summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/syscalls
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry/syscalls')
-rw-r--r--pkg/sentry/syscalls/linux/BUILD2
-rw-r--r--pkg/sentry/syscalls/linux/linux64.go4
-rw-r--r--pkg/sentry/syscalls/linux/sys_aio.go169
-rw-r--r--pkg/sentry/syscalls/linux/sys_file.go91
-rw-r--r--pkg/sentry/syscalls/linux/sys_futex.go48
-rw-r--r--pkg/sentry/syscalls/linux/sys_socket.go17
-rw-r--r--pkg/sentry/syscalls/linux/sys_splice.go6
-rw-r--r--pkg/sentry/syscalls/linux/sys_xattr.go12
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/BUILD11
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/aio.go216
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/fd.go187
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/filesystem.go16
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/inotify.go137
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/ioctl.go72
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/lock.go64
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/mount.go150
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/read_write.go69
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/setstat.go56
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/socket.go17
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/splice.go486
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/sync.go42
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/vfs2.go132
22 files changed, 1786 insertions, 218 deletions
diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD
index 217fcfef2..4a9b04fd0 100644
--- a/pkg/sentry/syscalls/linux/BUILD
+++ b/pkg/sentry/syscalls/linux/BUILD
@@ -99,5 +99,7 @@ go_library(
"//pkg/syserror",
"//pkg/usermem",
"//pkg/waiter",
+ "//tools/go_marshal/marshal",
+ "//tools/go_marshal/primitive",
],
)
diff --git a/pkg/sentry/syscalls/linux/linux64.go b/pkg/sentry/syscalls/linux/linux64.go
index ea4f9b1a7..80c65164a 100644
--- a/pkg/sentry/syscalls/linux/linux64.go
+++ b/pkg/sentry/syscalls/linux/linux64.go
@@ -325,8 +325,8 @@ var AMD64 = &kernel.SyscallTable{
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),
+ 273: syscalls.Supported("set_robust_list", SetRobustList),
+ 274: syscalls.Supported("get_robust_list", GetRobustList),
275: syscalls.Supported("splice", Splice),
276: syscalls.Supported("tee", Tee),
277: syscalls.PartiallySupported("sync_file_range", SyncFileRange, "Full data flush is not guaranteed at this time.", nil),
diff --git a/pkg/sentry/syscalls/linux/sys_aio.go b/pkg/sentry/syscalls/linux/sys_aio.go
index d781d6a04..ba2557c52 100644
--- a/pkg/sentry/syscalls/linux/sys_aio.go
+++ b/pkg/sentry/syscalls/linux/sys_aio.go
@@ -15,8 +15,8 @@
package linux
import (
- "encoding/binary"
-
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/sentry/arch"
"gvisor.dev/gvisor/pkg/sentry/fs"
"gvisor.dev/gvisor/pkg/sentry/kernel"
@@ -27,59 +27,6 @@ import (
"gvisor.dev/gvisor/pkg/usermem"
)
-// I/O commands.
-const (
- _IOCB_CMD_PREAD = 0
- _IOCB_CMD_PWRITE = 1
- _IOCB_CMD_FSYNC = 2
- _IOCB_CMD_FDSYNC = 3
- _IOCB_CMD_NOOP = 6
- _IOCB_CMD_PREADV = 7
- _IOCB_CMD_PWRITEV = 8
-)
-
-// I/O flags.
-const (
- _IOCB_FLAG_RESFD = 1
-)
-
-// ioCallback describes an I/O request.
-//
-// The priority field is currently ignored in the implementation below. Also
-// note that the IOCB_FLAG_RESFD feature is not supported.
-type ioCallback struct {
- Data uint64
- Key uint32
- Reserved1 uint32
-
- OpCode uint16
- ReqPrio int16
- FD int32
-
- Buf uint64
- Bytes uint64
- Offset int64
-
- Reserved2 uint64
- Flags uint32
-
- // eventfd to signal if IOCB_FLAG_RESFD is set in flags.
- ResFD int32
-}
-
-// ioEvent describes an I/O result.
-//
-// +stateify savable
-type ioEvent struct {
- Data uint64
- Obj uint64
- Result int64
- Result2 int64
-}
-
-// ioEventSize is the size of an ioEvent encoded.
-var ioEventSize = binary.Size(ioEvent{})
-
// IoSetup implements linux syscall io_setup(2).
func IoSetup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
nrEvents := args[0].Int()
@@ -192,7 +139,7 @@ func IoGetevents(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S
}
}
- ev := v.(*ioEvent)
+ ev := v.(*linux.IOEvent)
// Copy out the result.
if _, err := t.CopyOut(eventsAddr, ev); err != nil {
@@ -204,7 +151,7 @@ func IoGetevents(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S
}
// Keep rolling.
- eventsAddr += usermem.Addr(ioEventSize)
+ eventsAddr += usermem.Addr(linux.IOEventSize)
}
// Everything finished.
@@ -231,7 +178,7 @@ func waitForRequest(ctx *mm.AIOContext, t *kernel.Task, haveDeadline bool, deadl
}
// memoryFor returns appropriate memory for the given callback.
-func memoryFor(t *kernel.Task, cb *ioCallback) (usermem.IOSequence, error) {
+func memoryFor(t *kernel.Task, cb *linux.IOCallback) (usermem.IOSequence, error) {
bytes := int(cb.Bytes)
if bytes < 0 {
// Linux also requires that this field fit in ssize_t.
@@ -242,17 +189,17 @@ func memoryFor(t *kernel.Task, cb *ioCallback) (usermem.IOSequence, error) {
// we have no guarantee that t's AddressSpace will be active during the
// I/O.
switch cb.OpCode {
- case _IOCB_CMD_PREAD, _IOCB_CMD_PWRITE:
+ case linux.IOCB_CMD_PREAD, linux.IOCB_CMD_PWRITE:
return t.SingleIOSequence(usermem.Addr(cb.Buf), bytes, usermem.IOOpts{
AddressSpaceActive: false,
})
- case _IOCB_CMD_PREADV, _IOCB_CMD_PWRITEV:
+ case linux.IOCB_CMD_PREADV, linux.IOCB_CMD_PWRITEV:
return t.IovecsIOSequence(usermem.Addr(cb.Buf), bytes, usermem.IOOpts{
AddressSpaceActive: false,
})
- case _IOCB_CMD_FSYNC, _IOCB_CMD_FDSYNC, _IOCB_CMD_NOOP:
+ case linux.IOCB_CMD_FSYNC, linux.IOCB_CMD_FDSYNC, linux.IOCB_CMD_NOOP:
return usermem.IOSequence{}, nil
default:
@@ -261,54 +208,62 @@ func memoryFor(t *kernel.Task, cb *ioCallback) (usermem.IOSequence, error) {
}
}
-func performCallback(t *kernel.Task, file *fs.File, cbAddr usermem.Addr, cb *ioCallback, ioseq usermem.IOSequence, ctx *mm.AIOContext, eventFile *fs.File) {
- if ctx.Dead() {
- ctx.CancelPendingRequest()
- return
- }
- ev := &ioEvent{
- Data: cb.Data,
- Obj: uint64(cbAddr),
- }
+// IoCancel implements linux syscall io_cancel(2).
+//
+// It is not presently supported (ENOSYS indicates no support on this
+// architecture).
+func IoCancel(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ return 0, nil, syserror.ENOSYS
+}
- // Construct a context.Context that will not be interrupted if t is
- // interrupted.
- c := t.AsyncContext()
+// LINT.IfChange
- var err error
- switch cb.OpCode {
- case _IOCB_CMD_PREAD, _IOCB_CMD_PREADV:
- ev.Result, err = file.Preadv(c, ioseq, cb.Offset)
- case _IOCB_CMD_PWRITE, _IOCB_CMD_PWRITEV:
- ev.Result, err = file.Pwritev(c, ioseq, cb.Offset)
- case _IOCB_CMD_FSYNC:
- err = file.Fsync(c, 0, fs.FileMaxOffset, fs.SyncAll)
- case _IOCB_CMD_FDSYNC:
- err = file.Fsync(c, 0, fs.FileMaxOffset, fs.SyncData)
- }
+func getAIOCallback(t *kernel.Task, file *fs.File, cbAddr usermem.Addr, cb *linux.IOCallback, ioseq usermem.IOSequence, actx *mm.AIOContext, eventFile *fs.File) kernel.AIOCallback {
+ return func(ctx context.Context) {
+ if actx.Dead() {
+ actx.CancelPendingRequest()
+ return
+ }
+ ev := &linux.IOEvent{
+ Data: cb.Data,
+ Obj: uint64(cbAddr),
+ }
- // Update the result.
- if err != nil {
- err = handleIOError(t, ev.Result != 0 /* partial */, err, nil /* never interrupted */, "aio", file)
- ev.Result = -int64(kernel.ExtractErrno(err, 0))
- }
+ var err error
+ switch cb.OpCode {
+ case linux.IOCB_CMD_PREAD, linux.IOCB_CMD_PREADV:
+ ev.Result, err = file.Preadv(ctx, ioseq, cb.Offset)
+ case linux.IOCB_CMD_PWRITE, linux.IOCB_CMD_PWRITEV:
+ ev.Result, err = file.Pwritev(ctx, ioseq, cb.Offset)
+ case linux.IOCB_CMD_FSYNC:
+ err = file.Fsync(ctx, 0, fs.FileMaxOffset, fs.SyncAll)
+ case linux.IOCB_CMD_FDSYNC:
+ err = file.Fsync(ctx, 0, fs.FileMaxOffset, fs.SyncData)
+ }
+
+ // Update the result.
+ if err != nil {
+ err = handleIOError(t, ev.Result != 0 /* partial */, err, nil /* never interrupted */, "aio", file)
+ ev.Result = -int64(kernel.ExtractErrno(err, 0))
+ }
- file.DecRef()
+ file.DecRef()
- // Queue the result for delivery.
- ctx.FinishRequest(ev)
+ // Queue the result for delivery.
+ actx.FinishRequest(ev)
- // Notify the event file if one was specified. This needs to happen
- // *after* queueing the result to avoid racing with the thread we may
- // wake up.
- if eventFile != nil {
- eventFile.FileOperations.(*eventfd.EventOperations).Signal(1)
- eventFile.DecRef()
+ // Notify the event file if one was specified. This needs to happen
+ // *after* queueing the result to avoid racing with the thread we may
+ // wake up.
+ if eventFile != nil {
+ eventFile.FileOperations.(*eventfd.EventOperations).Signal(1)
+ eventFile.DecRef()
+ }
}
}
// submitCallback processes a single callback.
-func submitCallback(t *kernel.Task, id uint64, cb *ioCallback, cbAddr usermem.Addr) error {
+func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr usermem.Addr) error {
file := t.GetFile(cb.FD)
if file == nil {
// File not found.
@@ -318,7 +273,7 @@ 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 {
+ if cb.Flags&linux.IOCB_FLAG_RESFD != 0 {
eventFile = t.GetFile(cb.ResFD)
if eventFile == nil {
// Bad FD.
@@ -340,7 +295,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *ioCallback, cbAddr usermem.Ad
// Check offset for reads/writes.
switch cb.OpCode {
- case _IOCB_CMD_PREAD, _IOCB_CMD_PREADV, _IOCB_CMD_PWRITE, _IOCB_CMD_PWRITEV:
+ case linux.IOCB_CMD_PREAD, linux.IOCB_CMD_PREADV, linux.IOCB_CMD_PWRITE, linux.IOCB_CMD_PWRITEV:
if cb.Offset < 0 {
return syserror.EINVAL
}
@@ -366,7 +321,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *ioCallback, cbAddr usermem.Ad
// Perform the request asynchronously.
file.IncRef()
- fs.Async(func() { performCallback(t, file, cbAddr, cb, ioseq, ctx, eventFile) })
+ t.QueueAIO(getAIOCallback(t, file, cbAddr, cb, ioseq, ctx, eventFile))
// All set.
return nil
@@ -395,7 +350,7 @@ func IoSubmit(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc
}
// Copy in this callback.
- var cb ioCallback
+ var cb linux.IOCallback
cbAddr := usermem.Addr(t.Arch().Value(cbAddrNative))
if _, err := t.CopyIn(cbAddr, &cb); err != nil {
@@ -424,10 +379,4 @@ func IoSubmit(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc
return uintptr(nrEvents), nil, nil
}
-// IoCancel implements linux syscall io_cancel(2).
-//
-// It is not presently supported (ENOSYS indicates no support on this
-// architecture).
-func IoCancel(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
- return 0, nil, syserror.ENOSYS
-}
+// LINT.ThenChange(vfs2/aio.go)
diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go
index 35a98212a..8cf6401e7 100644
--- a/pkg/sentry/syscalls/linux/sys_file.go
+++ b/pkg/sentry/syscalls/linux/sys_file.go
@@ -900,14 +900,20 @@ func fGetOwn(t *kernel.Task, file *fs.File) int32 {
//
// If who is positive, it represents a PID. If negative, it represents a PGID.
