Merge release-20200622.1-185-g39525d64c (automated)

3 files changed, 85 insertions, 20 deletions

diff --git a/pkg/abi/linux/linux_abi_autogen_unsafe.go b/pkg/abi/linux/linux_abi_autogen_unsafe.go
index 832f260f0..59bccbb99 100644
--- a/pkg/abi/linux/linux_abi_autogen_unsafe.go
+++ b/pkg/abi/linux/linux_abi_autogen_unsafe.go
@@ -124,12 +124,12 @@ func (s *Statx) UnmarshalBytes(src []byte) {
 // Packed implements marshal.Marshallable.Packed.
 //go:nosplit
 func (s *Statx) Packed() bool {
-    return s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() && s.Atime.Packed()
+    return s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed()
 }
 
 // MarshalUnsafe implements marshal.Marshallable.MarshalUnsafe.
 func (s *Statx) MarshalUnsafe(dst []byte) {
-    if s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() {
+    if s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() && s.Atime.Packed() {
         safecopy.CopyIn(dst, unsafe.Pointer(s))
     } else {
         s.MarshalBytes(dst)
@@ -138,7 +138,7 @@ func (s *Statx) MarshalUnsafe(dst []byte) {
 
 // UnmarshalUnsafe implements marshal.Marshallable.UnmarshalUnsafe.
 func (s *Statx) UnmarshalUnsafe(src []byte) {
-    if s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() {
+    if s.Btime.Packed() && s.Ctime.Packed() && s.Mtime.Packed() && s.Atime.Packed() {
         safecopy.CopyOut(unsafe.Pointer(s), src)
     } else {
         s.UnmarshalBytes(src)
@@ -178,7 +178,7 @@ func (s *Statx) CopyOut(task marshal.Task, addr usermem.Addr) (int, error) {
 // CopyIn implements marshal.Marshallable.CopyIn.
 //go:nosplit
 func (s *Statx) CopyIn(task marshal.Task, addr usermem.Addr) (int, error) {
-    if !s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() && s.Ctime.Packed() {
+    if !s.Ctime.Packed() && s.Mtime.Packed() && s.Atime.Packed() && s.Btime.Packed() {
         // Type Statx doesn't have a packed layout in memory, fall back to UnmarshalBytes.
         buf := task.CopyScratchBuffer(s.SizeBytes()) // escapes: okay.
         length, err := task.CopyInBytes(addr, buf) // escapes: okay.
diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go
index f10350c97..02317a133 100644
--- a/pkg/sentry/fsimpl/gofer/regular_file.go
+++ b/pkg/sentry/fsimpl/gofer/regular_file.go
@@ -155,26 +155,53 @@ func (fd *regularFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts
 
 // PWrite implements vfs.FileDescriptionImpl.PWrite.
 func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) {
+	n, _, err := fd.pwrite(ctx, src, offset, opts)
+	return n, err
+}
+
+// pwrite returns the number of bytes written, final offset, error. The final
+// offset should be ignored by PWrite.
+func (fd *regularFileFD) pwrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (written, finalOff int64, err error) {
 	if offset < 0 {
-		return 0, syserror.EINVAL
+		return 0, offset, syserror.EINVAL
 	}
 
 	// Check that flags are supported.
 	//
 	// TODO(gvisor.dev/issue/2601): Support select pwritev2 flags.
 	if opts.Flags&^linux.RWF_HIPRI != 0 {
-		return 0, syserror.EOPNOTSUPP
+		return 0, offset, syserror.EOPNOTSUPP
+	}
+
+	d := fd.dentry()
+	// If the fd was opened with O_APPEND, make sure the file size is updated.
+	// There is a possible race here if size is modified externally after
+	// metadata cache is updated.
+	if fd.vfsfd.StatusFlags()&linux.O_APPEND != 0 && !d.cachedMetadataAuthoritative() {
+		if err := d.updateFromGetattr(ctx); err != nil {
+			return 0, offset, err
+		}
 	}
 
+	d.metadataMu.Lock()
+	defer d.metadataMu.Unlock()
+	// Set offset to file size if the fd was opened with O_APPEND.
+	if fd.vfsfd.StatusFlags()&linux.O_APPEND != 0 {
+		// Holding d.metadataMu is sufficient for reading d.size.
+		offset = int64(d.size)
+	}
 	limit, err := vfs.CheckLimit(ctx, offset, src.NumBytes())
 	if err != nil {
-		return 0, err
+		return 0, offset, err
 	}
 	src = src.TakeFirst64(limit)
+	n, err := fd.pwriteLocked(ctx, src, offset, opts)
+	return n, offset + n, err
+}
 
