diff options
Diffstat (limited to 'pkg/sentry')
102 files changed, 1691 insertions, 3716 deletions
diff --git a/pkg/sentry/control/proc.go b/pkg/sentry/control/proc.go index 66a506584..6ae60c5cb 100644 --- a/pkg/sentry/control/proc.go +++ b/pkg/sentry/control/proc.go @@ -28,7 +28,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/host" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/usage" @@ -123,9 +122,8 @@ func ExecAsync(proc *Proc, args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID // TTYFileOperations that wraps the TTY is also returned. func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID, *host.TTYFileOperations, error) { // Import file descriptors. - l := limits.NewLimitSet() - fdm := proc.Kernel.NewFDMap() - defer fdm.DecRef() + fdTable := proc.Kernel.NewFDTable() + defer fdTable.DecRef() // No matter what happens, we should close all files in the FilePayload // before returning. Any files that are imported will be duped. @@ -149,9 +147,9 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI WorkingDirectory: args.WorkingDirectory, Root: args.Root, Credentials: creds, - FDMap: fdm, + FDTable: fdTable, Umask: 0022, - Limits: l, + Limits: limits.NewLimitSet(), MaxSymlinkTraversals: linux.MaxSymlinkTraversals, UTSNamespace: proc.Kernel.RootUTSNamespace(), IPCNamespace: proc.Kernel.RootIPCNamespace(), @@ -212,7 +210,7 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI } // Add the file to the FD map. - if err := fdm.NewFDAt(kdefs.FD(appFD), appFile, kernel.FDFlags{}, l); err != nil { + if err := fdTable.NewFDAt(ctx, int32(appFD), appFile, kernel.FDFlags{}); err != nil { return nil, 0, nil, err } } diff --git a/pkg/sentry/fs/ashmem/area.go b/pkg/sentry/fs/ashmem/area.go deleted file mode 100644 index 3b8d6ca89..000000000 --- a/pkg/sentry/fs/ashmem/area.go +++ /dev/null @@ -1,308 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ashmem - -import ( - "sync" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/waiter" -) - -const ( - // namePrefix is the name prefix assumed and forced by the Linux implementation. - namePrefix = "dev/ashmem" - - // nameLen is the maximum name length. - nameLen = 256 -) - -// Area implements fs.FileOperations. -// -// +stateify savable -type Area struct { - fsutil.FileNoFsync `state:"nosave"` - fsutil.FileNoSplice `state:"nosave"` - fsutil.FileNoopFlush `state:"nosave"` - fsutil.FileNotDirReaddir `state:"nosave"` - fsutil.FileUseInodeUnstableAttr `state:"nosave"` - waiter.AlwaysReady `state:"nosave"` - - ad *Device - - // mu protects fields below. - mu sync.Mutex `state:"nosave"` - tmpfsFile *fs.File - name string - size uint64 - perms usermem.AccessType - pb *PinBoard -} - -// Release implements fs.FileOperations.Release. -func (a *Area) Release() { - a.mu.Lock() - defer a.mu.Unlock() - if a.tmpfsFile != nil { - a.tmpfsFile.DecRef() - a.tmpfsFile = nil - } -} - -// Seek implements fs.FileOperations.Seek. -func (a *Area) Seek(ctx context.Context, file *fs.File, whence fs.SeekWhence, offset int64) (int64, error) { - a.mu.Lock() - defer a.mu.Unlock() - if a.size == 0 { - return 0, syserror.EINVAL - } - if a.tmpfsFile == nil { - return 0, syserror.EBADF - } - return a.tmpfsFile.FileOperations.Seek(ctx, file, whence, offset) -} - -// Read implements fs.FileOperations.Read. -func (a *Area) Read(ctx context.Context, file *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { - a.mu.Lock() - defer a.mu.Unlock() - if a.size == 0 { - return 0, nil - } - if a.tmpfsFile == nil { - return 0, syserror.EBADF - } - return a.tmpfsFile.FileOperations.Read(ctx, file, dst, offset) -} - -// Write implements fs.FileOperations.Write. -func (a *Area) Write(ctx context.Context, file *fs.File, src usermem.IOSequence, offset int64) (int64, error) { - return 0, syserror.ENOSYS -} - -// ConfigureMMap implements fs.FileOperations.ConfigureMMap. -func (a *Area) ConfigureMMap(ctx context.Context, file *fs.File, opts *memmap.MMapOpts) error { - a.mu.Lock() - defer a.mu.Unlock() - if a.size == 0 { - return syserror.EINVAL - } - - if !a.perms.SupersetOf(opts.Perms) { - return syserror.EPERM - } - opts.MaxPerms = opts.MaxPerms.Intersect(a.perms) - - if a.tmpfsFile == nil { - tmpfsInodeOps := tmpfs.NewInMemoryFile(ctx, usage.Tmpfs, fs.UnstableAttr{}) - tmpfsInode := fs.NewInode(ctx, tmpfsInodeOps, fs.NewPseudoMountSource(ctx), fs.StableAttr{}) - dirent := fs.NewDirent(ctx, tmpfsInode, namePrefix+"/"+a.name) - tmpfsFile, err := tmpfsInode.GetFile(ctx, dirent, fs.FileFlags{Read: true, Write: true}) - // Drop the extra reference on the Dirent. - dirent.DecRef() - - if err != nil { - return err - } - - // Truncate to the size set by ASHMEM_SET_SIZE ioctl. - err = tmpfsInodeOps.Truncate(ctx, tmpfsInode, int64(a.size)) - if err != nil { - return err - } - a.tmpfsFile = tmpfsFile - a.pb = NewPinBoard() - } - - return a.tmpfsFile.ConfigureMMap(ctx, opts) -} - -// Ioctl implements fs.FileOperations.Ioctl. -func (a *Area) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { - // Switch on ioctl request. - switch args[1].Uint() { - case linux.AshmemSetNameIoctl: - name, err := usermem.CopyStringIn(ctx, io, args[2].Pointer(), nameLen-1, usermem.IOOpts{ - AddressSpaceActive: true, - }) - if err != nil { - return 0, err - } - - a.mu.Lock() - defer a.mu.Unlock() - - // Cannot set name for already mapped ashmem. - if a.tmpfsFile != nil { - return 0, syserror.EINVAL - } - a.name = name - return 0, nil - - case linux.AshmemGetNameIoctl: - a.mu.Lock() - var local []byte - if a.name != "" { - nameLen := len([]byte(a.name)) - local = make([]byte, nameLen, nameLen+1) - copy(local, []byte(a.name)) - local = append(local, 0) - } else { - nameLen := len([]byte(namePrefix)) - local = make([]byte, nameLen, nameLen+1) - copy(local, []byte(namePrefix)) - local = append(local, 0) - } - a.mu.Unlock() - - if _, err := io.CopyOut(ctx, args[2].Pointer(), local, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { - return 0, syserror.EFAULT - } - return 0, nil - - case linux.AshmemSetSizeIoctl: - a.mu.Lock() - defer a.mu.Unlock() - - // Cannot set size for already mapped ashmem. - if a.tmpfsFile != nil { - return 0, syserror.EINVAL - } - a.size = uint64(args[2].SizeT()) - return 0, nil - - case linux.AshmemGetSizeIoctl: - return uintptr(a.size), nil - - case linux.AshmemPinIoctl, linux.AshmemUnpinIoctl, linux.AshmemGetPinStatusIoctl: - // Locking and unlocking is ok since once tmpfsFile is set, it won't be nil again - // even after unmapping! Unlocking is needed in order to avoid a deadlock on - // usermem.CopyObjectIn. - - // Cannot execute pin-related ioctls before mapping. - a.mu.Lock() - if a.tmpfsFile == nil { - a.mu.Unlock() - return 0, syserror.EINVAL - } - a.mu.Unlock() - - var pin linux.AshmemPin - _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &pin, usermem.IOOpts{ - AddressSpaceActive: true, - }) - if err != nil { - return 0, syserror.EFAULT - } - - a.mu.Lock() - defer a.mu.Unlock() - return a.pinOperation(pin, args[1].Uint()) - - case linux.AshmemPurgeAllCachesIoctl: - return 0, nil - - case linux.AshmemSetProtMaskIoctl: - prot := uint64(args[2].ModeT()) - perms := usermem.AccessType{ - Read: prot&linux.PROT_READ != 0, - Write: prot&linux.PROT_WRITE != 0, - Execute: prot&linux.PROT_EXEC != 0, - } - - a.mu.Lock() - defer a.mu.Unlock() - - // Can only narrow prot mask. - if !a.perms.SupersetOf(perms) { - return 0, syserror.EINVAL - } - - // TODO(b/30946773,gvisor.dev/issue/153): If personality flag - // READ_IMPLIES_EXEC is set, set PROT_EXEC if PORT_READ is set. - - a.perms = perms - return 0, nil - - case linux.AshmemGetProtMaskIoctl: - return uintptr(a.perms.Prot()), nil - default: - // Ioctls irrelevant to Ashmem. - return 0, syserror.EINVAL - } -} - -// pinOperation should only be called while holding a.mu. -func (a *Area) pinOperation(pin linux.AshmemPin, op uint32) (uintptr, error) { - // Page-align a.size for checks. - pageAlignedSize, ok := usermem.Addr(a.size).RoundUp() - if !ok { - return 0, syserror.EINVAL - } - // Len 0 means everything onward. - if pin.Len == 0 { - pin.Len = uint32(pageAlignedSize) - pin.Offset - } - // Both Offset and Len have to be page-aligned. - if pin.Offset%uint32(usermem.PageSize) != 0 { - return 0, syserror.EINVAL - } - if pin.Len%uint32(usermem.PageSize) != 0 { - return 0, syserror.EINVAL - } - // Adding Offset and Len must not cause an uint32 overflow. - if end := pin.Offset + pin.Len; end < pin.Offset { - return 0, syserror.EINVAL - } - // Pin range must not exceed a's size. - if uint32(pageAlignedSize) < pin.Offset+pin.Len { - return 0, syserror.EINVAL - } - // Handle each operation. - r := RangeFromAshmemPin(pin) - switch op { - case linux.AshmemPinIoctl: - if a.pb.PinRange(r) { - return linux.AshmemWasPurged, nil - } - return linux.AshmemNotPurged, nil - - case linux.AshmemUnpinIoctl: - // TODO(b/30946773): Implement purge on unpin. - a.pb.UnpinRange(r) - return 0, nil - - case linux.AshmemGetPinStatusIoctl: - if a.pb.RangePinnedStatus(r) { - return linux.AshmemIsPinned, nil - } - return linux.AshmemIsUnpinned, nil - - default: - panic("unreachable") - } - -} diff --git a/pkg/sentry/fs/ashmem/ashmem_state_autogen.go b/pkg/sentry/fs/ashmem/ashmem_state_autogen.go deleted file mode 100755 index 13defb033..000000000 --- a/pkg/sentry/fs/ashmem/ashmem_state_autogen.go +++ /dev/null @@ -1,123 +0,0 @@ -// automatically generated by stateify. - -package ashmem - -import ( - "gvisor.dev/gvisor/pkg/state" -) - -func (x *Area) beforeSave() {} -func (x *Area) save(m state.Map) { - x.beforeSave() - m.Save("ad", &x.ad) - m.Save("tmpfsFile", &x.tmpfsFile) - m.Save("name", &x.name) - m.Save("size", &x.size) - m.Save("perms", &x.perms) - m.Save("pb", &x.pb) -} - -func (x *Area) afterLoad() {} -func (x *Area) load(m state.Map) { - m.Load("ad", &x.ad) - m.Load("tmpfsFile", &x.tmpfsFile) - m.Load("name", &x.name) - m.Load("size", &x.size) - m.Load("perms", &x.perms) - m.Load("pb", &x.pb) -} - -func (x *Device) beforeSave() {} -func (x *Device) save(m state.Map) { - x.beforeSave() - m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) -} - -func (x *Device) afterLoad() {} -func (x *Device) load(m state.Map) { - m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) -} - -func (x *PinBoard) beforeSave() {} -func (x *PinBoard) save(m state.Map) { - x.beforeSave() - m.Save("Set", &x.Set) -} - -func (x *PinBoard) afterLoad() {} -func (x *PinBoard) load(m state.Map) { - m.Load("Set", &x.Set) -} - -func (x *Range) beforeSave() {} -func (x *Range) save(m state.Map) { - x.beforeSave() - m.Save("Start", &x.Start) - m.Save("End", &x.End) -} - -func (x *Range) afterLoad() {} -func (x *Range) load(m state.Map) { - m.Load("Start", &x.Start) - m.Load("End", &x.End) -} - -func (x *Set) beforeSave() {} -func (x *Set) save(m state.Map) { - x.beforeSave() - var root *SegmentDataSlices = x.saveRoot() - m.SaveValue("root", root) -} - -func (x *Set) afterLoad() {} -func (x *Set) load(m state.Map) { - m.LoadValue("root", new(*SegmentDataSlices), func(y interface{}) { x.loadRoot(y.(*SegmentDataSlices)) }) -} - -func (x *node) beforeSave() {} -func (x *node) save(m state.Map) { - x.beforeSave() - m.Save("nrSegments", &x.nrSegments) - m.Save("parent", &x.parent) - m.Save("parentIndex", &x.parentIndex) - m.Save("hasChildren", &x.hasChildren) - m.Save("keys", &x.keys) - m.Save("values", &x.values) - m.Save("children", &x.children) -} - -func (x *node) afterLoad() {} -func (x *node) load(m state.Map) { - m.Load("nrSegments", &x.nrSegments) - m.Load("parent", &x.parent) - m.Load("parentIndex", &x.parentIndex) - m.Load("hasChildren", &x.hasChildren) - m.Load("keys", &x.keys) - m.Load("values", &x.values) - m.Load("children", &x.children) -} - -func (x *SegmentDataSlices) beforeSave() {} -func (x *SegmentDataSlices) save(m state.Map) { - x.beforeSave() - m.Save("Start", &x.Start) - m.Save("End", &x.End) - m.Save("Values", &x.Values) -} - -func (x *SegmentDataSlices) afterLoad() {} -func (x *SegmentDataSlices) load(m state.Map) { - m.Load("Start", &x.Start) - m.Load("End", &x.End) - m.Load("Values", &x.Values) -} - -func init() { - state.Register("ashmem.Area", (*Area)(nil), state.Fns{Save: (*Area).save, Load: (*Area).load}) - state.Register("ashmem.Device", (*Device)(nil), state.Fns{Save: (*Device).save, Load: (*Device).load}) - state.Register("ashmem.PinBoard", (*PinBoard)(nil), state.Fns{Save: (*PinBoard).save, Load: (*PinBoard).load}) - state.Register("ashmem.Range", (*Range)(nil), state.Fns{Save: (*Range).save, Load: (*Range).load}) - state.Register("ashmem.Set", (*Set)(nil), state.Fns{Save: (*Set).save, Load: (*Set).load}) - state.Register("ashmem.node", (*node)(nil), state.Fns{Save: (*node).save, Load: (*node).load}) - state.Register("ashmem.SegmentDataSlices", (*SegmentDataSlices)(nil), state.Fns{Save: (*SegmentDataSlices).save, Load: (*SegmentDataSlices).load}) -} diff --git a/pkg/sentry/fs/ashmem/device.go b/pkg/sentry/fs/ashmem/device.go deleted file mode 100644 index 776f54abe..000000000 --- a/pkg/sentry/fs/ashmem/device.go +++ /dev/null @@ -1,61 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package ashmem implements Android ashmem module (Anonymus Shared Memory). -package ashmem - -import ( - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/usermem" -) - -// Device implements fs.InodeOperations. -// -// +stateify savable -type Device struct { - fsutil.InodeGenericChecker `state:"nosave"` - fsutil.InodeNoExtendedAttributes `state:"nosave"` - fsutil.InodeNoopAllocate `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 = (*Device)(nil) - -// NewDevice creates and initializes a Device structure. -func NewDevice(ctx context.Context, owner fs.FileOwner, fp fs.FilePermissions) *Device { - return &Device{ - InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, owner, fp, linux.ANON_INODE_FS_MAGIC), - } -} - -// 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{ - ad: ad, - tmpfsFile: nil, - perms: usermem.AnyAccess, - }), nil -} diff --git a/pkg/sentry/fs/ashmem/pin_board.go b/pkg/sentry/fs/ashmem/pin_board.go deleted file mode 100644 index c5400dd94..000000000 --- a/pkg/sentry/fs/ashmem/pin_board.go +++ /dev/null @@ -1,127 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package ashmem - -import "gvisor.dev/gvisor/pkg/abi/linux" - -const maxUint64 = ^uint64(0) - -// setFunctions implements segment.Functions generated from segment.Functions for -// uint64 Key and noValue Value. For more information, see the build file and -// segment set implementation at pkg/segment/set.go. -type setFunctions struct{} - -// noValue is a type of range attached value, which is irrelevant here. -type noValue struct{} - -// MinKey implements segment.Functions.MinKey. -func (setFunctions) MinKey() uint64 { - return 0 -} - -// MaxKey implements segment.Functions.MaxKey. -func (setFunctions) MaxKey() uint64 { - return maxUint64 -} - -// ClearValue implements segment.Functions.ClearValue. -func (setFunctions) ClearValue(*noValue) { - return -} - -// Merge implements segment.Functions.Merge. -func (setFunctions) Merge(Range, noValue, Range, noValue) (noValue, bool) { - return noValue{}, true -} - -// Split implements segment.Functions.Split. -func (setFunctions) Split(Range, noValue, uint64) (noValue, noValue) { - return noValue{}, noValue{} -} - -// PinBoard represents a set of pinned ranges in ashmem. -// -// segment.Set is used for implementation where segments represent -// ranges of pinned bytes, while gaps represent ranges of unpinned -// bytes. All ranges are page-aligned. -// -// +stateify savable -type PinBoard struct { - Set -} - -// NewPinBoard creates a new pin board with all pages pinned. -func NewPinBoard() *PinBoard { - var pb PinBoard - pb.PinRange(Range{0, maxUint64}) - return &pb -} - -// PinRange pins all pages in the specified range and returns true -// if there are any newly pinned pages. -func (pb *PinBoard) PinRange(r Range) bool { - pinnedPages := false - for gap := pb.LowerBoundGap(r.Start); gap.Ok() && gap.Start() < r.End; { - common := gap.Range().Intersect(r) - if common.Length() == 0 { - gap = gap.NextGap() - continue - } - pinnedPages = true - gap = pb.Insert(gap, common, noValue{}).NextGap() - } - return pinnedPages -} - -// UnpinRange unpins all pages in the specified range. -func (pb *PinBoard) UnpinRange(r Range) { - for seg := pb.LowerBoundSegment(r.Start); seg.Ok() && seg.Start() < r.End; { - common := seg.Range().Intersect(r) - if common.Length() == 0 { - seg = seg.NextSegment() - continue - } - seg = pb.RemoveRange(common).NextSegment() - } -} - -// RangePinnedStatus returns false if there's at least one unpinned page in the -// specified range. -func (pb *PinBoard) RangePinnedStatus(r Range) bool { - for gap := pb.LowerBoundGap(r.Start); gap.Ok() && gap.Start() < r.End; { - common := gap.Range().Intersect(r) - if common.Length() == 0 { - gap = gap.NextGap() - continue - } - return false - } - return true -} - -// RangeFromAshmemPin converts ashmem's original pin structure -// to Range. -func RangeFromAshmemPin(ap linux.AshmemPin) Range { - if ap.Len == 0 { - return Range{ - uint64(ap.Offset), - maxUint64, - } - } - return Range{ - uint64(ap.Offset), - uint64(ap.Offset) + uint64(ap.Len), - } -} diff --git a/pkg/sentry/fs/ashmem/uint64_range.go b/pkg/sentry/fs/ashmem/uint64_range.go deleted file mode 100755 index d71a10b16..000000000 --- a/pkg/sentry/fs/ashmem/uint64_range.go +++ /dev/null @@ -1,62 +0,0 @@ -package ashmem - -// A Range represents a contiguous range of T. -// -// +stateify savable -type Range struct { - // Start is the inclusive start of the range. - Start uint64 - - // End is the exclusive end of the range. - End uint64 -} - -// WellFormed returns true if r.Start <= r.End. All other methods on a Range -// require that the Range is well-formed. -func (r Range) WellFormed() bool { - return r.Start <= r.End -} - -// Length returns the length of the range. -func (r Range) Length() uint64 { - return r.End - r.Start -} - -// Contains returns true if r contains x. -func (r Range) Contains(x uint64) bool { - return r.Start <= x && x < r.End -} - -// Overlaps returns true if r and r2 overlap. -func (r Range) Overlaps(r2 Range) bool { - return r.Start < r2.End && r2.Start < r.End -} - -// IsSupersetOf returns true if r is a superset of r2; that is, the range r2 is -// contained within r. -func (r Range) IsSupersetOf(r2 Range) bool { - return r.Start <= r2.Start && r.End >= r2.End -} - -// Intersect returns a range consisting of the intersection between r and r2. -// If r and r2 do not overlap, Intersect returns a range with unspecified -// bounds, but for which Length() == 0. -func (r Range) Intersect(r2 Range) Range { - if r.Start < r2.Start { - r.Start = r2.Start - } - if r.End > r2.End { - r.End = r2.End - } - if r.End < r.Start { - r.End = r.Start - } - return r -} - -// CanSplitAt returns true if it is legal to split a segment spanning the range -// r at x; that is, splitting at x would produce two ranges, both of which have -// non-zero length. -func (r Range) CanSplitAt(x uint64) bool { - return r.Contains(x) && r.Start < x -} diff --git a/pkg/sentry/fs/ashmem/uint64_set.go b/pkg/sentry/fs/ashmem/uint64_set.go deleted file mode 100755 index a4860175a..000000000 --- a/pkg/sentry/fs/ashmem/uint64_set.go +++ /dev/null @@ -1,1270 +0,0 @@ -package ashmem - -import ( - "bytes" - "fmt" -) - -const ( - // minDegree is the minimum degree of an internal node in a Set B-tree. - // - // - Any non-root node has at least minDegree-1 segments. - // - // - Any non-root internal (non-leaf) node has at least minDegree children. - // - // - The root node may have fewer than minDegree-1 segments, but it may - // only have 0 segments if the tree is empty. - // - // Our implementation requires minDegree >= 3. Higher values of minDegree - // usually improve performance, but increase memory usage for small sets. - minDegree = 3 - - maxDegree = 2 * minDegree -) - -// A Set is a mapping of segments with non-overlapping Range keys. The zero -// value for a Set is an empty set. Set values are not safely movable nor -// copyable. Set is thread-compatible. -// -// +stateify savable -type Set struct { - root node `state:".(*SegmentDataSlices)"` -} - -// IsEmpty returns true if the set contains no segments. -func (s *Set) IsEmpty() bool { - return s.root.nrSegments == 0 -} - -// IsEmptyRange returns true iff no segments in the set overlap the given -// range. This is semantically equivalent to s.SpanRange(r) == 0, but may be -// more efficient. -func (s *Set) IsEmptyRange(r Range) bool { - switch { - case r.Length() < 0: - panic(fmt.Sprintf("invalid range %v", r)) - case r.Length() == 0: - return true - } - _, gap := s.Find(r.Start) - if !gap.Ok() { - return false - } - return r.End <= gap.End() -} - -// Span returns the total size of all segments in the set. -func (s *Set) Span() uint64 { - var sz uint64 - for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { - sz += seg.Range().Length() - } - return sz -} - -// SpanRange returns the total size of the intersection of segments in the set -// with the given range. -func (s *Set) SpanRange(r Range) uint64 { - switch { - case r.Length() < 0: - panic(fmt.Sprintf("invalid range %v", r)) - case r.Length() == 0: - return 0 - } - var sz uint64 - for seg := s.LowerBoundSegment(r.Start); seg.Ok() && seg.Start() < r.End; seg = seg.NextSegment() { - sz += seg.Range().Intersect(r).Length() - } - return sz -} - -// FirstSegment returns the first segment in the set. If the set is empty, -// FirstSegment returns a terminal iterator. -func (s *Set) FirstSegment() Iterator { - if s.root.nrSegments == 0 { - return Iterator{} - } - return s.root.firstSegment() -} - -// LastSegment returns the last segment in the set. If the set is empty, -// LastSegment returns a terminal iterator. -func (s *Set) LastSegment() Iterator { - if s.root.nrSegments == 0 { - return Iterator{} - } - return s.root.lastSegment() -} - -// FirstGap returns the first gap in the set. -func (s *Set) FirstGap() GapIterator { - n := &s.root - for n.hasChildren { - n = n.children[0] - } - return GapIterator{n, 0} -} - -// LastGap returns the last gap in the set. -func (s *Set) LastGap() GapIterator { - n := &s.root - for n.hasChildren { - n = n.children[n.nrSegments] - } - return GapIterator{n, n.nrSegments} -} - -// Find returns the segment or gap whose range contains the given key. If a -// segment is found, the returned Iterator is non-terminal and the -// returned GapIterator is terminal. Otherwise, the returned Iterator is -// terminal and the returned GapIterator is non-terminal. -func (s *Set) Find(key uint64) (Iterator, GapIterator) { - n := &s.root - for { - - lower := 0 - upper := n.nrSegments - for lower < upper { - i := lower + (upper-lower)/2 - if r := n.keys[i]; key < r.End { - if key >= r.Start { - return Iterator{n, i}, GapIterator{} - } - upper = i - } else { - lower = i + 1 - } - } - i := lower - if !n.hasChildren { - return Iterator{}, GapIterator{n, i} - } - n = n.children[i] - } -} - -// FindSegment returns the segment whose range contains the given key. If no -// such segment exists, FindSegment returns a terminal iterator. -func (s *Set) FindSegment(key uint64) Iterator { - seg, _ := s.Find(key) - return seg -} - -// LowerBoundSegment returns the segment with the lowest range that contains a -// key greater than or equal to min. If no such segment exists, -// LowerBoundSegment returns a terminal iterator. -func (s *Set) LowerBoundSegment(min uint64) Iterator { - seg, gap := s.Find(min) - if seg.Ok() { - return seg - } - return gap.NextSegment() -} - -// UpperBoundSegment returns the segment with the highest range that contains a -// key less than or equal to max. If no such segment exists, UpperBoundSegment -// returns a terminal iterator. -func (s *Set) UpperBoundSegment(max uint64) Iterator { - seg, gap := s.Find(max) - if seg.Ok() { - return seg - } - return gap.PrevSegment() -} - -// FindGap returns the gap containing the given key. If no such gap exists -// (i.e. the set contains a segment containing that key), FindGap returns a -// terminal iterator. -func (s *Set) FindGap(key uint64) GapIterator { - _, gap := s.Find(key) - return gap -} - -// LowerBoundGap returns the gap with the lowest range that is greater than or -// equal to min. -func (s *Set) LowerBoundGap(min uint64) GapIterator { - seg, gap := s.Find(min) - if gap.Ok() { - return gap - } - return seg.NextGap() -} - -// UpperBoundGap returns the gap with the highest range that is less than or -// equal to max. -func (s *Set) UpperBoundGap(max uint64) GapIterator { - seg, gap := s.Find(max) - if gap.Ok() { - return gap - } - return seg.PrevGap() -} - -// Add inserts the given segment into the set and returns true. If the new -// segment can be merged with adjacent segments, Add will do so. If the new -// segment would overlap an existing segment, Add returns false. If Add -// succeeds, all existing iterators are invalidated. -func (s *Set) Add(r Range, val noValue) bool { - if r.Length() <= 0 { - panic(fmt.Sprintf("invalid segment range %v", r)) - } - gap := s.FindGap(r.Start) - if !gap.Ok() { - return false - } - if r.End > gap.End() { - return false - } - s.Insert(gap, r, val) - return true -} - -// AddWithoutMerging inserts the given segment into the set and returns true. -// If it would overlap an existing segment, AddWithoutMerging does nothing and -// returns false. If AddWithoutMerging succeeds, all existing iterators are -// invalidated. -func (s *Set) AddWithoutMerging(r Range, val noValue) bool { - if r.Length() <= 0 { - panic(fmt.Sprintf("invalid segment range %v", r)) - } - gap := s.FindGap(r.Start) - if !gap.Ok() { - return false - } - if r.End > gap.End() { - return false - } - s.InsertWithoutMergingUnchecked(gap, r, val) - return true -} - -// Insert inserts the given segment into the given gap. If the new segment can -// be merged with adjacent segments, Insert will do so. Insert returns an -// iterator to the segment containing the inserted value (which may have been -// merged with other values). All existing iterators (including gap, but not -// including the returned iterator) are invalidated. -// -// If the gap cannot accommodate the segment, or if r is invalid, Insert panics. -// -// Insert is semantically equivalent to a InsertWithoutMerging followed by a -// Merge, but may be more efficient. Note that there is no unchecked variant of -// Insert since Insert must retrieve and inspect gap's predecessor and -// successor segments regardless. -func (s *Set) Insert(gap GapIterator, r Range, val noValue) Iterator { - if r.Length() <= 0 { - panic(fmt.Sprintf("invalid segment range %v", r)) - } - prev, next := gap.PrevSegment(), gap.NextSegment() - if prev.Ok() && prev.End() > r.Start { - panic(fmt.Sprintf("new segment %v overlaps predecessor %v", r, prev.Range())) - } - if next.Ok() && next.Start() < r.End { - panic(fmt.Sprintf("new segment %v overlaps successor %v", r, next.Range())) - } - if prev.Ok() && prev.End() == r.Start { - if mval, ok := (setFunctions{}).Merge(prev.Range(), prev.Value(), r, val); ok { - prev.SetEndUnchecked(r.End) - prev.SetValue(mval) - if next.Ok() && next.Start() == r.End { - val = mval - if mval, ok := (setFunctions{}).Merge(prev.Range(), val, next.Range(), next.Value()); ok { - prev.SetEndUnchecked(next.End()) - prev.SetValue(mval) - return s.Remove(next).PrevSegment() - } - } - return prev - } - } - if next.Ok() && next.Start() == r.End { - if mval, ok := (setFunctions{}).Merge(r, val, next.Range(), next.Value()); ok { - next.SetStartUnchecked(r.Start) - next.SetValue(mval) - return next - } - } - return s.InsertWithoutMergingUnchecked(gap, r, val) -} - -// InsertWithoutMerging inserts the given segment into the given gap and -// returns an iterator to the inserted segment. All existing iterators -// (including gap, but not including the returned iterator) are invalidated. -// -// If the gap cannot accommodate the segment, or if r is invalid, -// InsertWithoutMerging panics. -func (s *Set) InsertWithoutMerging(gap GapIterator, r Range, val noValue) Iterator { - if r.Length() <= 0 { - panic(fmt.Sprintf("invalid segment range %v", r)) - } - if gr := gap.Range(); !gr.IsSupersetOf(r) { - panic(fmt.Sprintf("cannot insert segment range %v into gap range %v", r, gr)) - } - return s.InsertWithoutMergingUnchecked(gap, r, val) -} - -// InsertWithoutMergingUnchecked inserts the given segment into the given gap -// and returns an iterator to the inserted segment. All existing iterators -// (including gap, but not including the returned iterator) are invalidated. -// -// Preconditions: r.Start >= gap.Start(); r.End <= gap.End(). -func (s *Set) InsertWithoutMergingUnchecked(gap GapIterator, r Range, val noValue) Iterator { - gap = gap.node.rebalanceBeforeInsert(gap) - copy(gap.node.keys[gap.index+1:], gap.node.keys[gap.index:gap.node.nrSegments]) - copy(gap.node.values[gap.index+1:], gap.node.values[gap.index:gap.node.nrSegments]) - gap.node.keys[gap.index] = r - gap.node.values[gap.index] = val - gap.node.nrSegments++ - return Iterator{gap.node, gap.index} -} - -// Remove removes the given segment and returns an iterator to the vacated gap. -// All existing iterators (including seg, but not including the returned -// iterator) are invalidated. -func (s *Set) Remove(seg Iterator) GapIterator { - - if seg.node.hasChildren { - - victim := seg.PrevSegment() - - seg.SetRangeUnchecked(victim.Range()) - seg.SetValue(victim.Value()) - return s.Remove(victim).NextGap() - } - copy(seg.node.keys[seg.index:], seg.node.keys[seg.index+1:seg.node.nrSegments]) - copy(seg.node.values[seg.index:], seg.node.values[seg.index+1:seg.node.nrSegments]) - setFunctions{}.ClearValue(&seg.node.values[seg.node.nrSegments-1]) - seg.node.nrSegments-- - return seg.node.rebalanceAfterRemove(GapIterator{seg.node, seg.index}) -} - -// RemoveAll removes all segments from the set. All existing iterators are -// invalidated. -func (s *Set) RemoveAll() { - s.root = node{} -} - -// RemoveRange removes all segments in the given range. An iterator to the -// newly formed gap is returned, and all existing iterators are invalidated. -func (s *Set) RemoveRange(r Range) GapIterator { - seg, gap := s.Find(r.Start) - if seg.Ok() { - seg = s.Isolate(seg, r) - gap = s.Remove(seg) - } - for seg = gap.NextSegment(); seg.Ok() && seg.Start() < r.End; seg = gap.NextSegment() { - seg = s.Isolate(seg, r) - gap = s.Remove(seg) - } - return gap -} - -// Merge attempts to merge two neighboring segments. If successful, Merge -// returns an iterator to the merged segment, and all existing iterators are -// invalidated. Otherwise, Merge returns a terminal iterator. -// -// If first is not the predecessor of second, Merge panics. -func (s *Set) Merge(first, second Iterator) Iterator { - if first.NextSegment() != second { - panic(fmt.Sprintf("attempt to merge non-neighboring segments %v, %v", first.Range(), second.Range())) - } - return s.MergeUnchecked(first, second) -} - -// MergeUnchecked attempts to merge two neighboring segments. If successful, -// MergeUnchecked returns an iterator to the merged segment, and all existing -// iterators are invalidated. Otherwise, MergeUnchecked returns a terminal -// iterator. -// -// Precondition: first is the predecessor of second: first.NextSegment() == -// second, first == second.PrevSegment(). -func (s *Set) MergeUnchecked(first, second Iterator) Iterator { - if first.End() == second.Start() { - if mval, ok := (setFunctions{}).Merge(first.Range(), first.Value(), second.Range(), second.Value()); ok { - - first.SetEndUnchecked(second.End()) - first.SetValue(mval) - return s.Remove(second).PrevSegment() - } - } - return Iterator{} -} - -// MergeAll attempts to merge all adjacent segments in the set. All existing -// iterators are invalidated. -func (s *Set) MergeAll() { - seg := s.FirstSegment() - if !seg.Ok() { - return - } - next := seg.NextSegment() - for next.Ok() { - if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { - seg, next = mseg, mseg.NextSegment() - } else { - seg, next = next, next.NextSegment() - } - } -} - -// MergeRange attempts to merge all adjacent segments that contain a key in the -// specific range. All existing iterators are invalidated. -func (s *Set) MergeRange(r Range) { - seg := s.LowerBoundSegment(r.Start) - if !seg.Ok() { - return - } - next := seg.NextSegment() - for next.Ok() && next.Range().Start < r.End { - if mseg := s.MergeUnchecked(seg, next); mseg.Ok() { - seg, next = mseg, mseg.NextSegment() - } else { - seg, next = next, next.NextSegment() - } - } -} - -// MergeAdjacent attempts to merge the segment containing r.Start with its -// predecessor, and the segment containing r.End-1 with its successor. -func (s *Set) MergeAdjacent(r Range) { - first := s.FindSegment(r.Start) - if first.Ok() { - if prev := first.PrevSegment(); prev.Ok() { - s.Merge(prev, first) - } - } - last := s.FindSegment(r.End - 1) - if last.Ok() { - if next := last.NextSegment(); next.Ok() { - s.Merge(last, next) - } - } -} - -// Split splits the given segment at the given key and returns iterators to the -// two resulting segments. All existing iterators (including seg, but not -// including the returned iterators) are invalidated. -// -// If the segment cannot be split at split (because split is at the start or -// end of the segment's range, so splitting would produce a segment with zero -// length, or because split falls outside the segment's range altogether), -// Split panics. -func (s *Set) Split(seg Iterator, split uint64) (Iterator, Iterator) { - if !seg.Range().CanSplitAt(split) { - panic(fmt.Sprintf("can't split %v at %v", seg.Range(), split)) - } - return s.SplitUnchecked(seg, split) -} - -// SplitUnchecked splits the given segment at the given key and returns -// iterators to the two resulting segments. All existing iterators (including -// seg, but not including the returned iterators) are invalidated. -// -// Preconditions: seg.Start() < key < seg.End(). -func (s *Set) SplitUnchecked(seg Iterator, split uint64) (Iterator, Iterator) { - val1, val2 := (setFunctions{}).Split(seg.Range(), seg.Value(), split) - end2 := seg.End() - seg.SetEndUnchecked(split) - seg.SetValue(val1) - seg2 := s.InsertWithoutMergingUnchecked(seg.NextGap(), Range{split, end2}, val2) - - return seg2.PrevSegment(), seg2 -} - -// SplitAt splits the segment straddling split, if one exists. SplitAt returns -// true if a segment was split and false otherwise. If SplitAt splits a -// segment, all existing iterators are invalidated. -func (s *Set) SplitAt(split uint64) bool { - if seg := s.FindSegment(split); seg.Ok() && seg.Range().CanSplitAt(split) { - s.SplitUnchecked(seg, split) - return true - } - return false -} - -// Isolate ensures that the given segment's range does not escape r by -// splitting at r.Start and r.End if necessary, and returns an updated iterator -// to the bounded segment. All existing iterators (including seg, but not -// including the returned iterators) are invalidated. -func (s *Set) Isolate(seg Iterator, r Range) Iterator { - if seg.Range().CanSplitAt(r.Start) { - _, seg = s.SplitUnchecked(seg, r.Start) - } - if seg.Range().CanSplitAt(r.End) { - seg, _ = s.SplitUnchecked(seg, r.End) - } - return seg -} - -// ApplyContiguous applies a function to a contiguous range of segments, -// splitting if necessary. The function is applied until the first gap is -// encountered, at which point the gap is returned. If the function is applied -// across the entire range, a terminal gap is returned. All existing iterators -// are invalidated. -// -// N.B. The Iterator must not be invalidated by the function. -func (s *Set) ApplyContiguous(r Range, fn func(seg Iterator)) GapIterator { - seg, gap := s.Find(r.Start) - if !seg.Ok() { - return gap - } - for { - seg = s.Isolate(seg, r) - fn(seg) - if seg.End() >= r.End { - return GapIterator{} - } - gap = seg.NextGap() - if !gap.IsEmpty() { - return gap - } - seg = gap.NextSegment() - if !seg.Ok() { - - return GapIterator{} - } - } -} - -// +stateify savable -type node struct { - // An internal binary tree node looks like: - // - // K - // / \ - // Cl Cr - // - // where all keys in the subtree rooted by Cl (the left subtree) are less - // than K (the key of the parent node), and all keys in the subtree rooted - // by Cr (the right subtree) are greater than K. - // - // An internal B-tree node's indexes work out to look like: - // - // K0 K1 K2 ... Kn-1 - // / \/ \/ \ ... / \ - // C0 C1 C2 C3 ... Cn-1 Cn - // - // where n is nrSegments. - nrSegments int - - // parent is a pointer to this node's parent. If this node is root, parent - // is nil. - parent *node - - // parentIndex is the index of this node in parent.children. - parentIndex int - - // Flag for internal nodes that is technically redundant with "children[0] - // != nil", but is stored in the first cache line. "hasChildren" rather - // than "isLeaf" because false must be the correct value for an empty root. - hasChildren bool - - // Nodes store keys and values in separate arrays to maximize locality in - // the common case (scanning keys for lookup). - keys [maxDegree - 1]Range - values [maxDegree - 1]noValue - children [maxDegree]*node -} - -// firstSegment returns the first segment in the subtree rooted by