// If the PID or PGID is invalid, the owner is silently unset.
-func fSetOwn(t *kernel.Task, file *fs.File, who int32) {
+func fSetOwn(t *kernel.Task, file *fs.File, who int32) error {
a := file.Async(fasync.New).(*fasync.FileAsync)
if who < 0 {
+ // Check for overflow before flipping the sign.
+ if who-1 > who {
+ return syserror.EINVAL
+ }
pg := t.PIDNamespace().ProcessGroupWithID(kernel.ProcessGroupID(-who))
a.SetOwnerProcessGroup(t, pg)
+ } else {
+ tg := t.PIDNamespace().ThreadGroupWithID(kernel.ThreadID(who))
+ a.SetOwnerThreadGroup(t, tg)
}
- tg := t.PIDNamespace().ThreadGroupWithID(kernel.ThreadID(who))
- a.SetOwnerThreadGroup(t, tg)
+ return nil
}
// Fcntl implements linux syscall fcntl(2).
@@ -935,10 +941,10 @@ 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.FDTable().SetFlags(fd, kernel.FDFlags{
+ err := t.FDTable().SetFlags(fd, kernel.FDFlags{
CloseOnExec: flags&linux.FD_CLOEXEC != 0,
})
- return 0, nil, nil
+ return 0, nil, err
case linux.F_GETFL:
return uintptr(file.Flags().ToLinux()), nil, nil
case linux.F_SETFL:
@@ -998,9 +1004,6 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
return 0, nil, err
}
- // 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 {
@@ -1010,12 +1013,12 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
}
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) {
+ if !file.Dirent.Inode.LockCtx.Posix.LockRegion(t.FDTable(), lock.ReadLock, rng, nil) {
return 0, nil, syserror.EAGAIN
}
} else {
// Blocking lock, pass in the task to satisfy the lock.Blocker interface.
- if !file.Dirent.Inode.LockCtx.Posix.LockRegion(lockUniqueID, lock.ReadLock, rng, t) {
+ if !file.Dirent.Inode.LockCtx.Posix.LockRegion(t.FDTable(), lock.ReadLock, rng, t) {
return 0, nil, syserror.EINTR
}
}
@@ -1026,18 +1029,18 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
}
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) {
+ if !file.Dirent.Inode.LockCtx.Posix.LockRegion(t.FDTable(), lock.WriteLock, rng, nil) {
return 0, nil, syserror.EAGAIN
}
} else {
// Blocking lock, pass in the task to satisfy the lock.Blocker interface.
- if !file.Dirent.Inode.LockCtx.Posix.LockRegion(lockUniqueID, lock.WriteLock, rng, t) {
+ if !file.Dirent.Inode.LockCtx.Posix.LockRegion(t.FDTable(), lock.WriteLock, rng, t) {
return 0, nil, syserror.EINTR
}
}
return 0, nil, nil
case linux.F_UNLCK:
- file.Dirent.Inode.LockCtx.Posix.UnlockRegion(lockUniqueID, rng)
+ file.Dirent.Inode.LockCtx.Posix.UnlockRegion(t.FDTable(), rng)
return 0, nil, nil
default:
return 0, nil, syserror.EINVAL
@@ -1045,8 +1048,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
case linux.F_GETOWN:
return uintptr(fGetOwn(t, file)), nil, nil
case linux.F_SETOWN:
- fSetOwn(t, file, args[2].Int())
- return 0, nil, nil
+ return 0, nil, fSetOwn(t, file, args[2].Int())
case linux.F_GETOWN_EX:
addr := args[2].Pointer()
owner := fGetOwnEx(t, file)
@@ -1055,7 +1057,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
case linux.F_SETOWN_EX:
addr := args[2].Pointer()
var owner linux.FOwnerEx
- n, err := t.CopyIn(addr, &owner)
+ _, err := t.CopyIn(addr, &owner)
if err != nil {
return 0, nil, err
}
@@ -1067,21 +1069,21 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
return 0, nil, syserror.ESRCH
}
a.SetOwnerTask(t, task)
- return uintptr(n), nil, nil
+ return 0, nil, nil
case linux.F_OWNER_PID:
tg := t.PIDNamespace().ThreadGroupWithID(kernel.ThreadID(owner.PID))
if tg == nil {
return 0, nil, syserror.ESRCH
}
a.SetOwnerThreadGroup(t, tg)
- return uintptr(n), nil, nil
+ return 0, nil, nil
case linux.F_OWNER_PGRP:
pg := t.PIDNamespace().ProcessGroupWithID(kernel.ProcessGroupID(owner.PID))
if pg == nil {
return 0, nil, syserror.ESRCH
}
a.SetOwnerProcessGroup(t, pg)
- return uintptr(n), nil, nil
+ return 0, nil, nil
default:
return 0, nil, syserror.EINVAL
}
@@ -1114,17 +1116,6 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
}
}
-// LINT.ThenChange(vfs2/fd.go)
-
-const (
- _FADV_NORMAL = 0
- _FADV_RANDOM = 1
- _FADV_SEQUENTIAL = 2
- _FADV_WILLNEED = 3
- _FADV_DONTNEED = 4
- _FADV_NOREUSE = 5
-)
-
// 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) {
@@ -1149,12 +1140,12 @@ func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys
}
switch advice {
- case _FADV_NORMAL:
- case _FADV_RANDOM:
- case _FADV_SEQUENTIAL:
- case _FADV_WILLNEED:
- case _FADV_DONTNEED:
- case _FADV_NOREUSE:
+ case linux.POSIX_FADV_NORMAL:
+ case linux.POSIX_FADV_RANDOM:
+ case linux.POSIX_FADV_SEQUENTIAL:
+ case linux.POSIX_FADV_WILLNEED:
+ case linux.POSIX_FADV_DONTNEED:
+ case linux.POSIX_FADV_NOREUSE:
default:
return 0, nil, syserror.EINVAL
}
@@ -1163,8 +1154,6 @@ func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys
return 0, nil, nil
}
-// LINT.IfChange
-
func mkdirAt(t *kernel.Task, dirFD int32, addr usermem.Addr, mode linux.FileMode) error {
path, _, err := copyInPath(t, addr, false /* allowEmpty */)
if err != nil {
@@ -2157,22 +2146,6 @@ func Flock(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
nonblocking := operation&linux.LOCK_NB != 0
operation &^= linux.LOCK_NB
- // flock(2):
- // Locks created by flock() are associated with an open file table entry. This means that
- // duplicate file descriptors (created by, for example, fork(2) or dup(2)) refer to the
- // same lock, and this lock may be modified or released using any of these descriptors. Furthermore,
- // the lock is released either by an explicit LOCK_UN operation on any of these duplicate
- // descriptors, or when all such descriptors have been closed.
- //
- // If a process uses open(2) (or similar) to obtain more than one descriptor for the same file,
- // these descriptors are treated independently by flock(). An attempt to lock the file using
- // one of these file descriptors may be denied by a lock that the calling process has already placed via
- // another descriptor.
- //
- // We use the File UniqueID as the lock UniqueID because it needs to reference the same lock across dup(2)
- // and fork(2).
- lockUniqueID := lock.UniqueID(file.UniqueID)
-
// A BSD style lock spans the entire file.
rng := lock.LockRange{
Start: 0,
@@ -2183,29 +2156,29 @@ func Flock(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
case linux.LOCK_EX:
if nonblocking {
// Since we're nonblocking we pass a nil lock.Blocker implementation.
- if !file.Dirent.Inode.LockCtx.BSD.LockRegion(lockUniqueID, lock.WriteLock, rng, nil) {
+ if !file.Dirent.Inode.LockCtx.BSD.LockRegion(file, lock.WriteLock, rng, nil) {
return 0, nil, syserror.EWOULDBLOCK
}
} else {
// Because we're blocking we will pass the task to satisfy the lock.Blocker interface.
- if !file.Dirent.Inode.LockCtx.BSD.LockRegion(lockUniqueID, lock.WriteLock, rng, t) {
+ if !file.Dirent.Inode.LockCtx.BSD.LockRegion(file, lock.WriteLock, rng, t) {
return 0, nil, syserror.EINTR
}
}
case linux.LOCK_SH:
if nonblocking {
// Since we're nonblocking we pass a nil lock.Blocker implementation.
- if !file.Dirent.Inode.LockCtx.BSD.LockRegion(lockUniqueID, lock.ReadLock, rng, nil) {
+ if !file.Dirent.Inode.LockCtx.BSD.LockRegion(file, lock.ReadLock, rng, nil) {
return 0, nil, syserror.EWOULDBLOCK
}
} else {
// Because we're blocking we will pass the task to satisfy the lock.Blocker interface.
- if !file.Dirent.Inode.LockCtx.BSD.LockRegion(lockUniqueID, lock.ReadLock, rng, t) {
+ if !file.Dirent.Inode.LockCtx.BSD.LockRegion(file, lock.ReadLock, rng, t) {
return 0, nil, syserror.EINTR
}
}
case linux.LOCK_UN:
- file.Dirent.Inode.LockCtx.BSD.UnlockRegion(lockUniqueID, rng)
+ file.Dirent.Inode.LockCtx.BSD.UnlockRegion(file, rng)
default:
// flock(2): EINVAL operation is invalid.
return 0, nil, syserror.EINVAL
diff --git a/pkg/sentry/syscalls/linux/sys_futex.go b/pkg/sentry/syscalls/linux/sys_futex.go
index b68261f72..f04d78856 100644
--- a/pkg/sentry/syscalls/linux/sys_futex.go
+++ b/pkg/sentry/syscalls/linux/sys_futex.go
@@ -198,7 +198,7 @@ func Futex(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
switch cmd {
case linux.FUTEX_WAIT:
// WAIT uses a relative timeout.
- mask = ^uint32(0)
+ mask = linux.FUTEX_BITSET_MATCH_ANY
var timeoutDur time.Duration
if !forever {
timeoutDur = time.Duration(timespec.ToNsecCapped()) * time.Nanosecond
@@ -286,3 +286,49 @@ func Futex(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
return 0, nil, syserror.ENOSYS
}
}
+
+// SetRobustList implements linux syscall set_robust_list(2).
+func SetRobustList(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ // Despite the syscall using the name 'pid' for this variable, it is
+ // very much a tid.
+ head := args[0].Pointer()
+ length := args[1].SizeT()
+
+ if length != uint(linux.SizeOfRobustListHead) {
+ return 0, nil, syserror.EINVAL
+ }
+ t.SetRobustList(head)
+ return 0, nil, nil
+}
+
+// GetRobustList implements linux syscall get_robust_list(2).
+func GetRobustList(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ // Despite the syscall using the name 'pid' for this variable, it is
+ // very much a tid.
+ tid := args[0].Int()
+ head := args[1].Pointer()
+ size := args[2].Pointer()
+
+ if tid < 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ ot := t
+ if tid != 0 {
+ if ot = t.PIDNamespace().TaskWithID(kernel.ThreadID(tid)); ot == nil {
+ return 0, nil, syserror.ESRCH
+ }
+ }
+
+ // Copy out head pointer.
+ if _, err := t.CopyOut(head, uint64(ot.GetRobustList())); err != nil {
+ return 0, nil, err
+ }
+
+ // Copy out size, which is a constant.
+ if _, err := t.CopyOut(size, uint64(linux.SizeOfRobustListHead)); err != nil {
+ return 0, nil, err
+ }
+
+ return 0, nil, nil
+}
diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go
index 0760af77b..414fce8e3 100644
--- a/pkg/sentry/syscalls/linux/sys_socket.go
+++ b/pkg/sentry/syscalls/linux/sys_socket.go
@@ -29,6 +29,8 @@ import (
"gvisor.dev/gvisor/pkg/syserr"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/usermem"
+ "gvisor.dev/gvisor/tools/go_marshal/marshal"
+ "gvisor.dev/gvisor/tools/go_marshal/primitive"
)
// LINT.IfChange
@@ -474,7 +476,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy
}
if v != nil {
- if _, err := t.CopyOut(optValAddr, v); err != nil {
+ if _, err := v.CopyOut(t, optValAddr); err != nil {
return 0, nil, err
}
}
@@ -484,7 +486,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy
// getSockOpt tries to handle common socket options, or dispatches to a specific
// socket implementation.
