regular tmpfs file. It is // guaranteed to be the same as file.Dirent.Inode.InodeOperations, diff --git a/pkg/sentry/fs/tmpfs/file_test.go b/pkg/sentry/fs/tmpfs/file_test.go index 02da9af82..e7bbdc404 100644 --- a/pkg/sentry/fs/tmpfs/file_test.go +++ b/pkg/sentry/fs/tmpfs/file_test.go @@ -52,19 +52,19 @@ func TestGrow(t *testing.T) { abuf := bytes.Repeat([]byte{'a'}, 68) n, err := f.Pwritev(ctx, usermem.BytesIOSequence(abuf), 0) if n != int64(len(abuf)) || err != nil { - t.Fatalf("DeprecatedPwritev got (%d, %v) want (%d, nil)", n, err, len(abuf)) + t.Fatalf("Pwritev got (%d, %v) want (%d, nil)", n, err, len(abuf)) } bbuf := bytes.Repeat([]byte{'b'}, 856) n, err = f.Pwritev(ctx, usermem.BytesIOSequence(bbuf), 68) if n != int64(len(bbuf)) || err != nil { - t.Fatalf("DeprecatedPwritev got (%d, %v) want (%d, nil)", n, err, len(bbuf)) + t.Fatalf("Pwritev got (%d, %v) want (%d, nil)", n, err, len(bbuf)) } rbuf := make([]byte, len(abuf)+len(bbuf)) n, err = f.Preadv(ctx, usermem.BytesIOSequence(rbuf), 0) if n != int64(len(rbuf)) || err != nil { - t.Fatalf("DeprecatedPreadv got (%d, %v) want (%d, nil)", n, err, len(rbuf)) + t.Fatalf("Preadv got (%d, %v) want (%d, nil)", n, err, len(rbuf)) } if want := append(abuf, bbuf...); !bytes.Equal(rbuf, want) { diff --git a/pkg/sentry/fs/tmpfs/fs.go b/pkg/sentry/fs/tmpfs/fs.go index 88f85b85a..caa3220ee 100644 --- a/pkg/sentry/fs/tmpfs/fs.go +++ b/pkg/sentry/fs/tmpfs/fs.go @@ -50,6 +50,8 @@ const ( // +stateify savable type Filesystem struct{} +var _ fs.Filesystem = (*Filesystem)(nil) + func init() { fs.RegisterFilesystem(&Filesystem{}) } diff --git a/pkg/sentry/fs/tmpfs/inode_file.go b/pkg/sentry/fs/tmpfs/inode_file.go index ca2b4aabb..42d4bc76f 100644 --- a/pkg/sentry/fs/tmpfs/inode_file.go +++ b/pkg/sentry/fs/tmpfs/inode_file.go @@ -22,6 +22,7 @@ import ( "gvisor.googlesource.com/gvisor/pkg/sentry/fs" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" + ktime "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/time" "gvisor.googlesource.com/gvisor/pkg/sentry/memmap" "gvisor.googlesource.com/gvisor/pkg/sentry/safemem" "gvisor.googlesource.com/gvisor/pkg/sentry/usage" @@ -46,11 +47,13 @@ import ( // // +stateify savable type fileInodeOperations struct { - fsutil.DeprecatedFileOperations `state:"nosave"` - fsutil.InodeNotDirectory `state:"nosave"` - fsutil.InodeNotSocket `state:"nosave"` - fsutil.InodeNotSymlink `state:"nosave"` - fsutil.NoopWriteOut `state:"nosave"` + fsutil.InodeGenericChecker `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` + fsutil.InodeNotDirectory `state:"nosave"` + fsutil.InodeNotSocket `state:"nosave"` + fsutil.InodeNotSymlink `state:"nosave"` + + fsutil.InodeSimpleExtendedAttributes // kernel is used to allocate platform memory that stores the file's contents. kernel *kernel.Kernel @@ -62,10 +65,10 @@ type fileInodeOperations struct { // attr contains the unstable metadata for the file. // - // attr is protected by attrMu. attr.Unstable.Size is protected by both - // attrMu and dataMu; reading it requires locking either mutex, while - // mutating it requires locking both. - attr fsutil.InMemoryAttributes + // attr is protected by attrMu. attr.Size is protected by both attrMu + // and dataMu; reading it requires locking either mutex, while mutating + // it requires locking both. + attr fs.UnstableAttr mapsMu sync.Mutex `state:"nosave"` @@ -83,12 +86,12 @@ type fileInodeOperations struct { data fsutil.FileRangeSet } +var _ fs.InodeOperations = (*fileInodeOperations)(nil) + // NewInMemoryFile returns a new file backed by p.Memory(). func NewInMemoryFile(ctx context.Context, usage usage.MemoryKind, uattr fs.UnstableAttr, k *kernel.Kernel) fs.InodeOperations { return &fileInodeOperations{ - attr: fsutil.InMemoryAttributes{ - Unstable: uattr, - }, + attr: uattr, kernel: k, memUsage: usage, } @@ -121,71 +124,56 @@ func (f *fileInodeOperations) GetFile(ctx context.Context, d *fs.Dirent, flags f // UnstableAttr returns unstable attributes of this tmpfs