n. -// -// Preconditions: n.nrSegments != 0. -func (n *node) firstSegment() Iterator { - for n.hasChildren { - n = n.children[0] - } - return Iterator{n, 0} -} - -// lastSegment returns the last segment in the subtree rooted by n. -// -// Preconditions: n.nrSegments != 0. -func (n *node) lastSegment() Iterator { - for n.hasChildren { - n = n.children[n.nrSegments] - } - return Iterator{n, n.nrSegments - 1} -} - -func (n *node) prevSibling() *node { - if n.parent == nil || n.parentIndex == 0 { - return nil - } - return n.parent.children[n.parentIndex-1] -} - -func (n *node) nextSibling() *node { - if n.parent == nil || n.parentIndex == n.parent.nrSegments { - return nil - } - return n.parent.children[n.parentIndex+1] -} - -// rebalanceBeforeInsert splits n and its ancestors if they are full, as -// required for insertion, and returns an updated iterator to the position -// represented by gap. -func (n *node) rebalanceBeforeInsert(gap GapIterator) GapIterator { - if n.parent != nil { - gap = n.parent.rebalanceBeforeInsert(gap) - } - if n.nrSegments < maxDegree-1 { - return gap - } - if n.parent == nil { - - left := &node{ - nrSegments: minDegree - 1, - parent: n, - parentIndex: 0, - hasChildren: n.hasChildren, - } - right := &node{ - nrSegments: minDegree - 1, - parent: n, - parentIndex: 1, - hasChildren: n.hasChildren, - } - copy(left.keys[:minDegree-1], n.keys[:minDegree-1]) - copy(left.values[:minDegree-1], n.values[:minDegree-1]) - copy(right.keys[:minDegree-1], n.keys[minDegree:]) - copy(right.values[:minDegree-1], n.values[minDegree:]) - n.keys[0], n.values[0] = n.keys[minDegree-1], n.values[minDegree-1] - zeroValueSlice(n.values[1:]) - if n.hasChildren { - copy(left.children[:minDegree], n.children[:minDegree]) - copy(right.children[:minDegree], n.children[minDegree:]) - zeroNodeSlice(n.children[2:]) - for i := 0; i < minDegree; i++ { - left.children[i].parent = left - left.children[i].parentIndex = i - right.children[i].parent = right - right.children[i].parentIndex = i - } - } - n.nrSegments = 1 - n.hasChildren = true - n.children[0] = left - n.children[1] = right - if gap.node != n { - return gap - } - if gap.index < minDegree { - return GapIterator{left, gap.index} - } - return GapIterator{right, gap.index - minDegree} - } - - copy(n.parent.keys[n.parentIndex+1:], n.parent.keys[n.parentIndex:n.parent.nrSegments]) - copy(n.parent.values[n.parentIndex+1:], n.parent.values[n.parentIndex:n.parent.nrSegments]) - n.parent.keys[n.parentIndex], n.parent.values[n.parentIndex] = n.keys[minDegree-1], n.values[minDegree-1] - copy(n.parent.children[n.parentIndex+2:], n.parent.children[n.parentIndex+1:n.parent.nrSegments+1]) - for i := n.parentIndex + 2; i < n.parent.nrSegments+2; i++ { - n.parent.children[i].parentIndex = i - } - sibling := &node{ - nrSegments: minDegree - 1, - parent: n.parent, - parentIndex: n.parentIndex + 1, - hasChildren: n.hasChildren, - } - n.parent.children[n.parentIndex+1] = sibling - n.parent.nrSegments++ - copy(sibling.keys[:minDegree-1], n.keys[minDegree:]) - copy(sibling.values[:minDegree-1], n.values[minDegree:]) - zeroValueSlice(n.values[minDegree-1:]) - if n.hasChildren { - copy(sibling.children[:minDegree], n.children[minDegree:]) - zeroNodeSlice(n.children[minDegree:]) - for i := 0; i < minDegree; i++ { - sibling.children[i].parent = sibling - sibling.children[i].parentIndex = i - } - } - n.nrSegments = minDegree - 1 - - if gap.node != n { - return gap - } - if gap.index < minDegree { - return gap - } - return GapIterator{sibling, gap.index - minDegree} -} - -// rebalanceAfterRemove "unsplits" n and its ancestors if they are deficient -// (contain fewer segments than required by B-tree invariants), as required for -// removal, and returns an updated iterator to the position represented by gap. -// -// Precondition: n is the only node in the tree that may currently violate a -// B-tree invariant. -func (n *node) rebalanceAfterRemove(gap GapIterator) GapIterator { - for { - if n.nrSegments >= minDegree-1 { - return gap - } - if n.parent == nil { - - return gap - } - - if sibling := n.prevSibling(); sibling != nil && sibling.nrSegments >= minDegree { - copy(n.keys[1:], n.keys[:n.nrSegments]) - copy(n.values[1:], n.values[:n.nrSegments]) - n.keys[0] = n.parent.keys[n.parentIndex-1] - n.values[0] = n.parent.values[n.parentIndex-1] - n.parent.keys[n.parentIndex-1] = sibling.keys[sibling.nrSegments-1] - n.parent.values[n.parentIndex-1] = sibling.values[sibling.nrSegments-1] - setFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) - if n.hasChildren { - copy(n.children[1:], n.children[:n.nrSegments+1]) - n.children[0] = sibling.children[sibling.nrSegments] - sibling.children[sibling.nrSegments] = nil - n.children[0].parent = n - n.children[0].parentIndex = 0 - for i := 1; i < n.nrSegments+2; i++ { - n.children[i].parentIndex = i - } - } - n.nrSegments++ - sibling.nrSegments-- - if gap.node == sibling && gap.index == sibling.nrSegments { - return GapIterator{n, 0} - } - if gap.node == n { - return GapIterator{n, gap.index + 1} - } - return gap - } - if sibling := n.nextSibling(); sibling != nil && sibling.nrSegments >= minDegree { - n.keys[n.nrSegments] = n.parent.keys[n.parentIndex] - n.values[n.nrSegments] = n.parent.values[n.parentIndex] - n.parent.keys[n.parentIndex] = sibling.keys[0] - n.parent.values[n.parentIndex] = sibling.values[0] - copy(sibling.keys[:sibling.nrSegments-1], sibling.keys[1:]) - copy(sibling.values[:sibling.nrSegments-1], sibling.values[1:]) - setFunctions{}.ClearValue(&sibling.values[sibling.nrSegments-1]) - if n.hasChildren { - n.children[n.nrSegments+1] = sibling.children[0] - copy(sibling.children[:sibling.nrSegments], sibling.children[1:]) - sibling.children[sibling.nrSegments] = nil - n.children[n.nrSegments+1].parent = n - n.children[n.nrSegments+1].parentIndex = n.nrSegments + 1 - for i := 0; i < sibling.nrSegments; i++ { - sibling.children[i].parentIndex = i - } - } - n.nrSegments++ - sibling.nrSegments-- - if gap.node == sibling { - if gap.index == 0 { - return GapIterator{n, n.nrSegments} - } - return GapIterator{sibling, gap.index - 1} - } - return gap - } - - p := n.parent - if p.nrSegments == 1 { - - left, right := p.children[0], p.children[1] - p.nrSegments = left.nrSegments + right.nrSegments + 1 - p.hasChildren = left.hasChildren - p.keys[left.nrSegments] = p.keys[0] - p.values[left.nrSegments] = p.values[0] - copy(p.keys[:left.nrSegments], left.keys[:left.nrSegments]) - copy(p.values[:left.nrSegments], left.values[:left.nrSegments]) - copy(p.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) - copy(p.values[left.nrSegments+1:], right.values[:right.nrSegments]) - if left.hasChildren { - copy(p.children[:left.nrSegments+1], left.children[:left.nrSegments+1]) - copy(p.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) - for i := 0; i < p.nrSegments+1; i++ { - p.children[i].parent = p - p.children[i].parentIndex = i - } - } else { - p.children[0] = nil - p.children[1] = nil - } - if gap.node == left { - return GapIterator{p, gap.index} - } - if gap.node == right { - return GapIterator{p, gap.index + left.nrSegments + 1} - } - return gap - } - // Merge n and either sibling, along with the segment separating the - // two, into whichever of the two nodes comes first. This is the - // reverse of the non-root splitting case in - // node.rebalanceBeforeInsert. - var left, right *node - if n.parentIndex > 0 { - left = n.prevSibling() - right = n - } else { - left = n - right = n.nextSibling() - } - - if gap.node == right { - gap = GapIterator{left, gap.index + left.nrSegments + 1} - } - left.keys[left.nrSegments] = p.keys[left.parentIndex] - left.values[left.nrSegments] = p.values[left.parentIndex] - copy(left.keys[left.nrSegments+1:], right.keys[:right.nrSegments]) - copy(left.values[left.nrSegments+1:], right.values[:right.nrSegments]) - if left.hasChildren { - copy(left.children[left.nrSegments+1:], right.children[:right.nrSegments+1]) - for i := left.nrSegments + 1; i < left.nrSegments+right.nrSegments+2; i++ { - left.children[i].parent = left - left.children[i].parentIndex = i - } - } - left.nrSegments += right.nrSegments + 1 - copy(p.keys[left.parentIndex:], p.keys[left.parentIndex+1:p.nrSegments]) - copy(p.values[left.parentIndex:], p.values[left.parentIndex+1:p.nrSegments]) - setFunctions{}.ClearValue(&p.values[p.nrSegments-1]) - copy(p.children[left.parentIndex+1:], p.children[left.parentIndex+2:p.nrSegments+1]) - for i := 0; i < p.nrSegments; i++ { - p.children[i].parentIndex = i - } - p.children[p.nrSegments] = nil - p.nrSegments-- - - n = p - } -} - -// A Iterator is conceptually one of: -// -// - A pointer to a segment in a set; or -// -// - A terminal iterator, which is a sentinel indicating that the end of -// iteration has been reached. -// -// Iterators are copyable values and are meaningfully equality-comparable. The -// zero value of Iterator is a terminal iterator. -// -// Unless otherwise specified, any mutation of a set invalidates all existing -// iterators into the set. -type Iterator struct { - // node is the node containing the iterated segment. If the iterator is - // terminal, node is nil. - node *node - - // index is the index of the segment in node.keys/values. - index int -} - -// Ok returns true if the iterator is not terminal. All other methods are only -// valid for non-terminal iterators. -func (seg Iterator) Ok() bool { - return seg.node != nil -} - -// Range returns the iterated segment's range key. -func (seg Iterator) Range() Range { - return seg.node.keys[seg.index] -} - -// Start is equivalent to Range().Start, but should be preferred if only the -// start of the range is needed. -func (seg Iterator) Start() uint64 { - return seg.node.keys[seg.index].Start -} - -// End is equivalent to Range().End, but should be preferred if only the end of -// the range is needed. -func (seg Iterator) End() uint64 { - return seg.node.keys[seg.index].End -} - -// SetRangeUnchecked mutates the iterated segment's range key. This operation -// does not invalidate any iterators. -// -// Preconditions: -// -// - r.Length() > 0. -// -// - The new range must not overlap an existing one: If seg.NextSegment().Ok(), -// then r.end <= seg.NextSegment().Start(); if seg.PrevSegment().Ok(), then -// r.start >= seg.PrevSegment().End(). -func (seg Iterator) SetRangeUnchecked(r Range) { - seg.node.keys[seg.index] = r -} - -// SetRange mutates the iterated segment's range key. If the new range would -// cause the iterated segment to overlap another segment, or if the new range -// is invalid, SetRange panics. This operation does not invalidate any -// iterators. -func (seg Iterator) SetRange(r Range) { - if r.Length() <= 0 { - panic(fmt.Sprintf("invalid segment range %v", r)) - } - if prev := seg.PrevSegment(); prev.Ok() && r.Start < prev.End() { - panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, prev.Range())) - } - if next := seg.NextSegment(); next.Ok() && r.End > next.Start() { - panic(fmt.Sprintf("new segment range %v overlaps segment range %v", r, next.Range())) - } - seg.SetRangeUnchecked(r) -} - -// SetStartUnchecked mutates the iterated segment's start. This operation does -// not invalidate any iterators. -// -// Preconditions: The new start must be valid: start < seg.End(); if -// seg.PrevSegment().Ok(), then start >= seg.PrevSegment().End(). -func (seg Iterator) SetStartUnchecked(start uint64) { - seg.node.keys[seg.index].Start = start -} - -// SetStart mutates the iterated segment's start. If the new start value would -// cause the iterated segment to overlap another segment, or would result in an -// invalid range, SetStart panics. This operation does not invalidate any -// iterators. -func (seg Iterator) SetStart(start uint64) { - if start >= seg.End() { - panic(fmt.Sprintf("new start %v would invalidate segment range %v", start, seg.Range())) - } - if prev := seg.PrevSegment(); prev.Ok() && start < prev.End() { - panic(fmt.Sprintf("new start %v would cause segment range %v to overlap segment range %v", start, seg.Range(), prev.Range())) - } - seg.SetStartUnchecked(start) -} - -// SetEndUnchecked mutates the iterated segment's end. This operation does not -// invalidate any iterators. -// -// Preconditions: The new end must be valid: end > seg.Start(); if -// seg.NextSegment().Ok(), then end <= seg.NextSegment().Start(). -func (seg Iterator) SetEndUnchecked(end uint64) { - seg.node.keys[seg.index].End = end -} - -// SetEnd mutates the iterated segment's end. If the new end value would cause -// the iterated segment to overlap another segment, or would result in an -// invalid range, SetEnd panics. This operation does not invalidate any -// iterators. -func (seg Iterator) SetEnd(end uint64) { - if end <= seg.Start() { - panic(fmt.Sprintf("new end %v would invalidate segment range %v", end, seg.Range())) - } - if next := seg.NextSegment(); next.Ok() && end > next.Start() { - panic(fmt.Sprintf("new end %v would cause segment range %v to overlap segment range %v", end, seg.Range(), next.Range())) - } - seg.SetEndUnchecked(end) -} - -// Value returns a copy of the iterated segment's value. -func (seg Iterator) Value() noValue { - return seg.node.values[seg.index] -} - -// ValuePtr returns a pointer to the iterated segment's value. The pointer is -// invalidated if the iterator is invalidated. This operation does not -// invalidate any iterators. -func (seg Iterator) ValuePtr() *noValue { - return &seg.node.values[seg.index] -} - -// SetValue mutates the iterated segment's value. This operation does not -// invalidate any iterators. -func (seg Iterator) SetValue(val noValue) { - seg.node.values[seg.index] = val -} - -// PrevSegment returns the iterated segment's predecessor. If there is no -// preceding segment, PrevSegment returns a terminal iterator. -func (seg Iterator) PrevSegment() Iterator { - if seg.node.hasChildren { - return seg.node.children[seg.index].lastSegment() - } - if seg.index > 0 { - return Iterator{seg.node, seg.index - 1} - } - if seg.node.parent == nil { - return Iterator{} - } - return segmentBeforePosition(seg.node.parent, seg.node.parentIndex) -} - -// NextSegment returns the iterated segment's successor. If there is no -// succeeding segment, NextSegment returns a terminal iterator. -func (seg Iterator) NextSegment() Iterator { - if seg.node.hasChildren { - return seg.node.children[seg.index+1].firstSegment() - } - if seg.index < seg.node.nrSegments-1 { - return Iterator{seg.node, seg.index + 1} - } - if seg.node.parent == nil { - return Iterator{} - } - return segmentAfterPosition(seg.node.parent, seg.node.parentIndex) -} - -// PrevGap returns the gap immediately before the iterated segment. -func (seg Iterator) PrevGap() GapIterator { - if seg.node.hasChildren { - - return seg.node.children[seg.index].lastSegment().NextGap() - } - return GapIterator{seg.node, seg.index} -} - -// NextGap returns the gap immediately after the iterated segment. -func (seg Iterator) NextGap() GapIterator { - if seg.node.hasChildren { - return seg.node.children[seg.index+1].firstSegment().PrevGap() - } - return GapIterator{seg.node, seg.index + 1} -} - -// PrevNonEmpty returns the iterated segment's predecessor if it is adjacent, -// or the gap before the iterated segment otherwise. If seg.Start() == -// Functions.MinKey(), PrevNonEmpty will return two terminal iterators. -// Otherwise, exactly one of the iterators returned by PrevNonEmpty will be -// non-terminal. -func (seg Iterator) PrevNonEmpty() (Iterator, GapIterator) { - gap := seg.PrevGap() - if gap.Range().Length() != 0 { - return Iterator{}, gap - } - return gap.PrevSegment(), GapIterator{} -} - -// NextNonEmpty returns the iterated segment's successor if it is adjacent, or -// the gap after the iterated segment otherwise. If seg.End() == -// Functions.MaxKey(), NextNonEmpty will return two terminal iterators. -// Otherwise, exactly one of the iterators returned by NextNonEmpty will be -// non-terminal. -func (seg Iterator) NextNonEmpty() (Iterator, GapIterator) { - gap := seg.NextGap() - if gap.Range().Length() != 0 { - return Iterator{}, gap - } - return gap.NextSegment(), GapIterator{} -} - -// A GapIterator is conceptually one of: -// -// - A pointer to a position between two segments, before the first segment, or -// after the last segment in a set, called a *gap*; or -// -// - A terminal iterator, which is a sentinel indicating that the end of -// iteration has been reached. -// -// Note that the gap between two adjacent segments exists (iterators to it are -// non-terminal), but has a length of zero. GapIterator.IsEmpty returns true -// for such gaps. An empty set contains a single gap, spanning the entire range -// of the set's keys. -// -// GapIterators are copyable values and are meaningfully equality-comparable. -// The zero value of GapIterator is a terminal iterator. -// -// Unless otherwise specified, any mutation of a set invalidates all existing -// iterators into the set. -type GapIterator struct { - // The representation of a GapIterator is identical to that of an Iterator, - // except that index corresponds to positions between segments in the same - // way as for node.children (see comment for node.nrSegments). - node *node - index int -} - -// Ok returns true if the iterator is not terminal. All other methods are only -// valid for non-terminal iterators. -func (gap GapIterator) Ok() bool { - return gap.node != nil -} - -// Range returns the range spanned by the iterated gap. -func (gap GapIterator) Range() Range { - return Range{gap.Start(), gap.End()} -} - -// Start is equivalent to Range().Start, but should be preferred if only the -// start of the range is needed. -func (gap GapIterator) Start() uint64 { - if ps := gap.PrevSegment(); ps.Ok() { - return ps.End() - } - return setFunctions{}.MinKey() -} - -// End is equivalent to Range().End, but should be preferred if only the end of -// the range is needed. -func (gap GapIterator) End() uint64 { - if ns := gap.NextSegment(); ns.Ok() { - return ns.Start() - } - return setFunctions{}.MaxKey() -} - -// IsEmpty returns true if the iterated gap is empty (that is, the "gap" is -// between two adjacent segments.) -func (gap GapIterator) IsEmpty() bool { - return gap.Range().Length() == 0 -} - -// PrevSegment returns the segment immediately before the iterated gap. If no -// such segment exists, PrevSegment returns a terminal iterator. -func (gap GapIterator) PrevSegment() Iterator { - return segmentBeforePosition(gap.node, gap.index) -} - -// NextSegment returns the segment immediately after the iterated gap. If no -// such segment exists, NextSegment returns a terminal iterator. -func (gap GapIterator) NextSegment() Iterator { - return segmentAfterPosition(gap.node, gap.index) -} - -// PrevGap returns the iterated gap's predecessor. If no such gap exists, -// PrevGap returns a terminal iterator. -func (gap GapIterator) PrevGap() GapIterator { - seg := gap.PrevSegment() - if !seg.Ok() { - return GapIterator{} - } - return seg.PrevGap() -} - -// NextGap returns the iterated gap's successor. If no such gap exists, NextGap -// returns a terminal iterator. -func (gap GapIterator) NextGap() GapIterator { - seg := gap.NextSegment() - if !seg.Ok() { - return GapIterator{} - } - return seg.NextGap() -} - -// segmentBeforePosition returns the predecessor segment of the position given -// by n.children[i], which may or may not contain a child. If no such segment -// exists, segmentBeforePosition returns a terminal iterator. -func segmentBeforePosition(n *node, i int) Iterator { - for i == 0 { - if n.parent == nil { - return Iterator{} - } - n, i = n.parent, n.parentIndex - } - return Iterator{n, i - 1} -} - -// segmentAfterPosition returns the successor segment of the position given by -// n.children[i], which may or may not contain a child. If no such segment -// exists, segmentAfterPosition returns a terminal iterator. -func segmentAfterPosition(n *node, i int) Iterator { - for i == n.nrSegments { - if n.parent == nil { - return Iterator{} - } - n, i = n.parent, n.parentIndex - } - return Iterator{n, i} -} - -func zeroValueSlice(slice []noValue) { - - for i := range slice { - setFunctions{}.ClearValue(&slice[i]) - } -} - -func zeroNodeSlice(slice []*node) { - for i := range slice { - slice[i] = nil - } -} - -// String stringifies a Set for debugging. -func (s *Set) String() string { - return s.root.String() -} - -// String stringifies a node (and all of its children) for debugging. -func (n *node) String() string { - var buf bytes.Buffer - n.writeDebugString(&buf, "") - return buf.String() -} - -func (n *node) writeDebugString(buf *bytes.Buffer, prefix string) { - if n.hasChildren != (n.nrSegments > 0 && n.children[0] != nil) { - buf.WriteString(prefix) - buf.WriteString(fmt.Sprintf("WARNING: inconsistent value of hasChildren: got %v, want %v\n", n.hasChildren, !n.hasChildren)) - } - for i := 0; i < n.nrSegments; i++ { - if child := n.children[i]; child != nil { - cprefix := fmt.Sprintf("%s- % 3d ", prefix, i) - if child.parent != n || child.parentIndex != i { - buf.WriteString(cprefix) - buf.WriteString(fmt.Sprintf("WARNING: inconsistent linkage to parent: got (%p, %d), want (%p, %d)\n", child.parent, child.parentIndex, n, i)) - } - child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, i)) - } - buf.WriteString(prefix) - buf.WriteString(fmt.Sprintf("- % 3d: %v => %v\n", i, n.keys[i], n.values[i])) - } - if child := n.children[n.nrSegments]; child != nil { - child.writeDebugString(buf, fmt.Sprintf("%s- % 3d ", prefix, n.nrSegments)) - } -} - -// SegmentDataSlices represents segments from a set as slices of start, end, and -// values. SegmentDataSlices is primarily used as an intermediate representation -// for save/restore and the layout here is optimized for that. -// -// +stateify savable -type SegmentDataSlices struct { - Start []uint64 - End []uint64 - Values []noValue -} - -// ExportSortedSlice returns a copy of all segments in the given set, in ascending -// key order. -func (s *Set) ExportSortedSlices() *SegmentDataSlices { - var sds SegmentDataSlices - for seg := s.FirstSegment(); seg.Ok(); seg = seg.NextSegment() { - sds.Start = append(sds.Start, seg.Start()) - sds.End = append(sds.End, seg.End()) - sds.Values = append(sds.Values, seg.Value()) - } - sds.Start = sds.Start[:len(sds.Start):len(sds.Start)] - sds.End = sds.End[:len(sds.End):len(sds.End)] - sds.Values = sds.Values[:len(sds.Values):len(sds.Values)] - return &sds -} - -// ImportSortedSlice initializes the given set from the given slice. -// -// Preconditions: s must be empty. sds must represent a valid set (the segments -// in sds must have valid lengths that do not overlap). The segments in sds -// must be sorted in ascending key order. -func (s *Set) ImportSortedSlices(sds *SegmentDataSlices) error { - if !s.IsEmpty() { - return fmt.Errorf("cannot import into non-empty set %v", s) - } - gap := s.FirstGap() - for i := range sds.Start { - r := Range{sds.Start[i], sds.End[i]} - if !gap.Range().IsSupersetOf(r) { - return fmt.Errorf("segment overlaps a preceding segment or is incorrectly sorted: [%d, %d) => %v", sds.Start[i], sds.End[i], sds.Values[i]) - } - gap = s.InsertWithoutMerging(gap, r, sds.Values[i]).NextGap() - } - return nil -} -func (s *Set) saveRoot() *SegmentDataSlices { - return s.ExportSortedSlices() -} - -func (s *Set) loadRoot(sds *SegmentDataSlices) { - if err := s.ImportSortedSlices(sds); err != nil { - panic(err) - } -} diff --git a/pkg/sentry/fs/binder/binder.go b/pkg/sentry/fs/binder/binder.go deleted file mode 100644 index eef54d787..000000000 --- a/pkg/sentry/fs/binder/binder.go +++ /dev/null @@ -1,260 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package binder implements Android Binder IPC module. -package binder - -import ( - "sync" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/pgalloc" - "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/waiter" -) - -const ( - currentProtocolVersion = 8 - - // mmapSizeLimit is the upper limit for mapped memory size in Binder. - mmapSizeLimit = 4 * 1024 * 1024 // 4MB -) - -// Device implements fs.InodeOperations. -// -// +stateify savable -type Device struct { - fsutil.InodeGenericChecker `state:"nosave"` - fsutil.InodeNoExtendedAttributes `state:"nosave"` - fsutil.InodeNoopAllocate `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 = (*Device)(nil) - -// NewDevice creates and initializes a Device structure. -func NewDevice(ctx context.Context, owner fs.FileOwner, fp fs.FilePermissions) *Device { - return &Device{ - InodeSimpleAttributes: fsutil.NewInodeSimpleAttributes(ctx, owner, fp, 0), - } -} - -// GetFile implements fs.InodeOperations.GetFile. -// -// TODO(b/30946773): Add functionality to GetFile: Additional fields will be -// needed in the Device structure, initialize them here. Also, Device will need -// to keep track of the created Procs in order to implement BINDER_READ_WRITE -// ioctl. -func (bd *Device) GetFile(ctx context.Context, d *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { - return fs.NewFile(ctx, d, flags, &Proc{ - bd: bd, - task: kernel.TaskFromContext(ctx), - mfp: pgalloc.MemoryFileProviderFromContext(ctx), - }), nil -} - -// Proc implements fs.FileOperations and fs.IoctlGetter. -// -// +stateify savable -type Proc struct { - fsutil.FileNoFsync `state:"nosave"` - fsutil.FileNoSplice `state:"nosave"` - fsutil.FileNotDirReaddir `state:"nosave"` - fsutil.FileUseInodeUnstableAttr `state:"nosave"` - waiter.AlwaysReady `state:"nosave"` - - bd *Device - task *kernel.Task - mfp pgalloc.MemoryFileProvider - - // mu protects fr. - mu sync.Mutex `state:"nosave"` - - // mapped is memory allocated from mfp.MemoryFile() by AddMapping. - mapped platform.FileRange -} - -// Release implements fs.FileOperations.Release. -func (bp *Proc) Release() { - bp.mu.Lock() - defer bp.mu.Unlock() - if bp.mapped.Length() != 0 { - bp.mfp.MemoryFile().DecRef(bp.mapped) - } -} - -// Seek implements fs.FileOperations.Seek. -// -// Binder doesn't support seek operation (unless in debug mode). -func (bp *Proc) Seek(ctx context.Context, file *fs.File, whence fs.SeekWhence, offset int64) (int64, error) { - return offset, syserror.EOPNOTSUPP -} - -// Read implements fs.FileOperations.Read. -// -// Binder doesn't support read operation (unless in debug mode). -func (bp *Proc) Read(ctx context.Context, file *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { - return 0, syserror.EOPNOTSUPP -} - -// Write implements fs.FileOperations.Write. -// -// Binder doesn't support write operation. -func (bp *Proc) Write(ctx context.Context, file *fs.File, src usermem.IOSequence, offset int64) (int64, error) { - return 0, syserror.EOPNOTSUPP -} - -// Flush implements fs.FileOperations.Flush. -// -// TODO(b/30946773): Implement. -func (bp *Proc) Flush(ctx context.Context, file *fs.File) error { - return nil -} - -// ConfigureMMap implements fs.FileOperations.ConfigureMMap. -func (bp *Proc) ConfigureMMap(ctx context.Context, file *fs.File, opts *memmap.MMapOpts) error { - // Compare drivers/android/binder.c:binder_mmap(). - if caller := kernel.TaskFromContext(ctx); caller != bp.task { - return syserror.EINVAL - } - if opts.Length > mmapSizeLimit { - opts.Length = mmapSizeLimit - } - opts.MaxPerms.Write = false - - // TODO(b/30946773): Binder sets VM_DONTCOPY, preventing the created vma - // from being copied across fork(), but we don't support this yet. As - // a result, MMs containing a Binder mapping cannot be forked (MM.Fork will - // fail when AddMapping returns EBUSY). - - return fsutil.GenericConfigureMMap(file, bp, opts) -} - -// Ioctl implements fs.FileOperations.Ioctl. -// -// TODO(b/30946773): Implement. -func (bp *Proc) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { - // Switch on ioctl request. - switch uint32(args[1].Int()) { - case linux.BinderVersionIoctl: - ver := &linux.BinderVersion{ - ProtocolVersion: currentProtocolVersion, - } - // Copy result to user-space. - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), ver, usermem.IOOpts{ - AddressSpaceActive: true, - }) - return 0, err - case linux.BinderWriteReadIoctl: - // TODO(b/30946773): Implement. - fallthrough - case linux.BinderSetIdleTimeoutIoctl: - // TODO(b/30946773): Implement. - fallthrough - case linux.BinderSetMaxThreadsIoctl: - // TODO(b/30946773): Implement. - fallthrough - case linux.BinderSetIdlePriorityIoctl: - // TODO(b/30946773): Implement. - fallthrough - case linux.BinderSetContextMgrIoctl: - // TODO(b/30946773): Implement. - fallthrough - case linux.BinderThreadExitIoctl: - // TODO(b/30946773): Implement. - return 0, syserror.ENOSYS - default: - // Ioctls irrelevant to Binder. - return 0, syserror.EINVAL - } -} - -// AddMapping implements memmap.Mappable.AddMapping. -func (bp *Proc) AddMapping(ctx context.Context, ms memmap.MappingSpace, ar usermem.AddrRange, offset uint64, _ bool) error { - bp.mu.Lock() - defer bp.mu.Unlock() - if bp.mapped.Length() != 0 { - // mmap has been called before, which binder_mmap() doesn't like. - return syserror.EBUSY - } - // Binder only allocates and maps a single page up-front - // (drivers/android/binder.c:binder_mmap() => binder_update_page_range()). - fr, err := bp.mfp.MemoryFile().Allocate(usermem.PageSize, usage.Anonymous) - if err != nil { - return err - } - bp.mapped = fr - return nil -} - -// RemoveMapping implements memmap.Mappable.RemoveMapping. -func (*Proc) RemoveMapping(context.Context, memmap.MappingSpace, usermem.AddrRange, uint64, bool) { - // Nothing to do. Notably, we don't free bp.mapped to allow another mmap. -} - -// CopyMapping implements memmap.Mappable.CopyMapping. -func (bp *Proc) CopyMapping(ctx context.Context, ms memmap.MappingSpace, srcAR, dstAR usermem.AddrRange, offset uint64, _ bool) error { - // Nothing to do. Notably, this is one case where CopyMapping isn't - // equivalent to AddMapping, as AddMapping would return EBUSY. - return nil -} - -// Translate implements memmap.Mappable.Translate. -func (bp *Proc) Translate(ctx context.Context, required, optional memmap.MappableRange, at usermem.AccessType) ([]memmap.Translation, error) { - // TODO(b/30946773): In addition to the page initially allocated and mapped - // in AddMapping (Linux: binder_mmap), Binder allocates and maps pages for - // each transaction (Linux: binder_ioctl => binder_ioctl_write_read => - // binder_thread_write => binder_transaction => binder_alloc_buf => - // binder_update_page_range). Since we don't actually implement - // BinderWriteReadIoctl (Linux: BINDER_WRITE_READ), we only ever have the - // first page. - var err error - if required.End > usermem.PageSize { - err = &memmap.BusError{syserror.EFAULT} - } - if required.Start == 0 { - return []memmap.Translation{ - { - Source: memmap.MappableRange{0, usermem.PageSize}, - File: bp.mfp.MemoryFile(), - Offset: bp.mapped.Start, - Perms: usermem.AnyAccess, - }, - }, err - } - return nil, err -} - -// InvalidateUnsavable implements memmap.Mappable.InvalidateUnsavable. -func (bp *Proc) InvalidateUnsavable(ctx context.Context) error { - return nil -} diff --git a/pkg/sentry/fs/binder/binder_state_autogen.go b/pkg/sentry/fs/binder/binder_state_autogen.go deleted file mode 100755 index 1f321e3b6..000000000 --- a/pkg/sentry/fs/binder/binder_state_autogen.go +++ /dev/null @@ -1,40 +0,0 @@ -// automatically generated by stateify. - -package binder - -import ( - "gvisor.dev/gvisor/pkg/state" -) - -func (x *Device) beforeSave() {} -func (x *Device) save(m state.Map) { - x.beforeSave() - m.Save("InodeSimpleAttributes", &x.InodeSimpleAttributes) -} - -func (x *Device) afterLoad() {} -func (x *Device) load(m state.Map) { - m.Load("InodeSimpleAttributes", &x.InodeSimpleAttributes) -} - -func (x *Proc) beforeSave() {} -func (x *Proc) save(m state.Map) { - x.beforeSave() - m.Save("bd", &x.bd) - m.Save("task", &x.task) - m.Save("mfp", &x.mfp) - m.Save("mapped", &x.mapped) -} - -func (x *Proc) afterLoad() {} -func (x *Proc) load(m state.Map) { - m.Load("bd", &x.bd) - m.Load("task", &x.task) - m.Load("mfp", &x.mfp) - m.Load("mapped", &x.mapped) -} - -func init() { - state.Register("binder.Device", (*Device)(nil), state.Fns{Save: (*Device).save, Load: (*Device).load}) - state.Register("binder.Proc", (*Proc)(nil), state.Fns{Save: (*Proc).save, Load: (*Proc).load}) -} diff --git a/pkg/sentry/fs/dev/dev.go b/pkg/sentry/fs/dev/dev.go index fb6c30ff0..d4bbd9807 100644 --- a/pkg/sentry/fs/dev/dev.go +++ b/pkg/sentry/fs/dev/dev.go @@ -20,8 +20,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/ashmem" - "gvisor.dev/gvisor/pkg/sentry/fs/binder" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" "gvisor.dev/gvisor/pkg/sentry/fs/tmpfs" "gvisor.dev/gvisor/pkg/sentry/usermem" @@ -81,7 +79,7 @@ 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 { +func New(ctx context.Context, msrc *fs.MountSource) *fs.Inode { contents := map[string]*fs.Inode{ "fd": newSymlink(ctx, "/proc/self/fd", msrc), "stdin": newSymlink(ctx, "/proc/self/fd/0", msrc), @@ -118,16 +116,6 @@ func New(ctx context.Context, msrc *fs.MountSource, binderEnabled bool, ashmemEn "ptmx": newSymlink(ctx, "pts/ptmx", msrc), } - if binderEnabled { - binder := binder.NewDevice(ctx, fs.RootOwner, fs.FilePermsFromMode(0666)) - contents["binder"] = newCharacterDevice(ctx, binder, msrc) - } - - if ashmemEnabled { - ashmem := ashmem.NewDevice(ctx, fs.RootOwner, fs.FilePermsFromMode(0666)) - contents["ashmem"] = newCharacterDevice(ctx, ashmem, msrc) - } - iops := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) return fs.NewInode(ctx, iops, msrc, fs.StableAttr{ DeviceID: devDevice.DeviceID(), diff --git a/pkg/sentry/fs/dev/fs.go b/pkg/sentry/fs/dev/fs.go index cbc2c2f9b..55f8af704 100644 --- a/pkg/sentry/fs/dev/fs.go +++ b/pkg/sentry/fs/dev/fs.go @@ -15,19 +15,10 @@ package dev import ( - "strconv" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/syserror" ) -// Optional key containing boolean flag which specifies if Android Binder IPC should be enabled. -const binderEnabledKey = "binder_enabled" - -// Optional key containing boolean flag which specifies if Android ashmem should be enabled. -const ashmemEnabledKey = "ashmem_enabled" - // filesystem is a devtmpfs. // // +stateify savable @@ -67,33 +58,7 @@ func (*filesystem) Flags() fs.FilesystemFlags { // Mount returns a devtmpfs root that can be positioned in the vfs. func (f *filesystem) Mount(ctx context.Context, device string, flags fs.MountSourceFlags, data string, _ interface{}) (*fs.Inode, error) { - // device is always ignored. // devtmpfs backed by ramfs ignores bad options. See fs/ramfs/inode.c:ramfs_parse_options. // -> we should consider parsing the mode and backing devtmpfs by this. - - // Parse generic comma-separated key=value options. - options := fs.GenericMountSourceOptions(data) - - // binerEnabledKey is optional and binder is disabled by default. - binderEnabled := false - if beStr, exists := options[binderEnabledKey]; exists { - var err error - binderEnabled, err = strconv.ParseBool(beStr) - if err != nil { - return nil, syserror.EINVAL - } - } - - // ashmemEnabledKey is optional and ashmem is disabled by default. - ashmemEnabled := false - if aeStr, exists := options[ashmemEnabledKey]; exists { - var err error - ashmemEnabled, err = strconv.ParseBool(aeStr) - if err != nil { - return nil, syserror.EINVAL - } - } - - // Construct the devtmpfs root. - return New(ctx, fs.NewNonCachingMountSource(ctx, f, flags), binderEnabled, ashmemEnabled), nil + return New(ctx, fs.NewNonCachingMountSource(ctx, f, flags)), nil } diff --git a/pkg/sentry/fs/dirent.go b/pkg/sentry/fs/dirent.go index 28651e58b..fbca06761 100644 --- a/pkg/sentry/fs/dirent.go +++ b/pkg/sentry/fs/dirent.go @@ -229,11 +229,13 @@ func newDirent(inode *Inode, name string) *Dirent { if inode != nil { inode.MountSource.IncDirentRefs() } - return &Dirent{ + d := Dirent{ Inode: inode, name: name, children: make(map[string]*refs.WeakRef), } + d.EnableLeakCheck("fs.Dirent") + return &d } // NewNegativeDirent returns a new root negative Dirent. Otherwise same as NewDirent. diff --git a/pkg/sentry/fs/file.go b/pkg/sentry/fs/file.go index 8e1f5674d..bb8117f89 100644 --- a/pkg/sentry/fs/file.go +++ b/pkg/sentry/fs/file.go @@ -130,14 +130,15 @@ type File struct { // to false respectively. func NewFile(ctx context.Context, dirent *Dirent, flags FileFlags, fops FileOperations) *File { dirent.IncRef() - f := &File{ + f := File{ UniqueID: uniqueid.GlobalFromContext(ctx), Dirent: dirent, FileOperations: fops, flags: flags, } f.mu.Init() - return f + f.EnableLeakCheck("fs.File") + return &f } // DecRef destroys the File when it is no longer referenced. diff --git a/pkg/sentry/fs/file_overlay.go b/pkg/sentry/fs/file_overlay.go index c6cbd5631..9820f0b13 100644 --- a/pkg/sentry/fs/file_overlay.go +++ b/pkg/sentry/fs/file_overlay.go @@ -347,13 +347,14 @@ func (*overlayFileOperations) ConfigureMMap(ctx context.Context, file *File, opt // preventing us from saving a proper inode mapping for the // file. file.IncRef() - id := &overlayMappingIdentity{ + id := overlayMappingIdentity{ id: opts.MappingIdentity, overlayFile: file, } + id.EnableLeakCheck("fs.overlayMappingIdentity") // Swap out the old MappingIdentity for the wrapped one. - opts.MappingIdentity = id + opts.MappingIdentity = &id return nil } diff --git a/pkg/sentry/fs/gofer/handles.go b/pkg/sentry/fs/gofer/handles.go index b87c4f150..27eeae3d9 100644 --- a/pkg/sentry/fs/gofer/handles.go +++ b/pkg/sentry/fs/gofer/handles.go @@ -79,11 +79,12 @@ func newHandles(ctx context.Context, file contextFile, flags fs.FileFlags) (*han newFile.close(ctx) return nil, err } - h := &handles{ + h := handles{ File: newFile, Host: hostFile, } - return h, nil + h.EnableLeakCheck("gofer.handles") + return &h, nil } type handleReadWriter struct { diff --git a/pkg/sentry/fs/gofer/path.go b/pkg/sentry/fs/gofer/path.go index b91386909..8c17603f8 100644 --- a/pkg/sentry/fs/gofer/path.go +++ b/pkg/sentry/fs/gofer/path.go @@ -145,16 +145,17 @@ func (i *inodeOperations) Create(ctx context.Context, dir *fs.Inode, name string defer d.DecRef() // Construct the new file, caching the handles if allowed. - h := &handles{ + h := handles{ File: newFile, Host: hostFile, } + h.EnableLeakCheck("gofer.handles") if iops.fileState.canShareHandles() { iops.fileState.handlesMu.Lock() - iops.fileState.setSharedHandlesLocked(flags, h) + iops.fileState.setSharedHandlesLocked(flags, &h) iops.fileState.handlesMu.Unlock() } - return NewFile(ctx, d, name, flags, iops, h), nil + return NewFile(ctx, d, name, flags, iops, &h), nil } // CreateLink uses Create to create a symlink between oldname and newname. diff --git a/pkg/sentry/fs/gofer/session.go b/pkg/sentry/fs/gofer/session.go index 9f7660ed1..69d08a627 100644 --- a/pkg/sentry/fs/gofer/session.go +++ b/pkg/sentry/fs/gofer/session.go @@ -241,7 +241,7 @@ func Root(ctx context.Context, dev string, filesystem fs.Filesystem, superBlockF } // Construct the session. - s := &session{ + s := session{ connID: dev, msize: o.msize, version: o.version, @@ -250,13 +250,14 @@ func Root(ctx context.Context, dev string, filesystem fs.Filesystem, superBlockF superBlockFlags: superBlockFlags, mounter: mounter, } + s.EnableLeakCheck("gofer.session") if o.privateunixsocket { s.endpoints = newEndpointMaps() } // Construct the MountSource with the session and superBlockFlags. - m := fs.NewMountSource(ctx, s, filesystem, superBlockFlags) + m := fs.NewMountSource(ctx, &s, filesystem, superBlockFlags) // Given that gofer files can consume host FDs, restrict the number // of files that can be held by the cache. @@ -290,7 +291,7 @@ func Root(ctx context.Context, dev string, filesystem fs.Filesystem, superBlockF return nil, err } - sattr, iops := newInodeOperations(ctx, s, s.attach, qid, valid, attr, false) + sattr, iops := newInodeOperations(ctx, &s, s.attach, qid, valid, attr, false) return fs.NewInode(ctx, iops, m, sattr), nil } diff --git a/pkg/sentry/fs/gofer/session_state.go b/pkg/sentry/fs/gofer/session_state.go index 29a79441e..d045e04ff 100644 --- a/pkg/sentry/fs/gofer/session_state.go +++ b/pkg/sentry/fs/gofer/session_state.go @@ -111,5 +111,4 @@ func (s *session) afterLoad() { panic("failed to restore endpoint maps: " + err.Error()) } } - } diff --git a/pkg/sentry/fs/host/host_state_autogen.go b/pkg/sentry/fs/host/host_state_autogen.go index 9611da42a..f0e1c4b88 100755 --- a/pkg/sentry/fs/host/host_state_autogen.go +++ b/pkg/sentry/fs/host/host_state_autogen.go @@ -95,17 +95,17 @@ func (x *inodeFileState) load(m state.Map) { func (x *ConnectedEndpoint) save(m state.Map) { x.beforeSave() + m.Save("ref", &x.ref) m.Save("queue", &x.queue) m.Save("path", &x.path) - m.Save("ref", &x.ref) m.Save("srfd", &x.srfd) m.Save("stype", &x.stype) } func (x *ConnectedEndpoint) load(m state.Map) { + m.Load("ref", &x.ref) m.Load("queue", &x.queue) m.Load("path", &x.path) - m.Load("ref", &x.ref) m.LoadWait("srfd", &x.srfd) m.Load("stype", &x.stype) m.AfterLoad(x.afterLoad) diff --git a/pkg/sentry/fs/host/socket.go b/pkg/sentry/fs/host/socket.go index 7fedc88bc..44c4ee5f2 100644 --- a/pkg/sentry/fs/host/socket.go +++ b/pkg/sentry/fs/host/socket.go @@ -47,12 +47,12 @@ const maxSendBufferSize = 8 << 20 // // +stateify savable type ConnectedEndpoint struct { - queue *waiter.Queue - path string - // ref keeps track of references to a connectedEndpoint. ref refs.AtomicRefCount + queue *waiter.Queue + path string + // If srfd >= 0, it is the host FD that file was imported from. srfd int `state:"wait"` @@ -133,6 +133,8 @@ func NewConnectedEndpoint(ctx context.Context, file *fd.FD, queue *waiter.Queue, // AtomicRefCounters start off with a single reference. We need two. e.ref.IncRef() + e.ref.EnableLeakCheck("host.ConnectedEndpoint") + return &e, nil } diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go index e4aae1135..f4ddfa406 100644 --- a/pkg/sentry/fs/inode.go +++ b/pkg/sentry/fs/inode.go @@ -86,12 +86,14 @@ type LockCtx struct { // NewInode takes a reference on msrc. func NewInode(ctx context.Context, iops InodeOperations, msrc *MountSource, sattr StableAttr) *Inode { msrc.IncRef() - return &Inode{ + i := Inode{ InodeOperations: iops, StableAttr: sattr, Watches: newWatches(), MountSource: msrc, } + i.EnableLeakCheck("fs.Inode") + return &i } // DecRef drops a reference on the Inode. diff --git a/pkg/sentry/fs/inode_overlay.go b/pkg/sentry/fs/inode_overlay.go index b247fa514..24b769cfc 100644 --- a/pkg/sentry/fs/inode_overlay.go +++ b/pkg/sentry/fs/inode_overlay.go @@ -567,6 +567,12 @@ func overlayCheck(ctx context.Context, o *overlayEntry, p PermMask) error { if o.upper != nil { err = o.upper.check(ctx, p) } else { + if p.Write { + // Since writes will be redirected to the upper filesystem, the lower + // filesystem need not be writable, but must be readable for copy-up. + p.Write = false + p.Read = true + } err = o.lower.check(ctx, p) } o.copyMu.RUnlock() diff --git a/pkg/sentry/fs/mount.go b/pkg/sentry/fs/mount.go index 912495528..7a9692800 100644 --- a/pkg/sentry/fs/mount.go +++ b/pkg/sentry/fs/mount.go @@ -138,12 +138,14 @@ func NewMountSource(ctx context.Context, mops MountSourceOperations, filesystem if filesystem != nil { fsType = filesystem.Name() } - return &MountSource{ + msrc := MountSource{ MountSourceOperations: mops, Flags: flags, FilesystemType: fsType, fscache: NewDirentCache(DefaultDirentCacheSize), } + msrc.EnableLeakCheck("fs.MountSource") + return &msrc } // DirentRefs returns the current mount direntRefs. diff --git a/pkg/sentry/fs/mounts.go b/pkg/sentry/fs/mounts.go index 281364dfc..693ffc760 100644 --- a/pkg/sentry/fs/mounts.go +++ b/pkg/sentry/fs/mounts.go @@ -181,12 +181,14 @@ func NewMountNamespace(ctx context.Context, root *Inode) (*MountNamespace, error d: newRootMount(1, d), } - return &MountNamespace{ + mns := MountNamespace{ userns: creds.UserNamespace, root: d, mounts: mnts, mountID: 2, - }, nil + } + mns.EnableLeakCheck("fs.MountNamespace") + return &mns, nil } // UserNamespace returns the user namespace associated with this mount manager. @@ -661,6 +663,11 @@ func (mns *MountNamespace) ResolveExecutablePath(ctx context.Context, wd, name s } defer d.DecRef() + // Check that it is a regular file. + if !IsRegular(d.Inode.StableAttr) { + continue + } + // Check whether we can read and execute the found file. if err := d.Inode.CheckPermission(ctx, PermMask{Read: true, Execute: true}); err != nil { log.Infof("Found executable at %q, but user cannot execute it: %v", binPath, err) diff --git a/pkg/sentry/fs/proc/fds.go b/pkg/sentry/fs/proc/fds.go index ea7aded9a..bee421d76 100644 --- a/pkg/sentry/fs/proc/fds.go +++ b/pkg/sentry/fs/proc/fds.go @@ -25,7 +25,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/proc/device" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" "gvisor.dev/gvisor/pkg/syserror" ) @@ -42,8 +41,8 @@ func walkDescriptors(t *kernel.Task, p string, toInode func(*fs.File, kernel.FDF var file *fs.File var fdFlags kernel.FDFlags t.WithMuLocked(func(t *kernel.Task) { - if fdm := t.FDMap(); fdm != nil { - file, fdFlags = fdm.GetDescriptor(kdefs.FD(n)) + if fdTable := t.FDTable(); fdTable != nil { + file, fdFlags = fdTable.Get(int32(n)) } }) if file == nil { @@ -56,36 +55,31 @@ func walkDescriptors(t *kernel.Task, p string, toInode func(*fs.File, kernel.FDF // 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 int64, toDentAttr func(int) fs.DentAttr) (int64, error) { - var fds kernel.FDs + var fds []int32 t.WithMuLocked(func(t *kernel.Task) { - if fdm := t.FDMap(); fdm != nil { - fds = fdm.GetFDs() + if fdTable := t.FDTable(); fdTable != nil { + fds = fdTable.GetFDs() } }) - fdInts := make([]int, 0, len(fds)) - for _, fd := range fds { - fdInts = append(fdInts, int(fd)) - } - - // Find the fd to start at. - idx := sort.SearchInts(fdInts, int(offset)) - if idx == len(fdInts) { + // Find the appropriate starting point. + idx := sort.Search(len(fds), func(i int) bool { return fds[i] >= int32(offset) }) + if idx == len(fds) { return offset, nil } - fdInts = fdInts[idx:] + fds = fds[idx:] - var fd int - for _, fd = range fdInts { + // Serialize all FDs. + for _, fd := range fds { name := strconv.FormatUint(uint64(fd), 10) - if err := c.DirEmit(name, toDentAttr(fd)); err != nil { + if err := c.DirEmit(name, toDentAttr(int(fd))); err != nil { // Returned offset is the next fd to serialize. return int64(fd), err } } // We serialized them all. Next offset should be higher than last // serialized fd. - return int64(fd + 1), nil + return int64(fds[len(fds)-1] + 1), nil } // fd implements fs.InodeOperations for a file in /proc/TID/fd/. @@ -154,9 +148,9 @@ func (f *fd) Close() error { type fdDir struct { ramfs.Dir - // We hold a reference on the task's fdmap but only keep an indirect - // task pointer to avoid Dirent loading circularity caused by fdmap's - // potential back pointers into the dirent tree. + // We hold a reference on the task's FDTable but only keep an indirect + // task pointer to avoid Dirent loading circularity caused by the + // table's back pointers into the dirent tree. t *kernel.Task } diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index b2e36aeee..ef0ca3301 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -580,8 +580,8 @@ func (s *statusData) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ( var fds int var vss, rss, data uint64 s.t.WithMuLocked(func(t *kernel.Task) { - if fdm := t.FDMap(); fdm != nil { - fds = fdm.Size() + if fdTable := t.FDTable(); fdTable != nil { + fds = fdTable.Size() } if mm := t.MemoryManager(); mm != nil { vss = mm.VirtualMemorySize() diff --git a/pkg/sentry/fs/tty/terminal.go b/pkg/sentry/fs/tty/terminal.go index 8290f2530..b7cecb2ed 100644 --- a/pkg/sentry/fs/tty/terminal.go +++ b/pkg/sentry/fs/tty/terminal.go @@ -38,9 +38,11 @@ type Terminal struct { func newTerminal(ctx context.Context, d *dirInodeOperations, n uint32) *Terminal { termios := linux.DefaultSlaveTermios - return &Terminal{ + t := Terminal{ d: d, n: n, ld: newLineDiscipline(termios), } + t.EnableLeakCheck("tty.Terminal") + return &t } diff --git a/pkg/sentry/kernel/auth/atomicptr_credentials.go b/pkg/sentry/kernel/auth/atomicptr_credentials_unsafe.go index 4535c958f..4535c958f 100755 --- a/pkg/sentry/kernel/auth/atomicptr_credentials.go +++ b/pkg/sentry/kernel/auth/atomicptr_credentials_unsafe.go diff --git a/pkg/sentry/kernel/epoll/epoll.go b/pkg/sentry/kernel/epoll/epoll.go index 33c7dccae..9c0a4e1b4 100644 --- a/pkg/sentry/kernel/epoll/epoll.go +++ b/pkg/sentry/kernel/epoll/epoll.go @@ -26,7 +26,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/anon" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/waiter" ) @@ -61,7 +60,7 @@ const ( // +stateify savable type FileIdentifier struct { File *fs.File `state:"wait"` - Fd kdefs.FD + Fd int32 } // pollEntry holds all the state associated with an event poll entry, that is, diff --git a/pkg/sentry/kernel/fd_map.go b/pkg/sentry/kernel/fd_map.go deleted file mode 100644 index 786936a7d..000000000 --- a/pkg/sentry/kernel/fd_map.go +++ /dev/null @@ -1,364 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package kernel - -import ( - "bytes" - "fmt" - "sort" - "sync" - "sync/atomic" - "syscall" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/lock" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" - "gvisor.dev/gvisor/pkg/sentry/limits" -) - -// FDs is an ordering of FD's that can be made stable. -type FDs []kdefs.FD - -func (f FDs) Len() int { - return len(f) -} - -func (f FDs) Swap(i, j int) { - f[i], f[j] = f[j], f[i] -} - -func (f FDs) Less(i, j int) bool { - return f[i] < f[j] -} - -// FDFlags define flags for an individual descriptor. -// -// +stateify savable -type FDFlags struct { - // CloseOnExec indicates the descriptor should be closed on exec. - CloseOnExec bool -} - -// ToLinuxFileFlags converts a kernel.FDFlags object to a Linux file flags -// representation. -func (f FDFlags) ToLinuxFileFlags() (mask uint) { - if f.CloseOnExec { - mask |= linux.O_CLOEXEC - } - return -} - -// ToLinuxFDFlags converts a kernel.FDFlags object to a Linux descriptor flags -// representation. -func (f FDFlags) ToLinuxFDFlags() (mask uint) { - if f.CloseOnExec { - mask |= linux.FD_CLOEXEC - } - return -} - -// descriptor holds the details about a file descriptor, namely a pointer the -// file itself and the descriptor flags. -// -// +stateify savable -type descriptor struct { - file *fs.File - flags FDFlags -} - -// FDMap is used to manage File references and flags. -// -// +stateify savable -type FDMap struct { - refs.AtomicRefCount - k *Kernel - files map[kdefs.FD]descriptor - mu sync.RWMutex `state:"nosave"` - uid uint64 -} - -// ID returns a unique identifier for this FDMap. -func (f *FDMap) ID() uint64 { - return f.uid -} - -// NewFDMap allocates a new FDMap that may be used by tasks in k. -func (k *Kernel) NewFDMap() *FDMap { - return &FDMap{ - k: k, - files: make(map[kdefs.FD]descriptor), - uid: atomic.AddUint64(&k.fdMapUids, 1), - } -} - -// destroy removes all of the file descriptors from the map. -func (f *FDMap) destroy() { - f.RemoveIf(func(*fs.File, FDFlags) bool { - return true - }) -} - -// DecRef implements RefCounter.DecRef with destructor f.destroy. -func (f *FDMap) DecRef() { - f.DecRefWithDestructor(f.destroy) -} - -// Size returns the number of file descriptor slots currently allocated. -func (f *FDMap) Size() int { - f.mu.RLock() - defer f.mu.RUnlock() - - return len(f.files) -} - -// String is a stringer for FDMap. -func (f *FDMap) String() string { - f.mu.RLock() - defer f.mu.RUnlock() - - var b bytes.Buffer - for k, v := range f.files { - n, _ := v.file.Dirent.FullName(nil /* root */) - b.WriteString(fmt.Sprintf("\tfd:%d => name %s\n", k, n)) - } - return b.String() -} - -// NewFDFrom allocates a new FD guaranteed to be the lowest number available -// greater than or equal to from. This property is important as Unix programs -// tend to count on this allocation order. -func (f *FDMap) NewFDFrom(fd kdefs.FD, file *fs.File, flags FDFlags, limitSet *limits.LimitSet) (kdefs.FD, error) { - if fd < 0 { - // Don't accept negative FDs. - return 0, syscall.EINVAL - } - - f.mu.Lock() - defer f.mu.Unlock() - - // Finds the lowest fd not in the handles map. - lim := limitSet.Get(limits.NumberOfFiles) - for i := fd; lim.Cur == limits.Infinity || i < kdefs.FD(lim.Cur); i++ { - if _, ok := f.files[i]; !ok { - file.IncRef() - f.files[i] = descriptor{file, flags} - return i, nil - } - } - - return -1, syscall.EMFILE -} - -// NewFDAt sets the file reference for the given FD. If there is an -// active reference for that FD, the ref count for that existing reference -// is decremented. -func (f *FDMap) NewFDAt(fd kdefs.FD, file *fs.File, flags FDFlags, limitSet *limits.LimitSet) error { - if fd < 0 { - // Don't accept negative FDs. - return syscall.EBADF - } - - // In this one case we do not do a defer of the Unlock. The - // reason is that we must have done all the work needed for - // discarding any old open file before we return to the - // caller. In other words, the DecRef(), below, must have - // completed by the time we return to the caller to ensure - // side effects are, in fact, effected. A classic example is - // dup2(fd1, fd2); if fd2 was already open, it must be closed, - // and we don't want to resume the caller until it is; we have - // to block on the DecRef(). Hence we can not just do a 'go - // oldfile.DecRef()', since there would be no guarantee that - // it would be done before we the caller resumed. Since we - // must wait for the DecRef() to finish, and that could take - // time, it's best to first call f.muUnlock beore so we are - // not blocking other uses of this FDMap on the DecRef() call. - f.mu.Lock() - oldDesc, oldExists := f.files[fd] - lim := limitSet.Get(limits.NumberOfFiles).Cur - // if we're closing one then the effective limit is one - // more than the actual limit. - if oldExists && lim != limits.Infinity { - lim++ - } - if lim != limits.Infinity && fd >= kdefs.FD(lim) { - f.mu.Unlock() - return syscall.EMFILE - } - - file.IncRef() - f.files[fd] = descriptor{file, flags} - f.mu.Unlock() - - if oldExists { - oldDesc.file.DecRef() - } - return nil -} - -// SetFlags sets the flags for the given file descriptor, if it is valid. -func (f *FDMap) SetFlags(fd kdefs.FD, flags FDFlags) { - f.mu.Lock() - defer f.mu.Unlock() - - desc, ok := f.files[fd] - if !ok { - return - } - - f.files[fd] = descriptor{desc.file, flags} -} - -// GetDescriptor returns a reference to the file and the flags for the FD. It -// bumps its reference count as well. It returns nil if there is no File -// for the FD, i.e. if the FD is invalid. The caller must use DecRef -// when they are done. -func (f *FDMap) GetDescriptor(fd kdefs.FD) (*fs.File, FDFlags) { - f.mu.RLock() - defer f.mu.RUnlock() - - if desc, ok := f.files[fd]; ok { - desc.file.IncRef() - return desc.file, desc.flags - } - return nil, FDFlags{} -} - -// GetFile returns a reference to the File for the FD and bumps -// its reference count as well. It returns nil if there is no File -// for the FD, i.e. if the FD is invalid. The caller must use DecRef -// when they are done. -func (f *FDMap) GetFile(fd kdefs.FD) *fs.File { - f.mu.RLock() - if desc, ok := f.files[fd]; ok { - desc.file.IncRef() - f.mu.RUnlock() - return desc.file - } - f.mu.RUnlock() - return nil -} - -// fds returns an ordering of FDs. -func (f *FDMap) fds() FDs { - fds := make(FDs, 0, len(f.files)) - for fd := range f.files { - fds = append(fds, fd) - } - sort.Sort(fds) - return fds -} - -// GetFDs returns a list of valid fds. -func (f *FDMap) GetFDs() FDs { - f.mu.RLock() - defer f.mu.RUnlock() - return f.fds() -} - -// GetRefs returns a stable slice of references to all files and bumps the -// reference count on each. The caller must use DecRef on each reference when -// they're done using the slice. -func (f *FDMap) GetRefs() []*fs.File { - f.mu.RLock() - defer f.mu.RUnlock() - - fds := f.fds() - fs := make([]*fs.File, 0, len(fds)) - for _, fd := range fds { - desc := f.files[fd] - desc.file.IncRef() - fs = append(fs, desc.file) - } - return fs -} - -// Fork returns an independent FDMap pointing to the same descriptors. -func (f *FDMap) Fork() *FDMap { - f.mu.RLock() - defer f.mu.RUnlock() - - clone := f.k.NewFDMap() - - // Grab a extra reference for every file. - for fd, desc := range f.files { - desc.file.IncRef() - clone.files[fd] = desc - } - - // That's it! - return clone -} - -// unlock releases all file locks held by this FDMap's uid. Must only be -// called on a non-nil *fs.File. -func (f *FDMap) unlock(file *fs.File) { - id := lock.UniqueID(f.ID()) - file.Dirent.Inode.LockCtx.Posix.UnlockRegion(id, lock.LockRange{0, lock.LockEOF}) -} - -// inotifyFileClose generates the appropriate inotify events for f being closed. -func inotifyFileClose(f *fs.File) { - var ev uint32 - d := f.Dirent - - if fs.IsDir(d.Inode.StableAttr) { - ev |= linux.IN_ISDIR - } - - if f.Flags().Write { - ev |= linux.IN_CLOSE_WRITE - } else { - ev |= linux.IN_CLOSE_NOWRITE - } - - d.InotifyEvent(ev, 0) -} - -// Remove removes an FD from the FDMap, and returns (File, true) if a File -// one was found. Callers are expected to decrement the reference count on -// the File. Otherwise returns (nil, false). -func (f *FDMap) Remove(fd kdefs.FD) (*fs.File, bool) { - f.mu.Lock() - desc := f.files[fd] - delete(f.files, fd) - f.mu.Unlock() - if desc.file != nil { - f.unlock(desc.file) - inotifyFileClose(desc.file) - return desc.file, true - } - return nil, false -} - -// RemoveIf removes all FDs where cond is true. -func (f *FDMap) RemoveIf(cond func(*fs.File, FDFlags) bool) { - var removed []*fs.File - f.mu.Lock() - for fd, desc := range f.files { - if desc.file != nil && cond(desc.file, desc.flags) { - delete(f.files, fd) - removed = append(removed, desc.file) - } - } - f.mu.Unlock() - - for _, file := range removed { - f.unlock(file) - inotifyFileClose(file) - file.DecRef() - } -} diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go new file mode 100644 index 000000000..1f3a57dc1 --- /dev/null +++ b/pkg/sentry/kernel/fd_table.go @@ -0,0 +1,380 @@ +// 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 kernel + +import ( + "bytes" + "fmt" + "math" + "sync" + "sync/atomic" + "syscall" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/refs" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fs/lock" + "gvisor.dev/gvisor/pkg/sentry/limits" +) + +// FDFlags define flags for an individual descriptor. +// +// +stateify savable +type FDFlags struct { + // CloseOnExec indicates the descriptor should be closed on exec. + CloseOnExec bool +} + +// ToLinuxFileFlags converts a kernel.FDFlags object to a Linux file flags +// representation. +func (f FDFlags) ToLinuxFileFlags() (mask uint) { + if f.CloseOnExec { + mask |= linux.O_CLOEXEC + } + return +} + +// ToLinuxFDFlags converts a kernel.FDFlags object to a Linux descriptor flags +// representation. +func (f FDFlags) ToLinuxFDFlags() (mask uint) { + if f.CloseOnExec { + mask |= linux.FD_CLOEXEC + } + return +} + +// descriptor holds the details about a file descriptor, namely a pointer to +// the file itself and the descriptor flags. +// +// Note that this is immutable and can only be changed via operations on the +// descriptorTable. +// +// +stateify savable +type descriptor struct { + file *fs.File + flags FDFlags +} + +// FDTable is used to manage File references and flags. +// +// +stateify savable +type FDTable struct { + refs.AtomicRefCount + k *Kernel + + // uid is a unique identifier. + uid uint64 + + // mu protects below. + mu sync.Mutex `state:"nosave"` + + // used contains the number of non-nil entries. + used int32 + + // descriptorTable holds descriptors. + descriptorTable `state:".(map[int32]descriptor)"` +} + +func (f *FDTable) saveDescriptorTable() map[int32]descriptor { + m := make(map[int32]descriptor) + f.forEach(func(fd int32, file *fs.File, flags FDFlags) { + m[fd] = descriptor{ + file: file, + flags: flags, + } + }) + return m +} + +func (f *FDTable) loadDescriptorTable(m map[int32]descriptor) { + f.init() // Initialize table. + for fd, d := range m { + f.set(fd, d.file, d.flags) + + // Note that we do _not_ need to acquire a extra table + // reference here. The table reference will already be + // accounted for in the file, so we drop the reference taken by + // set above. + d.file.DecRef() + } +} + +// drop drops the table reference. +func (f *FDTable) drop(file *fs.File) { + // Release locks. + file.Dirent.Inode.LockCtx.Posix.UnlockRegion(lock.UniqueID(f.uid), lock.LockRange{0, lock.LockEOF}) + + // Send inotify events. + d := file.Dirent + var ev uint32 + if fs.IsDir(d.Inode.StableAttr) { + ev |= linux.IN_ISDIR + } + if file.Flags().Write { + ev |= linux.IN_CLOSE_WRITE + } else { + ev |= linux.IN_CLOSE_NOWRITE + } + d.InotifyEvent(ev, 0) + + // Drop the table reference. + file.DecRef() +} + +// ID returns a unique identifier for this FDTable. +func (f *FDTable) ID() uint64 { + return f.uid +} + +// NewFDTable allocates a new FDTable that may be used by tasks in k. +func (k *Kernel) NewFDTable() *FDTable { + f := &FDTable{ + k: k, + uid: atomic.AddUint64(&k.fdMapUids, 1), + } + f.init() + return f +} + +// destroy removes all of the file descriptors from the map. +func (f *FDTable) destroy() { + f.RemoveIf(func(*fs.File, FDFlags) bool { + return true + }) +} + +// DecRef implements RefCounter.DecRef with destructor f.destroy. +func (f *FDTable) DecRef() { + f.DecRefWithDestructor(f.destroy) +} + +// Size returns the number of file descriptor slots currently allocated. +func (f *FDTable) Size() int { + size := atomic.LoadInt32(&f.used) + return int(size) +} + +// forEach iterates over all non-nil files. +// +// It is the caller's responsibility to acquire an appropriate lock. +func (f *FDTable) forEach(fn func(fd int32, file *fs.File, flags FDFlags)) { + fd := int32(0) + for { + file, flags, ok := f.get(fd) + if !ok { + break + } + if file != nil { + if !file.TryIncRef() { + continue // Race caught. + } + fn(int32(fd), file, flags) + file.DecRef() + } + fd++ + } +} + +// String is a stringer for FDTable. +func (f *FDTable) String() string { + var b bytes.Buffer + f.forEach(func(fd int32, file *fs.File, flags FDFlags) { + n, _ := file.Dirent.FullName(nil /* root */) + b.WriteString(fmt.Sprintf("\tfd:%d => name %s\n", fd, n)) + }) + return b.String() +} + +// NewFDs allocates new FDs guaranteed to be the lowest number available +// greater than or equal to the fd parameter. All files will share the set +// flags. Success is guaranteed to be all or none. +func (f *FDTable) NewFDs(ctx context.Context, fd int32, files []*fs.File, flags FDFlags) (fds []int32, err error) { + if fd < 0 { + // Don't accept negative FDs. + return nil, syscall.EINVAL + } + + // Default limit. + end := int32(math.MaxInt32) + + // Ensure we don't get past the provided limit. + if limitSet := limits.FromContext(ctx); limitSet != nil { + lim := limitSet.Get(limits.NumberOfFiles) + if lim.Cur != limits.Infinity { + end = int32(lim.Cur) + } + if fd >= end { + return nil, syscall.EMFILE + } + } + + f.mu.Lock() + defer f.mu.Unlock() + + // Install all entries. + for i := fd; i < end && len(fds) < len(files); i++ { + if d, _, _ := f.get(i); d == nil { + f.set(i, files[len(fds)], flags) // Set the descriptor. + fds = append(fds, i) // Record the file descriptor. + } + } + + // Failure? Unwind existing FDs. + if len(fds) < len(files) { + for _, i := range fds { + f.set(i, nil, FDFlags{}) // Zap entry. + } + return nil, syscall.EMFILE + } + + return fds, nil +} + +// NewFDAt sets the file reference for the given FD. If there is an active +// reference for that FD, the ref count for that existing reference is +// decremented. +func (f *FDTable) NewFDAt(ctx context.Context, fd int32, file *fs.File, flags FDFlags) error { + if fd < 0 { + // Don't accept negative FDs. + return syscall.EBADF + } + + f.mu.Lock() + defer f.mu.Unlock() + + // Check the limit for the provided file. + if limitSet := limits.FromContext(ctx); limitSet != nil { + if lim := limitSet.Get(limits.NumberOfFiles); lim.Cur != limits.Infinity && uint64(fd) >= lim.Cur { + return syscall.EMFILE + } + } + + // Install the entry. + f.set(fd, file, flags) + return nil +} + +// SetFlags sets the flags for the given file descriptor. +// +// True is returned iff flags were changed. +func (f *FDTable) SetFlags(fd int32, flags FDFlags) error { + if fd < 0 { + // Don't accept negative FDs. + return syscall.EBADF + } + + f.mu.Lock() + defer f.mu.Unlock() + + file, _, _ := f.get(fd) + if file == nil { + // No file found. + return syscall.EBADF + } + + // Update the flags. + f.set(fd, file, flags) + return nil +} + +// Get returns a reference to the file and the flags for the FD or nil if no +// file is defined for the given fd. +// +// N.B. Callers are required to use DecRef when they are done. +// +//go:nosplit +func (f *FDTable) Get(fd int32) (*fs.File, FDFlags) { + if fd < 0 { + return nil, FDFlags{} + } + + for { + file, flags, _ := f.get(fd) + if file != nil { + if !file.TryIncRef() { + continue // Race caught. + } + // Reference acquired. + return file, flags + } + // No file available. + return nil, FDFlags{} + } +} + +// GetFDs returns a list of valid fds. +func (f *FDTable) GetFDs() []int32 { + fds := make([]int32, 0, f.used) + f.forEach(func(fd int32, file *fs.File, flags FDFlags) { + fds = append(fds, fd) + }) + return fds +} + +// GetRefs returns a stable slice of references to all files and bumps the +// reference count on each. The caller must use DecRef on each reference when +// they're done using the slice. +func (f *FDTable) GetRefs() []*fs.File { + files := make([]*fs.File, 0, f.Size()) + f.forEach(func(_ int32, file *fs.File, flags FDFlags) { + file.IncRef() // Acquire a reference for caller. + files = append(files, file) + }) + return files +} + +// Fork returns an independent FDTable. +func (f *FDTable) Fork() *FDTable { + clone := f.k.NewFDTable() + + f.forEach(func(fd int32, file *fs.File, flags FDFlags) { + // The set function here will acquire an appropriate table + // reference for the clone. We don't need anything else. + clone.set(fd, file, flags) + }) + return clone +} + +// Remove removes an FD from and returns a non-file iff successful. +// +// N.B. Callers are required to use DecRef when they are done. +func (f *FDTable) Remove(fd int32) *fs.File { + if fd < 0 { + return nil + } + + f.mu.Lock() + defer f.mu.Unlock() + + orig, _, _ := f.get(fd) + if orig != nil { + orig.IncRef() // Reference for caller. + f.set(fd, nil, FDFlags{}) // Zap entry. + } + return orig +} + +// RemoveIf removes all FDs where cond is true. +func (f *FDTable) RemoveIf(cond func(*fs.File, FDFlags) bool) { + f.mu.Lock() + defer f.mu.Unlock() + + f.forEach(func(fd int32, file *fs.File, flags FDFlags) { + if cond(file, flags) { + f.set(fd, nil, FDFlags{}) // Clear from table. + } + }) +} diff --git a/pkg/sentry/kernel/fd_table_unsafe.go b/pkg/sentry/kernel/fd_table_unsafe.go new file mode 100644 index 000000000..e009df974 --- /dev/null +++ b/pkg/sentry/kernel/fd_table_unsafe.go @@ -0,0 +1,103 @@ +// 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 kernel + +import ( + "sync/atomic" + "unsafe" + + "gvisor.dev/gvisor/pkg/sentry/fs" +) + +type descriptorTable struct { + // slice is a *[]unsafe.Pointer, where each element is actually + // *descriptor object, updated atomically. + // + // Changes to the slice itself requiring holding FDTable.mu. + slice unsafe.Pointer `state:".(map[int32]*descriptor)"` +} + +// init initializes the table. +func (f *FDTable) init() { + var slice []unsafe.Pointer // Empty slice. + atomic.StorePointer(&f.slice, unsafe.Pointer(&slice)) +} + +// get gets a file entry. +// +// The boolean indicates whether this was in range. +// +//go:nosplit +func (f *FDTable) get(fd int32) (*fs.File, FDFlags, bool) { + slice := *(*[]unsafe.Pointer)(atomic.LoadPointer(&f.slice)) + if fd >= int32(len(slice)) { + return nil, FDFlags{}, false + } + d := (*descriptor)(atomic.LoadPointer(&slice[fd])) + if d == nil { + return nil, FDFlags{}, true + } + return d.file, d.flags, true +} + +// set sets an entry. +// +// This handles accounting changes, as well as acquiring and releasing the +// reference needed by the table iff the file is different. +// +// Precondition: mu must be held. +func (f *FDTable) set(fd int32, file *fs.File, flags FDFlags) { + slice := *(*[]unsafe.Pointer)(atomic.LoadPointer(&f.slice)) + + // Grow the table as required. + if last := int32(len(slice)); fd >= last { + end := fd + 1 + if end < 2*last { + end = 2 * last + } + slice = append(slice, make([]unsafe.Pointer, end-last)...) + atomic.StorePointer(&f.slice, unsafe.Pointer(&slice)) + } + + // Create the new element. + var d *descriptor + if file != nil { + d = &descriptor{ + file: file, + flags: flags, + } + } + + // Update the single element. + orig := (*descriptor)(atomic.SwapPointer(&slice[fd], unsafe.Pointer(d))) + + // Acquire a table reference. + if file != nil && (orig == nil || file != orig.file) { + file.IncRef() + } + + // Drop the table reference. + if orig != nil && file != orig.file { + f.drop(orig.file) + } + + // Adjust used. + switch { + case orig == nil && file != nil: + atomic.AddInt32(&f.used, 1) + case orig != nil && file == nil: + atomic.AddInt32(&f.used, -1) + } +} diff --git a/pkg/sentry/kernel/fs_context.go b/pkg/sentry/kernel/fs_context.go index 938239aeb..ded27d668 100644 --- a/pkg/sentry/kernel/fs_context.go +++ b/pkg/sentry/kernel/fs_context.go @@ -51,11 +51,13 @@ type FSContext struct { func newFSContext(root, cwd *fs.Dirent, umask uint) *FSContext { root.IncRef() cwd.IncRef() - return &FSContext{ + f := FSContext{ root: root, cwd: cwd, umask: umask, } + f.EnableLeakCheck("kernel.FSContext") + return &f } // destroy is the destructor for an FSContext. diff --git a/pkg/sentry/kernel/futex/atomicptr_bucket.go b/pkg/sentry/kernel/futex/atomicptr_bucket_unsafe.go index d3fdf09b0..d3fdf09b0 100755 --- a/pkg/sentry/kernel/futex/atomicptr_bucket.go +++ b/pkg/sentry/kernel/futex/atomicptr_bucket_unsafe.go diff --git a/pkg/sentry/kernel/futex/futex.go b/pkg/sentry/kernel/futex/futex.go index 3bd5c04af..278cc8143 100644 --- a/pkg/sentry/kernel/futex/futex.go +++ b/pkg/sentry/kernel/futex/futex.go @@ -729,14 +729,14 @@ func (m *Manager) UnlockPI(t Target, addr usermem.Addr, tid uint32, private bool } b := m.lockBucket(&k) - err = m.unlockPILocked(t, addr, tid, b) + err = m.unlockPILocked(t, addr, tid, b, &k) k.release() b.mu.Unlock() return err } -func (m *Manager) unlockPILocked(t Target, addr usermem.Addr, tid uint32, b *bucket) error { +func (m *Manager) unlockPILocked(t Target, addr usermem.Addr, tid uint32, b *bucket, key *Key) error { cur, err := t.LoadUint32(addr) if err != nil { return err @@ -746,7 +746,22 @@ func (m *Manager) unlockPILocked(t Target, addr usermem.Addr, tid uint32, b *buc return syserror.EPERM } - if b.waiters.Empty() { + var next *Waiter // Who's the next owner? + var next2 *Waiter // Who's the one after that? + for w := b.waiters.Front(); w != nil; w = w.Next() { + if !w.key.matches(key) { + continue + } + + if next == nil { + next = w + } else { + next2 = w + break + } + } + + if next == nil { // It's safe to set 0 because there are no waiters, no new owner, and the // executing task is the current owner (no owner died bit). prev, err := t.CompareAndSwapUint32(addr, cur, 0) @@ -761,12 +776,10 @@ func (m *Manager) unlockPILocked(t Target, addr usermem.Addr, tid uint32, b *buc return nil } - next := b.waiters.Front() - // Set next owner's TID, waiters if there are any. Resets owner died bit, if // set, because the executing task takes over as the owner. val := next.tid - if next.Next() != nil { + if next2 != nil { val |= linux.FUTEX_WAITERS } diff --git a/pkg/sentry/kernel/kdefs/kdefs_state_autogen.go b/pkg/sentry/kernel/kdefs/kdefs_state_autogen.go deleted file mode 100755 index cef77125b..000000000 --- a/pkg/sentry/kernel/kdefs/kdefs_state_autogen.go +++ /dev/null @@ -1,4 +0,0 @@ -// automatically generated by stateify. - -package kdefs - diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 47dadc43a..38b49cba2 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -155,7 +155,7 @@ type Kernel struct { // cpuClockTicker increments cpuClock. cpuClockTicker *ktime.Timer `state:"nosave"` - // fdMapUids is an ever-increasing counter for generating FDMap uids. + // fdMapUids is an ever-increasing counter for generating FDTable uids. // // fdMapUids is mutable, and is accessed using atomic memory operations. fdMapUids uint64 @@ -400,8 +400,8 @@ func (k *Kernel) flushMountSourceRefs() error { // There may be some open FDs whose filesystems have been unmounted. We // must flush those as well. - return k.tasks.forEachFDPaused(func(desc descriptor) error { - desc.file.Dirent.Inode.MountSource.FlushDirentRefs() + return k.tasks.forEachFDPaused(func(file *fs.File) error { + file.Dirent.Inode.MountSource.FlushDirentRefs() return nil }) } @@ -410,35 +410,35 @@ func (k *Kernel) flushMountSourceRefs() error { // task. // // Precondition: Must be called with the kernel paused. -func (ts *TaskSet) forEachFDPaused(f func(descriptor) error) error { +func (ts *TaskSet) forEachFDPaused(f func(*fs.File) error) (err error) { ts.mu.RLock() defer ts.mu.RUnlock() for t := range ts.Root.tids { // We can skip locking Task.mu here since the kernel is paused. - if t.fds == nil { + if t.fdTable == nil { continue } - for _, desc := range t.fds.files { - if err := f(desc); err != nil { - return err + t.fdTable.forEach(func(_ int32, file *fs.File, _ FDFlags) { + if lastErr := f(file); lastErr != nil && err == nil { + err = lastErr } - } + }) } - return nil + return err } func (ts *TaskSet) flushWritesToFiles(ctx context.Context) error { - return ts.forEachFDPaused(func(desc descriptor) error { - if flags := desc.file.Flags(); !flags.Write { + return ts.forEachFDPaused(func(file *fs.File) error { + if flags := file.Flags(); !flags.Write { return nil } - if sattr := desc.file.Dirent.Inode.StableAttr; !fs.IsFile(sattr) && !fs.IsDir(sattr) { + if sattr := file.Dirent.Inode.StableAttr; !fs.IsFile(sattr) && !fs.IsDir(sattr) { return nil } // Here we need all metadata synced. - syncErr := desc.file.Fsync(ctx, 0, fs.FileMaxOffset, fs.SyncAll) + syncErr := file.Fsync(ctx, 0, fs.FileMaxOffset, fs.SyncAll) if err := fs.SaveFileFsyncError(syncErr); err != nil { - name, _ := desc.file.Dirent.FullName(nil /* root */) + name, _ := file.Dirent.FullName(nil /* root */) // Wrap this error in ErrSaveRejection // so that it will trigger a save // error, rather than a panic. This @@ -483,14 +483,12 @@ func (ts *TaskSet) unregisterEpollWaiters() { defer ts.mu.RUnlock() for t := range ts.Root.tids { // We can skip locking Task.mu here since the kernel is paused. - if fdmap := t.fds; fdmap != nil { - for _, desc := range fdmap.files { - if desc.file != nil { - if e, ok := desc.file.FileOperations.