-func getSockOpt(t *kernel.Task, s socket.Socket, level, name int, optValAddr usermem.Addr, len int) (interface{}, *syserr.Error) {
+func getSockOpt(t *kernel.Task, s socket.Socket, level, name int, optValAddr usermem.Addr, len int) (marshal.Marshallable, *syserr.Error) {
if level == linux.SOL_SOCKET {
switch name {
case linux.SO_TYPE, linux.SO_DOMAIN, linux.SO_PROTOCOL:
@@ -496,13 +498,16 @@ func getSockOpt(t *kernel.Task, s socket.Socket, level, name int, optValAddr use
switch name {
case linux.SO_TYPE:
_, skType, _ := s.Type()
- return int32(skType), nil
+ v := primitive.Int32(skType)
+ return &v, nil
case linux.SO_DOMAIN:
family, _, _ := s.Type()
- return int32(family), nil
+ v := primitive.Int32(family)
+ return &v, nil
case linux.SO_PROTOCOL:
_, _, protocol := s.Type()
- return int32(protocol), nil
+ v := primitive.Int32(protocol)
+ return &v, nil
}
}
@@ -539,7 +544,7 @@ func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy
return 0, nil, syserror.EINVAL
}
buf := t.CopyScratchBuffer(int(optLen))
- if _, err := t.CopyIn(optValAddr, &buf); err != nil {
+ if _, err := t.CopyInBytes(optValAddr, buf); err != nil {
return 0, nil, err
}
diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go
index 39f2b79ec..77c78889d 100644
--- a/pkg/sentry/syscalls/linux/sys_splice.go
+++ b/pkg/sentry/syscalls/linux/sys_splice.go
@@ -80,6 +80,12 @@ func doSplice(t *kernel.Task, outFile, inFile *fs.File, opts fs.SpliceOpts, nonB
}
}
+ if total > 0 {
+ // On Linux, inotify behavior is not very consistent with splice(2). We try
+ // our best to emulate Linux for very basic calls to splice, where for some
+ // reason, events are generated for output files, but not input files.
+ outFile.Dirent.InotifyEvent(linux.IN_MODIFY, 0)
+ }
return total, err
}
diff --git a/pkg/sentry/syscalls/linux/sys_xattr.go b/pkg/sentry/syscalls/linux/sys_xattr.go
index 2de5e3422..c24946160 100644
--- a/pkg/sentry/syscalls/linux/sys_xattr.go
+++ b/pkg/sentry/syscalls/linux/sys_xattr.go
@@ -207,7 +207,11 @@ func setXattr(t *kernel.Task, d *fs.Dirent, nameAddr, valueAddr usermem.Addr, si
return syserror.EOPNOTSUPP
}
- return d.Inode.SetXattr(t, d, name, value, flags)
+ if err := d.Inode.SetXattr(t, d, name, value, flags); err != nil {
+ return err
+ }
+ d.InotifyEvent(linux.IN_ATTRIB, 0)
+ return nil
}
func copyInXattrName(t *kernel.Task, nameAddr usermem.Addr) (string, error) {
@@ -418,7 +422,11 @@ func removeXattr(t *kernel.Task, d *fs.Dirent, nameAddr usermem.Addr) error {
return syserror.EOPNOTSUPP
}
- return d.Inode.RemoveXattr(t, d, name)
+ if err := d.Inode.RemoveXattr(t, d, name); err != nil {
+ return err
+ }
+ d.InotifyEvent(linux.IN_ATTRIB, 0)
+ return nil
}
// LINT.ThenChange(vfs2/xattr.go)
diff --git a/pkg/sentry/syscalls/linux/vfs2/BUILD b/pkg/sentry/syscalls/linux/vfs2/BUILD
index f882ef840..64696b438 100644
--- a/pkg/sentry/syscalls/linux/vfs2/BUILD
+++ b/pkg/sentry/syscalls/linux/vfs2/BUILD
@@ -5,6 +5,7 @@ package(licenses = ["notice"])
go_library(
name = "vfs2",
srcs = [
+ "aio.go",
"epoll.go",
"eventfd.go",
"execve.go",
@@ -12,9 +13,12 @@ go_library(
"filesystem.go",
"fscontext.go",
"getdents.go",
+ "inotify.go",
"ioctl.go",
+ "lock.go",
"memfd.go",
"mmap.go",
+ "mount.go",
"path.go",
"pipe.go",
"poll.go",
@@ -22,6 +26,7 @@ go_library(
"setstat.go",
"signal.go",
"socket.go",
+ "splice.go",
"stat.go",
"stat_amd64.go",
"stat_arm64.go",
@@ -36,9 +41,11 @@ go_library(
"//pkg/abi/linux",
"//pkg/binary",
"//pkg/bits",
+ "//pkg/context",
"//pkg/fspath",
"//pkg/gohacks",
"//pkg/sentry/arch",
+ "//pkg/sentry/fs/lock",
"//pkg/sentry/fsbridge",
"//pkg/sentry/fsimpl/eventfd",
"//pkg/sentry/fsimpl/pipefs",
@@ -47,11 +54,13 @@ go_library(
"//pkg/sentry/fsimpl/tmpfs",
"//pkg/sentry/kernel",
"//pkg/sentry/kernel/auth",
+ "//pkg/sentry/kernel/fasync",
"//pkg/sentry/kernel/pipe",
"//pkg/sentry/kernel/time",
"//pkg/sentry/limits",
"//pkg/sentry/loader",
"//pkg/sentry/memmap",
+ "//pkg/sentry/mm",
"//pkg/sentry/socket",
"//pkg/sentry/socket/control",
"//pkg/sentry/socket/unix/transport",
@@ -63,5 +72,7 @@ go_library(
"//pkg/syserror",
"//pkg/usermem",
"//pkg/waiter",
+ "//tools/go_marshal/marshal",
+ "//tools/go_marshal/primitive",
],
)
diff --git a/pkg/sentry/syscalls/linux/vfs2/aio.go b/pkg/sentry/syscalls/linux/vfs2/aio.go
new file mode 100644
index 000000000..e5cdefc50
--- /dev/null
+++ b/pkg/sentry/syscalls/linux/vfs2/aio.go
@@ -0,0 +1,216 @@
+// 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 vfs2
+
+import (
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/context"
+ "gvisor.dev/gvisor/pkg/sentry/arch"
+ "gvisor.dev/gvisor/pkg/sentry/fsimpl/eventfd"
+ "gvisor.dev/gvisor/pkg/sentry/kernel"
+ "gvisor.dev/gvisor/pkg/sentry/mm"
+ slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
+ "gvisor.dev/gvisor/pkg/syserror"
+ "gvisor.dev/gvisor/pkg/usermem"
+)
+
+// IoSubmit implements linux syscall io_submit(2).
+func IoSubmit(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ id := args[0].Uint64()
+ nrEvents := args[1].Int()
+ addr := args[2].Pointer()
+
+ if nrEvents < 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ for i := int32(0); i < nrEvents; i++ {
+ // Copy in the address.
+ cbAddrNative := t.Arch().Native(0)
+ if _, err := t.CopyIn(addr, cbAddrNative); err != nil {
+ if i > 0 {
+ // Some successful.
+ return uintptr(i), nil, nil
+ }
+ // Nothing done.
+ return 0, nil, err
+ }
+
+ // Copy in this callback.
+ var cb linux.IOCallback
+ cbAddr := usermem.Addr(t.Arch().Value(cbAddrNative))
+ if _, err := t.CopyIn(cbAddr, &cb); err != nil {
+ if i > 0 {
+ // Some have been successful.
+ return uintptr(i), nil, nil
+ }
+ // Nothing done.
+ return 0, nil, err
+ }
+
+ // Process this callback.
+ if err := submitCallback(t, id, &cb, cbAddr); err != nil {
+ if i > 0 {
+ // Partial success.
+ return uintptr(i), nil, nil
+ }
+ // Nothing done.
+ return 0, nil, err
+ }
+
+ // Advance to the next one.
+ addr += usermem.Addr(t.Arch().Width())
+ }
+
+ return uintptr(nrEvents), nil, nil
+}
+
+// submitCallback processes a single callback.
+func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr usermem.Addr) error {
+ if cb.Reserved2 != 0 {
+ return syserror.EINVAL
+ }
+
+ fd := t.GetFileVFS2(cb.FD)
+ if fd == nil {
+ return syserror.EBADF
+ }
+ defer fd.DecRef()
+
+ // Was there an eventFD? Extract it.
+ var eventFD *vfs.FileDescription
+ if cb.Flags&linux.IOCB_FLAG_RESFD != 0 {
+ eventFD = t.GetFileVFS2(cb.ResFD)
+ if eventFD == nil {
+ return syserror.EBADF
+ }
+ defer eventFD.DecRef()
+
+ // Check that it is an eventfd.
+ if _, ok := eventFD.Impl().(*eventfd.EventFileDescription); !ok {
+ return syserror.EINVAL
+ }
+ }
+
+ ioseq, err := memoryFor(t, cb)
+ if err != nil {
+ return err
+ }
+
+ // Check offset for reads/writes.
+ switch cb.OpCode {
+ case linux.IOCB_CMD_PREAD, linux.IOCB_CMD_PREADV, linux.IOCB_CMD_PWRITE, linux.IOCB_CMD_PWRITEV:
+ if cb.Offset < 0 {
+ return syserror.EINVAL
+ }
+ }
+
+ // Prepare the request.
+ aioCtx, ok := t.MemoryManager().LookupAIOContext(t, id)
+ if !ok {
+ return syserror.EINVAL
+ }
+ if ready := aioCtx.Prepare(); !ready {
+ // Context is busy.
+ return syserror.EAGAIN
+ }
+
+ if eventFD != nil {
+ // The request is set. Make sure there's a ref on the file.
+ //
+ // This is necessary when the callback executes on completion,
+ // which is also what will release this reference.
+ eventFD.IncRef()
+ }
+
+ // Perform the request asynchronously.
+ fd.IncRef()
+ t.QueueAIO(getAIOCallback(t, fd, eventFD, cbAddr, cb, ioseq, aioCtx))
+ return nil
+}
+
+func getAIOCallback(t *kernel.Task, fd, eventFD *vfs.FileDescription, cbAddr usermem.Addr, cb *linux.IOCallback, ioseq usermem.IOSequence, aioCtx *mm.AIOContext) kernel.AIOCallback {
+ return func(ctx context.Context) {
+ if aioCtx.Dead() {
+ aioCtx.CancelPendingRequest()
+ return
+ }
+ ev := &linux.IOEvent{
+ Data: cb.Data,
+ Obj: uint64(cbAddr),
+ }
+
+ var err error
+ switch cb.OpCode {
+ case linux.IOCB_CMD_PREAD, linux.IOCB_CMD_PREADV:
+ ev.Result, err = fd.PRead(ctx, ioseq, cb.Offset, vfs.ReadOptions{})
+ case linux.IOCB_CMD_PWRITE, linux.IOCB_CMD_PWRITEV:
+ ev.Result, err = fd.PWrite(ctx, ioseq, cb.Offset, vfs.WriteOptions{})
+ case linux.IOCB_CMD_FSYNC, linux.IOCB_CMD_FDSYNC:
+ err = fd.Sync(ctx)
+ }
+
+ // Update the result.
+ if err != nil {
+ err = slinux.HandleIOErrorVFS2(t, ev.Result != 0 /* partial */, err, nil /* never interrupted */, "aio", fd)
+ ev.Result = -int64(kernel.ExtractErrno(err, 0))
+ }
+
+ fd.DecRef()
+
+ // Queue the result for delivery.
+ aioCtx.FinishRequest(ev)
+
+ // Notify the event file if one was specified. This needs to happen
+ // *after* queueing the result to avoid racing with the thread we may
+ // wake up.
+ if eventFD != nil {
+ eventFD.Impl().(*eventfd.EventFileDescription).Signal(1)
+ eventFD.DecRef()
+ }
+ }
+}
+
+// memoryFor returns appropriate memory for the given callback.
+func memoryFor(t *kernel.Task, cb *linux.IOCallback) (usermem.IOSequence, error) {
+ bytes := int(cb.Bytes)
+ if bytes < 0 {
+ // Linux also requires that this field fit in ssize_t.
+ return usermem.IOSequence{}, syserror.EINVAL
+ }
+
+ // Since this I/O will be asynchronous with respect to t's task goroutine,
+ // we have no guarantee that t's AddressSpace will be active during the
+ // I/O.
+ switch cb.OpCode {
+ case linux.IOCB_CMD_PREAD, linux.IOCB_CMD_PWRITE:
+ return t.SingleIOSequence(usermem.Addr(cb.Buf), bytes, usermem.IOOpts{
+ AddressSpaceActive: false,
+ })
+
+ case linux.IOCB_CMD_PREADV, linux.IOCB_CMD_PWRITEV:
+ return t.IovecsIOSequence(usermem.Addr(cb.Buf), bytes, usermem.IOOpts{
+ AddressSpaceActive: false,
+ })
+
+ case linux.IOCB_CMD_FSYNC, linux.IOCB_CMD_FDSYNC, linux.IOCB_CMD_NOOP:
+ return usermem.IOSequence{}, nil
+
+ default:
+ // Not a supported command.