file. func (f *fileInodeOperations) UnstableAttr(ctx context.Context, inode *fs.Inode) (fs.UnstableAttr, error) { f.attrMu.Lock() - defer f.attrMu.Unlock() f.dataMu.RLock() - defer f.dataMu.RUnlock() - attr := f.attr.Unstable + attr := f.attr attr.Usage = int64(f.data.Span()) + f.dataMu.RUnlock() + f.attrMu.Unlock() return attr, nil } -// Getxattr implements fs.InodeOperations.Getxattr. -func (f *fileInodeOperations) Getxattr(inode *fs.Inode, name string) ([]byte, error) { - f.attrMu.Lock() - defer f.attrMu.Unlock() - return f.attr.Getxattr(name) -} - -// Setxattr implements fs.InodeOperations.Setxattr. -func (f *fileInodeOperations) Setxattr(inode *fs.Inode, name string, value []byte) error { - f.attrMu.Lock() - defer f.attrMu.Unlock() - return f.attr.Setxattr(name, value) -} - -// Listxattr implements fs.InodeOperations.Listxattr. -func (f *fileInodeOperations) Listxattr(inode *fs.Inode) (map[string]struct{}, error) { - f.attrMu.Lock() - defer f.attrMu.Unlock() - return f.attr.Listxattr() -} - // Check implements fs.InodeOperations.Check. func (f *fileInodeOperations) Check(ctx context.Context, inode *fs.Inode, p fs.PermMask) bool { return fs.ContextCanAccessFile(ctx, inode, p) } // SetPermissions implements fs.InodeOperations.SetPermissions. -func (f *fileInodeOperations) SetPermissions(ctx context.Context, inode *fs.Inode, p fs.FilePermissions) bool { +func (f *fileInodeOperations) SetPermissions(ctx context.Context, _ *fs.Inode, p fs.FilePermissions) bool { f.attrMu.Lock() - defer f.attrMu.Unlock() - return f.attr.SetPermissions(ctx, p) + f.attr.SetPermissions(ctx, p) + f.attrMu.Unlock() + return true } // SetTimestamps implements fs.InodeOperations.SetTimestamps. -func (f *fileInodeOperations) SetTimestamps(ctx context.Context, inode *fs.Inode, ts fs.TimeSpec) error { +func (f *fileInodeOperations) SetTimestamps(ctx context.Context, _ *fs.Inode, ts fs.TimeSpec) error { f.attrMu.Lock() - defer f.attrMu.Unlock() - return f.attr.SetTimestamps(ctx, ts) + f.attr.SetTimestamps(ctx, ts) + f.attrMu.Unlock() + return nil } // SetOwner implements fs.InodeOperations.SetOwner. -func (f *fileInodeOperations) SetOwner(ctx context.Context, inode *fs.Inode, owner fs.FileOwner) error { +func (f *fileInodeOperations) SetOwner(ctx context.Context, _ *fs.Inode, owner fs.FileOwner) error { f.attrMu.Lock() - defer f.attrMu.Unlock() - return f.attr.SetOwner(ctx, owner) + f.attr.SetOwner(ctx, owner) + f.attrMu.Unlock() + return nil } // Truncate implements fs.InodeOperations.Truncate. -func (f *fileInodeOperations) Truncate(ctx context.Context, inode *fs.Inode, size int64) error { +func (f *fileInodeOperations) Truncate(ctx context.Context, _ *fs.Inode, size int64) error { f.attrMu.Lock() defer f.attrMu.Unlock() f.dataMu.Lock() - oldSize := f.attr.Unstable.Size + oldSize := f.attr.Size if oldSize != size { - f.attr.Unstable.Size = size - f.attr.TouchModificationTime(ctx) + f.attr.Size = size + // Update mtime and ctime. + now := ktime.NowFromContext(ctx) + f.attr.ModificationTime = now + f.attr.StatusChangeTime = now } f.dataMu.Unlock() @@ -220,21 +208,21 @@ func (f *fileInodeOperations) Truncate(ctx context.Context, inode *fs.Inode, siz // AddLink implements fs.InodeOperations.AddLink. func (f *fileInodeOperations) AddLink() { f.attrMu.Lock() - f.attr.Unstable.Links++ + f.attr.Links++ f.attrMu.Unlock() } // DropLink implements fs.InodeOperations.DropLink. func (f *fileInodeOperations) DropLink() { f.attrMu.Lock() - f.attr.Unstable.Links-- + f.attr.Links-- f.attrMu.Unlock() } // NotifyStatusChange implements fs.InodeOperations.NotifyStatusChange. func (f *fileInodeOperations) NotifyStatusChange(ctx context.Context) { f.attrMu.Lock() - f.attr.TouchStatusChangeTime(ctx) + f.attr.StatusChangeTime = ktime.NowFromContext(ctx) f.attrMu.Unlock() } @@ -264,7 +252,7 @@ func (f *fileInodeOperations) read(ctx context.Context, dst usermem.IOSequence, // TODO: Separate out f.attr.Size and use atomics instead of // f.dataMu. f.dataMu.RLock() - size := f.attr.Unstable.Size + size := f.attr.Size f.dataMu.RUnlock() if offset >= size { return 0, io.EOF @@ -273,7 +261,7 @@ func (f *fileInodeOperations) read(ctx context.Context, dst usermem.IOSequence, n, err := dst.CopyOutFrom(ctx, &fileReadWriter{f, offset}) // Compare Linux's mm/filemap.c:do_generic_file_read() => file_accessed(). f.attrMu.Lock() - f.attr.TouchAccessTime(ctx) + f.attr.AccessTime = ktime.NowFromContext(ctx) f.attrMu.Unlock() return n, err } @@ -287,7 +275,9 @@ func (f *fileInodeOperations) write(ctx context.Context, src usermem.IOSequence, f.attrMu.Lock() defer f.attrMu.Unlock() // Compare Linux's mm/filemap.c:__generic_file_write_iter() => file_update_time(). - f.attr.TouchModificationTime(ctx) + now := ktime.NowFromContext(ctx) + f.attr.ModificationTime = now + f.attr.StatusChangeTime = now return src.CopyInTo(ctx, &fileReadWriter{f, offset}) } @@ -302,10 +292,10 @@ func (rw *fileReadWriter) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { defer rw.f.dataMu.RUnlock() // Compute the range to read. - if rw.offset >= rw.f.attr.Unstable.Size { + if rw.offset >= rw.f.attr.Size { return 0, io.EOF } - end := fs.ReadEndOffset(rw.offset, int64(dsts.NumBytes()), rw.f.attr.Unstable.Size) + end := fs.ReadEndOffset(rw.offset, int64(dsts.NumBytes()), rw.f.attr.Size) if end == rw.offset { // dsts.NumBytes() == 0? return 0, nil } @@ -371,8 +361,8 @@ func (rw *fileReadWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) defer func() { // If the write ends beyond the file's previous size, it causes the // file to grow. - if rw.offset > rw.f.attr.Unstable.Size { - rw.f.attr.Unstable.Size = rw.offset + if rw.offset > rw.f.attr.Size { + rw.f.attr.Size = rw.offset } }() @@ -450,9 +440,9 @@ func (f *fileInodeOperations) Translate(ctx context.Context, required, optional f.dataMu.Lock() defer f.dataMu.Unlock() - // Constrain translations to f.attr.Unstable.Size (rounded up) to prevent + // Constrain translations to f.attr.Size (rounded up) to prevent // translation to pages that may be concurrently truncated. - pgend := fs.OffsetPageEnd(f.attr.Unstable.Size) + pgend := fs.OffsetPageEnd(f.attr.Size) var beyondEOF bool if required.End > pgend { if required.Start >= pgend { diff --git a/pkg/sentry/fs/tmpfs/tmpfs.go b/pkg/sentry/fs/tmpfs/tmpfs.go index 40a8c4b1e..a0277a132 100644 --- a/pkg/sentry/fs/tmpfs/tmpfs.go +++ b/pkg/sentry/fs/tmpfs/tmpfs.go @@ -19,12 +19,14 @@ import ( "gvisor.googlesource.com/gvisor/pkg/abi/linux" "gvisor.googlesource.com/gvisor/pkg/sentry/context" "gvisor.googlesource.com/gvisor/pkg/sentry/fs" + "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" "gvisor.googlesource.com/gvisor/pkg/sentry/fs/ramfs" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel" "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/pipe" "gvisor.googlesource.com/gvisor/pkg/sentry/socket/unix/transport" "gvisor.googlesource.com/gvisor/pkg/sentry/usage" "gvisor.googlesource.com/gvisor/pkg/sentry/usermem" + "gvisor.googlesource.com/gvisor/pkg/syserror" ) var fsInfo = fs.Info{ @@ -39,32 +41,54 @@ var fsInfo = fs.Info{ func rename(ctx context.Context, oldParent *fs.Inode, oldName string, newParent *fs.Inode, newName string) error { op, ok := oldParent.InodeOperations.(*Dir) if !ok { - return ramfs.ErrCrossDevice + return syserror.EXDEV } np, ok := newParent.InodeOperations.(*Dir) if !ok { - return ramfs.ErrCrossDevice + return syserror.EXDEV } - return ramfs.Rename(ctx, &op.Dir, oldName, &np.Dir, newName) + return ramfs.Rename(ctx, op.ramfsDir, oldName, np.ramfsDir, newName) } // Dir is a directory. // // +stateify savable type Dir struct { - ramfs.Dir + fsutil.InodeGenericChecker `state:"nosave"` + fsutil.InodeIsDirTruncate `state:"nosave"` + fsutil.InodeNoopRelease `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` + fsutil.InodeNotMappable `state:"nosave"` + fsutil.InodeNotSocket `state:"nosave"` + fsutil.InodeNotSymlink `state:"nosave"` + fsutil.InodeVirtual `state:"nosave"` + + // Ideally this would be