diff options
Diffstat (limited to 'pkg')
290 files changed, 2679 insertions, 1965 deletions
diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index a91f9f018..9c27f7bb2 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -59,7 +59,7 @@ var VerdictStrings = map[int32]string{ NF_RETURN: "RETURN", } -// Socket options. These correspond to values in +// Socket options for SOL_SOCKET. These correspond to values in // include/uapi/linux/netfilter_ipv4/ip_tables.h. const ( IPT_BASE_CTL = 64 @@ -74,6 +74,12 @@ const ( IPT_SO_GET_MAX = IPT_SO_GET_REVISION_TARGET ) +// Socket option for SOL_IP. This corresponds to the value in +// include/uapi/linux/netfilter_ipv4.h. +const ( + SO_ORIGINAL_DST = 80 +) + // Name lengths. These correspond to values in // include/uapi/linux/netfilter/x_tables.h. const ( diff --git a/pkg/abi/linux/socket.go b/pkg/abi/linux/socket.go index c24a8216e..693996c01 100644 --- a/pkg/abi/linux/socket.go +++ b/pkg/abi/linux/socket.go @@ -83,7 +83,6 @@ const ( MSG_MORE = 0x8000 MSG_WAITFORONE = 0x10000 MSG_SENDPAGE_NOTLAST = 0x20000 - MSG_REINJECT = 0x8000000 MSG_ZEROCOPY = 0x4000000 MSG_FASTOPEN = 0x20000000 MSG_CMSG_CLOEXEC = 0x40000000 @@ -239,11 +238,13 @@ const SockAddrMax = 128 type InetAddr [4]byte // SockAddrInet is struct sockaddr_in, from uapi/linux/in.h. +// +// +marshal type SockAddrInet struct { Family uint16 Port uint16 Addr InetAddr - Zero [8]uint8 // pad to sizeof(struct sockaddr). + _ [8]uint8 // pad to sizeof(struct sockaddr). } // InetMulticastRequest is struct ip_mreq, from uapi/linux/in.h. diff --git a/pkg/refs/BUILD b/pkg/refs/BUILD index 74affc887..9888cce9c 100644 --- a/pkg/refs/BUILD +++ b/pkg/refs/BUILD @@ -24,6 +24,7 @@ go_library( ], visibility = ["//:sandbox"], deps = [ + "//pkg/context", "//pkg/log", "//pkg/sync", ], @@ -34,5 +35,8 @@ go_test( size = "small", srcs = ["refcounter_test.go"], library = ":refs", - deps = ["//pkg/sync"], + deps = [ + "//pkg/context", + "//pkg/sync", + ], ) diff --git a/pkg/refs/refcounter.go b/pkg/refs/refcounter.go index c45ba8200..61790221b 100644 --- a/pkg/refs/refcounter.go +++ b/pkg/refs/refcounter.go @@ -23,6 +23,7 @@ import ( "runtime" "sync/atomic" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sync" ) @@ -38,7 +39,7 @@ type RefCounter interface { // Note that AtomicRefCounter.DecRef() does not support destructors. // If a type has a destructor, it must implement its own DecRef() // method and call AtomicRefCounter.DecRefWithDestructor(destructor). - DecRef() + DecRef(ctx context.Context) // TryIncRef attempts to increase the reference counter on the object, // but may fail if all references have already been dropped. This @@ -57,7 +58,7 @@ type RefCounter interface { // A WeakRefUser is notified when the last non-weak reference is dropped. type WeakRefUser interface { // WeakRefGone is called when the last non-weak reference is dropped. - WeakRefGone() + WeakRefGone(ctx context.Context) } // WeakRef is a weak reference. @@ -123,7 +124,7 @@ func (w *WeakRef) Get() RefCounter { // Drop drops this weak reference. You should always call drop when you are // finished with the weak reference. You may not use this object after calling // drop. -func (w *WeakRef) Drop() { +func (w *WeakRef) Drop(ctx context.Context) { rc, ok := w.get() if !ok { // We've been zapped already. When the refcounter has called @@ -145,7 +146,7 @@ func (w *WeakRef) Drop() { // And now aren't on the object's list of weak references. So it won't // zap us if this causes