Internal change.

PiperOrigin-RevId: 311046755

2 files changed, 39 insertions, 24 deletions

diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go
index 7ca0cf3f4..7f2181216 100644
--- a/pkg/sentry/fsimpl/gofer/filesystem.go
+++ b/pkg/sentry/fsimpl/gofer/filesystem.go
@@ -877,10 +877,15 @@ func (d *dentry) openSpecialFileLocked(ctx context.Context, mnt *vfs.Mount, opts
 	if err != nil {
 		return nil, err
 	}
+	seekable := d.fileType() == linux.S_IFREG
 	fd := &specialFileFD{
-		handle: h,
+		handle:   h,
+		seekable: seekable,
 	}
-	if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{}); err != nil {
+	if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &d.vfsd, &vfs.FileDescriptionOptions{
+		DenyPRead:  !seekable,
+		DenyPWrite: !seekable,
+	}); err != nil {
 		h.close(ctx)
 		return nil, err
 	}
@@ -911,7 +916,11 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving
 	}
 	creds := rp.Credentials()
 	name := rp.Component()
-	fdobj, openFile, createQID, _, err := dirfile.create(ctx, name, (p9.OpenFlags)(opts.Flags), (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID))
+	// Filter file creation flags and O_LARGEFILE out; the create RPC already
+	// has the semantics of O_CREAT|O_EXCL, while some servers will choke on
+	// O_LARGEFILE.
+	createFlags := p9.OpenFlags(opts.Flags &^ (linux.O_CREAT | linux.O_EXCL | linux.O_NOCTTY | linux.O_TRUNC | linux.O_LARGEFILE))
+	fdobj, openFile, createQID, _, err := dirfile.create(ctx, name, createFlags, (p9.FileMode)(opts.Mode), (p9.UID)(creds.EffectiveKUID), (p9.GID)(creds.EffectiveKGID))
 	if err != nil {
 		dirfile.close(ctx)
 		return nil, err
@@ -919,7 +928,7 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving
 	// Then we need to walk to the file we just created to get a non-open fid
 	// representing it, and to get its metadata. This must use d.file since, as
 	// explained above, dirfile was invalidated by dirfile.Create().
-	walkQID, nonOpenFile, attrMask, attr, err := d.file.walkGetAttrOne(ctx, name)
+	_, nonOpenFile, attrMask, attr, err := d.file.walkGetAttrOne(ctx, name)
 	if err != nil {
 		openFile.close(ctx)
 		if fdobj != nil {
@@ -927,17 +936,6 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving
 		}
 		return nil, err
 	}
-	// Sanity-check that we walked to the file we created.
-	if createQID.Path != walkQID.Path {
-		// Probably due to concurrent remote filesystem mutation?
-		ctx.Warningf("gofer.dentry.createAndOpenChildLocked: created file has QID %v before walk, QID %v after (interop=%v)", createQID, walkQID, d.fs.opts.interop)
-		nonOpenFile.close(ctx)
-		openFile.close(ctx)
-		if fdobj != nil {
-			fdobj.Close()
-		}
-		return nil, syserror.EAGAIN
-	}
 
 	// Construct the new dentry.
 	child, err := d.fs.newDentry(ctx, nonOpenFile, createQID, attrMask, &attr)
@@ -983,16 +981,21 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving
 		}
 		childVFSFD = &fd.vfsfd
 	} else {
+		seekable := child.fileType() == linux.S_IFREG
 		fd := &specialFileFD{
 			handle: handle{
 				file: openFile,
 				fd:   -1,
 			},
+			seekable: seekable,
 		}
 		if fdobj != nil {
 			fd.handle.fd = int32(fdobj.Release())
 		}
-		if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &child.vfsd, &vfs.FileDescriptionOptions{}); err != nil {
+		if err := fd.vfsfd.Init(fd, opts.Flags, mnt, &child.vfsd, &vfs.FileDescriptionOptions{
+			DenyPRead:  !seekable,
+			DenyPWrite: !seekable,
+		}); err != nil {
 			fd.handle.close(ctx)
 			return nil, err
 		}
diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go
index 507e0e276..a464e6a94 100644
--- a/pkg/sentry/fsimpl/gofer/special_file.go
+++ b/pkg/sentry/fsimpl/gofer/special_file.go
@@ -33,13 +33,14 @@ import (
 type specialFileFD struct {
 	fileDescription
 
-	// handle is immutable.
+	// handle is used for file I/O. handle is immutable.
 	handle handle
 
-	// off is the file offset. off is protected by mu. (POSIX 2.9.7 only
-	// requires operations using the file offset to be atomic for regular files
-	// and symlinks; however, since specialFileFD may be used for regular
-	// files, we apply this atomicity unconditionally.)
+	// seekable is true if this file description represents a file for which
+	// file offset is significant, i.e. a regular file. seekable is immutable.
+	seekable bool
+
+	// If seekable is true, off is the file offset. off is protected by mu.
 	mu  sync.Mutex
 	off int64
 }
@@ -63,7 +64,7 @@ func (fd *specialFileFD) OnClose(ctx context.Context) error {
 
 // PRead implements vfs.FileDescriptionImpl.PRead.
 func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) {
-	if offset < 0 {
+	if fd.seekable && offset < 0 {
 		return 0, syserror.EINVAL
 	}
 	if opts.Flags != 0 {
@@ -91,6 +92,10 @@ func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs
 
 // Read implements vfs.FileDescriptionImpl.Read.
 func (fd *specialFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) {
+	if !fd.seekable {
+		return fd.PRead(ctx, dst, -1, opts)
+	}
+
 	fd.mu.Lock()
 	n, err := fd.PRead(ctx, dst, fd.off, opts)
 	fd.off += n
@@ -100,14 +105,14 @@ 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) {
-	if offset < 0 {
+	if fd.seekable && offset < 0 {
 		return 0, syserror.EINVAL
 	}
 	if opts.Flags != 0 {
 		return 0, syserror.EOPNOTSUPP
 	}
 
-	if fd.dentry().fileType() == linux.S_IFREG {
+	if fd.seekable {
 		limit, err := vfs.CheckLimit(ctx, offset, src.NumBytes())
 		if err != nil {
 			return 0, err
@@ -130,6 +135,10 @@ func (fd *specialFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off
 
 // Write implements vfs.FileDescriptionImpl.Write.
 func (fd *specialFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) {
+	if !fd.seekable {
+		return fd.PWrite(ctx, src, -1, opts)
+	}
+
 	fd.mu.Lock()
 	n, err := fd.PWrite(ctx, src, fd.off, opts)
 	fd.off += n
@@ -139,6 +148,9 @@ func (fd *specialFileFD) Write(ctx context.Context, src usermem.IOSequence, opts
 
 // Seek implements vfs.FileDescriptionImpl.Seek.
 func (fd *specialFileFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) {
+	if !fd.seekable {
+		return 0, syserror.ESPIPE
+	}
 	fd.mu.Lock()
 	defer fd.mu.Unlock()
 	switch whence {