embedded, so that we "inherit" all of the + // InodeOperations implemented by ramfs.Dir for free. + // + // However, ramfs.dirFileOperations stores a pointer to a ramfs.Dir, + // and our save/restore package does not allow saving a pointer to an + // embedded field elsewhere. + // + // Thus, we must make the ramfs.Dir is a field, and we delegate all the + // InodeOperation methods to it. + ramfsDir *ramfs.Dir // kernel is used to allocate platform memory as storage for tmpfs Files. kernel *kernel.Kernel } +var _ fs.InodeOperations = (*Dir)(nil) + // NewDir returns a new directory. func NewDir(ctx context.Context, contents map[string]*fs.Inode, owner fs.FileOwner, perms fs.FilePermissions, msrc *fs.MountSource, kernel *kernel.Kernel) *fs.Inode { - d := &Dir{kernel: kernel} - d.InitDir(ctx, contents, owner, perms) + d := &Dir{ + ramfsDir: ramfs.NewDir(ctx, contents, owner, perms), + kernel: kernel, + } // Manually set the CreateOps. - d.CreateOps = d.newCreateOps() + d.ramfsDir.CreateOps = d.newCreateOps() return fs.NewInode(d, msrc, fs.StableAttr{ DeviceID: tmpfsDevice.DeviceID(), @@ -77,7 +101,107 @@ func NewDir(ctx context.Context, contents map[string]*fs.Inode, owner fs.FileOwn // afterLoad is invoked by stateify. func (d *Dir) afterLoad() { // Per NewDir, manually set the CreateOps. - d.Dir.CreateOps = d.newCreateOps() + d.ramfsDir.CreateOps = d.newCreateOps() +} + +// GetFile implements fs.InodeOperations.GetFile. +func (d *Dir) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { + return d.ramfsDir.GetFile(ctx, dirent, flags) +} + +// AddLink implements fs.InodeOperations.AddLink. +func (d *Dir) AddLink() { + d.ramfsDir.AddLink() +} + +// DropLink implements fs.InodeOperations.DropLink. +func (d *Dir) DropLink() { + d.ramfsDir.DropLink() +} + +// Bind implements fs.InodeOperations.Bind. +func (d *Dir) Bind(ctx context.Context, dir *fs.Inode, name string, ep transport.BoundEndpoint, perms fs.FilePermissions) (*fs.Dirent, error) { + return d.ramfsDir.Bind(ctx, dir, name, ep, perms) +} + +// Create implements fs.InodeOperations.Create. +func (d *Dir) Create(ctx context.Context, dir *fs.Inode, name string, flags fs.FileFlags, perms fs.FilePermissions) (*fs.File, error) { + return d.ramfsDir.Create(ctx, dir, name, flags, perms) +} + +// CreateLink implements fs.InodeOperations.CreateLink. +func (d *Dir) CreateLink(ctx context.Context, dir *fs.Inode, oldname, newname string) error { + return d.ramfsDir.CreateLink(ctx, dir, oldname, newname) +} + +// CreateHardLink implements fs.InodeOperations.CreateHardLink. +func (d *Dir) CreateHardLink(ctx context.Context, dir *fs.Inode, target *fs.Inode, name string) error { + return d.ramfsDir.CreateHardLink(ctx, dir, target, name) +} + +// CreateDirectory implements fs.InodeOperations.CreateDirectory. +func (d *Dir) CreateDirectory(ctx context.Context, dir *fs.Inode, name string, perms fs.FilePermissions) error { + return d.ramfsDir.CreateDirectory(ctx, dir, name, perms) +} + +// CreateFifo implements fs.InodeOperations.CreateFifo. +func (d *Dir) CreateFifo(ctx context.Context, dir *fs.Inode, name string, perms fs.FilePermissions) error { + return d.ramfsDir.CreateFifo(ctx, dir, name, perms) +} + +// Getxattr implements fs.InodeOperations.Getxattr. +func (d *Dir) Getxattr(i *fs.Inode, name string) ([]byte, error) { + return d.ramfsDir.Getxattr(i, name) +} + +// Setxattr implements fs.InodeOperations.Setxattr. +func (d *Dir) Setxattr(i *fs.Inode, name string, value []byte) error { + return d.ramfsDir.Setxattr(i, name, value) +} + +// Listxattr implements fs.InodeOperations.Listxattr. +func (d *Dir) Listxattr(i *fs.Inode) (map[string]struct{}, error) { + return d.ramfsDir.Listxattr(i) +} + +// Lookup implements fs.InodeOperations.Lookup. +func (d *Dir) Lookup(ctx context.Context, i *fs.Inode, p string) (*fs.Dirent, error) { + return d.ramfsDir.Lookup(ctx, i, p) +} + +// NotifyStatusChange implements fs.InodeOperations.NotifyStatusChange. +func (d *Dir) NotifyStatusChange(ctx