(*epoll.EventPoll); ok { - e.UnregisterEpollWaiters() - } + if t.fdTable != nil { + t.fdTable.forEach(func(_ int32, file *fs.File, _ FDFlags) { + if e, ok := file.FileOperations.(*epoll.EventPoll); ok { + e.UnregisterEpollWaiters() } - } + }) } } } @@ -538,6 +536,8 @@ func (k *Kernel) LoadFrom(r io.Reader, net inet.Stack) error { } log.Infof("Memory load took [%s].", time.Since(memoryStart)) + log.Infof("Overall load took [%s]", time.Since(loadStart)) + // Ensure that all pending asynchronous work is complete: // - namedpipe opening // - inode file opening @@ -602,9 +602,9 @@ type CreateProcessArgs struct { // Credentials is the initial credentials. Credentials *auth.Credentials - // FDMap is the initial set of file descriptors. If CreateProcess succeeds, - // it takes a reference on FDMap. - FDMap *FDMap + // FDTable is the initial set of file descriptors. If CreateProcess succeeds, + // it takes a reference on FDTable. + FDTable *FDTable // Umask is the initial umask. Umask uint @@ -789,9 +789,9 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, return nil, 0, errors.New(se.String()) } - // Take a reference on the FDMap, which will be transferred to + // Take a reference on the FDTable, which will be transferred to // TaskSet.NewTask(). - args.FDMap.IncRef() + args.FDTable.IncRef() // Create the task. config := &TaskConfig{ @@ -799,7 +799,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, ThreadGroup: tg, TaskContext: tc, FSContext: newFSContext(root, wd, args.Umask), - FDMap: args.FDMap, + FDTable: args.FDTable, Credentials: args.Credentials, AllowedCPUMask: sched.NewFullCPUSet(k.applicationCores), UTSNamespace: args.UTSNamespace, @@ -871,7 +871,7 @@ func (k *Kernel) pauseTimeLocked() { } // By precondition, nothing else can be interacting with PIDNamespace.tids - // or FDMap.files, so we can iterate them without synchronization. (We + // or FDTable.files, so we can iterate them without synchronization. (We // can't hold the TaskSet mutex when pausing thread group timers because // thread group timers call ThreadGroup.SendSignal, which takes the TaskSet // mutex, while holding the Timer mutex.) @@ -882,14 +882,14 @@ func (k *Kernel) pauseTimeLocked() { it.PauseTimer() } } - // This means we'll iterate FDMaps shared by multiple tasks repeatedly, + // This means we'll iterate FDTables shared by multiple tasks repeatedly, // but ktime.Timer.Pause is idempotent so this is harmless. - if fdm := t.fds; fdm != nil { - for _, desc := range fdm.files { - if tfd, ok := desc.file.FileOperations.(*timerfd.TimerOperations); ok { + if t.fdTable != nil { + t.fdTable.forEach(func(_ int32, file *fs.File, _ FDFlags) { + if tfd, ok := file.FileOperations.(*timerfd.TimerOperations); ok { tfd.PauseTimer() } - } + }) } } k.timekeeper.PauseUpdates() @@ -914,12 +914,12 @@ func (k *Kernel) resumeTimeLocked() { it.ResumeTimer() } } - if fdm := t.fds; fdm != nil { - for _, desc := range fdm.files { - if tfd, ok := desc.file.FileOperations.(*timerfd.TimerOperations); ok { + if t.fdTable != nil { + t.fdTable.forEach(func(_ int32, file *fs.File, _ FDFlags) { + if tfd, ok := file.FileOperations.(*timerfd.TimerOperations); ok { tfd.ResumeTimer() } - } + }) } } } diff --git a/pkg/sentry/kernel/kernel_state_autogen.go b/pkg/sentry/kernel/kernel_state_autogen.go index 56c59e76c..b24036d33 100755 --- a/pkg/sentry/kernel/kernel_state_autogen.go +++ b/pkg/sentry/kernel/kernel_state_autogen.go @@ -61,21 +61,24 @@ func (x *descriptor) load(m state.Map) { m.Load("flags", &x.flags) } -func (x *FDMap) beforeSave() {} -func (x *FDMap) save(m state.Map) { +func (x *FDTable) beforeSave() {} +func (x *FDTable) save(m state.Map) { x.beforeSave() + var descriptorTable map[int32]descriptor = x.saveDescriptorTable() + m.SaveValue("descriptorTable", descriptorTable) m.Save("AtomicRefCount", &x.AtomicRefCount) m.Save("k", &x.k) - m.Save("files", &x.files) m.Save("uid", &x.uid) + m.Save("used", &x.used) } -func (x *FDMap) afterLoad() {} -func (x *FDMap) load(m state.Map) { +func (x *FDTable) afterLoad() {} +func (x *FDTable) load(m state.Map) { m.Load("AtomicRefCount", &x.AtomicRefCount) m.Load("k", &x.k) - m.Load("files", &x.files) m.Load("uid", &x.uid) + m.Load("used", &x.used) + m.LoadValue("descriptorTable", new(map[int32]descriptor), func(y interface{}) { x.loadDescriptorTable(y.(map[int32]descriptor)) }) } func (x *FSContext) beforeSave() {} @@ -536,8 +539,8 @@ func (x *Task) save(m state.Map) { m.Save("k", &x.k) m.Save("containerID", &x.containerID) m.Save("tc", &x.tc) - m.Save("fsc", &x.fsc) - m.Save("fds", &x.fds) + m.Save("fsContext", &x.fsContext) + m.Save("fdTable", &x.fdTable) m.Save("vforkParent", &x.vforkParent) m.Save("exitState", &x.exitState) m.Save("exitTracerNotified", &x.exitTracerNotified) @@ -592,8 +595,8 @@ func (x *Task) load(m state.Map) { m.Load("k", &x.k) m.Load("containerID", &x.containerID) m.Load("tc", &x.tc) - m.Load("fsc", &x.fsc) - m.Load("fds", &x.fds) + m.Load("fsContext", &x.fsContext) + m.Load("fdTable", &x.fdTable) m.Load("vforkParent", &x.vforkParent) m.Load("exitState", &x.exitState) m.Load("exitTracerNotified", &x.exitTracerNotified) @@ -1118,7 +1121,7 @@ func init() { state.Register("kernel.AbstractSocketNamespace", (*AbstractSocketNamespace)(nil), state.Fns{Save: (*AbstractSocketNamespace).save, Load: (*AbstractSocketNamespace).load}) state.Register("kernel.FDFlags", (*FDFlags)(nil), state.Fns{Save: (*FDFlags).save, Load: (*FDFlags).load}) state.Register("kernel.descriptor", (*descriptor)(nil), state.Fns{Save: (*descriptor).save, Load: (*descriptor).load}) - state.Register("kernel.FDMap", (*FDMap)(nil), state.Fns{Save: (*FDMap).save, Load: (*FDMap).load}) + state.Register("kernel.FDTable", (*FDTable)(nil), state.Fns{Save: (*FDTable).save, Load: (*FDTable).load}) state.Register("kernel.FSContext", (*FSContext)(nil), state.Fns{Save: (*FSContext).save, Load: (*FSContext).load}) state.Register("kernel.IPCNamespace", (*IPCNamespace)(nil), state.Fns{Save: (*IPCNamespace).save, Load: (*IPCNamespace).load}) state.Register("kernel.Kernel", (*Kernel)(nil), state.Fns{Save: (*Kernel).save, Load: (*Kernel).load}) diff --git a/pkg/sentry/kernel/seqatomic_taskgoroutineschedinfo.go b/pkg/sentry/kernel/seqatomic_taskgoroutineschedinfo_unsafe.go index be6b07629..895abb129 100755 --- a/pkg/sentry/kernel/seqatomic_taskgoroutineschedinfo.go +++ b/pkg/sentry/kernel/seqatomic_taskgoroutineschedinfo_unsafe.go @@ -1,11 +1,12 @@ package kernel import ( + "unsafe" + "fmt" "gvisor.dev/gvisor/third_party/gvsync" "reflect" "strings" - "unsafe" ) // SeqAtomicLoad returns a copy of *ptr, ensuring that the read does not race diff --git a/pkg/sentry/kernel/sessions.go b/pkg/sentry/kernel/sessions.go index 355984140..81fcd8258 100644 --- a/pkg/sentry/kernel/sessions.go +++ b/pkg/sentry/kernel/sessions.go @@ -294,6 +294,7 @@ func (tg *ThreadGroup) createSession() error { id: SessionID(id), leader: tg, } + s.refs.EnableLeakCheck("kernel.Session") // Create a new ProcessGroup, belonging to that Session. // This also has a single reference (assigned below). @@ -307,6 +308,7 @@ func (tg *ThreadGroup) createSession() error { session: s, ancestors: 0, } + pg.refs.EnableLeakCheck("kernel.ProcessGroup") // Tie them and return the result. s.processGroups.PushBack(pg) @@ -378,11 +380,13 @@ func (tg *ThreadGroup) CreateProcessGroup() error { // We manually adjust the ancestors if the parent is in the same // session. tg.processGroup.session.incRef() - pg := &ProcessGroup{ + pg := ProcessGroup{ id: ProcessGroupID(id), originator: tg, session: tg.processGroup.session, } + pg.refs.EnableLeakCheck("kernel.ProcessGroup") + if tg.leader.parent != nil && tg.leader.parent.tg.processGroup.session == pg.session { pg.ancestors++ } @@ -390,20 +394,20 @@ func (tg *ThreadGroup) CreateProcessGroup() error { // Assign the new process group; adjust children. oldParentPG := tg.parentPG() tg.forEachChildThreadGroupLocked(func(childTG *ThreadGroup) { - childTG.processGroup.incRefWithParent(pg) + childTG.processGroup.incRefWithParent(&pg) childTG.processGroup.decRefWithParent(oldParentPG) }) tg.processGroup.decRefWithParent(oldParentPG) - tg.processGroup = pg + tg.processGroup = &pg // Add the new process group to the session. - pg.session.processGroups.PushBack(pg) + pg.session.processGroups.PushBack(&pg) // Ensure this translation is added to all namespaces. for ns := tg.pidns; ns != nil; ns = ns.parent { local := ns.tgids[tg] - ns.pgids[pg] = ProcessGroupID(local) - ns.processGroups[ProcessGroupID(local)] = pg + ns.pgids[&pg] = ProcessGroupID(local) + ns.processGroups[ProcessGroupID(local)] = &pg } return nil diff --git a/pkg/sentry/kernel/shm/shm.go b/pkg/sentry/kernel/shm/shm.go index 3e9fe70e2..5bd610f68 100644 --- a/pkg/sentry/kernel/shm/shm.go +++ b/pkg/sentry/kernel/shm/shm.go @@ -224,6 +224,7 @@ func (r *Registry) newShm(ctx context.Context, pid int32, key Key, creator fs.Fi creatorPID: pid, changeTime: ktime.NowFromContext(ctx), } + shm.EnableLeakCheck("kernel.Shm") // Find the next available ID. for id := r.lastIDUsed + 1; id != r.lastIDUsed; id++ { diff --git a/pkg/sentry/kernel/syslog.go b/pkg/sentry/kernel/syslog.go index 175d1b247..8227ecf1d 100644 --- a/pkg/sentry/kernel/syslog.go +++ b/pkg/sentry/kernel/syslog.go @@ -67,6 +67,7 @@ func (s *syslog) Log() []byte { "Creating process schedule...", "Generating random numbers by fair dice roll...", "Rewriting operating system in Javascript...", + "Reticulating splines...", "Consulting tar man page...", "Forking spaghetti code...", "Checking naughty and nice process list...", diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index 2e3a39d3b..e91f82bb3 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -236,15 +236,15 @@ type Task struct { // tc is protected by mu, and is owned by the task goroutine. tc TaskContext - // fsc is the task's filesystem context. + // fsContext is the task's filesystem context. // - // fsc is protected by mu, and is owned by the task goroutine. - fsc *FSContext + // fsContext is protected by mu, and is owned by the task goroutine. + fsContext *FSContext - // fds is the task's file descriptor table. + // fdTable is the task's file descriptor table. // - // fds is protected by mu, and is owned by the task goroutine. - fds *FDMap + // fdTable is protected by mu, and is owned by the task goroutine. + fdTable *FDTable // If vforkParent is not nil, it is the task that created this task with // vfork() or clone(CLONE_VFORK), and should have its vforkStop ended when @@ -602,7 +602,7 @@ func (t *Task) Value(key interface{}) interface{} { case context.CtxThreadGroupID: return int32(t.ThreadGroup().ID()) case fs.CtxRoot: - return t.fsc.RootDirectory() + return t.fsContext.RootDirectory() case fs.CtxDirentCacheLimiter: return t.k.DirentCacheLimiter case inet.CtxStack: @@ -668,7 +668,7 @@ func (t *Task) SyscallRestartBlock() SyscallRestartBlock { func (t *Task) IsChrooted() bool { realRoot := t.tg.mounts.Root() defer realRoot.DecRef() - root := t.fsc.RootDirectory() + root := t.fsContext.RootDirectory() if root != nil { defer root.DecRef() } @@ -689,16 +689,55 @@ func (t *Task) TaskContext() *TaskContext { // Precondition: The caller must be running on the task goroutine, or t.mu must // be locked. func (t *Task) FSContext() *FSContext { - return t.fsc + return t.fsContext } -// FDMap returns t's FDMap. FDMap does not take an additional reference on the -// returned FDMap. +// FDTable returns t's FDTable. FDMTable does not take an additional reference +// on the returned FDMap. // // Precondition: The caller must be running on the task goroutine, or t.mu must // be locked. -func (t *Task) FDMap() *FDMap { - return t.fds +func (t *Task) FDTable() *FDTable { + return t.fdTable +} + +// GetFile is a convenience wrapper t.FDTable().GetFile. +// +// Precondition: same as FDTable. +func (t *Task) GetFile(fd int32) *fs.File { + f, _ := t.fdTable.Get(fd) + return f +} + +// NewFDs is a convenience wrapper for t.FDTable().NewFDs. +// +// This automatically passes the task as the context. +// +// Precondition: same as FDTable. +func (t *Task) NewFDs(fd int32, files []*fs.File, flags FDFlags) ([]int32, error) { + return t.fdTable.NewFDs(t, fd, files, flags) +} + +// NewFDFrom is a convenience wrapper for t.FDTable().NewFDs with a single file. +// +// This automatically passes the task as the context. +// +// Precondition: same as FDTable. +func (t *Task) NewFDFrom(fd int32, file *fs.File, flags FDFlags) (int32, error) { + fds, err := t.fdTable.NewFDs(t, fd, []*fs.File{file}, flags) + if err != nil { + return 0, err + } + return fds[0], nil +} + +// NewFDAt is a convenience wrapper for t.FDTable().NewFDAt. +// +// This automatically passes the task as the context. +// +// Precondition: same as FDTable. +func (t *Task) NewFDAt(fd int32, file *fs.File, flags FDFlags) error { + return t.fdTable.NewFDAt(t, fd, file, flags) } // WithMuLocked executes f with t.mu locked. diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go index b5cc3860d..0916fd658 100644 --- a/pkg/sentry/kernel/task_clone.go +++ b/pkg/sentry/kernel/task_clone.go @@ -214,20 +214,20 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { } } - var fsc *FSContext + var fsContext *FSContext if opts.NewFSContext { - fsc = t.fsc.Fork() + fsContext = t.fsContext.Fork() } else { - fsc = t.fsc - fsc.IncRef() + fsContext = t.fsContext + fsContext.IncRef() } - var fds *FDMap + var fdTable *FDTable if opts.NewFiles { - fds = t.fds.Fork() + fdTable = t.fdTable.Fork() } else { - fds = t.fds - fds.IncRef() + fdTable = t.fdTable + fdTable.IncRef() } pidns := t.tg.pidns @@ -251,8 +251,8 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { ThreadGroup: tg, SignalMask: t.SignalMask(), TaskContext: tc, - FSContext: fsc, - FDMap: fds, + FSContext: fsContext, + FDTable: fdTable, Credentials: creds, Niceness: t.Niceness(), NetworkNamespaced: t.netns, @@ -485,22 +485,22 @@ func (t *Task) Unshare(opts *SharingOptions) error { // namespace" t.ipcns = NewIPCNamespace(creds.UserNamespace) } - var oldfds *FDMap + var oldFDTable *FDTable if opts.NewFiles { - oldfds = t.fds - t.fds = oldfds.Fork() + oldFDTable = t.fdTable + t.fdTable = oldFDTable.Fork() } - var oldfsc *FSContext + var oldFSContext *FSContext if opts.NewFSContext { - oldfsc = t.fsc - t.fsc = oldfsc.Fork() + oldFSContext = t.fsContext + t.fsContext = oldFSContext.Fork() } t.mu.Unlock() - if oldfds != nil { - oldfds.DecRef() + if oldFDTable != nil { + oldFDTable.DecRef() } - if oldfsc != nil { - oldfsc.DecRef() + if oldFSContext != nil { + oldFSContext.DecRef() } return nil } diff --git a/pkg/sentry/kernel/task_exec.go b/pkg/sentry/kernel/task_exec.go index cd85acaef..17a089b90 100644 --- a/pkg/sentry/kernel/task_exec.go +++ b/pkg/sentry/kernel/task_exec.go @@ -195,7 +195,7 @@ func (r *runSyscallAfterExecStop) execute(t *Task) taskRunState { t.tg.pidns.owner.mu.Unlock() // Remove FDs with the CloseOnExec flag set. - t.fds.RemoveIf(func(file *fs.File, flags FDFlags) bool { + t.fdTable.RemoveIf(func(file *fs.File, flags FDFlags) bool { return flags.CloseOnExec }) diff --git a/pkg/sentry/kernel/task_exit.go b/pkg/sentry/kernel/task_exit.go index b97d65185..535f03e50 100644 --- a/pkg/sentry/kernel/task_exit.go +++ b/pkg/sentry/kernel/task_exit.go @@ -265,8 +265,8 @@ func (*runExitMain) execute(t *Task) taskRunState { // Releasing the MM unblocks a blocked CLONE_VFORK parent. t.unstopVforkParent() - t.fsc.DecRef() - t.fds.DecRef() + t.fsContext.DecRef() + t.fdTable.DecRef() // If this is the last task to exit from the thread group, release the // thread group's resources. diff --git a/pkg/sentry/kernel/task_log.go b/pkg/sentry/kernel/task_log.go index cf48663b6..a29e9b9eb 100644 --- a/pkg/sentry/kernel/task_log.go +++ b/pkg/sentry/kernel/task_log.go @@ -63,7 +63,7 @@ func (t *Task) DebugDumpState() { if mm := t.MemoryManager(); mm != nil { t.Debugf("Mappings:\n%s", mm) } - t.Debugf("FDMap:\n%s", t.fds) + t.Debugf("FDTable:\n%s", t.fdTable) } // debugDumpRegisters logs register state at log level debug. diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go index 72caae537..a88bf3951 100644 --- a/pkg/sentry/kernel/task_start.go +++ b/pkg/sentry/kernel/task_start.go @@ -52,9 +52,10 @@ type TaskConfig struct { // succeeds. FSContext *FSContext - // FDMap is the FDMap of the new task. A reference must be held on FDMap, - // which is transferred to TaskSet.NewTask whether or not it succeeds. - FDMap *FDMap + // FDTable is the FDTableof the new task. A reference must be held on + // FDMap, which is transferred to TaskSet.NewTask whether or not it + // succeeds. + FDTable *FDTable // Credentials is the Credentials of the new task. Credentials *auth.Credentials @@ -90,7 +91,7 @@ func (ts *TaskSet) NewTask(cfg *TaskConfig) (*Task, error) { if err != nil { cfg.TaskContext.release() cfg.FSContext.DecRef() - cfg.FDMap.DecRef() + cfg.FDTable.DecRef() return nil, err } return t, nil @@ -112,8 +113,8 @@ func (ts *TaskSet) newTask(cfg *TaskConfig) (*Task, error) { signalMask: cfg.SignalMask, signalStack: arch.SignalStack{Flags: arch.SignalStackFlagDisable}, tc: *tc, - fsc: cfg.FSContext, - fds: cfg.FDMap, + fsContext: cfg.FSContext, + fdTable: cfg.FDTable, p: cfg.Kernel.Platform.NewContext(), k: cfg.Kernel, ptraceTracees: make(map[*Task]struct{}), diff --git a/pkg/sentry/loader/loader.go b/pkg/sentry/loader/loader.go index edfdac2a7..baa12d9a0 100644 --- a/pkg/sentry/loader/loader.go +++ b/pkg/sentry/loader/loader.go @@ -54,7 +54,7 @@ func readFull(ctx context.Context, f *fs.File, dst usermem.IOSequence, offset in // openPath opens name for loading. // // openPath returns the fs.Dirent and an *fs.File for name, which is not -// installed in the Task FDMap. The caller takes ownership of both. +// installed in the Task FDTable. The caller takes ownership of both. // // name must be a readable, executable, regular file. func openPath(ctx context.Context, mm *fs.MountNamespace, root, wd *fs.Dirent, maxTraversals *uint, name string) (*fs.Dirent, *fs.File, error) { diff --git a/pkg/sentry/mm/aio_context.go b/pkg/sentry/mm/aio_context.go index c9a942023..1b746d030 100644 --- a/pkg/sentry/mm/aio_context.go +++ b/pkg/sentry/mm/aio_context.go @@ -213,7 +213,9 @@ func newAIOMappable(mfp pgalloc.MemoryFileProvider) (*aioMappable, error) { if err != nil { return nil, err } - return &aioMappable{mfp: mfp, fr: fr}, nil + m := aioMappable{mfp: mfp, fr: fr} + m.EnableLeakCheck("mm.aioMappable") + return &m, nil } // DecRef implements refs.RefCounter.DecRef. diff --git a/pkg/sentry/mm/special_mappable.go b/pkg/sentry/mm/special_mappable.go index 2f8651a0a..ea2d7af74 100644 --- a/pkg/sentry/mm/special_mappable.go +++ b/pkg/sentry/mm/special_mappable.go @@ -45,7 +45,9 @@ type SpecialMappable struct { // // Preconditions: fr.Length() != 0. func NewSpecialMappable(name string, mfp pgalloc.MemoryFileProvider, fr platform.FileRange) *SpecialMappable { - return &SpecialMappable{mfp: mfp, fr: fr, name: name} + m := SpecialMappable{mfp: mfp, fr: fr, name: name} + m.EnableLeakCheck("mm.SpecialMappable") + return &m } // DecRef implements refs.RefCounter.DecRef. diff --git a/pkg/sentry/platform/kvm/filters.go b/pkg/sentry/platform/kvm/filters.go new file mode 100644 index 000000000..7d949f1dd --- /dev/null +++ b/pkg/sentry/platform/kvm/filters.go @@ -0,0 +1,33 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package kvm + +import ( + "syscall" + + "gvisor.dev/gvisor/pkg/seccomp" +) + +// SyscallFilters returns syscalls made exclusively by the KVM platform. +func (*KVM) SyscallFilters() seccomp.SyscallRules { + return seccomp.SyscallRules{ + syscall.SYS_ARCH_PRCTL: {}, + syscall.SYS_IOCTL: {}, + syscall.SYS_MMAP: {}, + syscall.SYS_RT_SIGSUSPEND: {}, + syscall.SYS_RT_SIGTIMEDWAIT: {}, + 0xffffffffffffffff: {}, // KVM uses syscall -1 to transition to host. + } +} diff --git a/pkg/sentry/platform/kvm/kvm.go b/pkg/sentry/platform/kvm/kvm.go index b49d7f3c4..ee4cd2f4d 100644 --- a/pkg/sentry/platform/kvm/kvm.go +++ b/pkg/sentry/platform/kvm/kvm.go @@ -141,3 +141,17 @@ func (k *KVM) NewContext() platform.Context { machine: k.machine, } } + +type constructor struct{} + +func (*constructor) New(f *os.File) (platform.Platform, error) { + return New(f) +} + +func (*constructor) OpenDevice() (*os.File, error) { + return OpenDevice() +} + +func init() { + platform.Register("kvm", &constructor{}) +} diff --git a/pkg/sentry/platform/kvm/machine.go b/pkg/sentry/platform/kvm/machine.go index 7d92e16cc..679087e25 100644 --- a/pkg/sentry/platform/kvm/machine.go +++ b/pkg/sentry/platform/kvm/machine.go @@ -426,7 +426,12 @@ func (c *vCPU) unlock() { // Normal state. case vCPUUser | vCPUGuest | vCPUWaiter: // Force a transition: this must trigger a notification when we - // return from guest mode. + // return from guest mode. We must clear vCPUWaiter here + // anyways, because BounceToKernel will force a transition only + // from ring3 to ring0, which will not clear this bit. Halt may + // workaround the issue, but if there is no exception or + // syscall in this period, BounceToKernel will hang. + atomicbitops.AndUint32(&c.state, ^vCPUWaiter) c.notify() case vCPUUser | vCPUWaiter: // Waiting for the lock to be released; the responsibility is diff --git a/pkg/sentry/platform/platform.go b/pkg/sentry/platform/platform.go index eccbe2336..ec22dbf87 100644 --- a/pkg/sentry/platform/platform.go +++ b/pkg/sentry/platform/platform.go @@ -19,8 +19,10 @@ package platform import ( "fmt" + "os" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/seccomp" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usermem" @@ -93,6 +95,9 @@ type Platform interface { // Platforms for which this does not hold may panic if PreemptAllCPUs is // called. PreemptAllCPUs() error + + // SyscallFilters returns syscalls made exclusively by this platform. + SyscallFilters() seccomp.SyscallRules } // NoCPUPreemptionDetection implements Platform.DetectsCPUPreemption and @@ -347,3 +352,26 @@ type File interface { func (fr FileRange) String() string { return fmt.Sprintf("[%#x, %#x)", fr.Start, fr.End) } + +// Constructor represents a platform type. +type Constructor interface { + New(deviceFile *os.File) (Platform, error) + OpenDevice() (*os.File, error) +} + +// platforms contains all available platform types. +var platforms = map[string]Constructor{} + +// Register registers a new platform type. +func Register(name string, platform Constructor) { + platforms[name] = platform +} + +// Lookup looks up the platform constructor by name. +func Lookup(name string) (Constructor, error) { + p, ok := platforms[name] + if !ok { + return nil, fmt.Errorf("unknown platform: %v", name) + } + return p, nil +} diff --git a/pkg/sentry/kernel/kdefs/kdefs.go b/pkg/sentry/platform/ptrace/filters.go index 304da2032..1e07cfd0d 100644 --- a/pkg/sentry/kernel/kdefs/kdefs.go +++ b/pkg/sentry/platform/ptrace/filters.go @@ -1,4 +1,4 @@ -// Copyright 2018 The gVisor Authors. +// Copyright 2019 The gVisor Authors. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. @@ -12,9 +12,22 @@ // See the License for the specific language governing permissions and // limitations under the License. -// Package kdefs defines common kernel definitions. -// -package kdefs +package ptrace + +import ( + "syscall" + + "golang.org/x/sys/unix" + "gvisor.dev/gvisor/pkg/seccomp" +) -// FD is a File Descriptor. -type FD int32 +// SyscallFilters returns syscalls made exclusively by the ptrace platform. +func (*PTrace) SyscallFilters() seccomp.SyscallRules { + return seccomp.SyscallRules{ + unix.SYS_GETCPU: {}, + unix.SYS_SCHED_SETAFFINITY: {}, + syscall.SYS_PTRACE: {}, + syscall.SYS_TGKILL: {}, + syscall.SYS_WAIT4: {}, + } +} diff --git a/pkg/sentry/platform/ptrace/ptrace.go b/pkg/sentry/platform/ptrace/ptrace.go index ee7e0640c..6fd30ed25 100644 --- a/pkg/sentry/platform/ptrace/ptrace.go +++ b/pkg/sentry/platform/ptrace/ptrace.go @@ -45,6 +45,7 @@ package ptrace import ( + "os" "sync" "gvisor.dev/gvisor/pkg/abi/linux" @@ -236,3 +237,17 @@ func (p *PTrace) NewAddressSpace(_ interface{}) (platform.AddressSpace, <-chan s func (*PTrace) NewContext() platform.Context { return &context{} } + +type constructor struct{} + +func (*constructor) New(*os.File) (platform.Platform, error) { + return New() +} + +func (*constructor) OpenDevice() (*os.File, error) { + return nil, nil +} + +func init() { + platform.Register("ptrace", &constructor{}) +} diff --git a/pkg/sentry/platform/ptrace/subprocess_linux.go b/pkg/sentry/platform/ptrace/subprocess_linux.go index dc3475494..87ded0bbd 100644 --- a/pkg/sentry/platform/ptrace/subprocess_linux.go +++ b/pkg/sentry/platform/ptrace/subprocess_linux.go @@ -306,7 +306,7 @@ func (s *subprocess) createStub() (*thread, error) { arch.SyscallArgument{Value: 0}, arch.SyscallArgument{Value: 0}) if err != nil { - return nil, err + return nil, fmt.Errorf("creating stub process: %v", err) } // Wait for child to enter group-stop, so we don't stop its @@ -325,7 +325,7 @@ func (s *subprocess) createStub() (*thread, error) { arch.SyscallArgument{Value: 0}, arch.SyscallArgument{Value: 0}) if err != nil { - return nil, err + return nil, fmt.Errorf("waiting on stub process: %v", err) } childT := &thread{ diff --git a/pkg/sentry/platform/ring0/defs_impl.go b/pkg/sentry/platform/ring0/defs_impl.go index 8efc3825f..d4bfc5a4a 100755 --- a/pkg/sentry/platform/ring0/defs_impl.go +++ b/pkg/sentry/platform/ring0/defs_impl.go @@ -1,14 +1,14 @@ package ring0 import ( + "syscall" + "fmt" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables" + "gvisor.dev/gvisor/pkg/sentry/usermem" "io" "reflect" - "syscall" - - "gvisor.dev/gvisor/pkg/sentry/usermem" ) var ( diff --git a/pkg/sentry/platform/ring0/kernel_amd64.go b/pkg/sentry/platform/ring0/kernel_amd64.go index 3577b5127..0feff8778 100644 --- a/pkg/sentry/platform/ring0/kernel_amd64.go +++ b/pkg/sentry/platform/ring0/kernel_amd64.go @@ -70,6 +70,14 @@ func (c *CPU) init() { c.tss.ist1Lo = uint32(stackAddr) c.tss.ist1Hi = uint32(stackAddr >> 32) + // Set the I/O bitmap base address beyond the last byte in the TSS + // to block access to the entire I/O address range. + // + // From section 18.5.2 "I/O Permission Bit Map" from Intel SDM vol1: + // I/O addresses not spanned by the map are treated as if they had set + // bits in the map. + c.tss.ioPerm = tssLimit + 1 + // Permanently set the kernel segments. c.registers.Cs = uint64(Kcode) c.registers.Ds = uint64(Kdata) diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index b646dc258..4f4a20dfe 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -23,7 +23,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" @@ -63,7 +62,7 @@ type RightsFiles []*fs.File func NewSCMRights(t *kernel.Task, fds []int32) (SCMRights, error) { files := make(RightsFiles, 0, len(fds)) for _, fd := range fds { - file, _ := t.FDMap().GetDescriptor(kdefs.FD(fd)) + file := t.GetFile(fd) if file == nil { files.Release() return nil, syserror.EBADF @@ -109,7 +108,9 @@ func rightsFDs(t *kernel.Task, rights SCMRights, cloexec bool, max int) ([]int32 files, trunc := rights.Files(t, max) fds := make([]int32, 0, len(files)) for i := 0; i < max && len(files) > 0; i++ { - fd, err := t.FDMap().NewFDFrom(0, files[0], kernel.FDFlags{cloexec}, t.ThreadGroup().Limits()) + fd, err := t.NewFDFrom(0, files[0], kernel.FDFlags{ + CloseOnExec: cloexec, + }) files[0].DecRef() files = files[1:] if err != nil { @@ -315,8 +316,7 @@ func PackTimestamp(t *kernel.Task, timestamp int64, buf []byte) []byte { // Parse parses a raw socket control message into portable objects. func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (transport.ControlMessages, error) { var ( - fds linux.ControlMessageRights - + fds linux.ControlMessageRights haveCreds bool creds linux.ControlMessageCredentials ) diff --git a/pkg/sentry/socket/epsocket/epsocket.go b/pkg/sentry/socket/epsocket/epsocket.go index 2a38e370a..9d1bcfd41 100644 --- a/pkg/sentry/socket/epsocket/epsocket.go +++ b/pkg/sentry/socket/epsocket/epsocket.go @@ -40,7 +40,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/socket" @@ -286,14 +285,14 @@ func bytesToIPAddress(addr []byte) tcpip.Address { // GetAddress reads an sockaddr struct from the given address and converts it // to the FullAddress format. It supports AF_UNIX, AF_INET and AF_INET6 // addresses. -func GetAddress(sfamily int, addr []byte) (tcpip.FullAddress, *syserr.Error) { +func GetAddress(sfamily int, addr []byte, strict bool) (tcpip.FullAddress, *syserr.Error) { // Make sure we have at least 2 bytes for the address family. if len(addr) < 2 { return tcpip.FullAddress{}, syserr.ErrInvalidArgument } family := usermem.ByteOrder.Uint16(addr) - if family != uint16(sfamily) { + if family != uint16(sfamily) && (!strict && family != linux.AF_UNSPEC) { return tcpip.FullAddress{}, syserr.ErrAddressFamilyNotSupported } @@ -318,7 +317,7 @@ func GetAddress(sfamily int, addr []byte) (tcpip.FullAddress, *syserr.Error) { case linux.AF_INET: var a linux.SockAddrInet if len(addr) < sockAddrInetSize { - return tcpip.FullAddress{}, syserr.ErrBadAddress + return tcpip.FullAddress{}, syserr.ErrInvalidArgument } binary.Unmarshal(addr[:sockAddrInetSize], usermem.ByteOrder, &a) @@ -331,7 +330,7 @@ func GetAddress(sfamily int, addr []byte) (tcpip.FullAddress, *syserr.Error) { case linux.AF_INET6: var a linux.SockAddrInet6 if len(addr) < sockAddrInet6Size { - return tcpip.FullAddress{}, syserr.ErrBadAddress + return tcpip.FullAddress{}, syserr.ErrInvalidArgument } binary.Unmarshal(addr[:sockAddrInet6Size], usermem.ByteOrder, &a) @@ -344,6 +343,9 @@ func GetAddress(sfamily int, addr []byte) (tcpip.FullAddress, *syserr.Error) { } return out, nil + case linux.AF_UNSPEC: + return tcpip.FullAddress{}, nil + default: return tcpip.FullAddress{}, syserr.ErrAddressFamilyNotSupported } @@ -466,7 +468,7 @@ func (s *SocketOperations) Readiness(mask waiter.EventMask) waiter.EventMask { // Connect implements the linux syscall connect(2) for sockets backed by // tpcip.Endpoint. func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error { - addr, err := GetAddress(s.family, sockaddr) + addr, err := GetAddress(s.family, sockaddr, false /* strict */) if err != nil { return err } @@ -499,7 +501,7 @@ func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo // Bind implements the linux syscall bind(2) for sockets backed by // tcpip.Endpoint. func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { - addr, err := GetAddress(s.family, sockaddr) + addr, err := GetAddress(s.family, sockaddr, true /* strict */) if err != nil { return err } @@ -537,7 +539,7 @@ func (s *SocketOperations) blockingAccept(t *kernel.Task) (tcpip.Endpoint, *wait // Accept implements the linux syscall accept(2) for sockets backed by // tcpip.Endpoint. -func (s *SocketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (kdefs.FD, interface{}, uint32, *syserr.Error) { +func (s *SocketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, interface{}, uint32, *syserr.Error) { // Issue the accept request to get the new endpoint. ep, wq, terr := s.Endpoint.Accept() if terr != nil { @@ -575,10 +577,9 @@ func (s *SocketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, } } - fdFlags := kernel.FDFlags{ + fd, e := t.NewFDFrom(0, ns, kernel.FDFlags{ CloseOnExec: flags&linux.SOCK_CLOEXEC != 0, - } - fd, e := t.FDMap().NewFDFrom(0, ns, fdFlags, t.ThreadGroup().Limits()) + }) t.Kernel().RecordSocket(ns) @@ -1924,7 +1925,7 @@ func (s *SocketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] var addr *tcpip.FullAddress if len(to) > 0 { - addrBuf, err := GetAddress(s.family, to) + addrBuf, err := GetAddress(s.family, to, true /* strict */) if err != nil { return 0, err } diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index c63f3aacf..7f69406b7 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -26,7 +26,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/socket" @@ -190,7 +189,7 @@ func (s *socketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo } // Accept implements socket.Socket.Accept. -func (s *socketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (kdefs.FD, interface{}, uint32, *syserr.Error) { +func (s *socketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, interface{}, uint32, *syserr.Error) { var peerAddr []byte var peerAddrlen uint32 var peerAddrPtr *byte @@ -236,11 +235,11 @@ func (s *socketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, } defer f.DecRef() - fdFlags := kernel.FDFlags{ + kfd, kerr := t.NewFDFrom(0, f, kernel.FDFlags{ CloseOnExec: flags&syscall.SOCK_CLOEXEC != 0, - } - kfd, kerr := t.FDMap().NewFDFrom(0, f, fdFlags, t.ThreadGroup().Limits()) + }) t.Kernel().RecordSocket(f) + return kfd, peerAddr, peerAddrlen, syserr.FromError(kerr) } diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go index ecc1e2d53..f3d6c1e9b 100644 --- a/pkg/sentry/socket/netlink/socket.go +++ b/pkg/sentry/socket/netlink/socket.go @@ -27,7 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/netlink/port" @@ -272,7 +271,7 @@ func (s *Socket) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr } // Accept implements socket.Socket.Accept. -func (s *Socket) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (kdefs.FD, interface{}, uint32, *syserr.Error) { +func (s *Socket) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, interface{}, uint32, *syserr.Error) { // Netlink sockets never support accept. return 0, nil, 0, syserr.ErrNotSupported } diff --git a/pkg/sentry/socket/rpcinet/socket.go b/pkg/sentry/socket/rpcinet/socket.go index cc7b964ea..ccaaddbfc 100644 --- a/pkg/sentry/socket/rpcinet/socket.go +++ b/pkg/sentry/socket/rpcinet/socket.go @@ -26,7 +26,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/conn" @@ -286,7 +285,7 @@ func rpcAccept(t *kernel.Task, fd uint32, peer bool) (*pb.AcceptResponse_ResultP } // Accept implements socket.Socket.Accept. -func (s *socketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (kdefs.FD, interface{}, uint32, *syserr.Error) { +func (s *socketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, interface{}, uint32, *syserr.Error) { payload, se := rpcAccept(t, s.fd, peerRequested) // Check if we need to block. @@ -336,10 +335,9 @@ func (s *socketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, }) defer file.DecRef() - fdFlags := kernel.FDFlags{ + fd, err := t.NewFDFrom(0, file, kernel.FDFlags{ CloseOnExec: flags&linux.SOCK_CLOEXEC != 0, - } - fd, err := t.FDMap().NewFDFrom(0, file, fdFlags, t.ThreadGroup().Limits()) + }) if err != nil { return 0, nil, 0, syserr.FromError(err) } diff --git a/pkg/sentry/socket/socket.go b/pkg/sentry/socket/socket.go index 933120f34..0efa58a58 100644 --- a/pkg/sentry/socket/socket.go +++ b/pkg/sentry/socket/socket.go @@ -27,7 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/usermem" @@ -53,7 +52,7 @@ type Socket interface { // Accept implements the accept4(2) linux syscall. // Returns fd, real peer address length and error. Real peer address // length is only set if len(peer) > 0. - Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (kdefs.FD, interface{}, uint32, *syserr.Error) + Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, interface{}, uint32, *syserr.Error) // Bind implements the bind(2) linux syscall. Bind(t *kernel.Task, sockaddr []byte) *syserr.Error diff --git a/pkg/sentry/socket/unix/transport/connectioned.go b/pkg/sentry/socket/unix/transport/connectioned.go index e4c416233..73d2df15d 100644 --- a/pkg/sentry/socket/unix/transport/connectioned.go +++ b/pkg/sentry/socket/unix/transport/connectioned.go @@ -143,7 +143,9 @@ func NewPair(ctx context.Context, stype linux.SockType, uid UniqueIDProvider) (E } q1 := &queue{ReaderQueue: a.Queue, WriterQueue: b.Queue, limit: initialLimit} + q1.EnableLeakCheck("transport.queue") q2 := &queue{ReaderQueue: b.Queue, WriterQueue: a.Queue, limit: initialLimit} + q2.EnableLeakCheck("transport.queue") if stype == linux.SOCK_STREAM { a.receiver = &streamQueueReceiver{queueReceiver: queueReceiver{q1}} @@ -294,12 +296,14 @@ func (e *connectionedEndpoint) BidirectionalConnect(ctx context.Context, ce Conn } readQueue := &queue{ReaderQueue: ce.WaiterQueue(), WriterQueue: ne.Queue, limit: initialLimit} + readQueue.EnableLeakCheck("transport.queue") ne.connected = &connectedEndpoint{ endpoint: ce, writeQueue: readQueue, } writeQueue := &queue{ReaderQueue: ne.Queue, WriterQueue: ce.WaiterQueue(), limit: initialLimit} + writeQueue.EnableLeakCheck("transport.queue") if e.stype == linux.SOCK_STREAM { ne.receiver = &streamQueueReceiver{queueReceiver: queueReceiver{readQueue: writeQueue}} } else { diff --git a/pkg/sentry/socket/unix/transport/connectionless.go b/pkg/sentry/socket/unix/transport/connectionless.go index e987519f0..c7f7c5b16 100644 --- a/pkg/sentry/socket/unix/transport/connectionless.go +++ b/pkg/sentry/socket/unix/transport/connectionless.go @@ -41,7 +41,9 @@ var ( // NewConnectionless creates a new unbound dgram endpoint. func NewConnectionless(ctx context.Context) Endpoint { ep := &connectionlessEndpoint{baseEndpoint{Queue: &waiter.Queue{}}} - ep.receiver = &queueReceiver{readQueue: &queue{ReaderQueue: ep.Queue, WriterQueue: &waiter.Queue{}, limit: initialLimit}} + q := queue{ReaderQueue: ep.Queue, WriterQueue: &waiter.Queue{}, limit: initialLimit} + q.EnableLeakCheck("transport.queue") + ep.receiver = &queueReceiver{readQueue: &q} return ep } @@ -52,29 +54,24 @@ func (e *connectionlessEndpoint) isBound() bool { // Close puts the endpoint in a closed state and frees all resources associated // with it. -// -// The socket will be a fresh state after a call to close and may be reused. -// That is, close may be used to "unbind" or "disconnect" the socket in error -// paths. func (e *connectionlessEndpoint) Close() { e.Lock() - var r Receiver - if e.Connected() { - e.receiver.CloseRecv() - r = e.receiver - e.receiver = nil - + if e.connected != nil { e.connected.Release() e.connected = nil } + if e.isBound() { e.path = "" } + + e.receiver.CloseRecv() + r := e.receiver + e.receiver = nil e.Unlock() - if r != nil { - r.CloseNotify() - r.Release() - } + + r.CloseNotify() + r.Release() } // BidirectionalConnect implements BoundEndpoint.BidirectionalConnect. @@ -137,6 +134,9 @@ func (e *connectionlessEndpoint) Connect(ctx context.Context, server BoundEndpoi } e.Lock() + if e.connected != nil { + e.connected.Release() + } e.connected = connected e.Unlock() diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 6190de0c5..637168714 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -28,7 +28,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/control" @@ -69,10 +68,13 @@ func New(ctx context.Context, endpoint transport.Endpoint, stype linux.SockType) // NewWithDirent creates a new unix socket using an existing dirent. func NewWithDirent(ctx context.Context, d *fs.Dirent, ep transport.Endpoint, stype linux.SockType, flags fs.FileFlags) *fs.File { - return fs.NewFile(ctx, d, flags, &SocketOperations{ + s := SocketOperations{ ep: ep, stype: stype, - }) + } + s.EnableLeakCheck("unix.SocketOperations") + + return fs.NewFile(ctx, d, flags, &s) } // DecRef implements RefCounter.DecRef. @@ -108,7 +110,7 @@ func (s *SocketOperations) Endpoint() transport.Endpoint { // extractPath extracts and validates the address. func extractPath(sockaddr []byte) (string, *syserr.Error) { - addr, err := epsocket.GetAddress(linux.AF_UNIX, sockaddr) + addr, err := epsocket.GetAddress(linux.AF_UNIX, sockaddr, true /* strict */) if err != nil { return "", err } @@ -191,7 +193,7 @@ func (s *SocketOperations) blockingAccept(t *kernel.Task) (transport.Endpoint, * // Accept implements the linux syscall accept(2) for sockets backed by // a transport.Endpoint. -func (s *SocketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (kdefs.FD, interface{}, uint32, *syserr.Error) { +func (s *SocketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, interface{}, uint32, *syserr.Error) { // Issue the accept request to get the new endpoint. ep, err := s.ep.Accept() if err != nil { @@ -226,10 +228,9 @@ func (s *SocketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, } } - fdFlags := kernel.FDFlags{ + fd, e := t.NewFDFrom(0, ns, kernel.FDFlags{ CloseOnExec: flags&linux.SOCK_CLOEXEC != 0, - } - fd, e := t.FDMap().NewFDFrom(0, ns, fdFlags, t.ThreadGroup().Limits()) + }) if e != nil { return 0, nil, 0, syserr.FromError(e) } diff --git a/pkg/sentry/strace/poll.go b/pkg/sentry/strace/poll.go index 57cf6b139..5187594a7 100644 --- a/pkg/sentry/strace/poll.go +++ b/pkg/sentry/strace/poll.go @@ -21,7 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/abi" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" "gvisor.dev/gvisor/pkg/sentry/usermem" ) @@ -50,7 +49,7 @@ func pollFD(t *kernel.Task, pfd *linux.PollFD, post bool) string { if post { revents = PollEventSet.Parse(uint64(pfd.REvents)) } - return fmt.Sprintf("{FD: %s, Events: %s, REvents: %s}", fd(t, kdefs.FD(pfd.FD)), PollEventSet.Parse(uint64(pfd.Events)), revents) + return fmt.Sprintf("{FD: %s, Events: %s, REvents: %s}", fd(t, pfd.FD), PollEventSet.Parse(uint64(pfd.Events)), revents) } func pollFDs(t *kernel.Task, addr usermem.Addr, nfds uint, post bool) string { diff --git a/pkg/sentry/strace/socket.go b/pkg/sentry/strace/socket.go index f9cf2eb21..386b40af7 100644 --- a/pkg/sentry/strace/socket.go +++ b/pkg/sentry/strace/socket.go @@ -332,7 +332,7 @@ func sockAddr(t *kernel.Task, addr usermem.Addr, length uint32) string { switch family { case linux.AF_INET, linux.AF_INET6, linux.AF_UNIX: - fa, err := epsocket.GetAddress(int(family), b) + fa, err := epsocket.GetAddress(int(family), b, true /* strict */) if err != nil { return fmt.Sprintf("%#x {Family: %s, error extracting address: %v}", addr, familyStr, err) } diff --git a/pkg/sentry/strace/strace.go b/pkg/sentry/strace/strace.go index 86e9c5690..311389547 100644 --- a/pkg/sentry/strace/strace.go +++ b/pkg/sentry/strace/strace.go @@ -31,7 +31,6 @@ import ( "gvisor.dev/gvisor/pkg/seccomp" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" pb "gvisor.dev/gvisor/pkg/sentry/strace/strace_go_proto" slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux" "gvisor.dev/gvisor/pkg/sentry/usermem" @@ -133,7 +132,7 @@ func path(t *kernel.Task, addr usermem.Addr) string { return fmt.Sprintf("%#x %s", addr, path) } -func fd(t *kernel.Task, fd kdefs.FD) string { +func fd(t *kernel.Task, fd int32) string { root := t.FSContext().RootDirectory() if root != nil { defer root.DecRef() @@ -151,7 +150,7 @@ func fd(t *kernel.Task, fd kdefs.FD) string { return fmt.Sprintf("AT_FDCWD %s", name) } - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { // Cast FD to uint64 to avoid printing negative hex. return fmt.Sprintf("%#x (bad FD)", uint64(fd)) @@ -375,7 +374,7 @@ func (i *SyscallInfo) pre(t *kernel.Task, args arch.SyscallArguments, maximumBlo } switch i.format[arg] { case FD: - output = append(output, fd(t, kdefs.FD(args[arg].Int()))) + output = append(output, fd(t, args[arg].Int())) case WriteBuffer: output = append(output, dump(t, args[arg].Pointer(), args[arg+1].SizeT(), maximumBlobSize)) case WriteIOVec: diff --git a/pkg/sentry/syscalls/epoll.go b/pkg/sentry/syscalls/epoll.go index c710ec9e3..87dcad18b 100644 --- a/pkg/sentry/syscalls/epoll.go +++ b/pkg/sentry/syscalls/epoll.go @@ -15,25 +15,23 @@ package syscalls import ( - "syscall" "time" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/epoll" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) // CreateEpoll implements the epoll_create(2) linux syscall. -func CreateEpoll(t *kernel.Task, closeOnExec bool) (kdefs.FD, error) { +func CreateEpoll(t *kernel.Task, closeOnExec bool) (int32, error) { file := epoll.NewEventPoll(t) defer file.DecRef() - flags := kernel.FDFlags{ + fd, err := t.NewFDFrom(0, file, kernel.FDFlags{ CloseOnExec: closeOnExec, - } - fd, err := t.FDMap().NewFDFrom(0, file, flags, t.ThreadGroup().Limits()) + }) if err != nil { return 0, err } @@ -42,25 +40,25 @@ func CreateEpoll(t *kernel.Task, closeOnExec bool) (kdefs.FD, error) { } // AddEpoll implements the epoll_ctl(2) linux syscall when op is EPOLL_CTL_ADD. -func AddEpoll(t *kernel.Task, epfd kdefs.FD, fd kdefs.FD, flags epoll.EntryFlags, mask waiter.EventMask, userData [2]int32) error { +func AddEpoll(t *kernel.Task, epfd int32, fd int32, flags epoll.EntryFlags, mask waiter.EventMask, userData [2]int32) error { // Get epoll from the file descriptor. - epollfile := t.FDMap().GetFile(epfd) + epollfile := t.GetFile(epfd) if epollfile == nil { - return syscall.EBADF + return syserror.EBADF } defer epollfile.DecRef() // Get the target file id. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return syscall.EBADF + return syserror.EBADF } defer file.DecRef() // Extract the epollPoll operations. e, ok := epollfile.FileOperations.(*epoll.EventPoll) if !ok { - return syscall.EBADF + return syserror.EBADF } // Try to add the entry. @@ -68,25 +66,25 @@ func AddEpoll(t *kernel.Task, epfd kdefs.FD, fd kdefs.FD, flags epoll.EntryFlags } // UpdateEpoll implements the epoll_ctl(2) linux syscall when op is EPOLL_CTL_MOD. -func UpdateEpoll(t *kernel.Task, epfd kdefs.FD, fd kdefs.FD, flags epoll.EntryFlags, mask waiter.EventMask, userData [2]int32) error { +func UpdateEpoll(t *kernel.Task, epfd int32, fd int32, flags epoll.EntryFlags, mask waiter.EventMask, userData [2]int32) error { // Get epoll from the file descriptor. - epollfile := t.FDMap().GetFile(epfd) + epollfile := t.GetFile(epfd) if epollfile == nil { - return syscall.EBADF + return syserror.EBADF } defer epollfile.DecRef() // Get the target file id. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return syscall.EBADF + return syserror.EBADF } defer file.DecRef() // Extract the epollPoll operations. e, ok := epollfile.FileOperations.(*epoll.EventPoll) if !ok { - return syscall.EBADF + return syserror.EBADF } // Try to update the entry. @@ -94,25 +92,25 @@ func UpdateEpoll(t *kernel.Task, epfd kdefs.FD, fd kdefs.FD, flags epoll.EntryFl } // RemoveEpoll implements the epoll_ctl(2) linux syscall when op is EPOLL_CTL_DEL. -func RemoveEpoll(t *kernel.Task, epfd kdefs.FD, fd kdefs.FD) error { +func RemoveEpoll(t *kernel.Task, epfd int32, fd int32) error { // Get epoll from the file descriptor. - epollfile := t.FDMap().GetFile(epfd) + epollfile := t.GetFile(epfd) if epollfile == nil { - return syscall.EBADF + return syserror.EBADF } defer epollfile.DecRef() // Get the target file id. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return syscall.EBADF + return syserror.EBADF } defer file.DecRef() // Extract the epollPoll operations. e, ok := epollfile.FileOperations.(*epoll.EventPoll) if !ok { - return syscall.EBADF + return syserror.EBADF } // Try to remove the entry. @@ -120,18 +118,18 @@ func RemoveEpoll(t *kernel.Task, epfd kdefs.FD, fd kdefs.FD) error { } // WaitEpoll implements the epoll_wait(2) linux syscall. -func WaitEpoll(t *kernel.Task, fd kdefs.FD, max int, timeout int) ([]epoll.Event, error) { +func WaitEpoll(t *kernel.Task, fd int32, max int, timeout int) ([]epoll.Event, error) { // Get epoll from the file descriptor. - epollfile := t.FDMap().GetFile(fd) + epollfile := t.GetFile(fd) if epollfile == nil { - return nil, syscall.EBADF + return nil, syserror.EBADF } defer epollfile.DecRef() // Extract the epollPoll operations. e, ok := epollfile.FileOperations.(*epoll.EventPoll) if !ok { - return nil, syscall.EBADF + return nil, syserror.EBADF } // Try to read events and return right away if we got them or if the @@ -164,7 +162,7 @@ func WaitEpoll(t *kernel.Task, fd kdefs.FD, max int, timeout int) ([]epoll.Event } if err := t.BlockWithDeadline(ch, haveDeadline, deadline); err != nil { - if err == syscall.ETIMEDOUT { + if err == syserror.ETIMEDOUT { return nil, nil } diff --git a/pkg/sentry/syscalls/linux/error.go b/pkg/sentry/syscalls/linux/error.go index ac3905c5c..264301bfa 100644 --- a/pkg/sentry/syscalls/linux/error.go +++ b/pkg/sentry/syscalls/linux/error.go @@ -17,7 +17,6 @@ package linux import ( "io" "sync" - "syscall" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/log" @@ -54,7 +53,7 @@ func handleIOError(t *kernel.Task, partialResult bool, err, intr error, op strin // Do not consume the error and return it as EFBIG. // Simultaneously send a SIGXFSZ per setrlimit(2). t.SendSignal(kernel.SignalInfoNoInfo(linux.SIGXFSZ, t, t)) - return syscall.EFBIG + return syserror.EFBIG case syserror.ErrInterrupted: // The syscall was interrupted. Return nil if it completed // partially, otherwise return the error code that the syscall diff --git a/pkg/sentry/syscalls/linux/linux64.go b/pkg/sentry/syscalls/linux/linux64.go index 7f18b1ac8..51db2d8f7 100644 --- a/pkg/sentry/syscalls/linux/linux64.go +++ b/pkg/sentry/syscalls/linux/linux64.go @@ -16,8 +16,6 @@ package linux import ( - "syscall" - "gvisor.dev/gvisor/pkg/abi" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" @@ -50,335 +48,335 @@ var AMD64 = &kernel.SyscallTable{ Table: map[uintptr]kernel.Syscall{ 0: syscalls.Supported("read", Read), 1: syscalls.Supported("write", Write), - 2: syscalls.Supported("open", Open), + 2: syscalls.PartiallySupported("open", Open, "Options O_DIRECT, O_NOATIME, O_PATH, O_TMPFILE, O_SYNC are not supported.", nil), 3: syscalls.Supported("close", Close), - 4: syscalls.Undocumented("stat", Stat), - 5: syscalls.Undocumented("fstat", Fstat), - 6: syscalls.Undocumented("lstat", Lstat), - 7: syscalls.Undocumented("poll", Poll), - 8: syscalls.Undocumented("lseek", Lseek), - 9: syscalls.Undocumented("mmap", Mmap), - 10: syscalls.Undocumented("mprotect", Mprotect), - 11: syscalls.Undocumented("munmap", Munmap), - 12: syscalls.Undocumented("brk", Brk), - 13: syscalls.Undocumented("rt_sigaction", RtSigaction), - 14: syscalls.Undocumented("rt_sigprocmask", RtSigprocmask), - 15: syscalls.Undocumented("rt_sigreturn", RtSigreturn), - 16: syscalls.Undocumented("ioctl", Ioctl), - 17: syscalls.Undocumented("pread64", Pread64), - 18: syscalls.Undocumented("pwrite64", Pwrite64), - 19: syscalls.Undocumented("readv", Readv), - 20: syscalls.Undocumented("writev", Writev), - 21: syscalls.Undocumented("access", Access), - 22: syscalls.Undocumented("pipe", Pipe), - 23: syscalls.Undocumented("select", Select), - 24: syscalls.Undocumented("sched_yield", SchedYield), - 25: syscalls.Undocumented("mremap", Mremap), - 26: syscalls.Undocumented("msync", Msync), - 27: syscalls.Undocumented("mincore", Mincore), - 28: syscalls.Undocumented("madvise", Madvise), - 29: syscalls.Undocumented("shmget", Shmget), - 30: syscalls.Undocumented("shmat", Shmat), - 31: syscalls.Undocumented("shmctl", Shmctl), - 32: syscalls.Undocumented("dup", Dup), - 33: syscalls.Undocumented("dup2", Dup2), - 34: syscalls.Undocumented("pause", Pause), - 35: syscalls.Undocumented("nanosleep", Nanosleep), - 36: syscalls.Undocumented("getitimer", Getitimer), - 37: syscalls.Undocumented("alarm", Alarm), - 38: syscalls.Undocumented("setitimer", Setitimer), - 39: syscalls.Undocumented("getpid", Getpid), - 40: syscalls.Undocumented("sendfile", Sendfile), - 41: syscalls.Undocumented("socket", Socket), - 42: syscalls.Undocumented("connect", Connect), - 43: syscalls.Undocumented("accept", Accept), - 44: syscalls.Undocumented("sendto", SendTo), - 45: syscalls.Undocumented("recvfrom", RecvFrom), - 46: syscalls.Undocumented("sendmsg", SendMsg), - 47: syscalls.Undocumented("recvmsg", RecvMsg), - 48: syscalls.Undocumented("shutdown", Shutdown), - 49: syscalls.Undocumented("bind", Bind), - 50: syscalls.Undocumented("listen", Listen), - 51: syscalls.Undocumented("getsockname", GetSockName), - 52: syscalls.Undocumented("getpeername", GetPeerName), - 53: syscalls.Undocumented("socketpair", SocketPair), - 54: syscalls.Undocumented("setsockopt", SetSockOpt), - 55: syscalls.Undocumented("getsockopt", GetSockOpt), - 56: syscalls.Undocumented("clone", Clone), - 57: syscalls.Undocumented("fork", Fork), - 58: syscalls.Undocumented("vfork", Vfork), - 59: syscalls.Undocumented("execve", Execve), - 60: syscalls.Undocumented("exit", Exit), - 61: syscalls.Undocumented("wait4", Wait4), - 62: syscalls.Undocumented("kill", Kill), - 63: syscalls.Undocumented("uname", Uname), - 64: syscalls.Undocumented("semget", Semget), - 65: syscalls.Undocumented("semop", Semop), - 66: syscalls.Undocumented("semctl", Semctl), - 67: syscalls.Undocumented("shmdt", Shmdt), - 68: syscalls.ErrorWithEvent("msgget", syscall.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) - 69: syscalls.ErrorWithEvent("msgsnd", syscall.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) - 70: syscalls.ErrorWithEvent("msgrcv", syscall.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) - 71: syscalls.ErrorWithEvent("msgctl", syscall.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) - 72: syscalls.Undocumented("fcntl", Fcntl), - 73: syscalls.Undocumented("flock", Flock), - 74: syscalls.Undocumented("fsync", Fsync), - 75: syscalls.Undocumented("fdatasync", Fdatasync), - 76: syscalls.Undocumented("truncate", Truncate), - 77: syscalls.Undocumented("ftruncate", Ftruncate), - 78: syscalls.Undocumented("getdents", Getdents), - 79: syscalls.Undocumented("getcwd", Getcwd), - 80: syscalls.Undocumented("chdir", Chdir), - 81: syscalls.Undocumented("fchdir", Fchdir), - 82: syscalls.Undocumented("rename", Rename), - 83: syscalls.Undocumented("mkdir", Mkdir), - 84: syscalls.Undocumented("rmdir", Rmdir), - 85: syscalls.Undocumented("creat", Creat), - 86: syscalls.Undocumented("link", Link), - 87: syscalls.Undocumented("link", Unlink), - 88: syscalls.Undocumented("symlink", Symlink), - 89: syscalls.Undocumented("readlink", Readlink), - 90: syscalls.Undocumented("chmod", Chmod), - 91: syscalls.Undocumented("fchmod", Fchmod), - 92: syscalls.Undocumented("chown", Chown), - 93: syscalls.Undocumented("fchown", Fchown), - 94: syscalls.Undocumented("lchown", Lchown), - 95: syscalls.Undocumented("umask", Umask), - 96: syscalls.Undocumented("gettimeofday", Gettimeofday), - 97: syscalls.Undocumented("getrlimit", Getrlimit), - 98: syscalls.Undocumented("getrusage", Getrusage), - 99: syscalls.Undocumented("sysinfo", Sysinfo), - 100: syscalls.Undocumented("times", Times), - 101: syscalls.Undocumented("ptrace", Ptrace), - 102: syscalls.Undocumented("getuid", Getuid), - 103: syscalls.Undocumented("syslog", Syslog), - 104: syscalls.Undocumented("getgid", Getgid), - 105: syscalls.Undocumented("setuid", Setuid), - 106: syscalls.Undocumented("setgid", Setgid), - 107: syscalls.Undocumented("geteuid", Geteuid), - 108: syscalls.Undocumented("getegid", Getegid), - 109: syscalls.Undocumented("setpgid", Setpgid), - 110: syscalls.Undocumented("getppid", Getppid), - 111: syscalls.Undocumented("getpgrp", Getpgrp), - 112: syscalls.Undocumented("setsid", Setsid), - 113: syscalls.Undocumented("setreuid", Setreuid), - 114: syscalls.Undocumented("setregid", Setregid), - 115: syscalls.Undocumented("getgroups", Getgroups), - 116: syscalls.Undocumented("setgroups", Setgroups), - 117: syscalls.Undocumented("setresuid", Setresuid), - 118: syscalls.Undocumented("getresuid", Getresuid), - 119: syscalls.Undocumented("setresgid", Setresgid), - 120: syscalls.Undocumented("setresgid", Getresgid), - 121: syscalls.Undocumented("getpgid", Getpgid), - 122: syscalls.ErrorWithEvent("setfsuid", syscall.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) - 123: syscalls.ErrorWithEvent("setfsgid", syscall.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) - 124: syscalls.Undocumented("getsid", Getsid), - 125: syscalls.Undocumented("capget", Capget), - 126: syscalls.Undocumented("capset", Capset), - 127: syscalls.Undocumented("rt_sigpending", RtSigpending), - 128: syscalls.Undocumented("rt_sigtimedwait", RtSigtimedwait), - 129: syscalls.Undocumented("rt_sigqueueinfo", RtSigqueueinfo), - 130: syscalls.Undocumented("rt_sigsuspend", RtSigsuspend), - 131: syscalls.Undocumented("sigaltstack", Sigaltstack), - 132: syscalls.Undocumented("utime", Utime), - 133: syscalls.Undocumented("mknod", Mknod), - 134: syscalls.Error("uselib", syscall.ENOSYS, "Obsolete", nil), - 135: syscalls.ErrorWithEvent("personality", syscall.EINVAL, "Unable to change personality.", nil), - 136: syscalls.ErrorWithEvent("ustat", syscall.ENOSYS, "Needs filesystem support.", nil), - 137: syscalls.Undocumented("statfs", Statfs), - 138: syscalls.Undocumented("fstatfs", Fstatfs), - 139: syscalls.ErrorWithEvent("sysfs", syscall.ENOSYS, "", []string{"gvisor.dev/issue/165"}), - 140: syscalls.Undocumented("getpriority", Getpriority), - 141: syscalls.Undocumented("setpriority", Setpriority), + 4: syscalls.Supported("stat", Stat), + 5: syscalls.Supported("fstat", Fstat), + 6: syscalls.Supported("lstat", Lstat), + 7: syscalls.Supported("poll", Poll), + 8: syscalls.Supported("lseek", Lseek), + 9: syscalls.PartiallySupported("mmap", Mmap, "Generally supported with exceptions. Options MAP_FIXED_NOREPLACE, MAP_SHARED_VALIDATE, MAP_SYNC MAP_GROWSDOWN, MAP_HUGETLB are not supported.", nil), + 10: syscalls.Supported("mprotect", Mprotect), + 11: syscalls.Supported("munmap", Munmap), + 12: syscalls.Supported("brk", Brk), + 13: syscalls.Supported("rt_sigaction", RtSigaction), + 14: syscalls.Supported("rt_sigprocmask", RtSigprocmask), + 15: syscalls.Supported("rt_sigreturn", RtSigreturn), + 16: syscalls.PartiallySupported("ioctl", Ioctl, "Only a few ioctls are implemented for backing devices and file systems.", nil), + 17: syscalls.Supported("pread64", Pread64), + 18: syscalls.Supported("pwrite64", Pwrite64), + 19: syscalls.Supported("readv", Readv), + 20: syscalls.Supported("writev", Writev), + 21: syscalls.Supported("access", Access), + 22: syscalls.Supported("pipe", Pipe), + 23: syscalls.Supported("select", Select), + 24: syscalls.Supported("sched_yield", SchedYield), + 25: syscalls.Supported("mremap", Mremap), + 26: syscalls.PartiallySupported("msync", Msync, "Full data flush is not guaranteed at this time.", nil), + 27: syscalls.PartiallySupported("mincore", Mincore, "Stub implementation. The sandbox does not have access to this information. Reports all mapped pages are resident.", nil), + 28: syscalls.PartiallySupported("madvise", Madvise, "Options MADV_DONTNEED, MADV_DONTFORK are supported. Other advice is ignored.", nil), + 29: syscalls.PartiallySupported("shmget", Shmget, "Option SHM_HUGETLB is not supported.", nil), + 30: syscalls.PartiallySupported("shmat", Shmat, "Option SHM_RND is not supported.", nil), + 31: syscalls.PartiallySupported("shmctl", Shmctl, "Options SHM_LOCK, SHM_UNLOCK are not supported.", nil), + 32: syscalls.Supported("dup", Dup), + 33: syscalls.Supported("dup2", Dup2), + 34: syscalls.Supported("pause", Pause), + 35: syscalls.Supported("nanosleep", Nanosleep), + 36: syscalls.Supported("getitimer", Getitimer), + 37: syscalls.Supported("alarm", Alarm), + 38: syscalls.Supported("setitimer", Setitimer), + 39: syscalls.Supported("getpid", Getpid), + 40: syscalls.Supported("sendfile", Sendfile), + 41: syscalls.PartiallySupported("socket", Socket, "Limited support for AF_NETLINK, NETLINK_ROUTE sockets. Limited support for SOCK_RAW.", nil), + 42: syscalls.Supported("connect", Connect), + 43: syscalls.Supported("accept", Accept), + 44: syscalls.Supported("sendto", SendTo), + 45: syscalls.Supported("recvfrom", RecvFrom), + 46: syscalls.Supported("sendmsg", SendMsg), + 47: syscalls.PartiallySupported("recvmsg", RecvMsg, "Not all flags and control messages are supported.", nil), + 48: syscalls.PartiallySupported("shutdown", Shutdown, "Not all flags and control messages are supported.", nil), + 49: syscalls.PartiallySupported("bind", Bind, "Autobind for abstract Unix sockets is not supported.", nil), + 50: syscalls.Supported("listen", Listen), + 51: syscalls.Supported("getsockname", GetSockName), + 52: syscalls.Supported("getpeername", GetPeerName), + 53: syscalls.Supported("socketpair", SocketPair), + 54: syscalls.PartiallySupported("setsockopt", SetSockOpt, "Not all socket options are supported.", nil), + 55: syscalls.PartiallySupported("getsockopt", GetSockOpt, "Not all socket options are supported.", nil), + 56: syscalls.PartiallySupported("clone", Clone, "Mount namespace (CLONE_NEWNS) not supported. Options CLONE_PARENT, CLONE_SYSVSEM not supported.", nil), + 57: syscalls.Supported("fork", Fork), + 58: syscalls.Supported("vfork", Vfork), + 59: syscalls.Supported("execve", Execve), + 60: syscalls.Supported("exit", Exit), + 61: syscalls.Supported("wait4", Wait4), + 62: syscalls.Supported("kill", Kill), + 63: syscalls.Supported("uname", Uname), + 64: syscalls.Supported("semget", Semget), + 65: syscalls.PartiallySupported("semop", Semop, "Option SEM_UNDO not supported.", nil), + 66: syscalls.PartiallySupported("semctl", Semctl, "Options IPC_INFO, SEM_INFO, IPC_STAT, SEM_STAT, SEM_STAT_ANY, GETNCNT, GETZCNT not supported.", nil), + 67: syscalls.Supported("shmdt", Shmdt), + 68: syscalls.ErrorWithEvent("msgget", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 69: syscalls.ErrorWithEvent("msgsnd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 70: syscalls.ErrorWithEvent("msgrcv", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 71: syscalls.ErrorWithEvent("msgctl", syserror.ENOSYS, "", []string{"gvisor.dev/issue/135"}), // TODO(b/29354921) + 72: syscalls.PartiallySupported("fcntl", Fcntl, "Not all options are supported.", nil), + 73: syscalls.PartiallySupported("flock", Flock, "Locks are held within the sandbox only.", nil), + 74: syscalls.PartiallySupported("fsync", Fsync, "Full data flush is not guaranteed at this time.", nil), + 75: syscalls.PartiallySupported("fdatasync", Fdatasync, "Full data flush is not guaranteed at this time.", nil), + 76: syscalls.Supported("truncate", Truncate), + 77: syscalls.Supported("ftruncate", Ftruncate), + 78: syscalls.Supported("getdents", Getdents), + 79: syscalls.Supported("getcwd", Getcwd), + 80: syscalls.Supported("chdir", Chdir), + 81: syscalls.Supported("fchdir", Fchdir), + 82: syscalls.Supported("rename", Rename), + 83: syscalls.Supported("mkdir", Mkdir), + 84: syscalls.Supported("rmdir", Rmdir), + 85: syscalls.Supported("creat", Creat), + 86: syscalls.Supported("link", Link), + 87: syscalls.Supported("unlink", Unlink), + 88: syscalls.Supported("symlink", Symlink), + 89: syscalls.Supported("readlink", Readlink), + 90: syscalls.Supported("chmod", Chmod), + 91: syscalls.PartiallySupported("fchmod", Fchmod, "Options S_ISUID and S_ISGID not supported.", nil), + 92: syscalls.Supported("chown", Chown), + 93: syscalls.Supported("fchown", Fchown), + 94: syscalls.Supported("lchown", Lchown), + 95: syscalls.Supported("umask", Umask), + 96: syscalls.Supported("gettimeofday", Gettimeofday), + 97: syscalls.Supported("getrlimit", Getrlimit), + 98: syscalls.PartiallySupported("getrusage", Getrusage, "Fields ru_maxrss, ru_minflt, ru_majflt, ru_inblock, ru_oublock are not supported. Fields ru_utime and ru_stime have low precision.", nil), + 99: syscalls.PartiallySupported("sysinfo", Sysinfo, "Fields loads, sharedram, bufferram, totalswap, freeswap, totalhigh, freehigh not supported.", nil), + 100: syscalls.Supported("times", Times), + 101: syscalls.PartiallySupported("ptrace", Ptrace, "Options PTRACE_PEEKSIGINFO, PTRACE_SECCOMP_GET_FILTER not supported.", nil), + 102: syscalls.Supported("getuid", Getuid), + 103: syscalls.PartiallySupported("syslog", Syslog, "Outputs a dummy message for security reasons.", nil), + 104: syscalls.Supported("getgid", Getgid), + 105: syscalls.Supported("setuid", Setuid), + 106: syscalls.Supported("setgid", Setgid), + 107: syscalls.Supported("geteuid", Geteuid), + 108: syscalls.Supported("getegid", Getegid), + 109: syscalls.Supported("setpgid", Setpgid), + 110: syscalls.Supported("getppid", Getppid), + 111: syscalls.Supported("getpgrp", Getpgrp), + 112: syscalls.Supported("setsid", Setsid), + 113: syscalls.Supported("setreuid", Setreuid), + 114: syscalls.Supported("setregid", Setregid), + 115: syscalls.Supported("getgroups", Getgroups), + 116: syscalls.Supported("setgroups", Setgroups), + 117: syscalls.Supported("setresuid", Setresuid), + 118: syscalls.Supported("getresuid", Getresuid), + 119: syscalls.Supported("setresgid", Setresgid), + 120: syscalls.Supported("getresgid", Getresgid), + 121: syscalls.Supported("getpgid", Getpgid), + 122: syscalls.ErrorWithEvent("setfsuid", syserror.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) + 123: syscalls.ErrorWithEvent("setfsgid", syserror.ENOSYS, "", []string{"gvisor.dev/issue/260"}), // TODO(b/112851702) + 124: syscalls.Supported("getsid", Getsid), + 125: syscalls.Supported("capget", Capget), + 126: syscalls.Supported("capset", Capset), + 127: syscalls.Supported("rt_sigpending", RtSigpending), + 128: syscalls.Supported("rt_sigtimedwait", RtSigtimedwait), + 129: syscalls.Supported("rt_sigqueueinfo", RtSigqueueinfo), + 130: syscalls.Supported("rt_sigsuspend", RtSigsuspend), + 131: syscalls.Supported("sigaltstack", Sigaltstack), + 132: syscalls.Supported("utime", Utime), + 133: syscalls.PartiallySupported("mknod", Mknod, "Device creation is not generally supported. Only regular file and FIFO creation are supported.", nil), + 134: syscalls.Error("uselib", syserror.ENOSYS, "Obsolete", nil), + 135: syscalls.ErrorWithEvent("personality", syserror.EINVAL, "Unable to change personality.", nil), + 136: syscalls.ErrorWithEvent("ustat", syserror.ENOSYS, "Needs filesystem support.", nil), + 137: syscalls.PartiallySupported("statfs", Statfs, "Depends on the backing file system implementation.", nil), + 138: syscalls.PartiallySupported("fstatfs", Fstatfs, "Depends on the backing file system implementation.", nil), + 139: syscalls.ErrorWithEvent("sysfs", syserror.ENOSYS, "", []string{"gvisor.dev/issue/165"}), + 140: syscalls.PartiallySupported("getpriority", Getpriority, "Stub implementation.", nil), + 141: syscalls.PartiallySupported("setpriority", Setpriority, "Stub implementation.", nil), 142: syscalls.CapError("sched_setparam", linux.CAP_SYS_NICE, "", nil), - 143: syscalls.Undocumented("sched_getparam", SchedGetparam), - 144: syscalls.Undocumented("sched_setscheduler", SchedSetscheduler), - 145: syscalls.Undocumented("sched_getscheduler", SchedGetscheduler), - 146: syscalls.Undocumented("sched_get_priority_max", SchedGetPriorityMax), - 147: syscalls.Undocumented("sched_get_priority_min", SchedGetPriorityMin), - 148: syscalls.ErrorWithEvent("sched_rr_get_interval", syscall.EPERM, "", nil), - 149: syscalls.Undocumented("mlock", Mlock), - 150: syscalls.Undocumented("munlock", Munlock), - 151: syscalls.Undocumented("mlockall", Mlockall), - 152: syscalls.Undocumented("munlockall", Munlockall), + 143: syscalls.PartiallySupported("sched_getparam", SchedGetparam, "Stub implementation.", nil), + 144: syscalls.PartiallySupported("sched_setscheduler", SchedSetscheduler, "Stub implementation.", nil), + 145: syscalls.PartiallySupported("sched_getscheduler", SchedGetscheduler, "Stub implementation.", nil), + 146: syscalls.PartiallySupported("sched_get_priority_max", SchedGetPriorityMax, "Stub implementation.", nil), + 147: syscalls.PartiallySupported("sched_get_priority_min", SchedGetPriorityMin, "Stub implementation.", nil), + 148: syscalls.ErrorWithEvent("sched_rr_get_interval", syserror.EPERM, "", nil), + 149: syscalls.PartiallySupported("mlock", Mlock, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 150: syscalls.PartiallySupported("munlock", Munlock, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 151: syscalls.PartiallySupported("mlockall", Mlockall, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + 152: syscalls.PartiallySupported("munlockall", Munlockall, "Stub implementation. The sandbox lacks appropriate permissions.", nil), 153: syscalls.CapError("vhangup", linux.CAP_SYS_TTY_CONFIG, "", nil), - 154: syscalls.Error("modify_ldt", syscall.EPERM, "", nil), - 155: syscalls.Error("pivot_root", syscall.EPERM, "", nil), - 156: syscalls.Error("sysctl", syscall.EPERM, `syscall is "worthless"`, nil), - 157: syscalls.Undocumented("prctl", Prctl), - 158: syscalls.Undocumented("arch_prctl", ArchPrctl), + 154: syscalls.Error("modify_ldt", syserror.EPERM, "", nil), + 155: syscalls.Error("pivot_root", syserror.EPERM, "", nil), + 156: syscalls.Error("sysctl", syserror.EPERM, "Deprecated. Use /proc/sys instead.", nil), + 157: syscalls.PartiallySupported("prctl", Prctl, "Not all options are supported.", nil), + 158: syscalls.PartiallySupported("arch_prctl", ArchPrctl, "Options ARCH_GET_GS, ARCH_SET_GS not supported.", nil), 159: syscalls.CapError("adjtimex", linux.CAP_SYS_TIME, "", nil), - 160: syscalls.Undocumented("setrlimit", Setrlimit), - 161: syscalls.Undocumented("chroot", Chroot), - 162: syscalls.Undocumented("sync", Sync), + 160: syscalls.PartiallySupported("setrlimit", Setrlimit, "Not all rlimits are enforced.", nil), + 161: syscalls.Supported("chroot", Chroot), + 162: syscalls.PartiallySupported("sync", Sync, "Full data flush is not guaranteed at this time.", nil), 163: syscalls.CapError("acct", linux.CAP_SYS_PACCT, "", nil), 164: syscalls.CapError("settimeofday", linux.CAP_SYS_TIME, "", nil), - 165: syscalls.Undocumented("mount", Mount), - 166: syscalls.Undocumented("umount2", Umount2), + 165: syscalls.PartiallySupported("mount", Mount, "Not all options or file systems are supported.", nil), + 166: syscalls.PartiallySupported("umount2", Umount2, "Not all options or file systems are supported.", nil), 167: syscalls.CapError("swapon", linux.CAP_SYS_ADMIN, "", nil), 168: syscalls.CapError("swapoff", linux.CAP_SYS_ADMIN, "", nil), 169: syscalls.CapError("reboot", linux.CAP_SYS_BOOT, "", nil), - 170: syscalls.Undocumented("sethostname", Sethostname), - 171: syscalls.Undocumented("setdomainname", Setdomainname), + 170: syscalls.Supported("sethostname", Sethostname), + 171: syscalls.Supported("setdomainname", Setdomainname), 172: syscalls.CapError("iopl", linux.CAP_SYS_RAWIO, "", nil), 173: syscalls.CapError("ioperm", linux.CAP_SYS_RAWIO, "", nil), 174: syscalls.CapError("create_module", linux.CAP_SYS_MODULE, "", nil), 175: syscalls.CapError("init_module", linux.CAP_SYS_MODULE, "", nil), 176: syscalls.CapError("delete_module", linux.CAP_SYS_MODULE, "", nil), - 177: syscalls.Error("get_kernel_syms", syscall.ENOSYS, "Not supported in > 2.6", nil), - 178: syscalls.Error("query_module", syscall.ENOSYS, "Not supported in > 2.6", nil), + 177: syscalls.Error("get_kernel_syms", syserror.ENOSYS, "Not supported in Linux > 2.6.", nil), + 178: syscalls.Error("query_module", syserror.ENOSYS, "Not supported in Linux > 2.6.", nil), 179: syscalls.CapError("quotactl", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_admin for most operations - 180: syscalls.Error("nfsservctl", syscall.ENOSYS, "Does not exist > 3.1", nil), - 181: syscalls.Error("getpmsg", syscall.ENOSYS, "Not implemented in Linux", nil), - 182: syscalls.Error("putpmsg", syscall.ENOSYS, "Not implemented in Linux", nil), - 183: syscalls.Error("afs_syscall", syscall.ENOSYS, "Not implemented in Linux", nil), - 184: syscalls.Error("tuxcall", syscall.ENOSYS, "Not implemented in Linux", nil), - 185: syscalls.Error("security", syscall.ENOSYS, "Not implemented in Linux", nil), - 186: syscalls.Undocumented("gettid", Gettid), - 187: syscalls.ErrorWithEvent("readahead", syscall.ENOSYS, "", []string{"gvisor.dev/issue/261"}), // TODO(b/29351341) - 188: syscalls.ErrorWithEvent("setxattr", syscall.ENOTSUP, "Requires filesystem support", nil), - 189: syscalls.ErrorWithEvent("lsetxattr", syscall.ENOTSUP, "Requires filesystem support", nil), - 190: syscalls.ErrorWithEvent("fsetxattr", syscall.ENOTSUP, "Requires filesystem support", nil), - 191: syscalls.ErrorWithEvent("getxattr", syscall.ENOTSUP, "Requires filesystem support", nil), - 192: syscalls.ErrorWithEvent("lgetxattr", syscall.ENOTSUP, "Requires filesystem support", nil), - 193: syscalls.ErrorWithEvent("fgetxattr", syscall.ENOTSUP, "Requires filesystem support", nil), - 194: syscalls.ErrorWithEvent("listxattr", syscall.ENOTSUP, "Requires filesystem support", nil), - 195: syscalls.ErrorWithEvent("llistxattr", syscall.ENOTSUP, "Requires filesystem support", nil), - 196: syscalls.ErrorWithEvent("flistxattr", syscall.ENOTSUP, "Requires filesystem support", nil), - 197: syscalls.ErrorWithEvent("removexattr", syscall.ENOTSUP, "Requires filesystem support", nil), - 198: syscalls.ErrorWithEvent("lremovexattr", syscall.ENOTSUP, "Requires filesystem support", nil), - 199: syscalls.ErrorWithEvent("fremovexattr", syscall.ENOTSUP, "Requires filesystem support", nil), - 200: syscalls.Undocumented("tkill", Tkill), - 201: syscalls.Undocumented("time", Time), - 202: syscalls.Undocumented("futex", Futex), - 203: syscalls.Undocumented("sched_setaffinity", SchedSetaffinity), - 204: syscalls.Undocumented("sched_getaffinity", SchedGetaffinity), - 205: syscalls.Error("set_thread_area", syscall.ENOSYS, "Expected to return ENOSYS on 64-bit", nil), - 206: syscalls.Undocumented("io_setup", IoSetup), - 207: syscalls.Undocumented("io_destroy", IoDestroy), - 208: syscalls.Undocumented("io_getevents", IoGetevents), - 209: syscalls.Undocumented("io_submit", IoSubmit), - 210: syscalls.Undocumented("io_cancel", IoCancel), - 211: syscalls.Error("get_thread_area", syscall.ENOSYS, "Expected to return ENOSYS on 64-bit", nil), + 180: syscalls.Error("nfsservctl", syserror.ENOSYS, "Removed after Linux 3.1.", nil), + 181: syscalls.Error("getpmsg", syserror.ENOSYS, "Not implemented in Linux.", nil), + 182: syscalls.Error("putpmsg", syserror.ENOSYS, "Not implemented in Linux.", nil), + 183: syscalls.Error("afs_syscall", syserror.ENOSYS, "Not implemented in Linux.", nil), + 184: syscalls.Error("tuxcall", syserror.ENOSYS, "Not implemented in Linux.", nil), + 185: syscalls.Error("security", syserror.ENOSYS, "Not implemented in Linux.", nil), + 186: syscalls.Supported("gettid", Gettid), + 187: syscalls.ErrorWithEvent("readahead", syserror.ENOSYS, "", []string{"gvisor.dev/issue/261"}), // TODO(b/29351341) + 188: syscalls.Error("setxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 189: syscalls.Error("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 190: syscalls.Error("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 191: syscalls.ErrorWithEvent("getxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 192: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 193: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 194: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 195: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 196: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 197: syscalls.ErrorWithEvent("removexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 198: syscalls.ErrorWithEvent("lremovexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 199: syscalls.ErrorWithEvent("fremovexattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 200: syscalls.Supported("tkill", Tkill), + 201: syscalls.Supported("time", Time), + 202: syscalls.PartiallySupported("futex", Futex, "Robust futexes not supported.", nil), + 203: syscalls.PartiallySupported("sched_setaffinity", SchedSetaffinity, "Stub implementation.", nil), + 204: syscalls.PartiallySupported("sched_getaffinity", SchedGetaffinity, "Stub implementation.", nil), + 205: syscalls.Error("set_thread_area", syserror.ENOSYS, "Expected to return ENOSYS on 64-bit", nil), + 206: syscalls.PartiallySupported("io_setup", IoSetup, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 207: syscalls.PartiallySupported("io_destroy", IoDestroy, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 208: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 209: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 210: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), + 211: syscalls.Error("get_thread_area", syserror.ENOSYS, "Expected to return ENOSYS on 64-bit", nil), 212: syscalls.CapError("lookup_dcookie", linux.CAP_SYS_ADMIN, "", nil), - 213: syscalls.Undocumented("epoll_create", EpollCreate), - 214: syscalls.ErrorWithEvent("epoll_ctl_old", syscall.ENOSYS, "Deprecated", nil), - 215: syscalls.ErrorWithEvent("epoll_wait_old", syscall.ENOSYS, "Deprecated", nil), - 216: syscalls.ErrorWithEvent("remap_file_pages", syscall.ENOSYS, "Deprecated since 3.16", nil), - 217: syscalls.Undocumented("getdents64", Getdents64), - 218: syscalls.Undocumented("set_tid_address", SetTidAddress), - 219: syscalls.Undocumented("restart_syscall", RestartSyscall), - 220: syscalls.ErrorWithEvent("semtimedop", syscall.ENOSYS, "", []string{"gvisor.dev/issue/137"}), // TODO(b/29354920) - 221: syscalls.Undocumented("fadvise64", Fadvise64), - 222: syscalls.Undocumented("timer_create", TimerCreate), - 223: syscalls.Undocumented("timer_settime", TimerSettime), - 224: syscalls.Undocumented("timer_gettime", TimerGettime), - 225: syscalls.Undocumented("timer_getoverrun", TimerGetoverrun), - 226: syscalls.Undocumented("timer_delete", TimerDelete), - 227: syscalls.Undocumented("clock_settime", ClockSettime), - 228: syscalls.Undocumented("clock_gettime", ClockGettime), - 229: syscalls.Undocumented("clock_getres", ClockGetres), - 230: syscalls.Undocumented("clock_nanosleep", ClockNanosleep), - 231: syscalls.Undocumented("exit_group", ExitGroup), - 232: syscalls.Undocumented("epoll_wait", EpollWait), - 233: syscalls.Undocumented("epoll_ctl", EpollCtl), - 234: syscalls.Undocumented("tgkill", Tgkill), - 235: syscalls.Undocumented("utimes", Utimes), - 236: syscalls.Error("vserver", syscall.ENOSYS, "Not implemented by Linux", nil), + 213: syscalls.Supported("epoll_create", EpollCreate), + 214: syscalls.ErrorWithEvent("epoll_ctl_old", syserror.ENOSYS, "Deprecated.", nil), + 215: syscalls.ErrorWithEvent("epoll_wait_old", syserror.ENOSYS, "Deprecated.", nil), + 216: syscalls.ErrorWithEvent("remap_file_pages", syserror.ENOSYS, "Deprecated since Linux 3.16.", nil), + 217: syscalls.Supported("getdents64", Getdents64), + 218: syscalls.Supported("set_tid_address", SetTidAddress), + 219: syscalls.Supported("restart_syscall", RestartSyscall), + 220: syscalls.ErrorWithEvent("semtimedop", syserror.ENOSYS, "", []string{"gvisor.dev/issue/137"}), // TODO(b/29354920) + 221: syscalls.PartiallySupported("fadvise64", Fadvise64, "Not all options are supported.", nil), + 222: syscalls.Supported("timer_create", TimerCreate), + 223: syscalls.Supported("timer_settime", TimerSettime), + 224: syscalls.Supported("timer_gettime", TimerGettime), + 225: syscalls.Supported("timer_getoverrun", TimerGetoverrun), + 226: syscalls.Supported("timer_delete", TimerDelete), + 227: syscalls.Supported("clock_settime", ClockSettime), + 228: syscalls.Supported("clock_gettime", ClockGettime), + 229: syscalls.Supported("clock_getres", ClockGetres), + 230: syscalls.Supported("clock_nanosleep", ClockNanosleep), + 231: syscalls.Supported("exit_group", ExitGroup), + 232: syscalls.Supported("epoll_wait", EpollWait), + 233: syscalls.Supported("epoll_ctl", EpollCtl), + 234: syscalls.Supported("tgkill", Tgkill), + 235: syscalls.Supported("utimes", Utimes), + 236: syscalls.Error("vserver", syserror.ENOSYS, "Not implemented by Linux", nil), 237: syscalls.PartiallySupported("mbind", Mbind, "Stub implementation. Only a single NUMA node is advertised, and mempolicy is ignored accordingly, but mbind() will succeed and has effects reflected by get_mempolicy.", []string{"gvisor.dev/issue/262"}), - 238: syscalls.Undocumented("set_mempolicy", SetMempolicy), - 239: syscalls.Undocumented("get_mempolicy", GetMempolicy), - 240: syscalls.ErrorWithEvent("mq_open", syscall.