+ return usermem.IOSequence{}, syserror.EINVAL
+ }
+}
diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go
index ca0f7fd1e..67f191551 100644
--- a/pkg/sentry/syscalls/linux/vfs2/fd.go
+++ b/pkg/sentry/syscalls/linux/vfs2/fd.go
@@ -17,10 +17,13 @@ package vfs2
import (
"gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/sentry/arch"
+ "gvisor.dev/gvisor/pkg/sentry/fs/lock"
"gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs"
"gvisor.dev/gvisor/pkg/sentry/kernel"
+ "gvisor.dev/gvisor/pkg/sentry/kernel/fasync"
"gvisor.dev/gvisor/pkg/sentry/kernel/pipe"
slinux "gvisor.dev/gvisor/pkg/sentry/syscalls/linux"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/syserror"
)
@@ -134,10 +137,10 @@ 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.FDTable().SetFlags(fd, kernel.FDFlags{
+ err := t.FDTable().SetFlagsVFS2(fd, kernel.FDFlags{
CloseOnExec: flags&linux.FD_CLOEXEC != 0,
})
- return 0, nil, nil
+ return 0, nil, err
case linux.F_GETFL:
return uintptr(file.StatusFlags()), nil, nil
case linux.F_SETFL:
@@ -152,6 +155,41 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
return 0, nil, err
}
return uintptr(n), nil, nil
+ case linux.F_GETOWN:
+ owner, hasOwner := getAsyncOwner(t, file)
+ if !hasOwner {
+ return 0, nil, nil
+ }
+ if owner.Type == linux.F_OWNER_PGRP {
+ return uintptr(-owner.PID), nil, nil
+ }
+ return uintptr(owner.PID), nil, nil
+ case linux.F_SETOWN:
+ who := args[2].Int()
+ ownerType := int32(linux.F_OWNER_PID)
+ if who < 0 {
+ // Check for overflow before flipping the sign.
+ if who-1 > who {
+ return 0, nil, syserror.EINVAL
+ }
+ ownerType = linux.F_OWNER_PGRP
+ who = -who
+ }
+ return 0, nil, setAsyncOwner(t, file, ownerType, who)
+ case linux.F_GETOWN_EX:
+ owner, hasOwner := getAsyncOwner(t, file)
+ if !hasOwner {
+ return 0, nil, nil
+ }
+ _, err := t.CopyOut(args[2].Pointer(), &owner)
+ return 0, nil, err
+ case linux.F_SETOWN_EX:
+ var owner linux.FOwnerEx
+ _, err := t.CopyIn(args[2].Pointer(), &owner)
+ if err != nil {
+ return 0, nil, err
+ }
+ return 0, nil, setAsyncOwner(t, file, owner.Type, owner.PID)
case linux.F_GETPIPE_SZ:
pipefile, ok := file.Impl().(*pipe.VFSPipeFD)
if !ok {
@@ -167,8 +205,151 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
}
err := tmpfs.AddSeals(file, args[2].Uint())
return 0, nil, err
+ case linux.F_SETLK, linux.F_SETLKW:
+ return 0, nil, posixLock(t, args, file, cmd)
+ default:
+ // TODO(gvisor.dev/issue/2920): Everything else is not yet supported.
+ return 0, nil, syserror.EINVAL
+ }
+}
+
+func getAsyncOwner(t *kernel.Task, fd *vfs.FileDescription) (ownerEx linux.FOwnerEx, hasOwner bool) {
+ a := fd.AsyncHandler()
+ if a == nil {
+ return linux.FOwnerEx{}, false
+ }
+
+ ot, otg, opg := a.(*fasync.FileAsync).Owner()
+ switch {
+ case ot != nil:
+ return linux.FOwnerEx{
+ Type: linux.F_OWNER_TID,
+ PID: int32(t.PIDNamespace().IDOfTask(ot)),
+ }, true
+ case otg != nil:
+ return linux.FOwnerEx{
+ Type: linux.F_OWNER_PID,
+ PID: int32(t.PIDNamespace().IDOfThreadGroup(otg)),
+ }, true
+ case opg != nil:
+ return linux.FOwnerEx{
+ Type: linux.F_OWNER_PGRP,
+ PID: int32(t.PIDNamespace().IDOfProcessGroup(opg)),
+ }, true
+ default:
+ return linux.FOwnerEx{}, true
+ }
+}
+
+func setAsyncOwner(t *kernel.Task, fd *vfs.FileDescription, ownerType, pid int32) error {
+ switch ownerType {
+ case linux.F_OWNER_TID, linux.F_OWNER_PID, linux.F_OWNER_PGRP:
+ // Acceptable type.
+ default:
+ return syserror.EINVAL
+ }
+
+ a := fd.SetAsyncHandler(fasync.NewVFS2).(*fasync.FileAsync)
+ if pid == 0 {
+ a.ClearOwner()
+ return nil
+ }
+
+ switch ownerType {
+ case linux.F_OWNER_TID:
+ task := t.PIDNamespace().TaskWithID(kernel.ThreadID(pid))
+ if task == nil {
+ return syserror.ESRCH
+ }
+ a.SetOwnerTask(t, task)
+ return nil
+ case linux.F_OWNER_PID:
+ tg := t.PIDNamespace().ThreadGroupWithID(kernel.ThreadID(pid))
+ if tg == nil {
+ return syserror.ESRCH
+ }
+ a.SetOwnerThreadGroup(t, tg)
+ return nil
+ case linux.F_OWNER_PGRP:
+ pg := t.PIDNamespace().ProcessGroupWithID(kernel.ProcessGroupID(pid))
+ if pg == nil {
+ return syserror.ESRCH
+ }
+ a.SetOwnerProcessGroup(t, pg)
+ return nil
+ default:
+ return syserror.EINVAL
+ }
+}
+
+func posixLock(t *kernel.Task, args arch.SyscallArguments, file *vfs.FileDescription, cmd int32) error {
+ // Copy in the lock request.
+ flockAddr := args[2].Pointer()
+ var flock linux.Flock
+ if _, err := t.CopyIn(flockAddr, &flock); err != nil {
+ return err
+ }
+
+ var blocker lock.Blocker
+ if cmd == linux.F_SETLKW {
+ blocker = t
+ }
+
+ switch flock.Type {
+ case linux.F_RDLCK:
+ if !file.IsReadable() {
+ return syserror.EBADF
+ }
+ return file.LockPOSIX(t, t.FDTable(), lock.ReadLock, uint64(flock.Start), uint64(flock.Len), flock.Whence, blocker)
+
+ case linux.F_WRLCK:
+ if !file.IsWritable() {
+ return syserror.EBADF
+ }
+ return file.LockPOSIX(t, t.FDTable(), lock.WriteLock, uint64(flock.Start), uint64(flock.Len), flock.Whence, blocker)
+
+ case linux.F_UNLCK:
+ return file.UnlockPOSIX(t, t.FDTable(), uint64(flock.Start), uint64(flock.Len), flock.Whence)
+
+ default:
+ return syserror.EINVAL
+ }
+}
+
+// Fadvise64 implements fadvise64(2).
+// This implementation currently ignores the provided advice.
+func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ length := args[2].Int64()
+ advice := args[3].Int()
+
+ // Note: offset is allowed to be negative.
+ if length < 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ // If the FD refers to a pipe or FIFO, return error.
+ if _, isPipe := file.Impl().(*pipe.VFSPipeFD); isPipe {
+ return 0, nil, syserror.ESPIPE
+ }
+
+ switch advice {
+ case linux.POSIX_FADV_NORMAL:
+ case linux.POSIX_FADV_RANDOM:
+ case linux.POSIX_FADV_SEQUENTIAL:
+ case linux.POSIX_FADV_WILLNEED:
+ case linux.POSIX_FADV_DONTNEED:
+ case linux.POSIX_FADV_NOREUSE:
default:
- // TODO(gvisor.dev/issue/1623): Everything else is not yet supported.
return 0, nil, syserror.EINVAL
}
+
+ // Sure, whatever.
+ return 0, nil, nil
}
diff --git a/pkg/sentry/syscalls/linux/vfs2/filesystem.go b/pkg/sentry/syscalls/linux/vfs2/filesystem.go
index 46d3e189c..b6d2ddd65 100644
--- a/pkg/sentry/syscalls/linux/vfs2/filesystem.go
+++ b/pkg/sentry/syscalls/linux/vfs2/filesystem.go
@@ -106,7 +106,7 @@ func Mknod(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
addr := args[0].Pointer()
mode := args[1].ModeT()
dev := args[2].Uint()
- return 0, nil, mknodat(t, linux.AT_FDCWD, addr, mode, dev)
+ return 0, nil, mknodat(t, linux.AT_FDCWD, addr, linux.FileMode(mode), dev)
}
// Mknodat implements Linux syscall mknodat(2).
@@ -115,10 +115,10 @@ func Mknodat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca
addr := args[1].Pointer()
mode := args[2].ModeT()
dev := args[3].Uint()
- return 0, nil, mknodat(t, dirfd, addr, mode, dev)
+ return 0, nil, mknodat(t, dirfd, addr, linux.FileMode(mode), dev)
}
-func mknodat(t *kernel.Task, dirfd int32, addr usermem.Addr, mode uint, dev uint32) error {
+func mknodat(t *kernel.Task, dirfd int32, addr usermem.Addr, mode linux.FileMode, dev uint32) error {
path, err := copyInPath(t, addr)
if err != nil {
return err
@@ -128,9 +128,14 @@ func mknodat(t *kernel.Task, dirfd int32, addr usermem.Addr, mode uint, dev uint
return err
}
defer tpop.Release()
+
+ // "Zero file type is equivalent to type S_IFREG." - mknod(2)
+ if mode.FileType() == 0 {
+ mode |= linux.ModeRegular
+ }
major, minor := linux.DecodeDeviceID(dev)
return t.Kernel().VFS().MknodAt(t, t.Credentials(), &tpop.pop, &vfs.MknodOptions{
- Mode: linux.FileMode(mode &^ t.FSContext().Umask()),
+ Mode: mode &^ linux.FileMode(t.FSContext().Umask()),
DevMajor: uint32(major),
DevMinor: minor,
})
@@ -313,6 +318,9 @@ func symlinkat(t *kernel.Task, targetAddr usermem.Addr, newdirfd int32, linkpath
if err != nil {
return err
}
+ if len(target) == 0 {
+ return syserror.ENOENT
+ }
linkpath, err := copyInPath(t, linkpathAddr)
if err != nil {
return err
diff --git a/pkg/sentry/syscalls/linux/vfs2/inotify.go b/pkg/sentry/syscalls/linux/vfs2/inotify.go
new file mode 100644
index 000000000..5d98134a5
--- /dev/null
+++ b/pkg/sentry/syscalls/linux/vfs2/inotify.go
@@ -0,0 +1,137 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package vfs2
+
+import (
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/sentry/arch"
+ "gvisor.dev/gvisor/pkg/sentry/kernel"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
+ "gvisor.dev/gvisor/pkg/syserror"
+)
+
+const allFlags = linux.IN_NONBLOCK | linux.IN_CLOEXEC
+
+// InotifyInit1 implements the inotify_init1() syscalls.
+func InotifyInit1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ flags := args[0].Int()
+ if flags&^allFlags != 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ ino, err := vfs.NewInotifyFD(t, t.Kernel().VFS(), uint32(flags))
+ if err != nil {
+ return 0, nil, err
+ }
+ defer ino.DecRef()
+
+ fd, err := t.NewFDFromVFS2(0, ino, kernel.FDFlags{
+ CloseOnExec: flags&linux.IN_CLOEXEC != 0,
+ })
+
+ if err != nil {
+ return 0, nil, err
+ }
+
+ return uintptr(fd), nil, nil
+}
+
+// InotifyInit implements the inotify_init() syscalls.
+func InotifyInit(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ args[0].Value = 0
+ return InotifyInit1(t, args)
+}
+
+// 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 int32) (*vfs.Inotify, *vfs.FileDescription, error) {
+ f := t.GetFileVFS2(fd)
+ if f == nil {
+ // Invalid fd.
+ return nil, nil, syserror.EBADF
+ }
+
+ ino, ok := f.Impl().(*vfs.Inotify)
+ if !ok {
+ // Not an inotify fd.
+ f.DecRef()
+ return nil, nil, syserror.EINVAL
+ }
+
+ return ino, f, nil
+}
+
+// InotifyAddWatch implements the inotify_add_watch() syscall.
+func InotifyAddWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ addr := args[1].Pointer()
+ mask := args[2].Uint()
+
+ // "EINVAL: The given event mask contains no valid events."
+ // -- inotify_add_watch(2)
+ if mask&linux.ALL_INOTIFY_BITS == 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ // "IN_DONT_FOLLOW: Don't dereference pathname if it is a symbolic link."
+ // -- inotify(7)
+ follow := followFinalSymlink
+ if mask&linux.IN_DONT_FOLLOW == 0 {
+ follow = nofollowFinalSymlink
+ }
+
+ ino, f, err := fdToInotify(t, fd)
+ if err != nil {
+ return 0, nil, err
+ }
+ defer f.DecRef()
+
+ path, err := copyInPath(t, addr)
+ if err != nil {
+ return 0, nil, err
+ }
+ if mask&linux.IN_ONLYDIR != 0 {
+ path.Dir = true
+ }
+ tpop, err := getTaskPathOperation(t, linux.AT_FDCWD, path, disallowEmptyPath, follow)
+ if err != nil {
+ return 0, nil, err
+ }
+ defer tpop.Release()
+ d, err := t.Kernel().VFS().GetDentryAt(t, t.Credentials(), &tpop.pop, &vfs.GetDentryOptions{})
+ if err != nil {
+ return 0, nil, err
+ }
+ defer d.DecRef()
+
+ fd, err = ino.AddWatch(d.Dentry(), mask)
+ if err != nil {
+ return 0, nil, err
+ }
+ return uintptr(fd), nil, nil
+}
+
+// InotifyRmWatch implements the inotify_rm_watch() syscall.