context.Context) { + d.ramfsDir.NotifyStatusChange(ctx) +} + +// Remove implements fs.InodeOperations.Remove. +func (d *Dir) Remove(ctx context.Context, i *fs.Inode, name string) error { + return d.ramfsDir.Remove(ctx, i, name) +} + +// RemoveDirectory implements fs.InodeOperations.RemoveDirectory. +func (d *Dir) RemoveDirectory(ctx context.Context, i *fs.Inode, name string) error { + return d.ramfsDir.RemoveDirectory(ctx, i, name) +} + +// UnstableAttr implements fs.InodeOperations.UnstableAttr. +func (d *Dir) UnstableAttr(ctx context.Context, i *fs.Inode) (fs.UnstableAttr, error) { + return d.ramfsDir.UnstableAttr(ctx, i) +} + +// SetPermissions implements fs.InodeOperations.SetPermissions. +func (d *Dir) SetPermissions(ctx context.Context, i *fs.Inode, p fs.FilePermissions) bool { + return d.ramfsDir.SetPermissions(ctx, i, p) +} + +// SetOwner implements fs.InodeOperations.SetOwner. +func (d *Dir) SetOwner(ctx context.Context, i *fs.Inode, owner fs.FileOwner) error { + return d.ramfsDir.SetOwner(ctx, i, owner) +} + +// SetTimestamps implements fs.InodeOperations.SetTimestamps. +func (d *Dir) SetTimestamps(ctx context.Context, i *fs.Inode, ts fs.TimeSpec) error { + return d.ramfsDir.SetTimestamps(ctx, i, ts) } // newCreateOps builds the custom CreateOps for this Dir. @@ -132,8 +256,7 @@ type Symlink struct { // NewSymlink returns a new symlink with the provided permissions. func NewSymlink(ctx context.Context, target string, owner fs.FileOwner, msrc *fs.MountSource) *fs.Inode { - s := &Symlink{} - s.InitSymlink(ctx, owner, target) + s := &Symlink{Symlink: *ramfs.NewSymlink(ctx, owner, target)} return fs.NewInode(s, msrc, fs.StableAttr{ DeviceID: tmpfsDevice.DeviceID(), InodeID: tmpfsDevice.NextIno(), @@ -157,12 +280,12 @@ func (s *Symlink) StatFS(context.Context) (fs.Info, error) { // +stateify savable type Socket struct { ramfs.Socket + fsutil.InodeNotTruncatable `state:"nosave"` } // NewSocket returns a new socket with the provided permissions. func NewSocket(ctx context.Context, socket transport.BoundEndpoint, owner fs.FileOwner, perms fs.FilePermissions, msrc *fs.MountSource) *fs.Inode { - s := &Socket{} - s.InitSocket(ctx, socket, owner, perms) + s := &Socket{Socket: *ramfs.NewSocket(ctx, socket, owner, perms)} return fs.NewInode(s, msrc, fs.StableAttr{ DeviceID: tmpfsDevice.DeviceID(), InodeID: tmpfsDevice.NextIno(), @@ -185,15 +308,22 @@ func (s *Socket) StatFS(context.Context) (fs.Info, error) { // // +stateify savable type Fifo struct { - ramfs.Entry + fs.InodeOperations } // NewFifo creates a new named pipe. func NewFifo(ctx context.Context, owner fs.FileOwner, perms fs.FilePermissions, msrc *fs.MountSource) *fs.Inode { - f := &Fifo{} - f.InitEntry(ctx, owner, perms) - iops := pipe.NewInodeOperations(f, pipe.NewPipe(ctx, true /* isNamed */, pipe.DefaultPipeSize, usermem.PageSize)) - return fs.NewInode(iops, msrc, fs.StableAttr{ + // First create a pipe. + p := pipe.NewPipe(ctx, true /* isNamed */, pipe.DefaultPipeSize, usermem.PageSize) + + // Build pipe InodeOperations. + iops := pipe.NewInodeOperations(ctx, perms, p) + + // Wrap the iops with our Fifo. + fifoIops := &Fifo{iops} + + // Build a new Inode. + return fs.NewInode(fifoIops, msrc, fs.StableAttr{ DeviceID: tmpfsDevice.DeviceID(), InodeID: tmpfsDevice.NextIno(), BlockSize: usermem.PageSize, diff --git a/pkg/sentry/fs/tty/BUILD b/pkg/sentry/fs/tty/BUILD index 2b45069a6..011cb6955 100644 --- a/pkg/sentry/fs/tty/BUILD +++ b/pkg/sentry/fs/tty/BUILD @@ -7,7 +7,6 @@ go_library( srcs = [ "dir.go", "fs.go", - "inode.go", "line_discipline.go", "master.go", "queue.go", @@ -25,7 +24,6 @@ go_library( "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", "//pkg/sentry/kernel/auth", - "//pkg/sentry/kernel/time", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/unimpl", "//pkg/sentry/usermem", diff --git a/pkg/sentry/fs/tty/dir.go b/pkg/sentry/fs/tty/dir.go index e32b05c1d..485cdb456 100644 --- a/pkg/sentry/fs/tty/dir.go +++ b/pkg/sentry/fs/tty/dir.go @@ -52,13 +52,17 @@ import ( // // +stateify savable type dirInodeOperations struct { - fsutil.DeprecatedFileOperations `state:"nosave"` - fsutil.InodeNotSocket `state:"nosave"` + fsutil.InodeGenericChecker `state:"nosave"` + fsutil.InodeNoExtendedAttributes `state:"nosave"` + fsutil.InodeNoopWriteOut `state:"nosave"` + fsutil.InodeNotMappable `state:"nosave"` fsutil.InodeNotRenameable `state:"nosave"` fsutil.InodeNotSymlink `state:"nosave"` - fsutil.InodeNoExtendedAttributes `state:"nosave"` - fsutil.NoMappable `state:"nosave"` - fsutil.NoopWriteOut `state:"nosave"` + fsutil.InodeNotSocket `state:"nosave"` + fsutil.InodeNotTruncatable `state:"nosave"` + fsutil.InodeVirtual `state:"nosave"` + + fsutil.InodeSimpleAttributes // msrc is the super block this directory is on. // @@ -68,9 +72,6 @@ type dirInodeOperations struct { // mu protects the fields below. mu sync.Mutex `state:"nosave"` - // attr contains the UnstableAttrs. - attr fsutil.InMemoryAttributes - // master is the master PTY inode. master *fs.Inode @@ -97,15 +98,10 @@ var _ fs.InodeOperations = (*dirInodeOperations)(nil) // newDir creates a new dir with a ptmx file and no terminals. func newDir(ctx context.Context, m *fs.MountSource) *fs.Inode { d := &dirInodeOperations{ - attr: fsutil.InMemoryAttributes{ - Unstable: fs.WithCurrentTime(ctx, fs.UnstableAttr{ - Owner: fs.RootOwner, - Perms: fs.FilePermsFromMode(0555), - }), - }, - msrc: m, - slaves: make(map[uint32]*fs.Inode), - dentryMap: fs.NewSortedDentryMap(nil), + InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, fs.RootOwner, fs.FilePermsFromMode(0555), linux.DEVPTS_SUPER_MAGIC), + msrc: m, + slaves: make(map[uint32]*fs.Inode), + dentryMap: fs.NewSortedDentryMap(nil), } // Linux devpts uses a default mode of 0000 for ptmx which can be // changed with the ptmxmode mount option. However, that default is not @@ -224,70 +220,6 @@ func (d *dirInodeOperations) GetFile(ctx context.Context, dirent *fs.Dirent, fla return fs.NewFile(ctx, dirent, flags, &dirFileOperations{di: d}), nil } -// UnstableAttr implements fs.InodeOperations.UnstableAttr. -func (d *dirInodeOperations) UnstableAttr(ctx context.Context, inode *fs.Inode) (fs.UnstableAttr, error) { - d.mu.Lock() - defer d.mu.Unlock() - return d.attr.Unstable, nil -} - -// Check implements fs.InodeOperations.Check. -func (d *dirInodeOperations) Check(ctx context.Context, inode *fs.Inode, p fs.PermMask) bool { - return fs.ContextCanAccessFile(ctx, inode, p) -} - -// SetPermissions implements fs.InodeOperations.SetPermissions. -func (d *dirInodeOperations) SetPermissions(ctx context.Context, inode *fs.Inode, p fs.FilePermissions) bool { - d.mu.Lock() - defer d.mu.Unlock() - return d.attr.SetPermissions(ctx, p) -} - -// SetOwner implements fs.InodeOperations.SetOwner. -func (d *dirInodeOperations) SetOwner(ctx context.Context, inode *fs.Inode, owner fs.FileOwner) error { - d.mu.Lock() - defer d.mu.Unlock() - return d.attr.SetOwner(ctx, owner) -} - -// SetTimestamps implements fs.InodeOperations.SetTimestamps. -func (d *dirInodeOperations) SetTimestamps(ctx context.Context, inode *fs.Inode, ts fs.TimeSpec) error { - d.mu.Lock() - defer d.mu.Unlock() - return d.attr.SetTimestamps(ctx, ts) -} - -// Truncate implements fs.InodeOperations.Truncate. -func (d *dirInodeOperations) Truncate(ctx context.Context, inode *fs.Inode, size int64) error { - return syserror.EINVAL -} - -// AddLink implements fs.InodeOperations.AddLink. -func (d *dirInodeOperations) AddLink() {} - -// DropLink implements fs.InodeOperations.DropLink. -func (d *dirInodeOperations) DropLink() {} - -// NotifyStatusChange implements fs.InodeOperations.NotifyStatusChange. -func (d *dirInodeOperations) NotifyStatusChange(ctx context.Context) { - d.mu.Lock() - defer d.mu.Unlock() - - d.attr.TouchStatusChangeTime(ctx) -} - -// IsVirtual implements fs.InodeOperations.IsVirtual. -func (d *dirInodeOperations) IsVirtual() bool { - return true -} - -// StatFS implements fs.InodeOperations.StatFS. -func (d *dirInodeOperations) StatFS(ctx context.Context) (fs.Info, error) { - return fs.Info{ - Type: linux.DEVPTS_SUPER_MAGIC, - }, nil -} - // allocateTerminal creates a new Terminal and installs a pts node for it. // // The caller must call DecRef when done with the returned Terminal. @@ -353,13 +285,13 @@ func (d *dirInodeOperations) masterClose(t *Terminal) { // // +stateify savable type dirFileOperations struct { - waiter.AlwaysReady `state:"nosave"` - fsutil.NoopRelease `state:"nosave"` - fsutil.GenericSeek `state:"nosave"` - fsutil.NoFsync `state:"nosave"` - fsutil.NoopFlush `state:"nosave"` - fsutil.NoMMap `state:"nosave"` - fsutil.NoIoctl `state:"nosave"` + waiter.AlwaysReady `state:"nosave"` + fsutil.FileNoopRelease `state:"nosave"` + fsutil.FileGenericSeek `state:"nosave"` + fsutil.FileNoFsync `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` + fsutil.FileNoIoctl `state:"nosave"` // di is the inode operations. di *dirInodeOperations diff --git a/pkg/sentry/fs/tty/inode.go b/pkg/sentry/fs/tty/inode.go deleted file mode 100644 index d5d1caafc..000000000 --- a/pkg/sentry/fs/tty/inode.go +++ /dev/null @@ -1,145 +0,0 @@ -// Copyright 2018 Google LLC -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package tty - -import ( - "sync" - - "gvisor.googlesource.com/gvisor/pkg/abi/linux" - "gvisor.googlesource.com/gvisor/pkg/sentry/context" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs" - "gvisor.googlesource.com/gvisor/pkg/sentry/fs/fsutil" - ktime "gvisor.googlesource.com/gvisor/pkg/sentry/kernel/time" - "gvisor.googlesource.com/gvisor/pkg/syserror" -) - -// inodeOperations are the base fs.InodeOperations for master and slave Inodes. -// -// inodeOperations does not implement: -// -// * fs.InodeOperations.Release -// * fs.InodeOperations.GetFile -// -// +stateify savable -type inodeOperations struct { - fsutil.DeprecatedFileOperations `state:"nosave"` - fsutil.InodeNoExtendedAttributes `state:"nosave"` - fsutil.InodeNotDirectory `state:"nosave"` - fsutil.InodeNotRenameable `state:"nosave"` - fsutil.InodeNotSocket `state:"nosave"` - fsutil.InodeNotSymlink `state:"nosave"` - fsutil.NoMappable `state:"nosave"` - fsutil.NoopWriteOut `state:"nosave"` - - // mu protects the fields below. - mu sync.Mutex `state:"nosave"` - - // uattr is the inode's UnstableAttr. - uattr fs.UnstableAttr -} - -// UnstableAttr implements fs.InodeOperations.UnstableAttr. -func (i *inodeOperations) UnstableAttr(ctx context.Context, inode *fs.Inode) (fs.UnstableAttr, error) { - i.mu.Lock() - defer i.mu.Unlock() - return i.uattr, nil -} - -// Check implements fs.InodeOperations.Check. -func (i *inodeOperations) Check(ctx context.Context, inode *fs.Inode, p fs.PermMask) bool { - return fs.ContextCanAccessFile(ctx, inode, p) -} - -// SetPermissions implements fs.InodeOperations.SetPermissions -func (i *inodeOperations) SetPermissions(ctx context.Context, inode *fs.Inode, p fs.FilePermissions) bool { - i.mu.Lock() - defer i.mu.Unlock() - i.uattr.Perms = p - i.uattr.StatusChangeTime = ktime.NowFromContext(ctx) - return true -} - -// SetOwner implements fs.InodeOperations.SetOwner. -func (i *inodeOperations) SetOwner(ctx context.Context, inode *fs.Inode, owner fs.FileOwner) error { - i.mu.Lock() - defer i.mu.Unlock() - if owner.UID.Ok() { - i.uattr.Owner.UID = owner.UID - } - if owner.GID.Ok() { - i.uattr.Owner.GID = owner.GID - } - return nil -} - -// SetTimestamps implements fs.InodeOperations.SetTimestamps. -func (i *inodeOperations) SetTimestamps(ctx context.Context, inode *fs.Inode, ts fs.TimeSpec) error { - if ts.ATimeOmit && ts.MTimeOmit { - return nil - } - - i.mu.Lock() - defer i.mu.Unlock() - - now := ktime.NowFromContext(ctx) - if !ts.ATimeOmit { - if ts.ATime.IsZero() { - i.uattr.AccessTime = now - } else { - i.uattr.AccessTime = ts.ATime - } - } - if !ts.MTimeOmit { - if ts.MTime.IsZero() { - i.uattr.ModificationTime = now - } else { - i.uattr.ModificationTime = ts.MTime - } - } - i.uattr.StatusChangeTime = now - return nil -} - -// Truncate implements fs.InodeOperations.Truncate. -func (i *inodeOperations) Truncate(ctx context.Context, inode *fs.Inode, size int64) error { - return syserror.EINVAL -} - -// AddLink implements fs.InodeOperations.AddLink. -func (i *inodeOperations) AddLink() { -} - -// DropLink implements fs.InodeOperations.DropLink. -func (i *inodeOperations) DropLink() { -} - -// NotifyStatusChange implements fs.InodeOperations.NotifyStatusChange. -func (i *inodeOperations) NotifyStatusChange(ctx context.Context) { - i.mu.Lock() - defer i.mu.Unlock() - i.uattr.StatusChangeTime = ktime.NowFromContext(ctx) -} - -// IsVirtual implements fs.InodeOperations.IsVirtual. -func (i *inodeOperations) IsVirtual() bool { - return true -} - -// StatFS implements fs.InodeOperations.StatFS. -func (i *inodeOperations) StatFS(ctx context.Context) (fs.Info, error) { - return fs.Info{ - Type: linux.DEVPTS_SUPER_MAGIC, - }, nil -} diff --git a/pkg/sentry/fs/tty/master.go b/pkg/sentry/fs/tty/master.go index 00bec4c2c..b5e13ab36 100644 --- a/pkg/sentry/fs/tty/master.go +++ b/pkg/sentry/fs/tty/master.go @@ -31,7 +31,7 @@ import ( // // +stateify savable type masterInodeOperations struct { - inodeOperations + fsutil.SimpleFileInode // d is the containing dir. d *dirInodeOperations @@ -42,15 +42,8 @@ var _ fs.InodeOperations = (*masterInodeOperations)(nil) // newMasterInode creates an Inode for the master end of a terminal. func newMasterInode(ctx context.Context, d *dirInodeOperations, owner fs.FileOwner, p fs.FilePermissions) *fs.Inode { iops := &masterInodeOperations{ - inodeOperations: inodeOperations{ - uattr: fs.WithCurrentTime(ctx, fs.UnstableAttr{ - Owner: owner, - Perms: p, - Links: 1, - // Size and Blocks are always 0. - }), - }, - d: d, + SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, owner, p, linux.DEVPTS_SUPER_MAGIC), + d: d, } return fs.NewInode(iops, d.msrc, fs.StableAttr{ @@ -102,11 +95,11 @@ func (mi *masterInodeOperations) GetFile(ctx context.Context, d *fs.Dirent, flag // // +stateify savable type masterFileOperations struct { - fsutil.PipeSeek `state:"nosave"` - fsutil.NotDirReaddir `state:"nosave"` - fsutil.NoFsync `state:"nosave"` - fsutil.NoopFlush `state:"nosave"` - fsutil.NoMMap `state:"nosave"` + fsutil.FilePipeSeek `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoFsync `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` // d is the containing dir. d *dirInodeOperations diff --git a/pkg/sentry/fs/tty/slave.go b/pkg/sentry/fs/tty/slave.go index a696fbb51..6dbce90b4 100644 --- a/pkg/sentry/fs/tty/slave.go +++ b/pkg/sentry/fs/tty/slave.go @@ -30,7 +30,7 @@ import ( // // +stateify savable type slaveInodeOperations struct { - inodeOperations + fsutil.SimpleFileInode // d is the containing dir. d *dirInodeOperations @@ -46,16 +46,9 @@ var _ fs.InodeOperations = (*slaveInodeOperations)(nil) // newSlaveInode takes ownership of t. func newSlaveInode(ctx context.Context, d *dirInodeOperations, t *Terminal, owner fs.FileOwner, p fs.FilePermissions) *fs.Inode { iops := &slaveInodeOperations{ - inodeOperations: inodeOperations{ - uattr: fs.WithCurrentTime(ctx, fs.UnstableAttr{ - Owner: owner, - Perms: p, - Links: 1, - // Size and Blocks are always 0. - }), - }, - d: d, - t: t, + SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, owner, p, linux.DEVPTS_SUPER_MAGIC), + d: d, + t: t, } return fs.NewInode(iops, d.msrc, fs.StableAttr{ @@ -91,11 +84,11 @@ func (si *slaveInodeOperations) GetFile(ctx context.Context, d *fs.Dirent, flags // // +stateify savable type slaveFileOperations struct { - fsutil.PipeSeek `state:"nosave"` - fsutil.NotDirReaddir `state:"nosave"` - fsutil.NoFsync `state:"nosave"` - fsutil.NoopFlush `state:"nosave"` - fsutil.NoMMap `state:"nosave"` + fsutil.FilePipeSeek `state:"nosave"` + fsutil.FileNotDirReaddir `state:"nosave"` + fsutil.FileNoFsync `state:"nosave"` + fsutil.FileNoopFlush `state:"nosave"` + fsutil.FileNoMMap `state:"nosave"` // si is the inode operations. si *slaveInodeOperations |