summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/fsimpl/gofer
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry/fsimpl/gofer')
-rw-r--r--pkg/sentry/fsimpl/gofer/filesystem.go35
-rw-r--r--pkg/sentry/fsimpl/gofer/gofer.go74
-rw-r--r--pkg/sentry/fsimpl/gofer/regular_file.go37
-rw-r--r--pkg/sentry/fsimpl/gofer/special_file.go13
-rw-r--r--pkg/sentry/fsimpl/gofer/time.go12
5 files changed, 121 insertions, 50 deletions
diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go
index 51082359d..7bcc99b29 100644
--- a/pkg/sentry/fsimpl/gofer/filesystem.go
+++ b/pkg/sentry/fsimpl/gofer/filesystem.go
@@ -869,11 +869,22 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf
if err := d.checkPermissions(rp.Credentials(), ats); err != nil {
return nil, err
}
+
+ trunc := opts.Flags&linux.O_TRUNC != 0 && d.fileType() == linux.S_IFREG
+ if trunc {
+ // Lock metadataMu *while* we open a regular file with O_TRUNC because
+ // open(2) will change the file size on server.
+ d.metadataMu.Lock()
+ defer d.metadataMu.Unlock()
+ }
+
+ var vfd *vfs.FileDescription
+ var err error
mnt := rp.Mount()
switch d.fileType() {
case linux.S_IFREG:
if !d.fs.opts.regularFilesUseSpecialFileFD {
- if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, opts.Flags&linux.O_TRUNC != 0); err != nil {
+ if err := d.ensureSharedHandle(ctx, ats&vfs.MayRead != 0, ats&vfs.MayWrite != 0, trunc); err != nil {
return nil, err
}
fd := &regularFileFD{}
@@ -883,7 +894,7 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf
}); err != nil {
return nil, err
}
- return &fd.vfsfd, nil
+ vfd = &fd.vfsfd
}
case linux.S_IFDIR:
// Can't open directories with O_CREAT.
@@ -923,7 +934,25 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf
return d.pipe.Open(ctx, mnt, &d.vfsd, opts.Flags, &d.locks)
}
}
- return d.openSpecialFileLocked(ctx, mnt, opts)
+
+ if vfd == nil {
+ if vfd, err = d.openSpecialFileLocked(ctx, mnt, opts); err != nil {
+ return nil, err
+ }
+ }
+
+ if trunc {
+ // If no errors occured so far then update file size in memory. This
+ // step is required even if !d.cachedMetadataAuthoritative() because
+ // d.mappings has to be updated.
+ // d.metadataMu has already been acquired if trunc == true.
+ d.updateFileSizeLocked(0)
+
+ if d.cachedMetadataAuthoritative() {
+ d.touchCMtimeLocked()
+ }
+ }
+ return vfd, err
}
func (d *dentry) connectSocketLocked(ctx context.Context, opts *vfs.OpenOptions) (*vfs.FileDescription, error) {
diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go
index 71c8d3ae1..8e74e60a5 100644
--- a/pkg/sentry/fsimpl/gofer/gofer.go
+++ b/pkg/sentry/fsimpl/gofer/gofer.go
@@ -794,9 +794,7 @@ func (d *dentry) updateFromP9Attrs(mask p9.AttrMask, attr *p9.Attr) {
atomic.StoreUint32(&d.nlink, uint32(attr.NLink))
}
if mask.Size {
- d.dataMu.Lock()
- atomic.StoreUint64(&d.size, attr.Size)
- d.dataMu.Unlock()
+ d.updateFileSizeLocked(attr.Size)
}
d.metadataMu.Unlock()
}
@@ -902,6 +900,12 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin
}
d.metadataMu.Lock()
defer d.metadataMu.Unlock()
+ if stat.Mask&linux.STATX_SIZE != 0 {
+ // The size needs to be changed even when
+ // !d.cachedMetadataAuthoritative() because d.mappings has to be
+ // updated.
+ d.updateFileSizeLocked(stat.Size)
+ }
if !d.isSynthetic() {
if stat.Mask != 0 {
if err := d.file.setAttr(ctx, p9.SetAttrMask{
@@ -963,40 +967,42 @@ func (d *dentry) setStat(ctx context.Context, creds *auth.Credentials, stat *lin
stat.Mask |= linux.STATX_MTIME
}
atomic.StoreInt64(&d.ctime, now)
- if stat.Mask&linux.STATX_SIZE != 0 {
+ return nil
+}
+
+// Preconditions: d.metadataMu must be locked.