+func InotifyRmWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ wd := args[1].Int()
+
+ ino, f, err := fdToInotify(t, fd)
+ if err != nil {
+ return 0, nil, err
+ }
+ defer f.DecRef()
+ return 0, nil, ino.RmWatch(wd)
+}
diff --git a/pkg/sentry/syscalls/linux/vfs2/ioctl.go b/pkg/sentry/syscalls/linux/vfs2/ioctl.go
index 5a2418da9..fd6ab94b2 100644
--- a/pkg/sentry/syscalls/linux/vfs2/ioctl.go
+++ b/pkg/sentry/syscalls/linux/vfs2/ioctl.go
@@ -15,6 +15,7 @@
package vfs2
import (
+ "gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/sentry/arch"
"gvisor.dev/gvisor/pkg/sentry/kernel"
"gvisor.dev/gvisor/pkg/syserror"
@@ -30,6 +31,77 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
}
defer file.DecRef()
+ // Handle ioctls that apply to all FDs.
+ switch args[1].Int() {
+ case linux.FIONCLEX:
+ t.FDTable().SetFlagsVFS2(fd, kernel.FDFlags{
+ CloseOnExec: false,
+ })
+ return 0, nil, nil
+
+ case linux.FIOCLEX:
+ t.FDTable().SetFlagsVFS2(fd, kernel.FDFlags{
+ CloseOnExec: true,
+ })
+ return 0, nil, nil
+
+ case linux.FIONBIO:
+ var set int32
+ if _, err := t.CopyIn(args[2].Pointer(), &set); err != nil {
+ return 0, nil, err
+ }
+ flags := file.StatusFlags()
+ if set != 0 {
+ flags |= linux.O_NONBLOCK
+ } else {
+ flags &^= linux.O_NONBLOCK
+ }
+ return 0, nil, file.SetStatusFlags(t, t.Credentials(), flags)
+
+ case linux.FIOASYNC:
+ var set int32
+ if _, err := t.CopyIn(args[2].Pointer(), &set); err != nil {
+ return 0, nil, err
+ }
+ flags := file.StatusFlags()
+ if set != 0 {
+ flags |= linux.O_ASYNC
+ } else {
+ flags &^= linux.O_ASYNC
+ }
+ file.SetStatusFlags(t, t.Credentials(), flags)
+ return 0, nil, nil
+
+ case linux.FIOGETOWN, linux.SIOCGPGRP:
+ var who int32
+ owner, hasOwner := getAsyncOwner(t, file)
+ if hasOwner {
+ if owner.Type == linux.F_OWNER_PGRP {
+ who = -owner.PID
+ } else {
+ who = owner.PID
+ }
+ }
+ _, err := t.CopyOut(args[2].Pointer(), &who)
+ return 0, nil, err
+
+ case linux.FIOSETOWN, linux.SIOCSPGRP:
+ var who int32
+ if _, err := t.CopyIn(args[2].Pointer(), &who); err != nil {
+ return 0, nil, err
+ }
+ ownerType := int32(linux.F_OWNER_PID)
+ if who < 0 {
+ // Check for overflow before flipping the sign.
+ if who-1 > who {
+ return 0, nil, syserror.EINVAL
+ }
+ ownerType = linux.F_OWNER_PGRP
+ who = -who
+ }
+ return 0, nil, setAsyncOwner(t, file, ownerType, who)
+ }
+
ret, err := file.Ioctl(t, t.MemoryManager(), args)
return ret, nil, err
}
diff --git a/pkg/sentry/syscalls/linux/vfs2/lock.go b/pkg/sentry/syscalls/linux/vfs2/lock.go
new file mode 100644
index 000000000..bf19028c4
--- /dev/null
+++ b/pkg/sentry/syscalls/linux/vfs2/lock.go
@@ -0,0 +1,64 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package vfs2
+
+import (
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/sentry/arch"
+ "gvisor.dev/gvisor/pkg/sentry/fs/lock"
+ "gvisor.dev/gvisor/pkg/sentry/kernel"
+ "gvisor.dev/gvisor/pkg/syserror"
+)
+
+// Flock implements linux syscall flock(2).
+func Flock(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ operation := args[1].Int()
+
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ // flock(2): EBADF fd is not an open file descriptor.
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ nonblocking := operation&linux.LOCK_NB != 0
+ operation &^= linux.LOCK_NB
+
+ var blocker lock.Blocker
+ if !nonblocking {
+ blocker = t
+ }
+
+ switch operation {
+ case linux.LOCK_EX:
+ if err := file.LockBSD(t, lock.WriteLock, blocker); err != nil {
+ return 0, nil, err
+ }
+ case linux.LOCK_SH:
+ if err := file.LockBSD(t, lock.ReadLock, blocker); err != nil {
+ return 0, nil, err
+ }
+ case linux.LOCK_UN:
+ if err := file.UnlockBSD(t); err != nil {
+ return 0, nil, err
+ }
+ default:
+ // flock(2): EINVAL operation is invalid.
+ return 0, nil, syserror.EINVAL
+ }
+
+ return 0, nil, nil
+}
diff --git a/pkg/sentry/syscalls/linux/vfs2/mount.go b/pkg/sentry/syscalls/linux/vfs2/mount.go
new file mode 100644
index 000000000..ea337de7c
--- /dev/null
+++ b/pkg/sentry/syscalls/linux/vfs2/mount.go
@@ -0,0 +1,150 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package vfs2
+
+import (
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/sentry/arch"
+ "gvisor.dev/gvisor/pkg/sentry/kernel"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
+ "gvisor.dev/gvisor/pkg/syserror"
+ "gvisor.dev/gvisor/pkg/usermem"
+)
+
+// Mount implements Linux syscall mount(2).
+func Mount(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ sourceAddr := args[0].Pointer()
+ targetAddr := args[1].Pointer()
+ typeAddr := args[2].Pointer()
+ flags := args[3].Uint64()
+ dataAddr := args[4].Pointer()
+
+ // For null-terminated strings related to mount(2), Linux copies in at most
+ // a page worth of data. See fs/namespace.c:copy_mount_string().
+ fsType, err := t.CopyInString(typeAddr, usermem.PageSize)
+ if err != nil {
+ return 0, nil, err
+ }
+ source, err := t.CopyInString(sourceAddr, usermem.PageSize)
+ if err != nil {
+ return 0, nil, err
+ }
+
+ targetPath, err := copyInPath(t, targetAddr)
+ if err != nil {
+ return 0, nil, err
+ }
+
+ data := ""
+ if dataAddr != 0 {
+ // In Linux, a full page is always copied in regardless of null
+ // character placement, and the address is passed to each file system.
+ // Most file systems always treat this data as a string, though, and so
+ // do all of the ones we implement.
+ data, err = t.CopyInString(dataAddr, usermem.PageSize)
+ if err != nil {
+ return 0, nil, err
+ }
+ }
+
+ // Ignore magic value that was required before Linux 2.4.
+ if flags&linux.MS_MGC_MSK == linux.MS_MGC_VAL {
+ flags = flags &^ linux.MS_MGC_MSK
+ }
+
+ // Must have CAP_SYS_ADMIN in the current mount namespace's associated user
+ // namespace.
+ creds := t.Credentials()
+ if !creds.HasCapabilityIn(linux.CAP_SYS_ADMIN, t.MountNamespaceVFS2().Owner) {
+ return 0, nil, syserror.EPERM
+ }
+
+ const unsupportedOps = linux.MS_REMOUNT | linux.MS_BIND |
+ linux.MS_SHARED | linux.MS_PRIVATE | linux.MS_SLAVE |
+ linux.MS_UNBINDABLE | linux.MS_MOVE
+
+ // Silently allow MS_NOSUID, since we don't implement set-id bits
+ // anyway.
+ const unsupportedFlags = linux.MS_NODIRATIME | linux.MS_STRICTATIME
+
+ // Linux just allows passing any flags to mount(2) - it won't fail when
+ // unknown or unsupported flags are passed. Since we don't implement
+ // everything, we fail explicitly on flags that are unimplemented.
+ if flags&(unsupportedOps|unsupportedFlags) != 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ var opts vfs.MountOptions
+ if flags&linux.MS_NOATIME == linux.MS_NOATIME {
+ opts.Flags.NoATime = true
+ }
+ if flags&linux.MS_NOEXEC == linux.MS_NOEXEC {
+ opts.Flags.NoExec = true
+ }
+ if flags&linux.MS_NODEV == linux.MS_NODEV {
+ opts.Flags.NoDev = true
+ }
+ if flags&linux.MS_NOSUID == linux.MS_NOSUID {
+ opts.Flags.NoSUID = true
+ }
+ if flags&linux.MS_RDONLY == linux.MS_RDONLY {
+ opts.ReadOnly = true
+ }
+ opts.GetFilesystemOptions.Data = data
+
+ target, err := getTaskPathOperation(t, linux.AT_FDCWD, targetPath, disallowEmptyPath, nofollowFinalSymlink)
+ if err != nil {
+ return 0, nil, err
+ }
+ defer target.Release()
+
+ return 0, nil, t.Kernel().VFS().MountAt(t, creds, source, &target.pop, fsType, &opts)
+}
+
+// Umount2 implements Linux syscall umount2(2).
+func Umount2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ addr := args[0].Pointer()
+ flags := args[1].Int()
+
+ // Must have CAP_SYS_ADMIN in the mount namespace's associated user
+ // namespace.
+ //
+ // Currently, this is always the init task's user namespace.
+ creds := t.Credentials()
+ if !creds.HasCapabilityIn(linux.CAP_SYS_ADMIN, t.MountNamespaceVFS2().Owner) {
+ return 0, nil, syserror.EPERM
+ }
+
+ const unsupported = linux.MNT_FORCE | linux.MNT_EXPIRE
+ if flags&unsupported != 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ path, err := copyInPath(t, addr)
+ if err != nil {
+ return 0, nil, err
+ }
+ tpop, err := getTaskPathOperation(t, linux.AT_FDCWD, path, disallowEmptyPath, nofollowFinalSymlink)
+ if err != nil {
+ return 0, nil, err
+ }
+ defer tpop.Release()
+
+ opts := vfs.UmountOptions{
+ Flags: uint32(flags),
+ }
+
+ return 0, nil, t.Kernel().VFS().UmountAt(t, creds, &tpop.pop, &opts)
+}
diff --git a/pkg/sentry/syscalls/linux/vfs2/read_write.go b/pkg/sentry/syscalls/linux/vfs2/read_write.go
index 3a7ef24f5..cd25597a7 100644
--- a/pkg/sentry/syscalls/linux/vfs2/read_write.go
+++ b/pkg/sentry/syscalls/linux/vfs2/read_write.go
@@ -93,11 +93,17 @@ func Readv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) {
n, err := file.Read(t, dst, opts)
if err != syserror.ErrWouldBlock {
+ if n > 0 {
+ file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent)
+ }
return n, err
}
allowBlock, deadline, hasDeadline := blockPolicy(t, file)
if !allowBlock {
+ if n > 0 {
+ file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent)
+ }
return n, err
}
@@ -128,6 +134,9 @@ func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opt
}
file.EventUnregister(&w)
+ if total > 0 {
+ file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent)
+ }
return total, err
}
@@ -248,11 +257,17 @@ func Preadv2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca
func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) {
n, err := file.PRead(t, dst, offset, opts)
if err != syserror.ErrWouldBlock {
+ if n > 0 {
+ file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent)
+ }
return n, err
}
allowBlock, deadline, hasDeadline := blockPolicy(t, file)
if !allowBlock {
+ if n > 0 {
+ file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent)
+ }
return n, err
}
@@ -283,6 +298,9 @@ func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, of
}
file.EventUnregister(&w)
+ if total > 0 {
+ file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent)
+ }
return total, err
}
@@ -345,11 +363,17 @@ func Writev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal
func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) {
n, err := file.Write(t, src, opts)
if err != syserror.ErrWouldBlock {
+ if n > 0 {
+ file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent)
+ }
return n, err
}
allowBlock, deadline, hasDeadline := blockPolicy(t, file)
if !allowBlock {
+ if n > 0 {
+ file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent)
+ }
return n, err
}
@@ -380,6 +404,9 @@ func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, op
}
file.EventUnregister(&w)
+ if total > 0 {
+ file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent)
+ }
return total, err
}
@@ -500,11 +527,17 @@ func Pwritev2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc
func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) {
n, err := file.PWrite(t, src, offset, opts)
if err != syserror.ErrWouldBlock {
+ if n > 0 {
+ file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent)
+ }
return n, err
}
allowBlock, deadline, hasDeadline := blockPolicy(t, file)
if !allowBlock {
+ if n > 0 {
+ file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent)
+ }
return n, err
}
@@ -535,6 +568,9 @@ func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, o
}
file.EventUnregister(&w)
+ if total > 0 {
+ file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent)
+ }
return total, err
}
@@ -570,3 +606,36 @@ func Lseek(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
newoff, err := file.Seek(t, offset, whence)
return uintptr(newoff), nil, err
}
+
+// Readahead implements readahead(2).