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 241: syscalls.ErrorWithEvent("mq_unlink", syscall.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 242: syscalls.ErrorWithEvent("mq_timedsend", syscall.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 243: syscalls.ErrorWithEvent("mq_timedreceive", syscall.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 244: syscalls.ErrorWithEvent("mq_notify", syscall.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) - 245: syscalls.ErrorWithEvent("mq_getsetattr", syscall.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 238: syscalls.PartiallySupported("set_mempolicy", SetMempolicy, "Stub implementation.", nil), + 239: syscalls.PartiallySupported("get_mempolicy", GetMempolicy, "Stub implementation.", nil), + 240: syscalls.ErrorWithEvent("mq_open", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 241: syscalls.ErrorWithEvent("mq_unlink", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 242: syscalls.ErrorWithEvent("mq_timedsend", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 243: syscalls.ErrorWithEvent("mq_timedreceive", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 244: syscalls.ErrorWithEvent("mq_notify", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) + 245: syscalls.ErrorWithEvent("mq_getsetattr", syserror.ENOSYS, "", []string{"gvisor.dev/issue/136"}), // TODO(b/29354921) 246: syscalls.CapError("kexec_load", linux.CAP_SYS_BOOT, "", nil), - 247: syscalls.Undocumented("waitid", Waitid), - 248: syscalls.Error("add_key", syscall.EACCES, "Not available to user", nil), - 249: syscalls.Error("request_key", syscall.EACCES, "Not available to user", nil), - 250: syscalls.Error("keyctl", syscall.EACCES, "Not available to user", nil), + 247: syscalls.Supported("waitid", Waitid), + 248: syscalls.Error("add_key", syserror.EACCES, "Not available to user.", nil), + 249: syscalls.Error("request_key", syserror.EACCES, "Not available to user.", nil), + 250: syscalls.Error("keyctl", syserror.EACCES, "Not available to user.", nil), 251: syscalls.CapError("ioprio_set", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) 252: syscalls.CapError("ioprio_get", linux.CAP_SYS_ADMIN, "", nil), // requires cap_sys_nice or cap_sys_admin (depending) - 253: syscalls.Undocumented("inotify_init", InotifyInit), - 254: syscalls.Undocumented("inotify_add_watch", InotifyAddWatch), - 255: syscalls.Undocumented("inotify_rm_watch", InotifyRmWatch), + 253: syscalls.PartiallySupported("inotify_init", InotifyInit, "inotify events are only available inside the sandbox.", nil), + 254: syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "inotify events are only available inside the sandbox.", nil), + 255: syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "inotify events are only available inside the sandbox.", nil), 256: syscalls.CapError("migrate_pages", linux.CAP_SYS_NICE, "", nil), - 257: syscalls.Undocumented("openat", Openat), - 258: syscalls.Undocumented("mkdirat", Mkdirat), - 259: syscalls.Undocumented("mknodat", Mknodat), - 260: syscalls.Undocumented("fchownat", Fchownat), - 261: syscalls.Undocumented("futimesat", Futimesat), - 262: syscalls.Undocumented("fstatat", Fstatat), - 263: syscalls.Undocumented("unlinkat", Unlinkat), - 264: syscalls.Undocumented("renameat", Renameat), - 265: syscalls.Undocumented("linkat", Linkat), - 266: syscalls.Undocumented("symlinkat", Symlinkat), - 267: syscalls.Undocumented("readlinkat", Readlinkat), - 268: syscalls.Undocumented("fchmodat", Fchmodat), - 269: syscalls.Undocumented("faccessat", Faccessat), - 270: syscalls.Undocumented("pselect", Pselect), - 271: syscalls.Undocumented("ppoll", Ppoll), - 272: syscalls.Undocumented("unshare", Unshare), - 273: syscalls.Error("set_robust_list", syscall.ENOSYS, "Obsolete", nil), - 274: syscalls.Error("get_robust_list", syscall.ENOSYS, "Obsolete", nil), - 275: syscalls.PartiallySupported("splice", Splice, "Stub implementation", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) - 276: syscalls.ErrorWithEvent("tee", syscall.ENOSYS, "", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) - 277: syscalls.Undocumented("sync_file_range", SyncFileRange), - 278: syscalls.ErrorWithEvent("vmsplice", syscall.ENOSYS, "", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) - 279: syscalls.CapError("move_pages", linux.CAP_SYS_NICE, "", nil), // requires cap_sys_nice (mostly) - 280: syscalls.Undocumented("utimensat", Utimensat), - 281: syscalls.Undocumented("epoll_pwait", EpollPwait), - 282: syscalls.ErrorWithEvent("signalfd", syscall.ENOSYS, "", []string{"gvisor.dev/issue/139"}), // TODO(b/19846426) - 283: syscalls.Undocumented("timerfd_create", TimerfdCreate), - 284: syscalls.Undocumented("eventfd", Eventfd), - 285: syscalls.Undocumented("fallocate", Fallocate), - 286: syscalls.Undocumented("timerfd_settime", TimerfdSettime), - 287: syscalls.Undocumented("timerfd_gettime", TimerfdGettime), - 288: syscalls.Undocumented("accept4", Accept4), - 289: syscalls.ErrorWithEvent("signalfd4", syscall.ENOSYS, "", []string{"gvisor.dev/issue/139"}), // TODO(b/19846426) - 290: syscalls.Undocumented("eventfd2", Eventfd2), - 291: syscalls.Undocumented("epoll_create1", EpollCreate1), - 292: syscalls.Undocumented("dup3", Dup3), - 293: syscalls.Undocumented("pipe2", Pipe2), - 294: syscalls.Undocumented("inotify_init1", InotifyInit1), - 295: syscalls.Undocumented("preadv", Preadv), - 296: syscalls.Undocumented("pwritev", Pwritev), - 297: syscalls.Undocumented("rt_tgsigqueueinfo", RtTgsigqueueinfo), - 298: syscalls.ErrorWithEvent("perf_event_open", syscall.ENODEV, "No support for perf counters", nil), - 299: syscalls.Undocumented("recvmmsg", RecvMMsg), - 300: syscalls.ErrorWithEvent("fanotify_init", syscall.ENOSYS, "Needs CONFIG_FANOTIFY", nil), - 301: syscalls.ErrorWithEvent("fanotify_mark", syscall.ENOSYS, "Needs CONFIG_FANOTIFY", nil), - 302: syscalls.Undocumented("prlimit64", Prlimit64), - 303: syscalls.ErrorWithEvent("name_to_handle_at", syscall.EOPNOTSUPP, "Needs filesystem support", nil), - 304: syscalls.ErrorWithEvent("open_by_handle_at", syscall.EOPNOTSUPP, "Needs filesystem support", nil), + 257: syscalls.Supported("openat", Openat), + 258: syscalls.Supported("mkdirat", Mkdirat), + 259: syscalls.Supported("mknodat", Mknodat), + 260: syscalls.Supported("fchownat", Fchownat), + 261: syscalls.Supported("futimesat", Futimesat), + 262: syscalls.Supported("fstatat", Fstatat), + 263: syscalls.Supported("unlinkat", Unlinkat), + 264: syscalls.Supported("renameat", Renameat), + 265: syscalls.Supported("linkat", Linkat), + 266: syscalls.Supported("symlinkat", Symlinkat), + 267: syscalls.Supported("readlinkat", Readlinkat), + 268: syscalls.Supported("fchmodat", Fchmodat), + 269: syscalls.Supported("faccessat", Faccessat), + 270: syscalls.Supported("pselect", Pselect), + 271: syscalls.Supported("ppoll", Ppoll), + 272: syscalls.PartiallySupported("unshare", Unshare, "Mount, cgroup namespaces not supported. Network namespaces supported but must be empty.", nil), + 273: syscalls.Error("set_robust_list", syserror.ENOSYS, "Obsolete.", nil), + 274: syscalls.Error("get_robust_list", syserror.ENOSYS, "Obsolete.", nil), + 275: syscalls.PartiallySupported("splice", Splice, "Stub implementation.", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) + 276: syscalls.ErrorWithEvent("tee", syserror.ENOSYS, "", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) + 277: syscalls.PartiallySupported("sync_file_range", SyncFileRange, "Full data flush is not guaranteed at this time.", nil), + 278: syscalls.ErrorWithEvent("vmsplice", syserror.ENOSYS, "", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) + 279: syscalls.CapError("move_pages", linux.CAP_SYS_NICE, "", nil), // requires cap_sys_nice (mostly) + 280: syscalls.Supported("utimensat", Utimensat), + 281: syscalls.Supported("epoll_pwait", EpollPwait), + 282: syscalls.ErrorWithEvent("signalfd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/139"}), // TODO(b/19846426) + 283: syscalls.Supported("timerfd_create", TimerfdCreate), + 284: syscalls.Supported("eventfd", Eventfd), + 285: syscalls.PartiallySupported("fallocate", Fallocate, "Not all options are supported.", nil), + 286: syscalls.Supported("timerfd_settime", TimerfdSettime), + 287: syscalls.Supported("timerfd_gettime", TimerfdGettime), + 288: syscalls.Supported("accept4", Accept4), + 289: syscalls.ErrorWithEvent("signalfd4", syserror.ENOSYS, "", []string{"gvisor.dev/issue/139"}), // TODO(b/19846426) + 290: syscalls.Supported("eventfd2", Eventfd2), + 291: syscalls.Supported("epoll_create1", EpollCreate1), + 292: syscalls.Supported("dup3", Dup3), + 293: syscalls.Supported("pipe2", Pipe2), + 294: syscalls.Supported("inotify_init1", InotifyInit1), + 295: syscalls.Supported("preadv", Preadv), + 296: syscalls.Supported("pwritev", Pwritev), + 297: syscalls.Supported("rt_tgsigqueueinfo", RtTgsigqueueinfo), + 298: syscalls.ErrorWithEvent("perf_event_open", syserror.ENODEV, "No support for perf counters", nil), + 299: syscalls.PartiallySupported("recvmmsg", RecvMMsg, "Not all flags and control messages are supported.", nil), + 300: syscalls.ErrorWithEvent("fanotify_init", syserror.ENOSYS, "Needs CONFIG_FANOTIFY", nil), + 301: syscalls.ErrorWithEvent("fanotify_mark", syserror.ENOSYS, "Needs CONFIG_FANOTIFY", nil), + 302: syscalls.Supported("prlimit64", Prlimit64), + 303: syscalls.Error("name_to_handle_at", syserror.EOPNOTSUPP, "Not supported by gVisor filesystems", nil), + 304: syscalls.Error("open_by_handle_at", syserror.EOPNOTSUPP, "Not supported by gVisor filesystems", nil), 305: syscalls.CapError("clock_adjtime", linux.CAP_SYS_TIME, "", nil), - 306: syscalls.Undocumented("syncfs", Syncfs), - 307: syscalls.Undocumented("sendmmsg", SendMMsg), - 308: syscalls.ErrorWithEvent("setns", syscall.EOPNOTSUPP, "Needs filesystem support", []string{"gvisor.dev/issue/140"}), // TODO(b/29354995) - 309: syscalls.Undocumented("getcpu", Getcpu), - 310: syscalls.ErrorWithEvent("process_vm_readv", syscall.ENOSYS, "", []string{"gvisor.dev/issue/158"}), - 311: syscalls.ErrorWithEvent("process_vm_writev", syscall.ENOSYS, "", []string{"gvisor.dev/issue/158"}), + 306: syscalls.PartiallySupported("syncfs", Syncfs, "Depends on backing file system.", nil), + 307: syscalls.PartiallySupported("sendmmsg", SendMMsg, "Not all flags and control messages are supported.", nil), + 308: syscalls.ErrorWithEvent("setns", syserror.EOPNOTSUPP, "Needs filesystem support", []string{"gvisor.dev/issue/140"}), // TODO(b/29354995) + 309: syscalls.Supported("getcpu", Getcpu), + 310: syscalls.ErrorWithEvent("process_vm_readv", syserror.ENOSYS, "", []string{"gvisor.dev/issue/158"}), + 311: syscalls.ErrorWithEvent("process_vm_writev", syserror.ENOSYS, "", []string{"gvisor.dev/issue/158"}), 312: syscalls.CapError("kcmp", linux.CAP_SYS_PTRACE, "", nil), 313: syscalls.CapError("finit_module", linux.CAP_SYS_MODULE, "", nil), - 314: syscalls.ErrorWithEvent("sched_setattr", syscall.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) - 315: syscalls.ErrorWithEvent("sched_getattr", syscall.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) - 316: syscalls.ErrorWithEvent("renameat2", syscall.ENOSYS, "", []string{"gvisor.dev/issue/263"}), // TODO(b/118902772) - 317: syscalls.Undocumented("seccomp", Seccomp), - 318: syscalls.Undocumented("getrandom", GetRandom), - 319: syscalls.Undocumented("memfd_create", MemfdCreate), + 314: syscalls.ErrorWithEvent("sched_setattr", syserror.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) + 315: syscalls.ErrorWithEvent("sched_getattr", syserror.ENOSYS, "gVisor does not implement a scheduler.", []string{"gvisor.dev/issue/264"}), // TODO(b/118902272) + 316: syscalls.ErrorWithEvent("renameat2", syserror.ENOSYS, "", []string{"gvisor.dev/issue/263"}), // TODO(b/118902772) + 317: syscalls.Supported("seccomp", Seccomp), + 318: syscalls.Supported("getrandom", GetRandom), + 319: syscalls.Supported("memfd_create", MemfdCreate), 320: syscalls.CapError("kexec_file_load", linux.CAP_SYS_BOOT, "", nil), 321: syscalls.CapError("bpf", linux.CAP_SYS_ADMIN, "", nil), - 322: syscalls.ErrorWithEvent("execveat", syscall.ENOSYS, "", []string{"gvisor.dev/issue/265"}), // TODO(b/118901836) - 323: syscalls.ErrorWithEvent("userfaultfd", syscall.ENOSYS, "", []string{"gvisor.dev/issue/266"}), // TODO(b/118906345) - 324: syscalls.ErrorWithEvent("membarrier", syscall.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(b/118904897) - 325: syscalls.Undocumented("mlock2", Mlock2), + 322: syscalls.ErrorWithEvent("execveat", syserror.ENOSYS, "", []string{"gvisor.dev/issue/265"}), // TODO(b/118901836) + 323: syscalls.ErrorWithEvent("userfaultfd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/266"}), // TODO(b/118906345) + 324: syscalls.ErrorWithEvent("membarrier", syserror.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(b/118904897) + 325: syscalls.PartiallySupported("mlock2", Mlock2, "Stub implementation. The sandbox lacks appropriate permissions.", nil), // Syscalls after 325 are "backports" from versions of Linux after 4.4. - 326: syscalls.ErrorWithEvent("copy_file_range", syscall.ENOSYS, "", nil), - 327: syscalls.Undocumented("preadv2", Preadv2), - 328: syscalls.Undocumented("pwritev2", Pwritev2), + 326: syscalls.ErrorWithEvent("copy_file_range", syserror.ENOSYS, "", nil), + 327: syscalls.Supported("preadv2", Preadv2), + 328: syscalls.PartiallySupported("pwritev2", Pwritev2, "Flag RWF_HIPRI is not supported.", nil), 332: syscalls.Supported("statx", Statx), }, diff --git a/pkg/sentry/syscalls/linux/sigset.go b/pkg/sentry/syscalls/linux/sigset.go index 00b7e7cf2..333013d8c 100644 --- a/pkg/sentry/syscalls/linux/sigset.go +++ b/pkg/sentry/syscalls/linux/sigset.go @@ -15,8 +15,6 @@ package linux import ( - "syscall" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/usermem" @@ -27,7 +25,7 @@ import ( // STOP are clear. func copyInSigSet(t *kernel.Task, sigSetAddr usermem.Addr, size uint) (linux.SignalSet, error) { if size != linux.SignalSetSize { - return 0, syscall.EINVAL + return 0, syserror.EINVAL } b := t.CopyScratchBuffer(8) if _, err := t.CopyInBytes(sigSetAddr, b); err != nil { diff --git a/pkg/sentry/syscalls/linux/sys_aio.go b/pkg/sentry/syscalls/linux/sys_aio.go index 7081d1a45..f56411bfe 100644 --- a/pkg/sentry/syscalls/linux/sys_aio.go +++ b/pkg/sentry/syscalls/linux/sys_aio.go @@ -21,7 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/eventfd" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/mm" "gvisor.dev/gvisor/pkg/sentry/usermem" @@ -55,7 +54,7 @@ type ioCallback struct { OpCode uint16 ReqPrio int16 - FD uint32 + FD int32 Buf uint64 Bytes uint64 @@ -65,7 +64,7 @@ type ioCallback struct { Flags uint32 // eventfd to signal if IOCB_FLAG_RESFD is set in flags. - ResFD uint32 + ResFD int32 } // ioEvent describes an I/O result. @@ -292,7 +291,7 @@ func performCallback(t *kernel.Task, file *fs.File, cbAddr usermem.Addr, cb *ioC // submitCallback processes a single callback. func submitCallback(t *kernel.Task, id uint64, cb *ioCallback, cbAddr usermem.Addr) error { - file := t.FDMap().GetFile(kdefs.FD(cb.FD)) + file := t.GetFile(cb.FD) if file == nil { // File not found. return syserror.EBADF @@ -302,7 +301,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *ioCallback, cbAddr usermem.Ad // Was there an eventFD? Extract it. var eventFile *fs.File if cb.Flags&_IOCB_FLAG_RESFD != 0 { - eventFile = t.FDMap().GetFile(kdefs.FD(cb.ResFD)) + eventFile = t.GetFile(cb.ResFD) if eventFile == nil { // Bad FD. return syserror.EBADF diff --git a/pkg/sentry/syscalls/linux/sys_epoll.go b/pkg/sentry/syscalls/linux/sys_epoll.go index 14a61cfa5..4a2b9f061 100644 --- a/pkg/sentry/syscalls/linux/sys_epoll.go +++ b/pkg/sentry/syscalls/linux/sys_epoll.go @@ -15,12 +15,10 @@ package linux import ( - "syscall" - + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/epoll" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" "gvisor.dev/gvisor/pkg/sentry/syscalls" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" @@ -30,11 +28,11 @@ import ( // EpollCreate1 implements the epoll_create1(2) linux syscall. func EpollCreate1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { flags := args[0].Int() - if flags & ^syscall.EPOLL_CLOEXEC != 0 { + if flags & ^linux.EPOLL_CLOEXEC != 0 { return 0, nil, syserror.EINVAL } - closeOnExec := flags&syscall.EPOLL_CLOEXEC != 0 + closeOnExec := flags&linux.EPOLL_CLOEXEC != 0 fd, err := syscalls.CreateEpoll(t, closeOnExec) if err != nil { return 0, nil, err @@ -61,46 +59,43 @@ func EpollCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S // EpollCtl implements the epoll_ctl(2) linux syscall. func EpollCtl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - epfd := kdefs.FD(args[0].Int()) + epfd := args[0].Int() op := args[1].Int() - fd := kdefs.FD(args[2].Int()) + fd := args[2].Int() eventAddr := args[3].Pointer() // Capture the event state if needed. flags := epoll.EntryFlags(0) mask := waiter.EventMask(0) var data [2]int32 - if op != syscall.EPOLL_CTL_DEL { - var e syscall.EpollEvent + if op != linux.EPOLL_CTL_DEL { + var e linux.EpollEvent if _, err := t.CopyIn(eventAddr, &e); err != nil { return 0, nil, err } - if e.Events&syscall.EPOLLONESHOT != 0 { + if e.Events&linux.EPOLLONESHOT != 0 { flags |= epoll.OneShot } - // syscall.EPOLLET is incorrectly generated as a negative number - // in Go, see https://github.com/golang/go/issues/5328 for - // details. - if e.Events&-syscall.EPOLLET != 0 { + if e.Events&linux.EPOLLET != 0 { flags |= epoll.EdgeTriggered } mask = waiter.EventMaskFromLinux(e.Events) data[0] = e.Fd - data[1] = e.Pad + data[1] = e.Data } // Perform the requested operations. switch op { - case syscall.EPOLL_CTL_ADD: + case linux.EPOLL_CTL_ADD: // See fs/eventpoll.c. mask |= waiter.EventHUp | waiter.EventErr return 0, nil, syscalls.AddEpoll(t, epfd, fd, flags, mask, data) - case syscall.EPOLL_CTL_DEL: + case linux.EPOLL_CTL_DEL: return 0, nil, syscalls.RemoveEpoll(t, epfd, fd) - case syscall.EPOLL_CTL_MOD: + case linux.EPOLL_CTL_MOD: // Same as EPOLL_CTL_ADD. mask |= waiter.EventHUp | waiter.EventErr return 0, nil, syscalls.UpdateEpoll(t, epfd, fd, flags, mask, data) @@ -132,7 +127,7 @@ func copyOutEvents(t *kernel.Task, addr usermem.Addr, e []epoll.Event) error { // EpollWait implements the epoll_wait(2) linux syscall. func EpollWait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - epfd := kdefs.FD(args[0].Int()) + epfd := args[0].Int() eventsAddr := args[1].Pointer() maxEvents := int(args[2].Int()) timeout := int(args[3].Int()) diff --git a/pkg/sentry/syscalls/linux/sys_eventfd.go b/pkg/sentry/syscalls/linux/sys_eventfd.go index 7dbe84884..8a34c4e99 100644 --- a/pkg/sentry/syscalls/linux/sys_eventfd.go +++ b/pkg/sentry/syscalls/linux/sys_eventfd.go @@ -15,12 +15,11 @@ package linux import ( - "syscall" - "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/eventfd" + "gvisor.dev/gvisor/pkg/syserror" ) const ( @@ -38,7 +37,7 @@ func Eventfd2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc allOps := uint(EFD_SEMAPHORE | EFD_NONBLOCK | EFD_CLOEXEC) if flags & ^allOps != 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } event := eventfd.New(t, uint64(initVal), flags&EFD_SEMAPHORE != 0) @@ -47,10 +46,9 @@ func Eventfd2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc }) defer event.DecRef() - fd, err := t.FDMap().NewFDFrom(0, event, kernel.FDFlags{ + fd, err := t.NewFDFrom(0, event, kernel.FDFlags{ CloseOnExec: flags&EFD_CLOEXEC != 0, - }, - t.ThreadGroup().Limits()) + }) if err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 3ef7441c2..2e00a91ce 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -26,7 +26,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/kernel/fasync" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/usermem" @@ -34,7 +33,7 @@ import ( ) // fileOpAt performs an operation on the second last component in the path. -func fileOpAt(t *kernel.Task, dirFD kdefs.FD, path string, fn func(root *fs.Dirent, d *fs.Dirent, name string, remainingTraversals uint) error) error { +func fileOpAt(t *kernel.Task, dirFD int32, path string, fn func(root *fs.Dirent, d *fs.Dirent, name string, remainingTraversals uint) error) error { // Extract the last component. dir, name := fs.SplitLast(path) if dir == "/" { @@ -60,7 +59,7 @@ func fileOpAt(t *kernel.Task, dirFD kdefs.FD, path string, fn func(root *fs.Dire } // fileOpOn performs an operation on the last entry of the path. -func fileOpOn(t *kernel.Task, dirFD kdefs.FD, path string, resolve bool, fn func(root *fs.Dirent, d *fs.Dirent, remainingTraversals uint) error) error { +func fileOpOn(t *kernel.Task, dirFD int32, path string, resolve bool, fn func(root *fs.Dirent, d *fs.Dirent, remainingTraversals uint) error) error { var ( d *fs.Dirent // The file. wd *fs.Dirent // The working directory (if required.) @@ -78,7 +77,7 @@ func fileOpOn(t *kernel.Task, dirFD kdefs.FD, path string, resolve bool, fn func rel = wd } else { // Need to extract the given FD. - f = t.FDMap().GetFile(dirFD) + f = t.GetFile(dirFD) if f == nil { return syserror.EBADF } @@ -131,7 +130,7 @@ func copyInPath(t *kernel.Task, addr usermem.Addr, allowEmpty bool) (path string return path, dirPath, nil } -func openAt(t *kernel.Task, dirFD kdefs.FD, addr usermem.Addr, flags uint) (fd uintptr, err error) { +func openAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint) (fd uintptr, err error) { path, dirPath, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { return 0, err @@ -184,8 +183,9 @@ func openAt(t *kernel.Task, dirFD kdefs.FD, addr usermem.Addr, flags uint) (fd u defer file.DecRef() // Success. - fdFlags := kernel.FDFlags{CloseOnExec: flags&linux.O_CLOEXEC != 0} - newFD, err := t.FDMap().NewFDFrom(0, file, fdFlags, t.ThreadGroup().Limits()) + newFD, err := t.NewFDFrom(0, file, kernel.FDFlags{ + CloseOnExec: flags&linux.O_CLOEXEC != 0, + }) if err != nil { return err } @@ -201,7 +201,7 @@ func openAt(t *kernel.Task, dirFD kdefs.FD, addr usermem.Addr, flags uint) (fd u return fd, err // Use result in frame. } -func mknodAt(t *kernel.Task, dirFD kdefs.FD, addr usermem.Addr, mode linux.FileMode) error { +func mknodAt(t *kernel.Task, dirFD int32, addr usermem.Addr, mode linux.FileMode) error { path, dirPath, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { return err @@ -285,7 +285,7 @@ func Mknod(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Mknodat implements the linux syscall mknodat(2). func Mknodat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - dirFD := kdefs.FD(args[0].Int()) + dirFD := args[0].Int() path := args[1].Pointer() mode := linux.FileMode(args[2].ModeT()) // We don't need this argument until we support creation of device nodes. @@ -294,7 +294,7 @@ func Mknodat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca return 0, nil, mknodAt(t, dirFD, path, mode) } -func createAt(t *kernel.Task, dirFD kdefs.FD, addr usermem.Addr, flags uint, mode linux.FileMode) (fd uintptr, err error) { +func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode linux.FileMode) (fd uintptr, err error) { path, dirPath, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { return 0, err @@ -326,6 +326,7 @@ func createAt(t *kernel.Task, dirFD kdefs.FD, addr usermem.Addr, flags uint, mod if err != nil { break } + defer found.DecRef() // We found something (possibly a symlink). If the // O_EXCL flag was passed, then we can immediately @@ -346,31 +347,41 @@ func createAt(t *kernel.Task, dirFD kdefs.FD, addr usermem.Addr, flags uint, mod } // Try to resolve the symlink directly to a Dirent. - resolved, err := found.Inode.Getlink(t) - if err == nil || err != fs.ErrResolveViaReadlink { + var resolved *fs.Dirent + resolved, err = found.Inode.Getlink(t) + if err == nil { // No more resolution necessary. - found.DecRef() - found = resolved + defer resolved.DecRef() break } + if err != fs.ErrResolveViaReadlink { + return err + } + + // Are we able to resolve further? + if remainingTraversals == 0 { + return syscall.ELOOP + } // Resolve the symlink to a path via Readlink. - path, err := found.Inode.Readlink(t) + var path string + path, err = found.Inode.Readlink(t) if err != nil { break } remainingTraversals-- // Get the new parent from the target path. + var newParent *fs.Dirent newParentPath, newName := fs.SplitLast(path) - newParent, err := t.MountNamespace().FindInode(t, root, parent, newParentPath, &remainingTraversals) + newParent, err = t.MountNamespace().FindInode(t, root, parent, newParentPath, &remainingTraversals) if err != nil { break } + defer newParent.DecRef() // Repeat the process with the parent and name of the // symlink target. - parent.DecRef() parent = newParent name = newName } @@ -378,9 +389,6 @@ func createAt(t *kernel.Task, dirFD kdefs.FD, addr usermem.Addr, flags uint, mod var newFile *fs.File switch err { case nil: - // The file existed. - defer found.DecRef() - // Like sys_open, check for a few things about the // filesystem before trying to get a reference to the // fs.File. The same constraints on Check apply. @@ -423,8 +431,9 @@ func createAt(t *kernel.Task, dirFD kdefs.FD, addr usermem.Addr, flags uint, mod } // Success. - fdFlags := kernel.FDFlags{CloseOnExec: flags&linux.O_CLOEXEC != 0} - newFD, err := t.FDMap().NewFDFrom(0, newFile, fdFlags, t.ThreadGroup().Limits()) + newFD, err := t.NewFDFrom(0, newFile, kernel.FDFlags{ + CloseOnExec: flags&linux.O_CLOEXEC != 0, + }) if err != nil { return err } @@ -458,7 +467,7 @@ func Open(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC // Openat implements linux syscall openat(2). func Openat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - dirFD := kdefs.FD(args[0].Int()) + dirFD := args[0].Int() addr := args[1].Pointer() flags := uint(args[2].Uint()) if flags&linux.O_CREAT != 0 { @@ -497,7 +506,7 @@ func (ac accessContext) Value(key interface{}) interface{} { } } -func accessAt(t *kernel.Task, dirFD kdefs.FD, addr usermem.Addr, resolve bool, mode uint) error { +func accessAt(t *kernel.Task, dirFD int32, addr usermem.Addr, resolve bool, mode uint) error { const rOK = 4 const wOK = 2 const xOK = 1 @@ -552,7 +561,7 @@ func Access(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Faccessat implements linux syscall faccessat(2). func Faccessat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - dirFD := kdefs.FD(args[0].Int()) + dirFD := args[0].Int() addr := args[1].Pointer() mode := args[2].ModeT() flags := args[3].Int() @@ -562,10 +571,10 @@ func Faccessat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys // Ioctl implements linux syscall ioctl(2). func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() request := int(args[1].Int()) - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -574,12 +583,12 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Shared flags between file and socket. switch request { case linux.FIONCLEX: - t.FDMap().SetFlags(fd, kernel.FDFlags{ + t.FDTable().SetFlags(fd, kernel.FDFlags{ CloseOnExec: false, }) return 0, nil, nil case linux.FIOCLEX: - t.FDMap().SetFlags(fd, kernel.FDFlags{ + t.FDTable().SetFlags(fd, kernel.FDFlags{ CloseOnExec: true, }) return 0, nil, nil @@ -723,9 +732,9 @@ func Chdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Fchdir implements the linux syscall fchdir(2). func Fchdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -747,44 +756,47 @@ func Fchdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Close implements linux syscall close(2). func Close(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() - file, ok := t.FDMap().Remove(fd) - if !ok { + // Note that Remove provides a reference on the file that we may use to + // flush. It is still active until we drop the final reference below + // (and other reference-holding operations complete). + file := t.FDTable().Remove(fd) + if file == nil { return 0, nil, syserror.EBADF } defer file.DecRef() err := file.Flush(t) - return 0, nil, handleIOError(t, false /* partial */, err, syscall.EINTR, "close", file) + return 0, nil, handleIOError(t, false /* partial */, err, syserror.EINTR, "close", file) } // Dup implements linux syscall dup(2). func Dup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } defer file.DecRef() - newfd, err := t.FDMap().NewFDFrom(0, file, kernel.FDFlags{}, t.ThreadGroup().Limits()) + newFD, err := t.NewFDFrom(0, file, kernel.FDFlags{}) if err != nil { return 0, nil, syserror.EMFILE } - return uintptr(newfd), nil, nil + return uintptr(newFD), nil, nil } // Dup2 implements linux syscall dup2(2). func Dup2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - oldfd := kdefs.FD(args[0].Int()) - newfd := kdefs.FD(args[1].Int()) + oldfd := args[0].Int() + newfd := args[1].Int() // If oldfd is a valid file descriptor, and newfd has the same value as oldfd, // then dup2() does nothing, and returns newfd. if oldfd == newfd { - oldFile := t.FDMap().GetFile(oldfd) + oldFile := t.GetFile(oldfd) if oldFile == nil { return 0, nil, syserror.EBADF } @@ -800,21 +812,21 @@ func Dup2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC // Dup3 implements linux syscall dup3(2). func Dup3(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - oldfd := kdefs.FD(args[0].Int()) - newfd := kdefs.FD(args[1].Int()) + oldfd := args[0].Int() + newfd := args[1].Int() flags := args[2].Uint() if oldfd == newfd { return 0, nil, syserror.EINVAL } - oldFile := t.FDMap().GetFile(oldfd) + oldFile := t.GetFile(oldfd) if oldFile == nil { return 0, nil, syserror.EBADF } defer oldFile.DecRef() - err := t.FDMap().NewFDAt(newfd, oldFile, kernel.FDFlags{CloseOnExec: flags&linux.O_CLOEXEC != 0}, t.ThreadGroup().Limits()) + err := t.NewFDAt(newfd, oldFile, kernel.FDFlags{CloseOnExec: flags&linux.O_CLOEXEC != 0}) if err != nil { return 0, nil, err } @@ -857,10 +869,10 @@ func fSetOwn(t *kernel.Task, file *fs.File, who int32) { // Fcntl implements linux syscall fcntl(2). func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() cmd := args[1].Int() - file, flags := t.FDMap().GetDescriptor(fd) + file, flags := t.FDTable().Get(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -868,9 +880,10 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall switch cmd { case linux.F_DUPFD, linux.F_DUPFD_CLOEXEC: - from := kdefs.FD(args[2].Int()) - fdFlags := kernel.FDFlags{CloseOnExec: cmd == linux.F_DUPFD_CLOEXEC} - fd, err := t.FDMap().NewFDFrom(from, file, fdFlags, t.ThreadGroup().Limits()) + from := args[2].Int() + fd, err := t.NewFDFrom(from, file, kernel.FDFlags{ + CloseOnExec: cmd == linux.F_DUPFD_CLOEXEC, + }) if err != nil { return 0, nil, err } @@ -879,7 +892,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return uintptr(flags.ToLinuxFDFlags()), nil, nil case linux.F_SETFD: flags := args[2].Uint() - t.FDMap().SetFlags(fd, kernel.FDFlags{ + t.FDTable().SetFlags(fd, kernel.FDFlags{ CloseOnExec: flags&linux.FD_CLOEXEC != 0, }) case linux.F_GETFL: @@ -897,7 +910,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Copy in the lock request. flockAddr := args[2].Pointer() - var flock syscall.Flock_t + var flock linux.Flock if _, err := t.CopyIn(flockAddr, &flock); err != nil { return 0, nil, err } @@ -940,17 +953,17 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return 0, nil, err } - // The lock uid is that of the Task's FDMap. - lockUniqueID := lock.UniqueID(t.FDMap().ID()) + // The lock uid is that of the Task's FDTable. + lockUniqueID := lock.UniqueID(t.FDTable().ID()) // These locks don't block; execute the non-blocking operation using the inode's lock // context directly. switch flock.Type { - case syscall.F_RDLCK: + case linux.F_RDLCK: if !file.Flags().Read { return 0, nil, syserror.EBADF } - if cmd == syscall.F_SETLK { + if cmd == linux.F_SETLK { // Non-blocking lock, provide a nil lock.Blocker. if !file.Dirent.Inode.LockCtx.Posix.LockRegion(lockUniqueID, lock.ReadLock, rng, nil) { return 0, nil, syserror.EAGAIN @@ -962,11 +975,11 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } } return 0, nil, nil - case syscall.F_WRLCK: + case linux.F_WRLCK: if !file.Flags().Write { return 0, nil, syserror.EBADF } - if cmd == syscall.F_SETLK { + if cmd == linux.F_SETLK { // Non-blocking lock, provide a nil lock.Blocker. if !file.Dirent.Inode.LockCtx.Posix.LockRegion(lockUniqueID, lock.WriteLock, rng, nil) { return 0, nil, syserror.EAGAIN @@ -978,7 +991,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } } return 0, nil, nil - case syscall.F_UNLCK: + case linux.F_UNLCK: file.Dirent.Inode.LockCtx.Posix.UnlockRegion(lockUniqueID, rng) return 0, nil, nil default: @@ -1031,7 +1044,7 @@ const ( // Fadvise64 implements linux syscall fadvise64(2). // This implementation currently ignores the provided advice. func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() length := args[2].Int64() advice := args[3].Int() @@ -1040,7 +1053,7 @@ func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, syserror.EINVAL } - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -1066,7 +1079,7 @@ func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, nil } -func mkdirAt(t *kernel.Task, dirFD kdefs.FD, addr usermem.Addr, mode linux.FileMode) error { +func mkdirAt(t *kernel.Task, dirFD int32, addr usermem.Addr, mode linux.FileMode) error { path, _, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { return err @@ -1111,14 +1124,14 @@ func Mkdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Mkdirat implements linux syscall mkdirat(2). func Mkdirat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - dirFD := kdefs.FD(args[0].Int()) + dirFD := args[0].Int() addr := args[1].Pointer() mode := linux.FileMode(args[2].ModeT()) return 0, nil, mkdirAt(t, dirFD, addr, mode) } -func rmdirAt(t *kernel.Task, dirFD kdefs.FD, addr usermem.Addr) error { +func rmdirAt(t *kernel.Task, dirFD int32, addr usermem.Addr) error { path, _, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { return err @@ -1158,7 +1171,7 @@ func Rmdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return 0, nil, rmdirAt(t, linux.AT_FDCWD, addr) } -func symlinkAt(t *kernel.Task, dirFD kdefs.FD, newAddr usermem.Addr, oldAddr usermem.Addr) error { +func symlinkAt(t *kernel.Task, dirFD int32, newAddr usermem.Addr, oldAddr usermem.Addr) error { newPath, dirPath, err := copyInPath(t, newAddr, false /* allowEmpty */) if err != nil { return err @@ -1201,7 +1214,7 @@ func Symlink(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca // Symlinkat implements linux syscall symlinkat(2). func Symlinkat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { oldAddr := args[0].Pointer() - dirFD := kdefs.FD(args[1].Int()) + dirFD := args[1].Int() newAddr := args[2].Pointer() return 0, nil, symlinkAt(t, dirFD, newAddr, oldAddr) @@ -1243,7 +1256,7 @@ func mayLinkAt(t *kernel.Task, target *fs.Inode) error { // linkAt creates a hard link to the target specified by oldDirFD and oldAddr, // specified by newDirFD and newAddr. If resolve is true, then the symlinks // will be followed when evaluating the target. -func linkAt(t *kernel.Task, oldDirFD kdefs.FD, oldAddr usermem.Addr, newDirFD kdefs.FD, newAddr usermem.Addr, resolve, allowEmpty bool) error { +func linkAt(t *kernel.Task, oldDirFD int32, oldAddr usermem.Addr, newDirFD int32, newAddr usermem.Addr, resolve, allowEmpty bool) error { oldPath, _, err := copyInPath(t, oldAddr, allowEmpty) if err != nil { return err @@ -1257,7 +1270,7 @@ func linkAt(t *kernel.Task, oldDirFD kdefs.FD, oldAddr usermem.Addr, newDirFD kd } if allowEmpty && oldPath == "" { - target := t.FDMap().GetFile(oldDirFD) + target := t.GetFile(oldDirFD) if target == nil { return syserror.EBADF } @@ -1319,9 +1332,9 @@ func Link(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC // Linkat implements linux syscall linkat(2). func Linkat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - oldDirFD := kdefs.FD(args[0].Int()) + oldDirFD := args[0].Int() oldAddr := args[1].Pointer() - newDirFD := kdefs.FD(args[2].Int()) + newDirFD := args[2].Int() newAddr := args[3].Pointer() // man linkat(2): @@ -1346,7 +1359,7 @@ func Linkat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return 0, nil, linkAt(t, oldDirFD, oldAddr, newDirFD, newAddr, resolve, allowEmpty) } -func readlinkAt(t *kernel.Task, dirFD kdefs.FD, addr usermem.Addr, bufAddr usermem.Addr, size uint) (copied uintptr, err error) { +func readlinkAt(t *kernel.Task, dirFD int32, addr usermem.Addr, bufAddr usermem.Addr, size uint) (copied uintptr, err error) { path, dirPath, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { return 0, err @@ -1396,7 +1409,7 @@ func Readlink(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // Readlinkat implements linux syscall readlinkat(2). func Readlinkat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - dirFD := kdefs.FD(args[0].Int()) + dirFD := args[0].Int() addr := args[1].Pointer() bufAddr := args[2].Pointer() size := args[3].SizeT() @@ -1405,7 +1418,7 @@ func Readlinkat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy return n, nil, err } -func unlinkAt(t *kernel.Task, dirFD kdefs.FD, addr usermem.Addr) error { +func unlinkAt(t *kernel.Task, dirFD int32, addr usermem.Addr) error { path, dirPath, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { return err @@ -1435,7 +1448,7 @@ func Unlink(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Unlinkat implements linux syscall unlinkat(2). func Unlinkat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - dirFD := kdefs.FD(args[0].Int()) + dirFD := args[0].Int() addr := args[1].Pointer() flags := args[2].Uint() if flags&linux.AT_REMOVEDIR != 0 { @@ -1463,7 +1476,7 @@ func Truncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if uint64(length) >= t.ThreadGroup().Limits().Get(limits.FileSize).Cur { t.SendSignal(&arch.SignalInfo{ - Signo: int32(syscall.SIGXFSZ), + Signo: int32(linux.SIGXFSZ), Code: arch.SignalInfoUser, }) return 0, nil, syserror.EFBIG @@ -1496,10 +1509,10 @@ func Truncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // Ftruncate implements linux syscall ftruncate(2). func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() length := args[1].Int64() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -1523,7 +1536,7 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if uint64(length) >= t.ThreadGroup().Limits().Get(limits.FileSize).Cur { t.SendSignal(&arch.SignalInfo{ - Signo: int32(syscall.SIGXFSZ), + Signo: int32(linux.SIGXFSZ), Code: arch.SignalInfoUser, }) return 0, nil, syserror.EFBIG @@ -1614,7 +1627,7 @@ func chown(t *kernel.Task, d *fs.Dirent, uid auth.UID, gid auth.GID) error { return nil } -func chownAt(t *kernel.Task, fd kdefs.FD, addr usermem.Addr, resolve, allowEmpty bool, uid auth.UID, gid auth.GID) error { +func chownAt(t *kernel.Task, fd int32, addr usermem.Addr, resolve, allowEmpty bool, uid auth.UID, gid auth.GID) error { path, _, err := copyInPath(t, addr, allowEmpty) if err != nil { return err @@ -1622,7 +1635,7 @@ func chownAt(t *kernel.Task, fd kdefs.FD, addr usermem.Addr, resolve, allowEmpty if path == "" { // Annoying. What's wrong with fchown? - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return syserror.EBADF } @@ -1656,11 +1669,11 @@ func Lchown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Fchown implements linux syscall fchown(2). func Fchown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() uid := auth.UID(args[1].Uint()) gid := auth.GID(args[2].Uint()) - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -1671,7 +1684,7 @@ func Fchown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Fchownat implements Linux syscall fchownat(2). func Fchownat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - dirFD := kdefs.FD(args[0].Int()) + dirFD := args[0].Int() addr := args[1].Pointer() uid := auth.UID(args[2].Uint()) gid := auth.GID(args[3].Uint()) @@ -1701,7 +1714,7 @@ func chmod(t *kernel.Task, d *fs.Dirent, mode linux.FileMode) error { return nil } -func chmodAt(t *kernel.Task, fd kdefs.FD, addr usermem.Addr, mode linux.FileMode) error { +func chmodAt(t *kernel.Task, fd int32, addr usermem.Addr, mode linux.FileMode) error { path, _, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { return err @@ -1722,10 +1735,10 @@ func Chmod(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Fchmod implements linux syscall fchmod(2). func Fchmod(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() mode := linux.FileMode(args[1].ModeT()) - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -1736,7 +1749,7 @@ func Fchmod(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Fchmodat implements linux syscall fchmodat(2). func Fchmodat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() mode := linux.FileMode(args[2].ModeT()) @@ -1752,7 +1765,7 @@ func defaultSetToSystemTimeSpec() fs.TimeSpec { } } -func utimes(t *kernel.Task, dirFD kdefs.FD, addr usermem.Addr, ts fs.TimeSpec, resolve bool) error { +func utimes(t *kernel.Task, dirFD int32, addr usermem.Addr, ts fs.TimeSpec, resolve bool) error { setTimestamp := func(root *fs.Dirent, d *fs.Dirent, _ uint) error { // Does the task own the file? if !d.Inode.CheckOwnership(t) { @@ -1785,7 +1798,7 @@ func utimes(t *kernel.Task, dirFD kdefs.FD, addr usermem.Addr, ts fs.TimeSpec, r // Linux returns EINVAL in this case. See utimes.c. return syserror.EINVAL } - f := t.FDMap().GetFile(dirFD) + f := t.GetFile(dirFD) if f == nil { return syserror.EBADF } @@ -1813,7 +1826,7 @@ func Utime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // No timesAddr argument will be interpreted as current system time. ts := defaultSetToSystemTimeSpec() if timesAddr != 0 { - var times syscall.Utimbuf + var times linux.Utime if _, err := t.CopyIn(timesAddr, ×); err != nil { return 0, nil, err } @@ -1853,7 +1866,7 @@ func timespecIsValid(ts linux.Timespec) bool { // Utimensat implements linux syscall utimensat(2). func Utimensat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - dirFD := kdefs.FD(args[0].Int()) + dirFD := args[0].Int() pathnameAddr := args[1].Pointer() timesAddr := args[2].Pointer() flags := args[3].Int() @@ -1888,7 +1901,7 @@ func Utimensat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys // Futimesat implements linux syscall futimesat(2). func Futimesat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - dirFD := kdefs.FD(args[0].Int()) + dirFD := args[0].Int() pathnameAddr := args[1].Pointer() timesAddr := args[2].Pointer() @@ -1912,7 +1925,7 @@ func Futimesat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, utimes(t, dirFD, pathnameAddr, ts, true) } -func renameAt(t *kernel.Task, oldDirFD kdefs.FD, oldAddr usermem.Addr, newDirFD kdefs.FD, newAddr usermem.Addr) error { +func renameAt(t *kernel.Task, oldDirFD int32, oldAddr usermem.Addr, newDirFD int32, newAddr usermem.Addr) error { newPath, _, err := copyInPath(t, newAddr, false /* allowEmpty */) if err != nil { return err @@ -1960,21 +1973,21 @@ func Rename(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Renameat implements linux syscall renameat(2). func Renameat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - oldDirFD := kdefs.FD(args[0].Int()) + oldDirFD := args[0].Int() oldPathAddr := args[1].Pointer() - newDirFD := kdefs.FD(args[2].Int()) + newDirFD := args[2].Int() newPathAddr := args[3].Pointer() return 0, nil, renameAt(t, oldDirFD, oldPathAddr, newDirFD, newPathAddr) } // Fallocate implements linux system call fallocate(2). func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() mode := args[1].Int64() offset := args[2].Int64() length := args[3].Int64() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -2005,7 +2018,7 @@ func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys } if uint64(size) >= t.ThreadGroup().Limits().Get(limits.FileSize).Cur { t.SendSignal(&arch.SignalInfo{ - Signo: int32(syscall.SIGXFSZ), + Signo: int32(linux.SIGXFSZ), Code: arch.SignalInfoUser, }) return 0, nil, syserror.EFBIG @@ -2023,10 +2036,10 @@ func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys // Flock implements linux syscall flock(2). func Flock(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() operation := args[1].Int() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { // flock(2): EBADF fd is not an open file descriptor. return 0, nil, syserror.EBADF @@ -2133,8 +2146,9 @@ func MemfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S defer dirent.DecRef() defer file.DecRef() - fdFlags := kernel.FDFlags{CloseOnExec: cloExec} - newFD, err := t.FDMap().NewFDFrom(0, file, fdFlags, t.ThreadGroup().Limits()) + newFD, err := t.NewFDFrom(0, file, kernel.FDFlags{ + CloseOnExec: cloExec, + }) if err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_getdents.go b/pkg/sentry/syscalls/linux/sys_getdents.go index dea872672..63e2c5a5d 100644 --- a/pkg/sentry/syscalls/linux/sys_getdents.go +++ b/pkg/sentry/syscalls/linux/sys_getdents.go @@ -17,20 +17,19 @@ package linux import ( "bytes" "io" - "syscall" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" ) // Getdents implements linux syscall getdents(2) for 64bit systems. func Getdents(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() size := int(args[2].Uint()) @@ -46,7 +45,7 @@ func Getdents(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // Getdents64 implements linux syscall getdents64(2). func Getdents64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() size := int(args[2].Uint()) @@ -62,8 +61,8 @@ func Getdents64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy // getdents implements the core of getdents(2)/getdents64(2). // f is the syscall implementation dirent serialization function. -func getdents(t *kernel.Task, fd kdefs.FD, addr usermem.Addr, size int, f func(*dirent, io.Writer) (int, error)) (uintptr, error) { - dir := t.FDMap().GetFile(fd) +func getdents(t *kernel.Task, fd int32, addr usermem.Addr, size int, f func(*dirent, io.Writer) (int, error)) (uintptr, error) { + dir := t.GetFile(fd) if dir == nil { return 0, syserror.EBADF } @@ -83,7 +82,7 @@ func getdents(t *kernel.Task, fd kdefs.FD, addr usermem.Addr, size int, f func(* switch err := handleIOError(t, ds.Written() > 0, rerr, kernel.ERESTARTSYS, "getdents", dir); err { case nil: - dir.Dirent.InotifyEvent(syscall.IN_ACCESS, 0) + dir.Dirent.InotifyEvent(linux.IN_ACCESS, 0) return uintptr(ds.Written()), nil case io.EOF: return 0, nil @@ -147,21 +146,21 @@ func smallestDirent64(a arch.Context) uint { func toType(nodeType fs.InodeType) uint8 { switch nodeType { case fs.RegularFile, fs.SpecialFile: - return syscall.DT_REG + return linux.DT_REG case fs.Symlink: - return syscall.DT_LNK + return linux.DT_LNK case fs.Directory, fs.SpecialDirectory: - return syscall.DT_DIR + return linux.DT_DIR case fs.Pipe: - return syscall.DT_FIFO + return linux.DT_FIFO case fs.CharacterDevice: - return syscall.DT_CHR + return linux.DT_CHR case fs.BlockDevice: - return syscall.DT_BLK + return linux.DT_BLK case fs.Socket: - return syscall.DT_SOCK + return linux.DT_SOCK default: - return syscall.DT_UNKNOWN + return linux.DT_UNKNOWN } } diff --git a/pkg/sentry/syscalls/linux/sys_inotify.go b/pkg/sentry/syscalls/linux/sys_inotify.go index 9cfa660fa..b2c7b3444 100644 --- a/pkg/sentry/syscalls/linux/sys_inotify.go +++ b/pkg/sentry/syscalls/linux/sys_inotify.go @@ -15,14 +15,12 @@ package linux import ( - "syscall" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/anon" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" + "gvisor.dev/gvisor/pkg/syserror" ) const allFlags = int(linux.IN_NONBLOCK | linux.IN_CLOEXEC) @@ -32,7 +30,7 @@ func InotifyInit1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. flags := int(args[0].Int()) if flags&^allFlags != 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } dirent := fs.NewDirent(t, anon.NewInode(t), "inotify") @@ -44,9 +42,9 @@ func InotifyInit1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. n := fs.NewFile(t, dirent, fileFlags, fs.NewInotify(t)) defer n.DecRef() - fd, err := t.FDMap().NewFDFrom(0, n, kernel.FDFlags{ + fd, err := t.NewFDFrom(0, n, kernel.FDFlags{ CloseOnExec: flags&linux.IN_CLOEXEC != 0, - }, t.ThreadGroup().Limits()) + }) if err != nil { return 0, nil, err @@ -63,18 +61,18 @@ func InotifyInit(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S // fdToInotify resolves an fd to an inotify object. If successful, the file will // have an extra ref and the caller is responsible for releasing the ref. -func fdToInotify(t *kernel.Task, fd kdefs.FD) (*fs.Inotify, *fs.File, error) { - file := t.FDMap().GetFile(fd) +func fdToInotify(t *kernel.Task, fd int32) (*fs.Inotify, *fs.File, error) { + file := t.GetFile(fd) if file == nil { // Invalid fd. - return nil, nil, syscall.EBADF + return nil, nil, syserror.EBADF } ino, ok := file.FileOperations.(*fs.Inotify) if !ok { // Not an inotify fd. file.DecRef() - return nil, nil, syscall.EINVAL + return nil, nil, syserror.EINVAL } return ino, file, nil @@ -82,7 +80,7 @@ func fdToInotify(t *kernel.Task, fd kdefs.FD) (*fs.Inotify, *fs.File, error) { // InotifyAddWatch implements the inotify_add_watch() syscall. func InotifyAddWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() mask := args[2].Uint() @@ -93,7 +91,7 @@ func InotifyAddWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kern // "EINVAL: The given event mask contains no valid events." // -- inotify_add_watch(2) if validBits := mask & linux.ALL_INOTIFY_BITS; validBits == 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } ino, file, err := fdToInotify(t, fd) @@ -110,11 +108,11 @@ func InotifyAddWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kern err = fileOpOn(t, linux.AT_FDCWD, path, resolve, func(root *fs.Dirent, dirent *fs.Dirent, _ uint) error { // "IN_ONLYDIR: Only watch pathname if it is a directory." -- inotify(7) if onlyDir := mask&linux.IN_ONLYDIR != 0; onlyDir && !fs.IsDir(dirent.Inode.StableAttr) { - return syscall.ENOTDIR + return syserror.ENOTDIR } // Copy out to the return frame. - fd = kdefs.FD(ino.AddWatch(dirent, mask)) + fd = ino.AddWatch(dirent, mask) return nil }) @@ -123,7 +121,7 @@ func InotifyAddWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kern // InotifyRmWatch implements the inotify_rm_watch() syscall. func InotifyRmWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() wd := args[1].Int() ino, file, err := fdToInotify(t, fd) diff --git a/pkg/sentry/syscalls/linux/sys_lseek.go b/pkg/sentry/syscalls/linux/sys_lseek.go index a3813b818..297e920c4 100644 --- a/pkg/sentry/syscalls/linux/sys_lseek.go +++ b/pkg/sentry/syscalls/linux/sys_lseek.go @@ -18,17 +18,16 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" "gvisor.dev/gvisor/pkg/syserror" ) // Lseek implements linux syscall lseek(2). func Lseek(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() offset := args[1].Int64() whence := args[2].Int() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } diff --git a/pkg/sentry/syscalls/linux/sys_mmap.go b/pkg/sentry/syscalls/linux/sys_mmap.go index d831833bc..58a05b5bb 100644 --- a/pkg/sentry/syscalls/linux/sys_mmap.go +++ b/pkg/sentry/syscalls/linux/sys_mmap.go @@ -20,7 +20,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/mm" "gvisor.dev/gvisor/pkg/sentry/usermem" @@ -40,7 +39,7 @@ func Brk(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { prot := args[2].Int() flags := args[3].Int() - fd := kdefs.FD(args[4].Int()) + fd := args[4].Int() fixed := flags&linux.MAP_FIXED != 0 private := flags&linux.MAP_PRIVATE != 0 shared := flags&linux.MAP_SHARED != 0 @@ -80,7 +79,7 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if !anon { // Convert the passed FD to a file reference. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } diff --git a/pkg/sentry/syscalls/linux/sys_pipe.go b/pkg/sentry/syscalls/linux/sys_pipe.go index 7c1bea43d..418d7fa5f 100644 --- a/pkg/sentry/syscalls/linux/sys_pipe.go +++ b/pkg/sentry/syscalls/linux/sys_pipe.go @@ -15,20 +15,19 @@ package linux import ( - "syscall" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/syserror" ) // pipe2 implements the actual system call with flags. func pipe2(t *kernel.Task, addr usermem.Addr, flags uint) (uintptr, error) { if flags&^(linux.O_NONBLOCK|linux.O_CLOEXEC) != 0 { - return 0, syscall.EINVAL + return 0, syserror.EINVAL } r, w := pipe.NewConnectedPipe(t, pipe.DefaultPipeSize, usermem.PageSize) @@ -38,25 +37,18 @@ func pipe2(t *kernel.Task, addr usermem.Addr, flags uint) (uintptr, error) { w.SetFlags(linuxToFlags(flags).Settable()) defer w.DecRef() - rfd, err := t.FDMap().NewFDFrom(0, r, kernel.FDFlags{ - CloseOnExec: flags&linux.O_CLOEXEC != 0}, - t.ThreadGroup().Limits()) - if err != nil { - return 0, err - } - - wfd, err := t.FDMap().NewFDFrom(0, w, kernel.FDFlags{ - CloseOnExec: flags&linux.O_CLOEXEC != 0}, - t.ThreadGroup().Limits()) + fds, err := t.NewFDs(0, []*fs.File{r, w}, kernel.FDFlags{ + CloseOnExec: flags&linux.O_CLOEXEC != 0, + }) if err != nil { - t.FDMap().Remove(rfd) return 0, err } - if _, err := t.CopyOut(addr, []kdefs.FD{rfd, wfd}); err != nil { - t.FDMap().Remove(rfd) - t.FDMap().Remove(wfd) - return 0, syscall.EFAULT + if _, err := t.CopyOut(addr, fds); err != nil { + // The files are not closed in this case, the exact semantics + // of this error case are not well defined, but they could have + // already been observed by user space. + return 0, syserror.EFAULT } return 0, nil } diff --git a/pkg/sentry/syscalls/linux/sys_poll.go b/pkg/sentry/syscalls/linux/sys_poll.go index ef6211218..7a13beac2 100644 --- a/pkg/sentry/syscalls/linux/sys_poll.go +++ b/pkg/sentry/syscalls/linux/sys_poll.go @@ -21,7 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/usermem" @@ -64,7 +63,7 @@ func initReadiness(t *kernel.Task, pfd *linux.PollFD, state *pollState, ch chan return } - file := t.FDMap().GetFile(kdefs.FD(pfd.FD)) + file := t.GetFile(pfd.FD) if file == nil { pfd.REvents = linux.POLLNVAL return @@ -265,7 +264,7 @@ func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Add // immediately to ensure we don't leak. Note, another thread // might be about to close fd. This is racy, but that's // OK. Linux is racy in the same way. - file := t.FDMap().GetFile(kdefs.FD(fd)) + file := t.GetFile(fd) if file == nil { return 0, syserror.EBADF } diff --git a/pkg/sentry/syscalls/linux/sys_prctl.go b/pkg/sentry/syscalls/linux/sys_prctl.go index 9d70881fd..98db32d77 100644 --- a/pkg/sentry/syscalls/linux/sys_prctl.go +++ b/pkg/sentry/syscalls/linux/sys_prctl.go @@ -16,15 +16,14 @@ package linux import ( "fmt" - "syscall" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" "gvisor.dev/gvisor/pkg/sentry/mm" + "gvisor.dev/gvisor/pkg/syserror" ) // Prctl implements linux syscall prctl(2). @@ -37,7 +36,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.PR_SET_PDEATHSIG: sig := linux.Signal(args[1].Int()) if sig != 0 && !sig.IsValid() { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } t.SetParentDeathSignal(sig) return 0, nil, nil @@ -68,7 +67,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall d = mm.UserDumpable default: // N.B. Userspace may not pass SUID_DUMP_ROOT. - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } t.MemoryManager().SetDumpability(d) return 0, nil, nil @@ -89,7 +88,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } else if val == 1 { t.SetKeepCaps(true) } else { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } return 0, nil, nil @@ -97,7 +96,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.PR_SET_NAME: addr := args[1].Pointer() name, err := t.CopyInString(addr, linux.TASK_COMM_LEN-1) - if err != nil && err != syscall.ENAMETOOLONG { + if err != nil && err != syserror.ENAMETOOLONG { return 0, nil, err } t.SetName(name) @@ -117,22 +116,22 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.PR_SET_MM: if !t.HasCapability(linux.CAP_SYS_RESOURCE) { - return 0, nil, syscall.EPERM + return 0, nil, syserror.EPERM } switch args[1].Int() { case linux.PR_SET_MM_EXE_FILE: - fd := kdefs.FD(args[2].Int()) + fd := args[2].Int() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return 0, nil, syscall.EBADF + return 0, nil, syserror.EBADF } defer file.DecRef() // They trying to set exe to a non-file? if !fs.IsFile(file.Dirent.Inode.StableAttr) { - return 0, nil, syscall.EBADF + return 0, nil, syserror.EBADF } // Set the underlying executable. @@ -154,12 +153,12 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall t.Kernel().EmitUnimplementedEvent(t) fallthrough default: - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } case linux.PR_SET_NO_NEW_PRIVS: if args[1].Int() != 1 || args[2].Int() != 0 || args[3].Int() != 0 || args[4].Int() != 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } // no_new_privs is assumed to always be set. See // kernel.Task.updateCredsForExec. @@ -167,14 +166,14 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.PR_GET_NO_NEW_PRIVS: if args[1].Int() != 0 || args[2].Int() != 0 || args[3].Int() != 0 || args[4].Int() != 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } return 1, nil, nil case linux.PR_SET_SECCOMP: if args[1].Int() != linux.SECCOMP_MODE_FILTER { // Unsupported mode. - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } return 0, nil, seccomp(t, linux.SECCOMP_SET_MODE_FILTER, 0, args[2].Pointer()) @@ -185,7 +184,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.PR_CAPBSET_READ: cp := linux.Capability(args[1].Uint64()) if !cp.Ok() { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } var rv uintptr if auth.CapabilitySetOf(cp)&t.Credentials().BoundingCaps != 0 { @@ -196,7 +195,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.PR_CAPBSET_DROP: cp := linux.Capability(args[1].Uint64()) if !cp.Ok() { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } return 0, nil, t.DropBoundingCapability(cp) @@ -221,7 +220,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall t.Kernel().EmitUnimplementedEvent(t) fallthrough default: - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } return 0, nil, nil diff --git a/pkg/sentry/syscalls/linux/sys_read.go b/pkg/sentry/syscalls/linux/sys_read.go index a1965f490..b2474e60d 100644 --- a/pkg/sentry/syscalls/linux/sys_read.go +++ b/pkg/sentry/syscalls/linux/sys_read.go @@ -21,7 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/usermem" @@ -39,11 +38,11 @@ const ( // they can do large reads all at once. Bug for bug. Same for other read // calls below. func Read(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() size := args[2].SizeT() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -75,12 +74,12 @@ func Read(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC // Pread64 implements linux syscall pread64(2). func Pread64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() size := args[2].SizeT() offset := args[3].Int64() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -122,11 +121,11 @@ func Pread64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca // Readv implements linux syscall readv(2). func Readv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() iovcnt := int(args[2].Int()) - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -152,12 +151,12 @@ func Readv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Preadv implements linux syscall preadv(2). func Preadv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() iovcnt := int(args[2].Int()) offset := args[3].Int64() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -201,13 +200,13 @@ func Preadv2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca // splits the offset argument into a high/low value for compatibility with // 32-bit architectures. The flags argument is the 5th argument. - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() iovcnt := int(args[2].Int()) offset := args[3].Int64() flags := int(args[5].Int()) - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } diff --git a/pkg/sentry/syscalls/linux/sys_sched.go b/pkg/sentry/syscalls/linux/sys_sched.go index 434bbb322..99f6993f5 100644 --- a/pkg/sentry/syscalls/linux/sys_sched.go +++ b/pkg/sentry/syscalls/linux/sys_sched.go @@ -15,11 +15,10 @@ package linux import ( - "syscall" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/syserror" ) const ( @@ -37,13 +36,13 @@ func SchedGetparam(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel pid := args[0].Int() param := args[1].Pointer() if param == 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } if pid < 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } if pid != 0 && t.PIDNamespace().TaskWithID(kernel.ThreadID(pid)) == nil { - return 0, nil, syscall.ESRCH + return 0, nil, syserror.ESRCH } r := SchedParam{schedPriority: onlyPriority} if _, err := t.CopyOut(param, r); err != nil { @@ -57,10 +56,10 @@ func SchedGetparam(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel func SchedGetscheduler(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { pid := args[0].Int() if pid < 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } if pid != 0 && t.PIDNamespace().TaskWithID(kernel.ThreadID(pid)) == nil { - return 0, nil, syscall.ESRCH + return 0, nil, syserror.ESRCH } return onlyScheduler, nil, nil } @@ -71,20 +70,20 @@ func SchedSetscheduler(t *kernel.Task, args arch.SyscallArguments) (uintptr, *ke policy := args[1].Int() param := args[2].Pointer() if pid < 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } if policy != onlyScheduler { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } if pid != 0 && t.PIDNamespace().TaskWithID(kernel.ThreadID(pid)) == nil { - return 0, nil, syscall.ESRCH + return 0, nil, syserror.ESRCH } var r SchedParam if _, err := t.CopyIn(param, &r); err != nil { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } if r.schedPriority != onlyPriority { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } return 0, nil, nil } diff --git a/pkg/sentry/syscalls/linux/sys_seccomp.go b/pkg/sentry/syscalls/linux/sys_seccomp.go index 4885b5e40..18510ead8 100644 --- a/pkg/sentry/syscalls/linux/sys_seccomp.go +++ b/pkg/sentry/syscalls/linux/sys_seccomp.go @@ -15,13 +15,12 @@ package linux import ( - "syscall" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/bpf" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/syserror" ) // userSockFprog is equivalent to Linux's struct sock_fprog on amd64. @@ -43,7 +42,7 @@ func seccomp(t *kernel.Task, mode, flags uint64, addr usermem.Addr) error { // We only support SECCOMP_SET_MODE_FILTER at the moment. if mode != linux.SECCOMP_SET_MODE_FILTER { // Unsupported mode. - return syscall.EINVAL + return syserror.EINVAL } tsync := flags&linux.SECCOMP_FILTER_FLAG_TSYNC != 0 @@ -51,7 +50,7 @@ func seccomp(t *kernel.Task, mode, flags uint64, addr usermem.Addr) error { // The only flag we support now is SECCOMP_FILTER_FLAG_TSYNC. if flags&^linux.SECCOMP_FILTER_FLAG_TSYNC != 0 { // Unsupported flag. - return syscall.EINVAL + return syserror.EINVAL } var fprog userSockFprog @@ -65,7 +64,7 @@ func seccomp(t *kernel.Task, mode, flags uint64, addr usermem.Addr) error { compiledFilter, err := bpf.Compile(filter) if err != nil { t.Debugf("Invalid seccomp-bpf filter: %v", err) - return syscall.EINVAL + return syserror.EINVAL } return t.AppendSyscallFilter(compiledFilter, tsync) diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index ccdb079bb..195734257 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -15,7 +15,6 @@ package linux import ( - "syscall" "time" "gvisor.dev/gvisor/pkg/abi/linux" @@ -23,7 +22,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/control" @@ -132,7 +130,7 @@ func CopyInMessageHeader64(t *kernel.Task, addr usermem.Addr, msg *MessageHeader // from the untrusted address space range. func CaptureAddress(t *kernel.Task, addr usermem.Addr, addrlen uint32) ([]byte, error) { if addrlen > maxAddrLen { - return nil, syscall.EINVAL + return nil, syserror.EINVAL } addrBuf := make([]byte, addrlen) @@ -154,7 +152,7 @@ func writeAddress(t *kernel.Task, addr interface{}, addrLen uint32, addrPtr user } if int32(bufLen) < 0 { - return syscall.EINVAL + return syserror.EINVAL } // Write the length unconditionally. @@ -187,7 +185,7 @@ func Socket(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Check and initialize the flags. if stype & ^(0xf|linux.SOCK_NONBLOCK|linux.SOCK_CLOEXEC) != 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } // Create the new socket. @@ -200,9 +198,9 @@ func Socket(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal }) defer s.DecRef() - fd, err := t.FDMap().NewFDFrom(0, s, kernel.FDFlags{ + fd, err := t.NewFDFrom(0, s, kernel.FDFlags{ CloseOnExec: stype&linux.SOCK_CLOEXEC != 0, - }, t.ThreadGroup().Limits()) + }) if err != nil { return 0, nil, err } @@ -219,15 +217,12 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy // Check and initialize the flags. if stype & ^(0xf|linux.SOCK_NONBLOCK|linux.SOCK_CLOEXEC) != 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } fileFlags := fs.SettableFileFlags{ NonBlocking: stype&linux.SOCK_NONBLOCK != 0, } - fdFlags := kernel.FDFlags{ - CloseOnExec: stype&linux.SOCK_CLOEXEC != 0, - } // Create the socket pair. s1, s2, e := socket.Pair(t, domain, linux.SockType(stype&0xf), protocol) @@ -240,20 +235,16 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy defer s2.DecRef() // Create the FDs for the sockets. - fd1, err := t.FDMap().NewFDFrom(0, s1, fdFlags, t.ThreadGroup().Limits()) - if err != nil { - return 0, nil, err - } - fd2, err := t.FDMap().NewFDFrom(0, s2, fdFlags, t.ThreadGroup().Limits()) + fds, err := t.NewFDs(0, []*fs.File{s1, s2}, kernel.FDFlags{ + CloseOnExec: stype&linux.SOCK_CLOEXEC != 0, + }) if err != nil { - t.FDMap().Remove(fd1) return 0, nil, err } // Copy the file descriptors out. - if _, err := t.CopyOut(socks, []int32{int32(fd1), int32(fd2)}); err != nil { - t.FDMap().Remove(fd1) - t.FDMap().Remove(fd2) + if _, err := t.CopyOut(socks, fds); err != nil { + // Note that we don't close files here; see pipe(2) also. return 0, nil, err } @@ -262,21 +253,21 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy // Connect implements the linux syscall connect(2). func Connect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() addrlen := args[2].Uint() // Get socket from the file descriptor. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return 0, nil, syscall.EBADF + return 0, nil, syserror.EBADF } defer file.DecRef() // Extract the socket. s, ok := file.FileOperations.(socket.Socket) if !ok { - return 0, nil, syscall.ENOTSOCK + return 0, nil, syserror.ENOTSOCK } // Capture address and call syscall implementation. @@ -291,23 +282,23 @@ func Connect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca // accept is the implementation of the accept syscall. It is called by accept // and accept4 syscall handlers. -func accept(t *kernel.Task, fd kdefs.FD, addr usermem.Addr, addrLen usermem.Addr, flags int) (uintptr, error) { +func accept(t *kernel.Task, fd int32, addr usermem.Addr, addrLen usermem.Addr, flags int) (uintptr, error) { // Check that no unsupported flags are passed in. if flags & ^(linux.SOCK_NONBLOCK|linux.SOCK_CLOEXEC) != 0 { - return 0, syscall.EINVAL + return 0, syserror.EINVAL } // Get socket from the file descriptor. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return 0, syscall.EBADF + return 0, syserror.EBADF } defer file.DecRef() // Extract the socket. s, ok := file.FileOperations.(socket.Socket) if !ok { - return 0, syscall.ENOTSOCK + return 0, syserror.ENOTSOCK } // Call the syscall implementation for this socket, then copy the @@ -322,7 +313,7 @@ func accept(t *kernel.Task, fd kdefs.FD, addr usermem.Addr, addrLen usermem.Addr if peerRequested { // NOTE(magi): Linux does not give you an error if it can't // write the data back out so neither do we. - if err := writeAddress(t, peer, peerLen, addr, addrLen); err == syscall.EINVAL { + if err := writeAddress(t, peer, peerLen, addr, addrLen); err == syserror.EINVAL { return 0, err } } @@ -331,7 +322,7 @@ func accept(t *kernel.Task, fd kdefs.FD, addr usermem.Addr, addrLen usermem.Addr // Accept4 implements the linux syscall accept4(2). func Accept4(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() addrlen := args[2].Pointer() flags := int(args[3].Int()) @@ -342,7 +333,7 @@ func Accept4(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca // Accept implements the linux syscall accept(2). func Accept(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() addrlen := args[2].Pointer() @@ -352,21 +343,21 @@ func Accept(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Bind implements the linux syscall bind(2). func Bind(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() addrlen := args[2].Uint() // Get socket from the file descriptor. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return 0, nil, syscall.EBADF + return 0, nil, syserror.EBADF } defer file.DecRef() // Extract the socket. s, ok := file.FileOperations.(socket.Socket) if !ok { - return 0, nil, syscall.ENOTSOCK + return 0, nil, syserror.ENOTSOCK } // Capture address and call syscall implementation. @@ -380,20 +371,20 @@ func Bind(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC // Listen implements the linux syscall listen(2). func Listen(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() backlog := args[1].Int() // Get socket from the file descriptor. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return 0, nil, syscall.EBADF + return 0, nil, syserror.EBADF } defer file.DecRef() // Extract the socket. s, ok := file.FileOperations.(socket.Socket) if !ok { - return 0, nil, syscall.ENOTSOCK + return 0, nil, syserror.ENOTSOCK } // Per Linux, the backlog is silently capped to reasonable values. @@ -409,27 +400,27 @@ func Listen(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Shutdown implements the linux syscall shutdown(2). func Shutdown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() how := args[1].Int() // Get socket from the file descriptor. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return 0, nil, syscall.EBADF + return 0, nil, syserror.EBADF } defer file.DecRef() // Extract the socket. s, ok := file.FileOperations.(socket.Socket) if !ok { - return 0, nil, syscall.ENOTSOCK + return 0, nil, syserror.ENOTSOCK } // Validate how, then call syscall implementation. switch how { case linux.SHUT_RD, linux.SHUT_WR, linux.SHUT_RDWR: default: - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } return 0, nil, s.Shutdown(t, int(how)).ToError() @@ -437,23 +428,23 @@ func Shutdown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // GetSockOpt implements the linux syscall getsockopt(2). func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() level := args[1].Int() name := args[2].Int() optValAddr := args[3].Pointer() optLenAddr := args[4].Pointer() // Get socket from the file descriptor. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return 0, nil, syscall.EBADF + return 0, nil, syserror.EBADF } defer file.DecRef() // Extract the socket. s, ok := file.FileOperations.(socket.Socket) if !ok { - return 0, nil, syscall.ENOTSOCK + return 0, nil, syserror.ENOTSOCK } // Read the length if present. Reject negative values. @@ -464,7 +455,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } if optLen < 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } } @@ -521,30 +512,30 @@ func getSockOpt(t *kernel.Task, s socket.Socket, level, name, len int) (interfac // // Note that unlike Linux, enabling SO_PASSCRED does not autobind the socket. func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() level := args[1].Int() name := args[2].Int() optValAddr := args[3].Pointer() optLen := args[4].Int() // Get socket from the file descriptor. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return 0, nil, syscall.EBADF + return 0, nil, syserror.EBADF } defer file.DecRef() // Extract the socket. s, ok := file.FileOperations.(socket.Socket) if !ok { - return 0, nil, syscall.ENOTSOCK + return 0, nil, syserror.ENOTSOCK } if optLen <= 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } if optLen > maxOptLen { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } buf := t.CopyScratchBuffer(int(optLen)) if _, err := t.CopyIn(optValAddr, &buf); err != nil { @@ -561,21 +552,21 @@ func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy // GetSockName implements the linux syscall getsockname(2). func GetSockName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() addrlen := args[2].Pointer() // Get socket from the file descriptor. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return 0, nil, syscall.EBADF + return 0, nil, syserror.EBADF } defer file.DecRef() // Extract the socket. s, ok := file.FileOperations.(socket.Socket) if !ok { - return 0, nil, syscall.ENOTSOCK + return 0, nil, syserror.ENOTSOCK } // Get the socket name and copy it to the caller. @@ -589,21 +580,21 @@ func GetSockName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S // GetPeerName implements the linux syscall getpeername(2). func GetPeerName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() addrlen := args[2].Pointer() // Get socket from the file descriptor. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return 0, nil, syscall.EBADF + return 0, nil, syserror.EBADF } defer file.DecRef() // Extract the socket. s, ok := file.FileOperations.(socket.Socket) if !ok { - return 0, nil, syscall.ENOTSOCK + return 0, nil, syserror.ENOTSOCK } // Get the socket peer name and copy it to the caller. @@ -617,31 +608,31 @@ func GetPeerName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S // RecvMsg implements the linux syscall recvmsg(2). func RecvMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() msgPtr := args[1].Pointer() flags := args[2].Int() if t.Arch().Width() != 8 { // We only handle 64-bit for now. - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } // Get socket from the file descriptor. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return 0, nil, syscall.EBADF + return 0, nil, syserror.EBADF } defer file.DecRef() // Extract the socket. s, ok := file.FileOperations.(socket.Socket) if !ok { - return 0, nil, syscall.ENOTSOCK + return 0, nil, syserror.ENOTSOCK } // Reject flags that we don't handle yet. if flags & ^(baseRecvFlags|linux.MSG_PEEK|linux.MSG_CMSG_CLOEXEC|linux.MSG_ERRQUEUE) != 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } if file.Flags().NonBlocking { @@ -663,7 +654,7 @@ func RecvMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca // RecvMMsg implements the linux syscall recvmmsg(2). func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() msgPtr := args[1].Pointer() vlen := args[2].Uint() flags := args[3].Int() @@ -671,25 +662,25 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if t.Arch().Width() != 8 { // We only handle 64-bit for now. - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } // Reject flags that we don't handle yet. if flags & ^(baseRecvFlags|linux.MSG_CMSG_CLOEXEC|linux.MSG_ERRQUEUE) != 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } // Get socket from the file descriptor. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return 0, nil, syscall.EBADF + return 0, nil, syserror.EBADF } defer file.DecRef() // Extract the socket. s, ok := file.FileOperations.(socket.Socket) if !ok { - return 0, nil, syscall.ENOTSOCK + return 0, nil, syserror.ENOTSOCK } if file.Flags().NonBlocking { @@ -704,7 +695,7 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc return 0, nil, err } if !ts.Valid() { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } deadline = t.Kernel().MonotonicClock().Now().Add(ts.ToDuration()) haveDeadline = true @@ -724,7 +715,7 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc for i := uint64(0); i < uint64(vlen); i++ { mp, ok := msgPtr.AddLength(i * multipleMessageHeader64Len) if !ok { - return 0, nil, syscall.EFAULT + return 0, nil, syserror.EFAULT } var n uintptr if n, err = recvSingleMsg(t, s, mp, flags, haveDeadline, deadline); err != nil { @@ -734,7 +725,7 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // Copy the received length to the caller. lp, ok := mp.AddLength(messageHeader64Len) if !ok { - return 0, nil, syscall.EFAULT + return 0, nil, syserror.EFAULT } if _, err = t.CopyOut(lp, uint32(n)); err != nil { break @@ -756,7 +747,7 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i } if msg.IovLen > linux.UIO_MAXIOV { - return 0, syscall.EMSGSIZE + return 0, syserror.EMSGSIZE } dst, err := t.IovecsIOSequence(usermem.Addr(msg.Iov), int(msg.IovLen), usermem.IOOpts{ AddressSpaceActive: true, @@ -767,7 +758,7 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i // FIXME(b/63594852): Pretend we have an empty error queue. if flags&linux.MSG_ERRQUEUE != 0 { - return 0, syscall.EAGAIN + return 0, syserror.EAGAIN } // Fast path when no control message nor name buffers are provided. @@ -792,7 +783,7 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i } if msg.ControlLen > maxControlLen { - return 0, syscall.ENOBUFS + return 0, syserror.ENOBUFS } n, mflags, sender, senderLen, cms, e := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, msg.NameLen != 0, msg.ControlLen) if e != nil { @@ -842,27 +833,27 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i // recvFrom is the implementation of the recvfrom syscall. It is called by // recvfrom and recv syscall handlers. -func recvFrom(t *kernel.Task, fd kdefs.FD, bufPtr usermem.Addr, bufLen uint64, flags int32, namePtr usermem.Addr, nameLenPtr usermem.Addr) (uintptr, error) { +func recvFrom(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flags int32, namePtr usermem.Addr, nameLenPtr usermem.Addr) (uintptr, error) { if int(bufLen) < 0 { - return 0, syscall.EINVAL + return 0, syserror.EINVAL } // Reject flags that we don't handle yet. if flags & ^(baseRecvFlags|linux.MSG_PEEK|linux.MSG_CONFIRM) != 0 { - return 0, syscall.EINVAL + return 0, syserror.EINVAL } // Get socket from the file descriptor. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return 0, syscall.EBADF + return 0, syserror.EBADF } defer file.DecRef() // Extract the socket. s, ok := file.FileOperations.(socket.Socket) if !ok { - return 0, syscall.ENOTSOCK + return 0, syserror.ENOTSOCK } if file.Flags().NonBlocking { @@ -903,7 +894,7 @@ func recvFrom(t *kernel.Task, fd kdefs.FD, bufPtr usermem.Addr, bufLen uint64, f // RecvFrom implements the linux syscall recvfrom(2). func RecvFrom(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() bufPtr := args[1].Pointer() bufLen := args[2].Uint64() flags := args[3].Int() @@ -916,31 +907,31 @@ func RecvFrom(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // SendMsg implements the linux syscall sendmsg(2). func SendMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() msgPtr := args[1].Pointer() flags := args[2].Int() if t.Arch().Width() != 8 { // We only handle 64-bit for now. - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } // Get socket from the file descriptor. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return 0, nil, syscall.EBADF + return 0, nil, syserror.EBADF } defer file.DecRef() // Extract the socket. s, ok := file.FileOperations.(socket.Socket) if !ok { - return 0, nil, syscall.ENOTSOCK + return 0, nil, syserror.ENOTSOCK } // Reject flags that we don't handle yet. if flags & ^(linux.MSG_DONTWAIT|linux.MSG_EOR|linux.MSG_MORE|linux.MSG_NOSIGNAL) != 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } if file.Flags().NonBlocking { @@ -953,32 +944,32 @@ func SendMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca // SendMMsg implements the linux syscall sendmmsg(2). func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() msgPtr := args[1].Pointer() vlen := args[2].Uint() flags := args[3].Int() if t.Arch().Width() != 8 { // We only handle 64-bit for now. - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } // Get socket from the file descriptor. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return 0, nil, syscall.EBADF + return 0, nil, syserror.EBADF } defer file.DecRef() // Extract the socket. s, ok := file.FileOperations.(socket.Socket) if !ok { - return 0, nil, syscall.ENOTSOCK + return 0, nil, syserror.ENOTSOCK } // Reject flags that we don't handle yet. if flags & ^(linux.MSG_DONTWAIT|linux.MSG_EOR|linux.MSG_MORE|linux.MSG_NOSIGNAL) != 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } if file.Flags().NonBlocking { @@ -990,7 +981,7 @@ func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc for i := uint64(0); i < uint64(vlen); i++ { mp, ok := msgPtr.AddLength(i * multipleMessageHeader64Len) if !ok { - return 0, nil, syscall.EFAULT + return 0, nil, syserror.EFAULT } var n uintptr if n, err = sendSingleMsg(t, s, file, mp, flags); err != nil { @@ -1000,7 +991,7 @@ func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // Copy the received length to the caller. lp, ok := mp.AddLength(messageHeader64Len) if !ok { - return 0, nil, syscall.EFAULT + return 0, nil, syserror.EFAULT } if _, err = t.CopyOut(lp, uint32(n)); err != nil { break @@ -1025,7 +1016,7 @@ func sendSingleMsg(t *kernel.Task, s socket.Socket, file *fs.File, msgPtr userme if msg.ControlLen > 0 { // Put an upper bound to prevent large allocations. if msg.ControlLen > maxControlLen { - return 0, syscall.ENOBUFS + return 0, syserror.ENOBUFS } controlData = make([]byte, msg.ControlLen) if _, err := t.CopyIn(usermem.Addr(msg.Control), &controlData); err != nil { @@ -1045,7 +1036,7 @@ func sendSingleMsg(t *kernel.Task, s socket.Socket, file *fs.File, msgPtr userme // Read data then call the sendmsg implementation. if msg.IovLen > linux.UIO_MAXIOV { - return 0, syscall.EMSGSIZE + return 0, syserror.EMSGSIZE } src, err := t.IovecsIOSequence(usermem.Addr(msg.Iov), int(msg.IovLen), usermem.IOOpts{ AddressSpaceActive: true, @@ -1079,23 +1070,23 @@ func sendSingleMsg(t *kernel.Task, s socket.Socket, file *fs.File, msgPtr userme // sendTo is the implementation of the sendto syscall. It is called by sendto // and send syscall handlers. -func sendTo(t *kernel.Task, fd kdefs.FD, bufPtr usermem.Addr, bufLen uint64, flags int32, namePtr usermem.Addr, nameLen uint32) (uintptr, error) { +func sendTo(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flags int32, namePtr usermem.Addr, nameLen uint32) (uintptr, error) { bl := int(bufLen) if bl < 0 { - return 0, syscall.EINVAL + return 0, syserror.EINVAL } // Get socket from the file descriptor. - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { - return 0, syscall.EBADF + return 0, syserror.EBADF } defer file.DecRef() // Extract the socket. s, ok := file.FileOperations.(socket.Socket) if !ok { - return 0, syscall.ENOTSOCK + return 0, syserror.ENOTSOCK } if file.Flags().NonBlocking { @@ -1135,7 +1126,7 @@ func sendTo(t *kernel.Task, fd kdefs.FD, bufPtr usermem.Addr, bufLen uint64, fla // SendTo implements the linux syscall sendto(2). func SendTo(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() bufPtr := args[1].Pointer() bufLen := args[2].Uint64() flags := args[3].Int() diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go index b6517313f..a7c98efcb 100644 --- a/pkg/sentry/syscalls/linux/sys_splice.go +++ b/pkg/sentry/syscalls/linux/sys_splice.go @@ -19,7 +19,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/waiter" ) @@ -81,8 +80,8 @@ func doSplice(t *kernel.Task, outFile, inFile *fs.File, opts fs.SpliceOpts, nonB // Sendfile implements linux system call sendfile(2). func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - outFD := kdefs.FD(args[0].Int()) - inFD := kdefs.FD(args[1].Int()) + outFD := args[0].Int() + inFD := args[1].Int() offsetAddr := args[2].Pointer() count := int64(args[3].SizeT()) @@ -92,13 +91,13 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } // Get files. - outFile := t.FDMap().GetFile(outFD) + outFile := t.GetFile(outFD) if outFile == nil { return 0, nil, syserror.EBADF } defer outFile.DecRef() - inFile := t.FDMap().GetFile(inFD) + inFile := t.GetFile(inFD) if inFile == nil { return 0, nil, syserror.EBADF } @@ -163,9 +162,9 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // Splice implements splice(2). func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - inFD := kdefs.FD(args[0].Int()) + inFD := args[0].Int() inOffset := args[1].Pointer() - outFD := kdefs.FD(args[2].Int()) + outFD := args[2].Int() outOffset := args[3].Pointer() count := int64(args[4].SizeT()) flags := args[5].Int() @@ -182,13 +181,13 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal nonBlocking := (flags & linux.SPLICE_F_NONBLOCK) != 0 // Get files. - outFile := t.FDMap().GetFile(outFD) + outFile := t.GetFile(outFD) if outFile == nil { return 0, nil, syserror.EBADF } defer outFile.DecRef() - inFile := t.FDMap().GetFile(inFD) + inFile := t.GetFile(inFD) if inFile == nil { return 0, nil, syserror.EBADF } @@ -251,8 +250,8 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Tee imlements tee(2). func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - inFD := kdefs.FD(args[0].Int()) - outFD := kdefs.FD(args[1].Int()) + inFD := args[0].Int() + outFD := args[1].Int() count := int64(args[2].SizeT()) flags := args[3].Int() @@ -265,13 +264,13 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo nonBlocking := (flags & linux.SPLICE_F_NONBLOCK) != 0 // Get files. - outFile := t.FDMap().GetFile(outFD) + outFile := t.GetFile(outFD) if outFile == nil { return 0, nil, syserror.EBADF } defer outFile.DecRef() - inFile := t.FDMap().GetFile(inFD) + inFile := t.GetFile(inFD) if inFile == nil { return 0, nil, syserror.EBADF } diff --git a/pkg/sentry/syscalls/linux/sys_stat.go b/pkg/sentry/syscalls/linux/sys_stat.go index 9a5657254..5556bc276 100644 --- a/pkg/sentry/syscalls/linux/sys_stat.go +++ b/pkg/sentry/syscalls/linux/sys_stat.go @@ -20,7 +20,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" ) @@ -42,7 +41,7 @@ func Stat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC // Fstatat implements linux syscall newfstatat, i.e. fstatat(2). func Fstatat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() statAddr := args[2].Pointer() flags := args[3].Int() @@ -54,7 +53,7 @@ func Fstatat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if path == "" { // Annoying. What's wrong with fstat? - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -93,10 +92,10 @@ func Lstat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Fstat implements linux syscall fstat(2). func Fstat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() statAddr := args[1].Pointer() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -178,7 +177,7 @@ func copyOutStat(t *kernel.Task, dst usermem.Addr, sattr fs.StableAttr, uattr fs // Statx implements linux syscall statx(2). func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() pathAddr := args[1].Pointer() flags := args[2].Int() mask := args[3].Uint() @@ -197,7 +196,7 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } if path == "" { - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -285,10 +284,10 @@ func Statfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Fstatfs implements linux syscall fstatfs(2). func Fstatfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() statfsAddr := args[1].Pointer() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } diff --git a/pkg/sentry/syscalls/linux/sys_sync.go b/pkg/sentry/syscalls/linux/sys_sync.go index 37225735f..3e55235bd 100644 --- a/pkg/sentry/syscalls/linux/sys_sync.go +++ b/pkg/sentry/syscalls/linux/sys_sync.go @@ -19,7 +19,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" "gvisor.dev/gvisor/pkg/syserror" ) @@ -32,9 +31,9 @@ func Sync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC // Syncfs implements linux system call syncfs(2). func Syncfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -47,9 +46,9 @@ func Syncfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Fsync implements linux syscall fsync(2). func Fsync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -63,9 +62,9 @@ func Fsync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // // At the moment, it just calls Fsync, which is a big hammer, but correct. func Fdatasync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -79,6 +78,7 @@ func Fdatasync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys func SyncFileRange(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { var err error + fd := args[0].Int() offset := args[1].Int64() nbytes := args[2].Int64() uflags := args[3].Uint() @@ -97,8 +97,7 @@ func SyncFileRange(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel nbytes = fs.FileMaxOffset } - fd := kdefs.FD(args[0].Int()) - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } diff --git a/pkg/sentry/syscalls/linux/sys_thread.go b/pkg/sentry/syscalls/linux/sys_thread.go index 9e037bd7b..595eb9155 100644 --- a/pkg/sentry/syscalls/linux/sys_thread.go +++ b/pkg/sentry/syscalls/linux/sys_thread.go @@ -123,29 +123,29 @@ func ExitGroup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys func clone(t *kernel.Task, flags int, stack usermem.Addr, parentTID usermem.Addr, childTID usermem.Addr, tls usermem.Addr) (uintptr, *kernel.SyscallControl, error) { opts := kernel.CloneOptions{ SharingOptions: kernel.SharingOptions{ - NewAddressSpace: flags&syscall.CLONE_VM == 0, - NewSignalHandlers: flags&syscall.CLONE_SIGHAND == 0, - NewThreadGroup: flags&syscall.CLONE_THREAD == 0, + NewAddressSpace: flags&linux.CLONE_VM == 0, + NewSignalHandlers: flags&linux.CLONE_SIGHAND == 0, + NewThreadGroup: flags&linux.CLONE_THREAD == 0, TerminationSignal: linux.Signal(flags & exitSignalMask), - NewPIDNamespace: flags&syscall.CLONE_NEWPID == syscall.CLONE_NEWPID, - NewUserNamespace: flags&syscall.CLONE_NEWUSER == syscall.CLONE_NEWUSER, - NewNetworkNamespace: flags&syscall.CLONE_NEWNET == syscall.CLONE_NEWNET, - NewFiles: flags&syscall.CLONE_FILES == 0, - NewFSContext: flags&syscall.CLONE_FS == 0, - NewUTSNamespace: flags&syscall.CLONE_NEWUTS == syscall.CLONE_NEWUTS, - NewIPCNamespace: flags&syscall.CLONE_NEWIPC == syscall.CLONE_NEWIPC, + NewPIDNamespace: flags&linux.CLONE_NEWPID == linux.CLONE_NEWPID, + NewUserNamespace: flags&linux.CLONE_NEWUSER == linux.CLONE_NEWUSER, + NewNetworkNamespace: flags&linux.CLONE_NEWNET == linux.CLONE_NEWNET, + NewFiles: flags&linux.CLONE_FILES == 0, + NewFSContext: flags&linux.CLONE_FS == 0, + NewUTSNamespace: flags&linux.CLONE_NEWUTS == linux.CLONE_NEWUTS, + NewIPCNamespace: flags&linux.CLONE_NEWIPC == linux.CLONE_NEWIPC, }, Stack: stack, - SetTLS: flags&syscall.CLONE_SETTLS == syscall.CLONE_SETTLS, + SetTLS: flags&linux.CLONE_SETTLS == linux.CLONE_SETTLS, TLS: tls, - ChildClearTID: flags&syscall.CLONE_CHILD_CLEARTID == syscall.CLONE_CHILD_CLEARTID, - ChildSetTID: flags&syscall.CLONE_CHILD_SETTID == syscall.CLONE_CHILD_SETTID, + ChildClearTID: flags&linux.CLONE_CHILD_CLEARTID == linux.CLONE_CHILD_CLEARTID, + ChildSetTID: flags&linux.CLONE_CHILD_SETTID == linux.CLONE_CHILD_SETTID, ChildTID: childTID, - ParentSetTID: flags&syscall.CLONE_PARENT_SETTID == syscall.CLONE_PARENT_SETTID, + ParentSetTID: flags&linux.CLONE_PARENT_SETTID == linux.CLONE_PARENT_SETTID, ParentTID: parentTID, - Vfork: flags&syscall.CLONE_VFORK == syscall.CLONE_VFORK, - Untraced: flags&syscall.CLONE_UNTRACED == syscall.CLONE_UNTRACED, - InheritTracer: flags&syscall.CLONE_PTRACE == syscall.CLONE_PTRACE, + Vfork: flags&linux.CLONE_VFORK == linux.CLONE_VFORK, + Untraced: flags&linux.CLONE_UNTRACED == linux.CLONE_UNTRACED, + InheritTracer: flags&linux.CLONE_PTRACE == linux.CLONE_PTRACE, } ntid, ctrl, err := t.Clone(&opts) return uintptr(ntid), ctrl, err @@ -168,7 +168,7 @@ func Clone(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall func Fork(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { // "A call to fork() is equivalent to a call to clone(2) specifying flags // as just SIGCHLD." - fork(2) - return clone(t, int(syscall.SIGCHLD), 0, 0, 0, 0) + return clone(t, int(linux.SIGCHLD), 0, 0, 0, 0) } // Vfork implements Linux syscall vfork(2). @@ -178,7 +178,7 @@ func Vfork(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // // CLONE_VM | CLONE_VFORK | SIGCHLD // """ - vfork(2) - return clone(t, syscall.CLONE_VM|syscall.CLONE_VFORK|int(syscall.SIGCHLD), 0, 0, 0, 0) + return clone(t, linux.CLONE_VM|linux.CLONE_VFORK|int(linux.SIGCHLD), 0, 0, 0, 0) } // parseCommonWaitOptions applies the options common to wait4 and waitid to @@ -193,7 +193,7 @@ func parseCommonWaitOptions(wopts *kernel.WaitOptions, options int) error { wopts.NonCloneTasks = true wopts.CloneTasks = true default: - return syscall.EINVAL + return syserror.EINVAL } if options&linux.WCONTINUED != 0 { wopts.Events |= kernel.EventGroupContinue @@ -210,7 +210,7 @@ func parseCommonWaitOptions(wopts *kernel.WaitOptions, options int) error { // wait4 waits for the given child process to exit. func wait4(t *kernel.Task, pid int, statusAddr usermem.Addr, options int, rusageAddr usermem.Addr) (uintptr, error) { if options&^(linux.WNOHANG|linux.WUNTRACED|linux.WCONTINUED|linux.WNOTHREAD|linux.WALL|linux.WCLONE) != 0 { - return 0, syscall.EINVAL + return 0, syserror.EINVAL } wopts := kernel.WaitOptions{ Events: kernel.EventExit | kernel.EventTraceeStop, @@ -291,10 +291,10 @@ func Waitid(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal rusageAddr := args[4].Pointer() if options&^(linux.WNOHANG|linux.WEXITED|linux.WSTOPPED|linux.WCONTINUED|linux.WNOWAIT|linux.WNOTHREAD|linux.WALL|linux.WCLONE) != 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } if options&(linux.WEXITED|linux.WSTOPPED|linux.WCONTINUED) == 0 { - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } wopts := kernel.WaitOptions{ Events: kernel.EventTraceeStop, @@ -307,7 +307,7 @@ func Waitid(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal case linux.P_PGID: wopts.SpecificPGID = kernel.ProcessGroupID(id) default: - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } if err := parseCommonWaitOptions(&wopts, options); err != nil { @@ -347,12 +347,12 @@ func Waitid(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return 0, nil, nil } si := arch.SignalInfo{ - Signo: int32(syscall.SIGCHLD), + Signo: int32(linux.SIGCHLD), } si.SetPid(int32(wr.TID)) si.SetUid(int32(wr.UID)) // TODO(b/73541790): convert kernel.ExitStatus to functions and make - // WaitResult.Status a linux.WaitStatus + // WaitResult.Status a linux.WaitStatus. s := syscall.WaitStatus(wr.Status) switch { case s.Exited(): @@ -374,7 +374,7 @@ func Waitid(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal } case s.Continued(): si.Code = arch.CLD_CONTINUED - si.SetStatus(int32(syscall.SIGCONT)) + si.SetStatus(int32(linux.SIGCONT)) default: t.Warningf("waitid got incomprehensible wait status %d", s) } @@ -395,16 +395,16 @@ func SetTidAddress(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel func Unshare(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { flags := args[0].Int() opts := kernel.SharingOptions{ - NewAddressSpace: flags&syscall.CLONE_VM == syscall.CLONE_VM, - NewSignalHandlers: flags&syscall.CLONE_SIGHAND == syscall.CLONE_SIGHAND, - NewThreadGroup: flags&syscall.CLONE_THREAD == syscall.CLONE_THREAD, - NewPIDNamespace: flags&syscall.CLONE_NEWPID == syscall.CLONE_NEWPID, - NewUserNamespace: flags&syscall.CLONE_NEWUSER == syscall.CLONE_NEWUSER, - NewNetworkNamespace: flags&syscall.CLONE_NEWNET == syscall.CLONE_NEWNET, - NewFiles: flags&syscall.CLONE_FILES == syscall.CLONE_FILES, - NewFSContext: flags&syscall.CLONE_FS == syscall.CLONE_FS, - NewUTSNamespace: flags&syscall.CLONE_NEWUTS == syscall.CLONE_NEWUTS, - NewIPCNamespace: flags&syscall.CLONE_NEWIPC == syscall.CLONE_NEWIPC, + NewAddressSpace: flags&linux.CLONE_VM == linux.CLONE_VM, + NewSignalHandlers: flags&linux.CLONE_SIGHAND == linux.CLONE_SIGHAND, + NewThreadGroup: flags&linux.CLONE_THREAD == linux.CLONE_THREAD, + NewPIDNamespace: flags&linux.CLONE_NEWPID == linux.CLONE_NEWPID, + NewUserNamespace: flags&linux.CLONE_NEWUSER == linux.CLONE_NEWUSER, + NewNetworkNamespace: flags&linux.CLONE_NEWNET == linux.CLONE_NEWNET, + NewFiles: flags&linux.CLONE_FILES == linux.CLONE_FILES, + NewFSContext: flags&linux.CLONE_FS == linux.CLONE_FS, + NewUTSNamespace: flags&linux.CLONE_NEWUTS == linux.CLONE_NEWUTS, + NewIPCNamespace: flags&linux.CLONE_NEWIPC == linux.CLONE_NEWIPC, } // "CLONE_NEWPID automatically implies CLONE_THREAD as well." - unshare(2) if opts.NewPIDNamespace { @@ -623,7 +623,7 @@ func Getpriority(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S who := kernel.ThreadID(args[1].Int()) switch which { - case syscall.PRIO_PROCESS: + case linux.PRIO_PROCESS: // Look for who, return ESRCH if not found. var task *kernel.Task if who == 0 { @@ -633,7 +633,7 @@ func Getpriority(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S } if task == nil { - return 0, nil, syscall.ESRCH + return 0, nil, syserror.ESRCH } // From kernel/sys.c:getpriority: @@ -641,13 +641,13 @@ func Getpriority(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S // will not return the normal nice-value, but a negated // value that has been offset by 20" return uintptr(20 - task.Niceness()), nil, nil - case syscall.PRIO_USER: + case linux.PRIO_USER: fallthrough - case syscall.PRIO_PGRP: + case linux.PRIO_PGRP: // PRIO_USER and PRIO_PGRP have no further implementation yet. return 0, nil, nil default: - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } } @@ -669,7 +669,7 @@ func Setpriority(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S } switch which { - case syscall.PRIO_PROCESS: + case linux.PRIO_PROCESS: // Look for who, return ESRCH if not found. var task *kernel.Task if who == 0 { @@ -679,17 +679,17 @@ func Setpriority(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S } if task == nil { - return 0, nil, syscall.ESRCH + return 0, nil, syserror.ESRCH } task.SetNiceness(niceval) - case syscall.PRIO_USER: + case linux.PRIO_USER: fallthrough - case syscall.PRIO_PGRP: + case linux.PRIO_PGRP: // PRIO_USER and PRIO_PGRP have no further implementation yet. return 0, nil, nil default: - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } return 0, nil, nil diff --git a/pkg/sentry/syscalls/linux/sys_timer.go b/pkg/sentry/syscalls/linux/sys_timer.go index ca5ccb7c3..d4134207b 100644 --- a/pkg/sentry/syscalls/linux/sys_timer.go +++ b/pkg/sentry/syscalls/linux/sys_timer.go @@ -15,13 +15,13 @@ package linux import ( - "syscall" "time" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/syserror" ) const nsecPerSec = int64(time.Second) @@ -47,7 +47,7 @@ func copyItimerValIn(t *kernel.Task, addr usermem.Addr) (linux.ItimerVal, error) return itv, nil default: - return linux.ItimerVal{}, syscall.ENOSYS + return linux.ItimerVal{}, syserror.ENOSYS } } @@ -65,7 +65,7 @@ func copyItimerValOut(t *kernel.Task, addr usermem.Addr, itv *linux.ItimerVal) e _, err := t.CopyOut(addr, itv) return err default: - return syscall.ENOSYS + return syserror.ENOSYS } } diff --git a/pkg/sentry/syscalls/linux/sys_timerfd.go b/pkg/sentry/syscalls/linux/sys_timerfd.go index ea6d44315..1ce5ce4c3 100644 --- a/pkg/sentry/syscalls/linux/sys_timerfd.go +++ b/pkg/sentry/syscalls/linux/sys_timerfd.go @@ -20,7 +20,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/timerfd" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/syserror" ) @@ -49,9 +48,9 @@ func TimerfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel NonBlocking: flags&linux.TFD_NONBLOCK != 0, }) - fd, err := t.FDMap().NewFDFrom(0, f, kernel.FDFlags{ + fd, err := t.NewFDFrom(0, f, kernel.FDFlags{ CloseOnExec: flags&linux.TFD_CLOEXEC != 0, - }, t.ThreadGroup().Limits()) + }) if err != nil { return 0, nil, err } @@ -61,7 +60,7 @@ func TimerfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel // TimerfdSettime implements Linux syscall timerfd_settime(2). func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() flags := args[1].Int() newValAddr := args[2].Pointer() oldValAddr := args[3].Pointer() @@ -70,7 +69,7 @@ func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne return 0, nil, syserror.EINVAL } - f := t.FDMap().GetFile(fd) + f := t.GetFile(fd) if f == nil { return 0, nil, syserror.EBADF } @@ -101,10 +100,10 @@ func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne // TimerfdGettime implements Linux syscall timerfd_gettime(2). func TimerfdGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() curValAddr := args[1].Pointer() - f := t.FDMap().GetFile(fd) + f := t.GetFile(fd) if f == nil { return 0, nil, syserror.EBADF } diff --git a/pkg/sentry/syscalls/linux/sys_tls.go b/pkg/sentry/syscalls/linux/sys_tls.go index e3d1c6201..b3eb96a1c 100644 --- a/pkg/sentry/syscalls/linux/sys_tls.go +++ b/pkg/sentry/syscalls/linux/sys_tls.go @@ -17,11 +17,10 @@ package linux import ( - "syscall" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/syserror" ) // ArchPrctl implements linux syscall arch_prctl(2). @@ -39,14 +38,14 @@ func ArchPrctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys case linux.ARCH_SET_FS: fsbase := args[1].Uint64() if !t.Arch().SetTLS(uintptr(fsbase)) { - return 0, nil, syscall.EPERM + return 0, nil, syserror.EPERM } case linux.ARCH_GET_GS, linux.ARCH_SET_GS: t.Kernel().EmitUnimplementedEvent(t) fallthrough default: - return 0, nil, syscall.EINVAL + return 0, nil, syserror.EINVAL } return 0, nil, nil diff --git a/pkg/sentry/syscalls/linux/sys_write.go b/pkg/sentry/syscalls/linux/sys_write.go index 3a5bf9ac4..5278c96a6 100644 --- a/pkg/sentry/syscalls/linux/sys_write.go +++ b/pkg/sentry/syscalls/linux/sys_write.go @@ -21,7 +21,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/kernel/kdefs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/usermem" @@ -39,11 +38,11 @@ const ( // Write implements linux syscall write(2). func Write(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() size := args[2].SizeT() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -75,12 +74,12 @@ func Write(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Pwrite64 implements linux syscall pwrite64(2). func Pwrite64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() size := args[2].SizeT() offset := args[3].Int64() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -122,11 +121,11 @@ func Pwrite64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // Writev implements linux syscall writev(2). func Writev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() iovcnt := int(args[2].Int()) - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -152,12 +151,12 @@ func Writev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Pwritev implements linux syscall pwritev(2). func Pwritev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() iovcnt := int(args[2].Int()) offset := args[3].Int64() - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } @@ -202,7 +201,7 @@ func Pwritev2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // splits the offset argument into a high/low value for compatibility with // 32-bit architectures. The flags argument is the 5th argument. - fd := kdefs.FD(args[0].Int()) + fd := args[0].Int() addr := args[1].Pointer() iovcnt := int(args[2].Int()) offset := args[3].Int64() @@ -212,7 +211,7 @@ func Pwritev2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc return 0, nil, syserror.EACCES } - file := t.FDMap().GetFile(fd) + file := t.GetFile(fd) if file == nil { return 0, nil, syserror.EBADF } diff --git a/pkg/sentry/syscalls/linux/timespec.go b/pkg/sentry/syscalls/linux/timespec.go index 9ba0eba7a..4ff8f9234 100644 --- a/pkg/sentry/syscalls/linux/timespec.go +++ b/pkg/sentry/syscalls/linux/timespec.go @@ -15,7 +15,6 @@ package linux import ( - "syscall" "time" "gvisor.dev/gvisor/pkg/abi/linux" @@ -70,7 +69,7 @@ func copyTimevalIn(t *kernel.Task, addr usermem.Addr) (linux.Timeval, error) { tv.Usec = int64(usermem.ByteOrder.Uint64(in[8:])) return tv, nil default: - return linux.Timeval{}, syscall.ENOSYS + return linux.Timeval{}, syserror.ENOSYS } } @@ -84,7 +83,7 @@ func copyTimevalOut(t *kernel.Task, addr usermem.Addr, tv *linux.Timeval) error _, err := t.CopyOutBytes(addr, out) return err default: - return syscall.ENOSYS + return syserror.ENOSYS } } @@ -104,7 +103,7 @@ func copyTimespecInToDuration(t *kernel.Task, timespecAddr usermem.Addr) (time.D return 0, err } if !timespec.Valid() { - return 0, syscall.EINVAL + return 0, syserror.EINVAL } timeout = time.Duration(timespec.ToNsecCapped()) } diff --git a/pkg/sentry/syscalls/syscalls.go b/pkg/sentry/syscalls/syscalls.go index a5f3d8407..f88055676 100644 --- a/pkg/sentry/syscalls/syscalls.go +++ b/pkg/sentry/syscalls/syscalls.go @@ -26,7 +26,6 @@ package syscalls import ( "fmt" - "syscall" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" @@ -40,16 +39,7 @@ func Supported(name string, fn kernel.SyscallFn) kernel.Syscall { Name: name, Fn: fn, SupportLevel: kernel.SupportFull, - Note: "Full Support", - } -} - -// Undocumented returns a syscall that is undocumented. -func Undocumented(name string, fn kernel.SyscallFn) kernel.Syscall { - return kernel.Syscall{ - Name: name, - Fn: fn, - SupportLevel: kernel.SupportUndocumented, + Note: "Fully Supported.", } } @@ -65,7 +55,7 @@ func PartiallySupported(name string, fn kernel.SyscallFn, note string, urls []st } // Error returns a syscall handler that will always give the passed error. -func Error(name string, err syscall.Errno, note string, urls []string) kernel.Syscall { +func Error(name string, err error, note string, urls []string) kernel.Syscall { if note != "" { note = note + "; " } @@ -75,14 +65,14 @@ func Error(name string, err syscall.Errno, note string, urls []string) kernel.Sy return 0, nil, err }, SupportLevel: kernel.SupportUnimplemented, - Note: fmt.Sprintf("%sReturns %q", note, err.Error()), + Note: fmt.Sprintf("%sReturns %q.", note, err.Error()), URLs: urls, } } // ErrorWithEvent gives a syscall function that sends an unimplemented // syscall event via the event channel and returns the passed error. -func ErrorWithEvent(name string, err syscall.Errno, note string, urls []string) kernel.Syscall { +func ErrorWithEvent(name string, err error, note string, urls []string) kernel.Syscall { if note != "" { note = note + "; " } @@ -93,7 +83,7 @@ func ErrorWithEvent(name string, err syscall.Errno, note string, urls []string) return 0, nil, err }, SupportLevel: kernel.SupportUnimplemented, - Note: fmt.Sprintf("%sReturns %q", note, err.Error()), + Note: fmt.Sprintf("%sReturns %q.", note, err.Error()), URLs: urls, } } @@ -115,7 +105,7 @@ func CapError(name string, c linux.Capability, note string, urls []string) kerne return 0, nil, syserror.ENOSYS }, SupportLevel: kernel.SupportUnimplemented, - Note: fmt.Sprintf("%sReturns %q if the process does not have %s; %q otherwise", note, syserror.EPERM, c.String(), syserror.ENOSYS), + Note: fmt.Sprintf("%sReturns %q if the process does not have %s; %q otherwise.", note, syserror.EPERM, c.String(), syserror.ENOSYS), URLs: urls, } } diff --git a/pkg/sentry/time/seqatomic_parameters.go b/pkg/sentry/time/seqatomic_parameters_unsafe.go index b4fb0a7f0..f6560d0bb 100755 --- a/pkg/sentry/time/seqatomic_parameters.go +++ b/pkg/sentry/time/seqatomic_parameters_unsafe.go @@ -1,11 +1,12 @@ package time import ( + "unsafe" + "fmt" "gvisor.dev/gvisor/third_party/gvsync" "reflect" "strings" - "unsafe" ) // SeqAtomicLoad returns a copy of *ptr, ensuring that the read does not race |