+func (d *dentry) updateFileSizeLocked(newSize uint64) {
+ d.dataMu.Lock()
+ oldSize := d.size
+ d.size = newSize
+ // d.dataMu must be unlocked to lock d.mapsMu and invalidate mappings
+ // below. This allows concurrent calls to Read/Translate/etc. These
+ // functions synchronize with truncation by refusing to use cache
+ // contents beyond the new d.size. (We are still holding d.metadataMu,
+ // so we can't race with Write or another truncate.)
+ d.dataMu.Unlock()
+ if d.size < oldSize {
+ oldpgend, _ := usermem.PageRoundUp(oldSize)
+ newpgend, _ := usermem.PageRoundUp(d.size)
+ if oldpgend != newpgend {
+ d.mapsMu.Lock()
+ d.mappings.Invalidate(memmap.MappableRange{newpgend, oldpgend}, memmap.InvalidateOpts{
+ // Compare Linux's mm/truncate.c:truncate_setsize() =>
+ // truncate_pagecache() =>
+ // mm/memory.c:unmap_mapping_range(evencows=1).
+ InvalidatePrivate: true,
+ })
+ d.mapsMu.Unlock()
+ }
+ // We are now guaranteed that there are no translations of
+ // truncated pages, and can remove them from the cache. Since
+ // truncated pages have been removed from the remote file, they
+ // should be dropped without being written back.
d.dataMu.Lock()
- oldSize := d.size
- d.size = stat.Size
- // d.dataMu must be unlocked to lock d.mapsMu and invalidate mappings
- // below. This allows concurrent calls to Read/Translate/etc. These
- // functions synchronize with truncation by refusing to use cache
- // contents beyond the new d.size. (We are still holding d.metadataMu,
- // so we can't race with Write or another truncate.)
+ d.cache.Truncate(d.size, d.fs.mfp.MemoryFile())
+ d.dirty.KeepClean(memmap.MappableRange{d.size, oldpgend})
d.dataMu.Unlock()
- if d.size < oldSize {
- oldpgend, _ := usermem.PageRoundUp(oldSize)
- newpgend, _ := usermem.PageRoundUp(d.size)
- if oldpgend != newpgend {
- d.mapsMu.Lock()
- d.mappings.Invalidate(memmap.MappableRange{newpgend, oldpgend}, memmap.InvalidateOpts{
- // Compare Linux's mm/truncate.c:truncate_setsize() =>
- // truncate_pagecache() =>
- // mm/memory.c:unmap_mapping_range(evencows=1).
- InvalidatePrivate: true,
- })
- d.mapsMu.Unlock()
- }
- // We are now guaranteed that there are no translations of
- // truncated pages, and can remove them from the cache. Since
- // truncated pages have been removed from the remote file, they
- // should be dropped without being written back.
- d.dataMu.Lock()
- d.cache.Truncate(d.size, d.fs.mfp.MemoryFile())
- d.dirty.KeepClean(memmap.MappableRange{d.size, oldpgend})
- d.dataMu.Unlock()
- }
}
- return nil
}
func (d *dentry) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error {
diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go
index 404a452c4..3d2d3530a 100644
--- a/pkg/sentry/fsimpl/gofer/regular_file.go
+++ b/pkg/sentry/fsimpl/gofer/regular_file.go
@@ -24,6 +24,7 @@ import (
"gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/log"
+ "gvisor.dev/gvisor/pkg/p9"
"gvisor.dev/gvisor/pkg/safemem"
"gvisor.dev/gvisor/pkg/sentry/fs/fsutil"
"gvisor.dev/gvisor/pkg/sentry/memmap"
@@ -67,13 +68,43 @@ func (fd *regularFileFD) OnClose(ctx context.Context) error {
return d.handle.file.flush(ctx)
}
+// Allocate implements vfs.FileDescriptionImpl.Allocate.
+func (fd *regularFileFD) Allocate(ctx context.Context, mode, offset, length uint64) error {
+
+ d := fd.dentry()
+ d.metadataMu.Lock()
+ defer d.metadataMu.Unlock()
+
+ size := offset + length
+
+ // Allocating a smaller size is a noop.