+func Readahead(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ offset := args[1].Int64()
+ size := args[2].SizeT()
+
+ file := t.GetFileVFS2(fd)
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ // Check that the file is readable.
+ if !file.IsReadable() {
+ return 0, nil, syserror.EBADF
+ }
+
+ // Check that the size is valid.
+ if int(size) < 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ // Check that the offset is legitimate and does not overflow.
+ if offset < 0 || offset+int64(size) < 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ // Return EINVAL; if the underlying file type does not support readahead,
+ // then Linux will return EINVAL to indicate as much. In the future, we
+ // may extend this function to actually support readahead hints.
+ return 0, nil, syserror.EINVAL
+}
diff --git a/pkg/sentry/syscalls/linux/vfs2/setstat.go b/pkg/sentry/syscalls/linux/vfs2/setstat.go
index 09ecfed26..25cdb7a55 100644
--- a/pkg/sentry/syscalls/linux/vfs2/setstat.go
+++ b/pkg/sentry/syscalls/linux/vfs2/setstat.go
@@ -20,6 +20,7 @@ import (
"gvisor.dev/gvisor/pkg/sentry/arch"
"gvisor.dev/gvisor/pkg/sentry/kernel"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+ "gvisor.dev/gvisor/pkg/sentry/limits"
"gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/usermem"
@@ -178,6 +179,7 @@ func Truncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc
Mask: linux.STATX_SIZE,
Size: uint64(length),
},
+ NeedWritePerm: true,
})
return 0, nil, handleSetSizeError(t, err)
}
@@ -197,6 +199,10 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys
}
defer file.DecRef()
+ if !file.IsWritable() {
+ return 0, nil, syserror.EINVAL
+ }
+
err := file.SetStat(t, vfs.SetStatOptions{
Stat: linux.Statx{
Mask: linux.STATX_SIZE,
@@ -206,6 +212,56 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys
return 0, nil, handleSetSizeError(t, err)
}
+// Fallocate implements linux system call fallocate(2).
+func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ mode := args[1].Uint64()
+ offset := args[2].Int64()
+ length := args[3].Int64()
+
+ file := t.GetFileVFS2(fd)
+
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ if !file.IsWritable() {
+ return 0, nil, syserror.EBADF
+ }
+
+ if mode != 0 {
+ return 0, nil, syserror.ENOTSUP
+ }
+
+ if offset < 0 || length <= 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ size := offset + length
+
+ if size < 0 {
+ return 0, nil, syserror.EFBIG
+ }
+
+ limit := limits.FromContext(t).Get(limits.FileSize).Cur
+
+ if uint64(size) >= limit {
+ t.SendSignal(&arch.SignalInfo{
+ Signo: int32(linux.SIGXFSZ),
+ Code: arch.SignalInfoUser,
+ })
+ return 0, nil, syserror.EFBIG
+ }
+
+ if err := file.Allocate(t, mode, uint64(offset), uint64(length)); err != nil {
+ return 0, nil, err
+ }
+
+ file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent)
+ return 0, nil, nil
+}
+
// Utime implements Linux syscall utime(2).
func Utime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
pathAddr := args[0].Pointer()
diff --git a/pkg/sentry/syscalls/linux/vfs2/socket.go b/pkg/sentry/syscalls/linux/vfs2/socket.go
index 10b668477..8096a8f9c 100644
--- a/pkg/sentry/syscalls/linux/vfs2/socket.go
+++ b/pkg/sentry/syscalls/linux/vfs2/socket.go
@@ -30,6 +30,8 @@ import (
"gvisor.dev/gvisor/pkg/syserr"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/usermem"
+ "gvisor.dev/gvisor/tools/go_marshal/marshal"
+ "gvisor.dev/gvisor/tools/go_marshal/primitive"
)
// minListenBacklog is the minimum reasonable backlog for listening sockets.
@@ -477,7 +479,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy
}
if v != nil {
- if _, err := t.CopyOut(optValAddr, v); err != nil {
+ if _, err := v.CopyOut(t, optValAddr); err != nil {
return 0, nil, err
}
}
@@ -487,7 +489,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy
// getSockOpt tries to handle common socket options, or dispatches to a specific
// socket implementation.
-func getSockOpt(t *kernel.Task, s socket.SocketVFS2, level, name int, optValAddr usermem.Addr, len int) (interface{}, *syserr.Error) {
+func getSockOpt(t *kernel.Task, s socket.SocketVFS2, level, name int, optValAddr usermem.Addr, len int) (marshal.Marshallable, *syserr.Error) {
if level == linux.SOL_SOCKET {
switch name {
case linux.SO_TYPE, linux.SO_DOMAIN, linux.SO_PROTOCOL:
@@ -499,13 +501,16 @@ func getSockOpt(t *kernel.Task, s socket.SocketVFS2, level, name int, optValAddr
switch name {
case linux.SO_TYPE:
_, skType, _ := s.Type()
- return int32(skType), nil
+ v := primitive.Int32(skType)
+ return &v, nil
case linux.SO_DOMAIN:
family, _, _ := s.Type()
- return int32(family), nil
+ v := primitive.Int32(family)
+ return &v, nil
case linux.SO_PROTOCOL:
_, _, protocol := s.Type()
- return int32(protocol), nil
+ v := primitive.Int32(protocol)
+ return &v, nil
}
}
@@ -542,7 +547,7 @@ func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy
return 0, nil, syserror.EINVAL
}
buf := t.CopyScratchBuffer(int(optLen))
- if _, err := t.CopyIn(optValAddr, &buf); err != nil {
+ if _, err := t.CopyInBytes(optValAddr, buf); err != nil {
return 0, nil, err
}
diff --git a/pkg/sentry/syscalls/linux/vfs2/splice.go b/pkg/sentry/syscalls/linux/vfs2/splice.go
new file mode 100644
index 000000000..63ab11f8c
--- /dev/null
+++ b/pkg/sentry/syscalls/linux/vfs2/splice.go
@@ -0,0 +1,486 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package vfs2
+
+import (
+ "io"
+
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/sentry/arch"
+ "gvisor.dev/gvisor/pkg/sentry/kernel"
+ "gvisor.dev/gvisor/pkg/sentry/kernel/pipe"
+ "gvisor.dev/gvisor/pkg/sentry/vfs"
+ "gvisor.dev/gvisor/pkg/syserror"
+ "gvisor.dev/gvisor/pkg/usermem"
+ "gvisor.dev/gvisor/pkg/waiter"
+)
+
+// Splice implements Linux syscall splice(2).
+func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ inFD := args[0].Int()
+ inOffsetPtr := args[1].Pointer()
+ outFD := args[2].Int()
+ outOffsetPtr := args[3].Pointer()
+ count := int64(args[4].SizeT())
+ flags := args[5].Int()
+
+ if count == 0 {
+ return 0, nil, nil
+ }
+ if count > int64(kernel.MAX_RW_COUNT) {
+ count = int64(kernel.MAX_RW_COUNT)
+ }
+
+ // Check for invalid flags.
+ if flags&^(linux.SPLICE_F_MOVE|linux.SPLICE_F_NONBLOCK|linux.SPLICE_F_MORE|linux.SPLICE_F_GIFT) != 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ // Get file descriptions.
+ inFile := t.GetFileVFS2(inFD)
+ if inFile == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer inFile.DecRef()
+ outFile := t.GetFileVFS2(outFD)
+ if outFile == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer outFile.DecRef()
+
+ // Check that both files support the required directionality.
+ if !inFile.IsReadable() || !outFile.IsWritable() {
+ return 0, nil, syserror.EBADF
+ }
+
+ // The operation is non-blocking if anything is non-blocking.
+ //
+ // N.B. This is a rather simplistic heuristic that avoids some
+ // poor edge case behavior since the exact semantics here are
+ // underspecified and vary between versions of Linux itself.
+ nonBlock := ((inFile.StatusFlags()|outFile.StatusFlags())&linux.O_NONBLOCK != 0) || (flags&linux.SPLICE_F_NONBLOCK != 0)
+
+ // At least one file description must represent a pipe.
+ inPipeFD, inIsPipe := inFile.Impl().(*pipe.VFSPipeFD)
+ outPipeFD, outIsPipe := outFile.Impl().(*pipe.VFSPipeFD)
+ if !inIsPipe && !outIsPipe {
+ return 0, nil, syserror.EINVAL
+ }
+
+ // Copy in offsets.
+ inOffset := int64(-1)
+ if inOffsetPtr != 0 {
+ if inIsPipe {
+ return 0, nil, syserror.ESPIPE
+ }
+ if inFile.Options().DenyPRead {
+ return 0, nil, syserror.EINVAL
+ }
+ if _, err := t.CopyIn(inOffsetPtr, &inOffset); err != nil {
+ return 0, nil, err
+ }
+ if inOffset < 0 {
+ return 0, nil, syserror.EINVAL
+ }
+ }
+ outOffset := int64(-1)
+ if outOffsetPtr != 0 {
+ if outIsPipe {
+ return 0, nil, syserror.ESPIPE
+ }
+ if outFile.Options().DenyPWrite {
+ return 0, nil, syserror.EINVAL
+ }
+ if _, err := t.CopyIn(outOffsetPtr, &outOffset); err != nil {
+ return 0, nil, err
+ }
+ if outOffset < 0 {
+ return 0, nil, syserror.EINVAL
+ }
+ }
+
+ // Move data.
+ var (
+ n int64
+ err error
+ )
+ dw := dualWaiter{
+ inFile: inFile,
+ outFile: outFile,
+ }
+ defer dw.destroy()
+ for {
+ // If both input and output are pipes, delegate to the pipe
+ // implementation. Otherwise, exactly one end is a pipe, which
+ // we ensure is consistently ordered after the non-pipe FD's
+ // locks by passing the pipe FD as usermem.IO to the non-pipe
+ // end.
+ switch {
+ case inIsPipe && outIsPipe:
+ n, err = pipe.Splice(t, outPipeFD, inPipeFD, count)
+ case inIsPipe:
+ if outOffset != -1 {
+ n, err = outFile.PWrite(t, inPipeFD.IOSequence(count), outOffset, vfs.WriteOptions{})
+ outOffset += n
+ } else {
+ n, err = outFile.Write(t, inPipeFD.IOSequence(count), vfs.WriteOptions{})
+ }
+ case outIsPipe:
+ if inOffset != -1 {
+ n, err = inFile.PRead(t, outPipeFD.IOSequence(count), inOffset, vfs.ReadOptions{})
+ inOffset += n
+ } else {
+ n, err = inFile.Read(t, outPipeFD.IOSequence(count), vfs.ReadOptions{})
+ }
+ default:
+ panic("not possible")
+ }
+
+ if n != 0 || err != syserror.ErrWouldBlock || nonBlock {
+ break
+ }
+ if err = dw.waitForBoth(t); err != nil {
+ break
+ }
+ }
+
+ // Copy updated offsets out.
+ if inOffsetPtr != 0 {
+ if _, err := t.CopyOut(inOffsetPtr, &inOffset); err != nil {
+ return 0, nil, err
+ }
+ }
+ if outOffsetPtr != 0 {
+ if _, err := t.CopyOut(outOffsetPtr, &outOffset); err != nil {
+ return 0, nil, err
+ }
+ }
+
+ if n == 0 {
+ return 0, nil, err
+ }
+
+ // On Linux, inotify behavior is not very consistent with splice(2). We try
+ // our best to emulate Linux for very basic calls to splice, where for some
+ // reason, events are generated for output files, but not input files.
+ outFile.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent)
+ return uintptr(n), nil, nil
+}
+
+// Tee implements Linux syscall tee(2).
+func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ inFD := args[0].Int()
+ outFD := args[1].Int()
+ count := int64(args[2].SizeT())
+ flags := args[3].Int()
+
+ if count == 0 {
+ return 0, nil, nil
+ }
+ if count > int64(kernel.MAX_RW_COUNT) {
+ count = int64(kernel.MAX_RW_COUNT)
+ }
+
+ // Check for invalid flags.
+ if flags&^(linux.SPLICE_F_MOVE|linux.SPLICE_F_NONBLOCK|linux.SPLICE_F_MORE|linux.SPLICE_F_GIFT) != 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ // Get file descriptions.
+ inFile := t.GetFileVFS2(inFD)
+ if inFile == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer inFile.DecRef()
+ outFile := t.GetFileVFS2(outFD)
+ if outFile == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer outFile.DecRef()
+
+ // Check that both files support the required directionality.