the reference count to drop to zero. - rc.DecRef() + rc.DecRef(ctx) // Return to the pool. weakRefPool.Put(w) @@ -427,7 +428,7 @@ func (r *AtomicRefCount) dropWeakRef(w *WeakRef) { // A: TryIncRef [transform speculative to real] // //go:nosplit -func (r *AtomicRefCount) DecRefWithDestructor(destroy func()) { +func (r *AtomicRefCount) DecRefWithDestructor(ctx context.Context, destroy func(context.Context)) { switch v := atomic.AddInt64(&r.refCount, -1); { case v < -1: panic("Decrementing non-positive ref count") @@ -448,7 +449,7 @@ func (r *AtomicRefCount) DecRefWithDestructor(destroy func()) { if user != nil { r.mu.Unlock() - user.WeakRefGone() + user.WeakRefGone(ctx) r.mu.Lock() } } @@ -456,7 +457,7 @@ func (r *AtomicRefCount) DecRefWithDestructor(destroy func()) { // Call the destructor. if destroy != nil { - destroy() + destroy(ctx) } } } @@ -464,6 +465,6 @@ func (r *AtomicRefCount) DecRefWithDestructor(destroy func()) { // DecRef decrements this object's reference count. // //go:nosplit -func (r *AtomicRefCount) DecRef() { - r.DecRefWithDestructor(nil) +func (r *AtomicRefCount) DecRef(ctx context.Context) { + r.DecRefWithDestructor(ctx, nil) } diff --git a/pkg/refs/refcounter_test.go b/pkg/refs/refcounter_test.go index 1ab4a4440..6d0dd1018 100644 --- a/pkg/refs/refcounter_test.go +++ b/pkg/refs/refcounter_test.go @@ -18,6 +18,7 @@ import ( "reflect" "testing" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sync" ) @@ -31,11 +32,11 @@ type testCounter struct { destroyed bool } -func (t *testCounter) DecRef() { - t.AtomicRefCount.DecRefWithDestructor(t.destroy) +func (t *testCounter) DecRef(ctx context.Context) { + t.AtomicRefCount.DecRefWithDestructor(ctx, t.destroy) } -func (t *testCounter) destroy() { +func (t *testCounter) destroy(context.Context) { t.mu.Lock() defer t.mu.Unlock() t.destroyed = true @@ -53,7 +54,7 @@ func newTestCounter() *testCounter { func TestOneRef(t *testing.T) { tc := newTestCounter() - tc.DecRef() + tc.DecRef(context.Background()) if !tc.IsDestroyed() { t.Errorf("object should have been destroyed") @@ -63,8 +64,9 @@ func TestOneRef(t *testing.T) { func TestTwoRefs(t *testing.T) { tc := newTestCounter() tc.IncRef() - tc.DecRef() - tc.DecRef() + ctx := context.Background() + tc.DecRef(ctx) + tc.DecRef(ctx) if !tc.IsDestroyed() { t.Errorf("object should have been destroyed") @@ -74,12 +76,13 @@ func TestTwoRefs(t *testing.T) { func TestMultiRefs(t *testing.T) { tc := newTestCounter() tc.IncRef() - tc.DecRef() + ctx := context.Background() + tc.DecRef(ctx) tc.IncRef() - tc.DecRef() + tc.DecRef(ctx) - tc.DecRef() + tc.DecRef(ctx) if !tc.IsDestroyed() { t.Errorf("object should have been destroyed") @@ -89,19 +92,20 @@ func TestMultiRefs(t *testing.T) { func TestWeakRef(t *testing.T) { tc := newTestCounter() w := NewWeakRef(tc, nil) + ctx := context.Background() // Try resolving. if x := w.Get(); x == nil { t.Errorf("weak reference didn't resolve: expected %v, got nil", tc) } else { - x.DecRef() + x.DecRef(ctx) } // Try resolving again. if x := w.Get(); x == nil { t.Errorf("weak reference didn't resolve: expected %v, got nil", tc) } else { - x.DecRef() + x.DecRef(ctx) } // Shouldn't be destroyed yet. (Can't continue if this fails.) @@ -110,7 +114,7 @@ func TestWeakRef(t *testing.T) { } // Drop the original reference. - tc.DecRef() + tc.DecRef(ctx) // Assert destroyed. if !tc.IsDestroyed() { @@ -126,7 +130,8 @@ func TestWeakRef(t *testing.T) { func TestWeakRefDrop(t *testing.T) { tc := newTestCounter() w := NewWeakRef(tc, nil) - w.Drop() + ctx := context.Background() + w.Drop(ctx) // Just assert the list is empty. if !tc.weakRefs.Empty() { @@ -134,14 +139,14 @@ func TestWeakRefDrop(t *testing.T) { } // Drop the original reference. - tc.DecRef() + tc.DecRef(ctx) } type testWeakRefUser struct { weakRefGone func() } -func (u *testWeakRefUser) WeakRefGone() { +func (u *testWeakRefUser) WeakRefGone(ctx context.Context) { u.weakRefGone() } @@ -165,7 +170,8 @@ func TestCallback(t *testing.T) { }}) // Drop the original reference, this must trigger the callback. - tc.DecRef() + ctx := context.Background() + tc.DecRef(ctx) if !called { t.Fatalf("Callback not called") diff --git a/pkg/sentry/control/proc.go b/pkg/sentry/control/proc.go index 1bae7cfaf..dfa936563 100644 --- a/pkg/sentry/control/proc.go +++ b/pkg/sentry/control/proc.go @@ -139,7 +139,6 @@ func ExecAsync(proc *Proc, args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadID, *host.TTYFileOperations, *hostvfs2.TTYFileDescription, error) { // Import file descriptors. fdTable := proc.Kernel.NewFDTable() - defer fdTable.DecRef() creds := auth.NewUserCredentials( args.KUID, @@ -177,6 +176,7 @@ func (proc *Proc) execAsync(args *ExecArgs) (*kernel.ThreadGroup, kernel.ThreadI initArgs.MountNamespaceVFS2.IncRef() } ctx := initArgs.NewContext(proc.Kernel) + defer fdTable.DecRef(ctx) if kernel.VFS2Enabled { // Get the full path to the filename from the PATH env variable. diff --git a/pkg/sentry/devices/memdev/full.go b/pkg/sentry/devices/memdev/full.go index af66fe4dc..511179e31 100644 --- a/pkg/sentry/devices/memdev/full.go +++ b/pkg/sentry/devices/memdev/full.go @@ -46,7 +46,7 @@ type fullFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *fullFD) Release() { +func (fd *fullFD) Release(context.Context) { // noop } diff --git a/pkg/sentry/devices/memdev/null.go b/pkg/sentry/devices/memdev/null.go index 92d3d71be..4918dbeeb 100644 --- a/pkg/sentry/devices/memdev/null.go +++ b/pkg/sentry/devices/memdev/null.go @@ -47,7 +47,7 @@ type nullFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *nullFD) Release() { +func (fd *nullFD) Release(context.Context) { // noop } diff --git a/pkg/sentry/devices/memdev/random.go b/pkg/sentry/devices/memdev/random.go index 6b81da5ef..5e7fe0280 100644 --- a/pkg/sentry/devices/memdev/random.go +++ b/pkg/sentry/devices/memdev/random.go @@ -56,7 +56,7 @@ type randomFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *randomFD) Release() { +func (fd *randomFD) Release(context.Context) { // noop } diff --git a/pkg/sentry/devices/memdev/zero.go b/pkg/sentry/devices/memdev/zero.go index c6f15054d..2e631a252 100644 --- a/pkg/sentry/devices/memdev/zero.go +++ b/pkg/sentry/devices/memdev/zero.go @@ -48,7 +48,7 @@ type zeroFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *zeroFD) Release() { +func (fd *zeroFD) Release(context.Context) { // noop } diff --git a/pkg/sentry/devices/ttydev/ttydev.go b/pkg/sentry/devices/ttydev/ttydev.go index fbb7fd92c..fd4b79c46 100644 --- a/pkg/sentry/devices/ttydev/ttydev.go +++ b/pkg/sentry/devices/ttydev/ttydev.go @@ -55,7 +55,7 @@ type ttyFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *ttyFD) Release() {} +func (fd *ttyFD) Release(context.Context) {} // PRead implements vfs.FileDescriptionImpl.PRead. func (fd *ttyFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { diff --git a/pkg/sentry/devices/tundev/tundev.go b/pkg/sentry/devices/tundev/tundev.go index dfbd069af..852ec3c5c 100644 --- a/pkg/sentry/devices/tundev/tundev.go +++ b/pkg/sentry/devices/tundev/tundev.go @@ -108,8 +108,8 @@ func (fd *tunFD) Ioctl(ctx context.Context, uio usermem.IO, args arch.SyscallArg } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *tunFD) Release() { - fd.device.Release() +func (fd *tunFD) Release(ctx context.Context) { + fd.device.Release(ctx) } // PRead implements vfs.FileDescriptionImpl.PRead. diff --git a/pkg/sentry/fdimport/fdimport.go b/pkg/sentry/fdimport/fdimport.go index a4199f9e9..1b7cb94c0 100644 --- a/pkg/sentry/fdimport/fdimport.go +++ b/pkg/sentry/fdimport/fdimport.go @@ -15,6 +15,8 @@ package fdimport import ( + "fmt" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/host" @@ -48,7 +50,7 @@ func importFS(ctx context.Context, fdTable *kernel.FDTable, console bool, fds [] if err != nil { return nil, err } - defer appFile.DecRef() + defer appFile.DecRef(ctx) // Remember this in the TTY file, as we will // use it for the other stdio FDs. @@ -67,7 +69,7 @@ func importFS(ctx context.Context, fdTable *kernel.FDTable, console bool, fds [] if err != nil { return nil, err } - defer appFile.DecRef() + defer appFile.DecRef(ctx) } // Add the file to the FD map. @@ -84,6 +86,9 @@ func importFS(ctx context.Context, fdTable *kernel.FDTable, console bool, fds [] func importVFS2(ctx context.Context, fdTable *kernel.FDTable, console bool, stdioFDs []int) (*hostvfs2.TTYFileDescription, error) { k := kernel.KernelFromContext(ctx) + if k == nil { + return nil, fmt.Errorf("cannot find kernel from context") + } var ttyFile *vfs.FileDescription for appFD, hostFD := range stdioFDs { @@ -97,7 +102,7 @@ func importVFS2(ctx context.Context, fdTable *kernel.FDTable, console bool, stdi if err != nil { return nil, err } - defer appFile.DecRef() + defer appFile.DecRef(ctx) // Remember this in the TTY file, as we will use it for the other stdio // FDs. @@ -114,7 +119,7 @@ func importVFS2(ctx context.Context, fdTable *kernel.FDTable, console bool, stdi if err != nil { return nil, err } - defer appFile.DecRef() + defer appFile.DecRef(ctx) } if err := fdTable.NewFDAtVFS2(ctx, int32(appFD), appFile, kernel.FDFlags{}); err != nil { diff --git a/pkg/sentry/fs/copy_up.go b/pkg/sentry/fs/copy_up.go index ab1424c95..735452b07 100644 --- a/pkg/sentry/fs/copy_up.go +++ b/pkg/sentry/fs/copy_up.go @@ -201,7 +201,7 @@ func copyUpLocked(ctx context.Context, parent *Dirent, next *Dirent) error { parentUpper := parent.Inode.overlay.upper root := RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } // Create the file in the upper filesystem and get an Inode for it. @@ -212,7 +212,7 @@ func copyUpLocked(ctx context.Context, parent *Dirent, next *Dirent) error { log.Warningf("copy up failed to create file: %v", err) return syserror.EIO } - defer childFile.DecRef() + defer childFile.DecRef(ctx) childUpperInode = childFile.Dirent.Inode case Directory: @@ -226,7 +226,7 @@ func copyUpLocked(ctx context.Context, parent *Dirent, next *Dirent) error { cleanupUpper(ctx, parentUpper, next.name, werr) return syserror.EIO } - defer childUpper.DecRef() + defer childUpper.DecRef(ctx) childUpperInode = childUpper.Inode case Symlink: @@ -246,7 +246,7 @@ func copyUpLocked(ctx context.Context, parent *Dirent, next *Dirent) error { cleanupUpper(ctx, parentUpper, next.name, werr) return syserror.EIO } - defer childUpper.DecRef() + defer childUpper.DecRef(ctx) childUpperInode = childUpper.Inode default: @@ -352,14 +352,14 @@ func copyContentsLocked(ctx context.Context, upper *Inode, lower *Inode, size in if err != nil { return err } - defer upperFile.DecRef() + defer upperFile.DecRef(ctx) // Get a handle to the lower filesystem, which we will read from. lowerFile, err := overlayFile(ctx, lower, FileFlags{Read: true}) if err != nil { return err } - defer lowerFile.DecRef() + defer lowerFile.DecRef(ctx) // Use a buffer pool to minimize allocations. buf := copyUpBuffers.Get().