+ if size <= d.size {
+ return nil
+ }
+
+ d.handleMu.Lock()
+ defer d.handleMu.Unlock()
+
+ err := d.handle.file.allocate(ctx, p9.ToAllocateMode(mode), offset, length)
+ if err != nil {
+ return err
+ }
+ d.size = size
+ if !d.cachedMetadataAuthoritative() {
+ d.touchCMtimeLocked()
+ }
+ return nil
+}
+
// PRead implements vfs.FileDescriptionImpl.PRead.
func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) {
if offset < 0 {
return 0, syserror.EINVAL
}
- // Check that flags are supported. Silently ignore RWF_HIPRI.
+ // Check that flags are supported.
+ //
+ // TODO(gvisor.dev/issue/2601): Support select preadv2 flags.
if opts.Flags&^linux.RWF_HIPRI != 0 {
return 0, syserror.EOPNOTSUPP
}
@@ -126,7 +157,9 @@ func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off
return 0, syserror.EINVAL
}
- // Check that flags are supported. Silently ignore RWF_HIPRI.
+ // 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
}
diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go
index a016cbae1..3c4e7e2e4 100644
--- a/pkg/sentry/fsimpl/gofer/special_file.go
+++ b/pkg/sentry/fsimpl/gofer/special_file.go
@@ -130,7 +130,9 @@ func (fd *specialFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offs
return 0, syserror.EINVAL
}
- // Check that flags are supported. Silently ignore RWF_HIPRI.
+ // Check that flags are supported.
+ //
+ // TODO(gvisor.dev/issue/2601): Support select preadv2 flags.
if opts.Flags&^linux.RWF_HIPRI != 0 {
return 0, syserror.EOPNOTSUPP
}
@@ -176,7 +178,9 @@ func (fd *specialFileFD) PWrite(ctx context.Context, src usermem.IOSequence, off
return 0, syserror.EINVAL
}
- // Check that flags are supported. Silently ignore RWF_HIPRI.
+ // 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
}
@@ -235,8 +239,5 @@ func (fd *specialFileFD) Seek(ctx context.Context, offset int64, whence int32) (
// Sync implements vfs.FileDescriptionImpl.Sync.
func (fd *specialFileFD) Sync(ctx context.Context) error {
- if !fd.vfsfd.IsWritable() {
- return nil
- }
- return fd.handle.sync(ctx)
+ return fd.dentry().syncSharedHandle(ctx)
}
diff --git a/pkg/sentry/fsimpl/gofer/time.go b/pkg/sentry/fsimpl/gofer/time.go
index 1d5aa82dc..0eef4e16e 100644
--- a/pkg/sentry/fsimpl/gofer/time.go
+++ b/pkg/sentry/fsimpl/gofer/time.go
@@ -36,7 +36,7 @@ func statxTimestampFromDentry(ns int64) linux.StatxTimestamp {
}
}
-// Preconditions: fs.interop != InteropModeShared.
+// Preconditions: d.cachedMetadataAuthoritative() == true.
func (d *dentry) touchAtime(mnt *vfs.Mount) {
if mnt.Flags.NoATime {
return
@@ -51,8 +51,8 @@ func (d *dentry) touchAtime(mnt *vfs.Mount) {
mnt.EndWrite()
}
-// Preconditions: fs.interop != InteropModeShared. The caller has successfully
-// called vfs.Mount.CheckBeginWrite().
+// Preconditions: d.cachedMetadataAuthoritative() == true. The caller has
+// successfully called vfs.Mount.CheckBeginWrite().
func (d *dentry) touchCtime() {
now := d.fs.clock.Now().Nanoseconds()
d.metadataMu.Lock()
@@ -60,8 +60,8 @@ func (d *dentry) touchCtime() {
d.metadataMu.Unlock()
}
-// Preconditions: fs.interop != InteropModeShared. The caller has successfully
-// called vfs.Mount.CheckBeginWrite().
+// Preconditions: d.cachedMetadataAuthoritative() == true. The caller has
+// successfully called vfs.Mount.CheckBeginWrite().
func (d *dentry) touchCMtime() {
now := d.fs.clock.Now().Nanoseconds()
d.metadataMu.Lock()
@@ -70,6 +70,8 @@ func (d *dentry) touchCMtime() {
d.metadataMu.Unlock()
}
+// Preconditions: d.cachedMetadataAuthoritative() == true. The caller has
+// locked d.metadataMu.
func (d *dentry) touchCMtimeLocked() {
now := d.fs.clock.Now().Nanoseconds()
atomic.StoreInt64(&d.mtime, now)
generated by cgit v1.2.3 (git 2.39.1) at 2025-05-08 20:25:34 +0000