+ if !inFile.IsReadable() || !outFile.IsWritable() {
+ return 0, nil, syserror.EBADF
+ }
+
+ // The operation is non-blocking if anything is non-blocking.
+ //
+ // N.B. This is a rather simplistic heuristic that avoids some
+ // poor edge case behavior since the exact semantics here are
+ // underspecified and vary between versions of Linux itself.
+ nonBlock := ((inFile.StatusFlags()|outFile.StatusFlags())&linux.O_NONBLOCK != 0) || (flags&linux.SPLICE_F_NONBLOCK != 0)
+
+ // Both file descriptions must represent pipes.
+ inPipeFD, inIsPipe := inFile.Impl().(*pipe.VFSPipeFD)
+ outPipeFD, outIsPipe := outFile.Impl().(*pipe.VFSPipeFD)
+ if !inIsPipe || !outIsPipe {
+ return 0, nil, syserror.EINVAL
+ }
+
+ // Copy data.
+ var (
+ n int64
+ err error
+ )
+ dw := dualWaiter{
+ inFile: inFile,
+ outFile: outFile,
+ }
+ defer dw.destroy()
+ for {
+ n, err = pipe.Tee(t, outPipeFD, inPipeFD, count)
+ if n != 0 || err != syserror.ErrWouldBlock || nonBlock {
+ break
+ }
+ if err = dw.waitForBoth(t); err != nil {
+ break
+ }
+ }
+ if n == 0 {
+ return 0, nil, err
+ }
+ outFile.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent)
+ return uintptr(n), nil, nil
+}
+
+// Sendfile implements linux system call sendfile(2).
+func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ outFD := args[0].Int()
+ inFD := args[1].Int()
+ offsetAddr := args[2].Pointer()
+ count := int64(args[3].SizeT())
+
+ inFile := t.GetFileVFS2(inFD)
+ if inFile == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer inFile.DecRef()
+ if !inFile.IsReadable() {
+ return 0, nil, syserror.EBADF
+ }
+
+ outFile := t.GetFileVFS2(outFD)
+ if outFile == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer outFile.DecRef()
+ if !outFile.IsWritable() {
+ return 0, nil, syserror.EBADF
+ }
+
+ // Verify that the outFile Append flag is not set.
+ if outFile.StatusFlags()&linux.O_APPEND != 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ // Verify that inFile is a regular file or block device. This is a
+ // requirement; the same check appears in Linux
+ // (fs/splice.c:splice_direct_to_actor).
+ if stat, err := inFile.Stat(t, vfs.StatOptions{Mask: linux.STATX_TYPE}); err != nil {
+ return 0, nil, err
+ } else if stat.Mask&linux.STATX_TYPE == 0 ||
+ (stat.Mode&linux.S_IFMT != linux.S_IFREG && stat.Mode&linux.S_IFMT != linux.S_IFBLK) {
+ return 0, nil, syserror.EINVAL
+ }
+
+ // Copy offset if it exists.
+ offset := int64(-1)
+ if offsetAddr != 0 {
+ if inFile.Options().DenyPRead {
+ return 0, nil, syserror.ESPIPE
+ }
+ if _, err := t.CopyIn(offsetAddr, &offset); err != nil {
+ return 0, nil, err
+ }
+ if offset < 0 {
+ return 0, nil, syserror.EINVAL
+ }
+ if offset+count < 0 {
+ return 0, nil, syserror.EINVAL
+ }
+ }
+
+ // Validate count. This must come after offset checks.
+ if count < 0 {
+ return 0, nil, syserror.EINVAL
+ }
+ if count == 0 {
+ return 0, nil, nil
+ }
+ if count > int64(kernel.MAX_RW_COUNT) {
+ count = int64(kernel.MAX_RW_COUNT)
+ }
+
+ // Copy data.
+ var (
+ n int64
+ err error
+ )
+ dw := dualWaiter{
+ inFile: inFile,
+ outFile: outFile,
+ }
+ defer dw.destroy()
+ outPipeFD, outIsPipe := outFile.Impl().(*pipe.VFSPipeFD)
+ // Reading from input file should never block, since it is regular or
+ // block device. We only need to check if writing to the output file
+ // can block.
+ nonBlock := outFile.StatusFlags()&linux.O_NONBLOCK != 0
+ if outIsPipe {
+ for n < count {
+ var spliceN int64
+ if offset != -1 {
+ spliceN, err = inFile.PRead(t, outPipeFD.IOSequence(count), offset, vfs.ReadOptions{})
+ offset += spliceN
+ } else {
+ spliceN, err = inFile.Read(t, outPipeFD.IOSequence(count), vfs.ReadOptions{})
+ }
+ n += spliceN
+ if err == syserror.ErrWouldBlock && !nonBlock {
+ err = dw.waitForBoth(t)
+ }
+ if err != nil {
+ break
+ }
+ }
+ } else {
+ // Read inFile to buffer, then write the contents to outFile.
+ buf := make([]byte, count)
+ for n < count {
+ var readN int64
+ if offset != -1 {
+ readN, err = inFile.PRead(t, usermem.BytesIOSequence(buf), offset, vfs.ReadOptions{})
+ offset += readN
+ } else {
+ readN, err = inFile.Read(t, usermem.BytesIOSequence(buf), vfs.ReadOptions{})
+ }
+ if readN == 0 && err == io.EOF {
+ // We reached the end of the file. Eat the
+ // error and exit the loop.
+ err = nil
+ break
+ }
+ n += readN
+ if err != nil {
+ break
+ }
+
+ // Write all of the bytes that we read. This may need
+ // multiple write calls to complete.
+ wbuf := buf[:n]
+ for len(wbuf) > 0 {
+ var writeN int64
+ writeN, err = outFile.Write(t, usermem.BytesIOSequence(wbuf), vfs.WriteOptions{})
+ wbuf = wbuf[writeN:]
+ if err == syserror.ErrWouldBlock && !nonBlock {
+ err = dw.waitForOut(t)
+ }
+ if err != nil {
+ // We didn't complete the write. Only
+ // report the bytes that were actually
+ // written, and rewind the offset.
+ notWritten := int64(len(wbuf))
+ n -= notWritten
+ if offset != -1 {
+ offset -= notWritten
+ }
+ break
+ }
+ }
+ if err == syserror.ErrWouldBlock && !nonBlock {
+ err = dw.waitForBoth(t)
+ }
+ if err != nil {
+ break
+ }
+ }
+ }
+
+ if offsetAddr != 0 {
+ // Copy out the new offset.
+ if _, err := t.CopyOut(offsetAddr, offset); err != nil {
+ return 0, nil, err
+ }
+ }
+
+ if n == 0 {
+ return 0, nil, err
+ }
+
+ inFile.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent)
+ outFile.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent)
+ return uintptr(n), nil, nil
+}
+
+// dualWaiter is used to wait on one or both vfs.FileDescriptions. It is not
+// thread-safe, and does not take a reference on the vfs.FileDescriptions.
+//
+// Users must call destroy() when finished.
+type dualWaiter struct {
+ inFile *vfs.FileDescription
+ outFile *vfs.FileDescription
+
+ inW waiter.Entry
+ inCh chan struct{}
+ outW waiter.Entry
+ outCh chan struct{}
+}
+
+// waitForBoth waits for both dw.inFile and dw.outFile to be ready.
+func (dw *dualWaiter) waitForBoth(t *kernel.Task) error {
+ if dw.inFile.Readiness(eventMaskRead)&eventMaskRead == 0 {
+ if dw.inCh == nil {
+ dw.inW, dw.inCh = waiter.NewChannelEntry(nil)
+ dw.inFile.EventRegister(&dw.inW, eventMaskRead)
+ // We might be ready now. Try again before blocking.
+ return nil
+ }
+ if err := t.Block(dw.inCh); err != nil {
+ return err
+ }
+ }
+ return dw.waitForOut(t)
+}
+
+// waitForOut waits for dw.outfile to be read.
+func (dw *dualWaiter) waitForOut(t *kernel.Task) error {
+ if dw.outFile.Readiness(eventMaskWrite)&eventMaskWrite == 0 {
+ if dw.outCh == nil {
+ dw.outW, dw.outCh = waiter.NewChannelEntry(nil)
+ dw.outFile.EventRegister(&dw.outW, eventMaskWrite)
+ // We might be ready now. Try again before blocking.
+ return nil
+ }
+ if err := t.Block(dw.outCh); err != nil {
+ return err
+ }
+ }
+ return nil
+}
+
+// destroy cleans up resources help by dw. No more calls to wait* can occur
+// after destroy is called.
+func (dw *dualWaiter) destroy() {
+ if dw.inCh != nil {
+ dw.inFile.EventUnregister(&dw.inW)
+ dw.inCh = nil
+ }
+ if dw.outCh != nil {
+ dw.outFile.EventUnregister(&dw.outW)
+ dw.outCh = nil
+ }
+ dw.inFile = nil
+ dw.outFile = nil
+}
diff --git a/pkg/sentry/syscalls/linux/vfs2/sync.go b/pkg/sentry/syscalls/linux/vfs2/sync.go
index 365250b0b..0d0ebf46a 100644
--- a/pkg/sentry/syscalls/linux/vfs2/sync.go
+++ b/pkg/sentry/syscalls/linux/vfs2/sync.go
@@ -65,10 +65,8 @@ func SyncFileRange(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel
nbytes := args[2].Int64()
flags := args[3].Uint()
- if offset < 0 {
- return 0, nil, syserror.EINVAL
- }
- if nbytes < 0 {
+ // Check for negative values and overflow.
+ if offset < 0 || offset+nbytes < 0 {
return 0, nil, syserror.EINVAL
}
if flags&^(linux.SYNC_FILE_RANGE_WAIT_BEFORE|linux.SYNC_FILE_RANGE_WRITE|linux.SYNC_FILE_RANGE_WAIT_AFTER) != 0 {
@@ -81,7 +79,37 @@ func SyncFileRange(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel
}
defer file.DecRef()
- // TODO(gvisor.dev/issue/1897): Avoid writeback of data ranges outside of
- // [offset, offset+nbytes).
- return 0, nil, file.Sync(t)
+ // TODO(gvisor.dev/issue/1897): Currently, the only file syncing we support
+ // is a full-file sync, i.e. fsync(2). As a result, there are severe
+ // limitations on how much we support sync_file_range:
+ // - In Linux, sync_file_range(2) doesn't write out the file's metadata, even
+ // if the file size is changed. We do.
+ // - We always sync the entire file instead of [offset, offset+nbytes).
+ // - We do not support the use of WAIT_BEFORE without WAIT_AFTER. For
+ // correctness, we would have to perform a write-out every time WAIT_BEFORE
+ // was used, but this would be much more expensive than expected if there
+ // were no write-out operations in progress.
+ // - Whenever WAIT_AFTER is used, we sync the file.
+ // - Ignore WRITE. If this flag is used with WAIT_AFTER, then the file will
+ // be synced anyway. If this flag is used without WAIT_AFTER, then it is
+ // safe (and less expensive) to do nothing, because the syscall will not
+ // wait for the write-out to complete--we only need to make sure that the
+ // next time WAIT_BEFORE or WAIT_AFTER are used, the write-out completes.
+ // - According to fs/sync.c, WAIT_BEFORE|WAIT_AFTER "will detect any I/O
+ // errors or ENOSPC conditions and will return those to the caller, after
+ // clearing the EIO and ENOSPC flags in the address_space." We don't do
+ // this.