+// Preconditions: fd.dentry().metatdataMu must be locked.
+func (fd *regularFileFD) pwriteLocked(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) {
 	d := fd.dentry()
-	d.metadataMu.Lock()
-	defer d.metadataMu.Unlock()
 	if d.fs.opts.interop != InteropModeShared {
 		// Compare Linux's mm/filemap.c:__generic_file_write_iter() =>
 		// file_update_time(). This is d.touchCMtime(), but without locking
@@ -237,8 +264,8 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off
 // Write implements vfs.FileDescriptionImpl.Write.
 func (fd *regularFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) {
 	fd.mu.Lock()
-	n, err := fd.PWrite(ctx, src, fd.off, opts)
-	fd.off += n
+	n, off, err := fd.pwrite(ctx, src, fd.off, opts)
+	fd.off = off
 	fd.mu.Unlock()
 	return n, err
 }
diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go
index a7b53b2d2..811528982 100644
--- a/pkg/sentry/fsimpl/gofer/special_file.go
+++ b/pkg/sentry/fsimpl/gofer/special_file.go
@@ -16,6 +16,7 @@ package gofer
 
 import (
 	"sync"
+	"sync/atomic"
 
 	"gvisor.dev/gvisor/pkg/abi/linux"
 	"gvisor.dev/gvisor/pkg/context"
@@ -144,7 +145,7 @@ func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs
 	// mmap due to lock ordering; MM locks precede dentry.dataMu. That doesn't
 	// hold here since specialFileFD doesn't client-cache data. Just buffer the
 	// read instead.
-	if d := fd.dentry(); d.fs.opts.interop != InteropModeShared {
+	if d := fd.dentry(); d.cachedMetadataAuthoritative() {
 		d.touchAtime(fd.vfsfd.Mount())
 	}
 	buf := make([]byte, dst.NumBytes())
@@ -176,39 +177,76 @@ func (fd *specialFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts
 
 // PWrite implements vfs.FileDescriptionImpl.PWrite.
 func (fd *specialFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) {
+	n, _, err := fd.pwrite(ctx, src, offset, opts)
+	return n, err
+}
+
+// pwrite returns the number of bytes written, final offset, error. The final
+// offset should be ignored by PWrite.
+func (fd *specialFileFD) pwrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (written, finalOff int64, err error) {
 	if fd.seekable && offset < 0 {
-		return 0, syserror.EINVAL
+		return 0, offset, syserror.EINVAL
 	}
 
 	// Check that flags are supported.
 	//
 	// TODO(gvisor.dev/issue/2601): Support select pwritev2 flags.
 	if opts.Flags&^linux.RWF_HIPRI != 0 {
-		return 0, syserror.EOPNOTSUPP
+		return 0, offset, syserror.EOPNOTSUPP
+	}
+
+	d := fd.dentry()
+	// If the regular file fd was opened with O_APPEND, make sure the file size
+	// is updated. There is a possible race here if size is modified externally
+	// after metadata cache is updated.
+	if fd.seekable && fd.vfsfd.StatusFlags()&linux.O_APPEND != 0 && !d.cachedMetadataAuthoritative() {
+		if err := d.updateFromGetattr(ctx); err != nil {
+			return 0, offset, err
+		}
 	}
 
 	if fd.seekable {
+		// We need to hold the metadataMu *while* writing to a regular file.
+		d.metadataMu.Lock()
+		defer d.metadataMu.Unlock()
+
+		// Set offset to file size if the regular file was opened with O_APPEND.
+		if fd.vfsfd.StatusFlags()&linux.O_APPEND != 0 {
+			// Holding d.metadataMu is sufficient for reading d.size.
+			offset = int64(d.size)
+		}
 		limit, err := vfs.CheckLimit(ctx, offset, src.NumBytes())
 		if err != nil {
-			return 0, err
+			return 0, offset, err
 		}
 		src = src.TakeFirst64(limit)
 	}
 
 	// Do a buffered write. See rationale in PRead.
-	if d := fd.dentry(); d.fs.opts.interop != InteropModeShared {
+	if d.cachedMetadataAuthoritative() {
 		d.touchCMtime()
 	}
 	buf := make([]byte, src.NumBytes())
 	// Don't do partial writes if we get a partial read from src.
 	if _, err := src.CopyIn(ctx, buf); err != nil {
-		return 0, err
+		return 0, offset, err
 	}
 	n, err := fd.handle.writeFromBlocksAt(ctx, safemem.BlockSeqOf(safemem.BlockFromSafeSlice(buf)), uint64(offset))
 	if err == syserror.EAGAIN {
 		err = syserror.ErrWouldBlock
 	}
-	return int64(n), err
+	finalOff = offset
+	// Update file size for regular files.
+	if fd.seekable {
+		finalOff += int64(n)
+		// d.metadataMu is already locked at this point.
+		if uint64(finalOff) > d.size {
+			d.dataMu.Lock()
+			defer d.dataMu.Unlock()
+			atomic.StoreUint64(&d.size, uint64(finalOff))
+		}
+	}
+	return int64(n), finalOff, err
 }
 
 // Write implements vfs.FileDescriptionImpl.Write.
@@ -218,8 +256,8 @@ func (fd *specialFileFD) Write(ctx context.Context, src usermem.IOSequence, opts
 	}
 
 	fd.mu.Lock()
-	n, err := fd.PWrite(ctx, src, fd.off, opts)
-	fd.off += n
+	n, off, err := fd.pwrite(ctx, src, fd.off, opts)
+	fd.off = off
 	fd.mu.Unlock()
 	return n, err
 }