+
+ if flags&linux.SYNC_FILE_RANGE_WAIT_BEFORE != 0 &&
+ flags&linux.SYNC_FILE_RANGE_WAIT_AFTER == 0 {
+ t.Kernel().EmitUnimplementedEvent(t)
+ return 0, nil, syserror.ENOSYS
+ }
+
+ if flags&linux.SYNC_FILE_RANGE_WAIT_AFTER != 0 {
+ if err := file.Sync(t); err != nil {
+ return 0, nil, syserror.ConvertIntr(err, kernel.ERESTARTSYS)
+ }
+ }
+ return 0, nil, nil
}
diff --git a/pkg/sentry/syscalls/linux/vfs2/vfs2.go b/pkg/sentry/syscalls/linux/vfs2/vfs2.go
index a332d01bd..c576d9475 100644
--- a/pkg/sentry/syscalls/linux/vfs2/vfs2.go
+++ b/pkg/sentry/syscalls/linux/vfs2/vfs2.go
@@ -44,7 +44,7 @@ func Override() {
s.Table[23] = syscalls.Supported("select", Select)
s.Table[32] = syscalls.Supported("dup", Dup)
s.Table[33] = syscalls.Supported("dup2", Dup2)
- delete(s.Table, 40) // sendfile
+ s.Table[40] = syscalls.Supported("sendfile", Sendfile)
s.Table[41] = syscalls.Supported("socket", Socket)
s.Table[42] = syscalls.Supported("connect", Connect)
s.Table[43] = syscalls.Supported("accept", Accept)
@@ -62,7 +62,7 @@ func Override() {
s.Table[55] = syscalls.Supported("getsockopt", GetSockOpt)
s.Table[59] = syscalls.Supported("execve", Execve)
s.Table[72] = syscalls.Supported("fcntl", Fcntl)
- delete(s.Table, 73) // flock
+ s.Table[73] = syscalls.Supported("flock", Flock)
s.Table[74] = syscalls.Supported("fsync", Fsync)
s.Table[75] = syscalls.Supported("fdatasync", Fdatasync)
s.Table[76] = syscalls.Supported("truncate", Truncate)
@@ -90,9 +90,9 @@ func Override() {
s.Table[138] = syscalls.Supported("fstatfs", Fstatfs)
s.Table[161] = syscalls.Supported("chroot", Chroot)
s.Table[162] = syscalls.Supported("sync", Sync)
- delete(s.Table, 165) // mount
- delete(s.Table, 166) // umount2
- delete(s.Table, 187) // readahead
+ s.Table[165] = syscalls.Supported("mount", Mount)
+ s.Table[166] = syscalls.Supported("umount2", Umount2)
+ s.Table[187] = syscalls.Supported("readahead", Readahead)
s.Table[188] = syscalls.Supported("setxattr", Setxattr)
s.Table[189] = syscalls.Supported("lsetxattr", Lsetxattr)
s.Table[190] = syscalls.Supported("fsetxattr", Fsetxattr)
@@ -105,20 +105,16 @@ func Override() {
s.Table[197] = syscalls.Supported("removexattr", Removexattr)
s.Table[198] = syscalls.Supported("lremovexattr", Lremovexattr)
s.Table[199] = syscalls.Supported("fremovexattr", Fremovexattr)
- delete(s.Table, 206) // io_setup
- delete(s.Table, 207) // io_destroy
- delete(s.Table, 208) // io_getevents
- delete(s.Table, 209) // io_submit
- delete(s.Table, 210) // io_cancel
+ s.Table[209] = syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"})
s.Table[213] = syscalls.Supported("epoll_create", EpollCreate)
s.Table[217] = syscalls.Supported("getdents64", Getdents64)
- delete(s.Table, 221) // fdavise64
+ s.Table[221] = syscalls.PartiallySupported("fadvise64", Fadvise64, "The syscall is 'supported', but ignores all provided advice.", nil)
s.Table[232] = syscalls.Supported("epoll_wait", EpollWait)
s.Table[233] = syscalls.Supported("epoll_ctl", EpollCtl)
s.Table[235] = syscalls.Supported("utimes", Utimes)
- delete(s.Table, 253) // inotify_init
- delete(s.Table, 254) // inotify_add_watch
- delete(s.Table, 255) // inotify_rm_watch
+ s.Table[253] = syscalls.PartiallySupported("inotify_init", InotifyInit, "inotify events are only available inside the sandbox.", nil)
+ s.Table[254] = syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "inotify events are only available inside the sandbox.", nil)
+ s.Table[255] = syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "inotify events are only available inside the sandbox.", nil)
s.Table[257] = syscalls.Supported("openat", Openat)
s.Table[258] = syscalls.Supported("mkdirat", Mkdirat)
s.Table[259] = syscalls.Supported("mknodat", Mknodat)
@@ -134,15 +130,15 @@ func Override() {
s.Table[269] = syscalls.Supported("faccessat", Faccessat)
s.Table[270] = syscalls.Supported("pselect", Pselect)
s.Table[271] = syscalls.Supported("ppoll", Ppoll)
- delete(s.Table, 275) // splice
- delete(s.Table, 276) // tee
+ s.Table[275] = syscalls.Supported("splice", Splice)
+ s.Table[276] = syscalls.Supported("tee", Tee)
s.Table[277] = syscalls.Supported("sync_file_range", SyncFileRange)
s.Table[280] = syscalls.Supported("utimensat", Utimensat)
s.Table[281] = syscalls.Supported("epoll_pwait", EpollPwait)
s.Table[282] = syscalls.Supported("signalfd", Signalfd)
s.Table[283] = syscalls.Supported("timerfd_create", TimerfdCreate)
s.Table[284] = syscalls.Supported("eventfd", Eventfd)
- delete(s.Table, 285) // fallocate
+ s.Table[285] = syscalls.PartiallySupported("fallocate", Fallocate, "Not all options are supported.", nil)
s.Table[286] = syscalls.Supported("timerfd_settime", TimerfdSettime)
s.Table[287] = syscalls.Supported("timerfd_gettime", TimerfdGettime)
s.Table[288] = syscalls.Supported("accept4", Accept4)
@@ -151,7 +147,7 @@ func Override() {
s.Table[291] = syscalls.Supported("epoll_create1", EpollCreate1)
s.Table[292] = syscalls.Supported("dup3", Dup3)
s.Table[293] = syscalls.Supported("pipe2", Pipe2)
- delete(s.Table, 294) // inotify_init1
+ s.Table[294] = syscalls.PartiallySupported("inotify_init1", InotifyInit1, "inotify events are only available inside the sandbox.", nil)
s.Table[295] = syscalls.Supported("preadv", Preadv)
s.Table[296] = syscalls.Supported("pwritev", Pwritev)
s.Table[299] = syscalls.Supported("recvmmsg", RecvMMsg)
@@ -167,6 +163,106 @@ func Override() {
// Override ARM64.
s = linux.ARM64
+ s.Table[5] = syscalls.Supported("setxattr", Setxattr)
+ s.Table[6] = syscalls.Supported("lsetxattr", Lsetxattr)
+ s.Table[7] = syscalls.Supported("fsetxattr", Fsetxattr)
+ s.Table[8] = syscalls.Supported("getxattr", Getxattr)
+ s.Table[9] = syscalls.Supported("lgetxattr", Lgetxattr)
+ s.Table[10] = syscalls.Supported("fgetxattr", Fgetxattr)
+ s.Table[11] = syscalls.Supported("listxattr", Listxattr)
+ s.Table[12] = syscalls.Supported("llistxattr", Llistxattr)
+ s.Table[13] = syscalls.Supported("flistxattr", Flistxattr)
+ s.Table[14] = syscalls.Supported("removexattr", Removexattr)
+ s.Table[15] = syscalls.Supported("lremovexattr", Lremovexattr)
+ s.Table[16] = syscalls.Supported("fremovexattr", Fremovexattr)
+ s.Table[17] = syscalls.Supported("getcwd", Getcwd)
+ s.Table[19] = syscalls.Supported("eventfd2", Eventfd2)
+ s.Table[20] = syscalls.Supported("epoll_create1", EpollCreate1)
+ s.Table[21] = syscalls.Supported("epoll_ctl", EpollCtl)
+ s.Table[22] = syscalls.Supported("epoll_pwait", EpollPwait)
+ s.Table[23] = syscalls.Supported("dup", Dup)
+ s.Table[24] = syscalls.Supported("dup3", Dup3)
+ s.Table[25] = syscalls.Supported("fcntl", Fcntl)
+ s.Table[26] = syscalls.PartiallySupported("inotify_init1", InotifyInit1, "inotify events are only available inside the sandbox.", nil)
+ s.Table[27] = syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "inotify events are only available inside the sandbox.", nil)
+ s.Table[28] = syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "inotify events are only available inside the sandbox.", nil)
+ s.Table[29] = syscalls.Supported("ioctl", Ioctl)
+ s.Table[32] = syscalls.Supported("flock", Flock)
+ s.Table[33] = syscalls.Supported("mknodat", Mknodat)
+ s.Table[34] = syscalls.Supported("mkdirat", Mkdirat)
+ s.Table[35] = syscalls.Supported("unlinkat", Unlinkat)
+ s.Table[36] = syscalls.Supported("symlinkat", Symlinkat)
+ s.Table[37] = syscalls.Supported("linkat", Linkat)
+ s.Table[38] = syscalls.Supported("renameat", Renameat)
+ s.Table[39] = syscalls.Supported("umount2", Umount2)
+ s.Table[40] = syscalls.Supported("mount", Mount)
+ s.Table[43] = syscalls.Supported("statfs", Statfs)
+ s.Table[44] = syscalls.Supported("fstatfs", Fstatfs)
+ s.Table[45] = syscalls.Supported("truncate", Truncate)
+ s.Table[46] = syscalls.Supported("ftruncate", Ftruncate)
+ s.Table[48] = syscalls.Supported("faccessat", Faccessat)
+ s.Table[49] = syscalls.Supported("chdir", Chdir)
+ s.Table[50] = syscalls.Supported("fchdir", Fchdir)
+ s.Table[51] = syscalls.Supported("chroot", Chroot)
+ s.Table[52] = syscalls.Supported("fchmod", Fchmod)
+ s.Table[53] = syscalls.Supported("fchmodat", Fchmodat)
+ s.Table[54] = syscalls.Supported("fchownat", Fchownat)
+ s.Table[55] = syscalls.Supported("fchown", Fchown)
+ s.Table[56] = syscalls.Supported("openat", Openat)
+ s.Table[57] = syscalls.Supported("close", Close)
+ s.Table[59] = syscalls.Supported("pipe2", Pipe2)
+ s.Table[61] = syscalls.Supported("getdents64", Getdents64)
+ s.Table[62] = syscalls.Supported("lseek", Lseek)
s.Table[63] = syscalls.Supported("read", Read)
+ s.Table[64] = syscalls.Supported("write", Write)
+ s.Table[65] = syscalls.Supported("readv", Readv)
+ s.Table[66] = syscalls.Supported("writev", Writev)
+ s.Table[67] = syscalls.Supported("pread64", Pread64)
+ s.Table[68] = syscalls.Supported("pwrite64", Pwrite64)
+ s.Table[69] = syscalls.Supported("preadv", Preadv)
+ s.Table[70] = syscalls.Supported("pwritev", Pwritev)
+ s.Table[72] = syscalls.Supported("pselect", Pselect)
+ s.Table[73] = syscalls.Supported("ppoll", Ppoll)
+ s.Table[74] = syscalls.Supported("signalfd4", Signalfd4)
+ s.Table[76] = syscalls.Supported("splice", Splice)
+ s.Table[77] = syscalls.Supported("tee", Tee)
+ s.Table[78] = syscalls.Supported("readlinkat", Readlinkat)
+ s.Table[80] = syscalls.Supported("fstat", Fstat)
+ s.Table[81] = syscalls.Supported("sync", Sync)
+ s.Table[82] = syscalls.Supported("fsync", Fsync)
+ s.Table[83] = syscalls.Supported("fdatasync", Fdatasync)
+ s.Table[84] = syscalls.Supported("sync_file_range", SyncFileRange)
+ s.Table[85] = syscalls.Supported("timerfd_create", TimerfdCreate)
+ s.Table[86] = syscalls.Supported("timerfd_settime", TimerfdSettime)
+ s.Table[87] = syscalls.Supported("timerfd_gettime", TimerfdGettime)
+ s.Table[88] = syscalls.Supported("utimensat", Utimensat)
+ s.Table[198] = syscalls.Supported("socket", Socket)
+ s.Table[199] = syscalls.Supported("socketpair", SocketPair)
+ s.Table[200] = syscalls.Supported("bind", Bind)
+ s.Table[201] = syscalls.Supported("listen", Listen)
+ s.Table[202] = syscalls.Supported("accept", Accept)
+ s.Table[203] = syscalls.Supported("connect", Connect)
+ s.Table[204] = syscalls.Supported("getsockname", GetSockName)
+ s.Table[205] = syscalls.Supported("getpeername", GetPeerName)
+ s.Table[206] = syscalls.Supported("sendto", SendTo)
+ s.Table[207] = syscalls.Supported("recvfrom", RecvFrom)
+ s.Table[208] = syscalls.Supported("setsockopt", SetSockOpt)
+ s.Table[209] = syscalls.Supported("getsockopt", GetSockOpt)
+ s.Table[210] = syscalls.Supported("shutdown", Shutdown)
+ s.Table[211] = syscalls.Supported("sendmsg", SendMsg)
+ s.Table[212] = syscalls.Supported("recvmsg", RecvMsg)
+ s.Table[221] = syscalls.Supported("execve", Execve)
+ s.Table[222] = syscalls.Supported("mmap", Mmap)
+ s.Table[242] = syscalls.Supported("accept4", Accept4)
+ s.Table[243] = syscalls.Supported("recvmmsg", RecvMMsg)
+ s.Table[267] = syscalls.Supported("syncfs", Syncfs)
+ s.Table[269] = syscalls.Supported("sendmmsg", SendMMsg)
+ s.Table[276] = syscalls.Supported("renameat2", Renameat2)
+ s.Table[279] = syscalls.Supported("memfd_create", MemfdCreate)
+ s.Table[281] = syscalls.Supported("execveat", Execveat)
+ s.Table[286] = syscalls.Supported("preadv2", Preadv2)
+ s.Table[287] = syscalls.Supported("pwritev2", Pwritev2)
+ s.Table[291] = syscalls.Supported("statx", Statx)
+
s.Init()
}