([]byte) diff --git a/pkg/sentry/fs/copy_up_test.go b/pkg/sentry/fs/copy_up_test.go index 91792d9fe..c7a11eec1 100644 --- a/pkg/sentry/fs/copy_up_test.go +++ b/pkg/sentry/fs/copy_up_test.go @@ -126,7 +126,7 @@ func makeOverlayTestFiles(t *testing.T) []*overlayTestFile { if err != nil { t.Fatalf("failed to create file %q: %v", name, err) } - defer f.DecRef() + defer f.DecRef(ctx) relname, _ := f.Dirent.FullName(lowerRoot) @@ -171,7 +171,7 @@ func makeOverlayTestFiles(t *testing.T) []*overlayTestFile { if err != nil { t.Fatalf("failed to find %q: %v", f.name, err) } - defer d.DecRef() + defer d.DecRef(ctx) f.File, err = d.Inode.GetFile(ctx, d, fs.FileFlags{Read: true}) if err != nil { diff --git a/pkg/sentry/fs/dev/net_tun.go b/pkg/sentry/fs/dev/net_tun.go index dc7ad075a..ec474e554 100644 --- a/pkg/sentry/fs/dev/net_tun.go +++ b/pkg/sentry/fs/dev/net_tun.go @@ -80,8 +80,8 @@ type netTunFileOperations struct { var _ fs.FileOperations = (*netTunFileOperations)(nil) // Release implements fs.FileOperations.Release. -func (fops *netTunFileOperations) Release() { - fops.device.Release() +func (fops *netTunFileOperations) Release(ctx context.Context) { + fops.device.Release(ctx) } // Ioctl implements fs.FileOperations.Ioctl. diff --git a/pkg/sentry/fs/dirent.go b/pkg/sentry/fs/dirent.go index 65be12175..a2f751068 100644 --- a/pkg/sentry/fs/dirent.go +++ b/pkg/sentry/fs/dirent.go @@ -325,7 +325,7 @@ func (d *Dirent) SyncAll(ctx context.Context) { for _, w := range d.children { if child := w.Get(); child != nil { child.(*Dirent).SyncAll(ctx) - child.DecRef() + child.DecRef(ctx) } } } @@ -451,7 +451,7 @@ func (d *Dirent) walk(ctx context.Context, root *Dirent, name string, walkMayUnl // which don't hold a hard reference on their parent (their parent holds a // hard reference on them, and they contain virtually no state). But this is // good house-keeping. - child.DecRef() + child.DecRef(ctx) return nil, syscall.ENOENT } @@ -468,20 +468,20 @@ func (d *Dirent) walk(ctx context.Context, root *Dirent, name string, walkMayUnl // their pins on the child. Inotify doesn't properly support filesystems that // revalidate dirents (since watches are lost on revalidation), but if we fail // to unpin the watches child will never be GCed. - cd.Inode.Watches.Unpin(cd) + cd.Inode.Watches.Unpin(ctx, cd) // This child needs to be revalidated, fallthrough to unhash it. Make sure // to not leak a reference from Get(). // // Note that previous lookups may still have a reference to this stale child; // this can't be helped, but we can ensure that *new* lookups are up-to-date. - child.DecRef() + child.DecRef(ctx) } // Either our weak reference expired or we need to revalidate it. Unhash child first, we're // about to replace it. delete(d.children, name) - w.Drop() + w.Drop(ctx) } // Slow path: load the InodeOperations into memory. Since this is a hot path and the lookup may be @@ -512,12 +512,12 @@ func (d *Dirent) walk(ctx context.Context, root *Dirent, name string, walkMayUnl // There are active references to the existing child, prefer it to the one we // retrieved from Lookup. Likely the Lookup happened very close to the insertion // of child, so considering one stale over the other is fairly arbitrary. - c.DecRef() + c.DecRef(ctx) // The child that was installed could be negative. if cd.IsNegative() { // If so, don't leak a reference and short circuit. - child.DecRef() + child.DecRef(ctx) return nil, syscall.ENOENT } @@ -531,7 +531,7 @@ func (d *Dirent) walk(ctx context.Context, root *Dirent, name string, walkMayUnl // we did the Inode.Lookup. Fully drop the weak reference and fallback to using the child // we looked up. delete(d.children, name) - w.Drop() + w.Drop(ctx) } // Give the looked up child a parent. We cannot kick out entries, since we just checked above @@ -587,7 +587,7 @@ func (d *Dirent) exists(ctx context.Context, root *Dirent, name string) bool { return false } // Child exists. - child.DecRef() + child.DecRef(ctx) return true } @@ -622,7 +622,7 @@ func (d *Dirent) Create(ctx context.Context, root *Dirent, name string, flags Fi } child := file.Dirent - d.finishCreate(child, name) + d.finishCreate(ctx, child, name) // Return the reference and the new file. When the last reference to // the file is dropped, file.Dirent may no longer be cached. @@ -631,7 +631,7 @@ func (d *Dirent) Create(ctx context.Context, root *Dirent, name string, flags Fi // finishCreate validates the created file, adds it as a child of this dirent, // and notifies any watchers. -func (d *Dirent) finishCreate(child *Dirent, name string) { +func (d *Dirent) finishCreate(ctx context.Context, child *Dirent, name string) { // Sanity check c, its name must be consistent. if child.name != name { panic(fmt.Sprintf("create from %q to %q returned unexpected name %q", d.name, name, child.name)) @@ -650,14 +650,14 @@ func (d *Dirent) finishCreate(child *Dirent, name string) { panic(fmt.Sprintf("hashed child %q over a positive child", child.name)) } // Don't leak a reference. - old.DecRef() + old.DecRef(ctx) // Drop d's reference. - old.DecRef() + old.DecRef(ctx) } // Finally drop the useless weak reference on the floor. - w.Drop() + w.Drop(ctx) } d.Inode.Watches.Notify(name, linux.IN_CREATE, 0) @@ -686,17 +686,17 @@ func (d *Dirent) genericCreate(ctx context.Context, root *Dirent, name string, c panic(fmt.Sprintf("hashed over a positive child %q", old.(*Dirent).name)) } // Don't leak a reference. - old.DecRef() + old.DecRef(ctx) // Drop d's reference. - old.DecRef() + old.DecRef(ctx) } // Unhash the negative Dirent, name needs to exist now. delete(d.children, name) // Finally drop the useless weak reference on the floor. - w.Drop() + w.Drop(ctx) } // Execute the create operation. @@ -756,7 +756,7 @@ func (d *Dirent) Bind(ctx context.Context, root *Dirent, name string, data trans if e != nil { return e } - d.finishCreate(childDir, name) + d.finishCreate(ctx, childDir, name) return nil }) if err == syscall.EEXIST { @@ -901,7 +901,7 @@ func direntReaddir(ctx context.Context, d *Dirent, it DirIterator, root *Dirent, // references to children. // // Preconditions: d.mu must be held. -func (d *Dirent) flush() { +func (d *Dirent) flush(ctx context.Context) { expired := make(map[string]*refs.WeakRef) for n, w := range d.children { // Call flush recursively on each child before removing our @@ -912,7 +912,7 @@ func (d *Dirent) flush() { if !cd.IsNegative() { // Flush the child. cd.mu.Lock() - cd.flush() + cd.flush(ctx) cd.mu.Unlock() // Allow the file system to drop extra references on child. @@ -920,13 +920,13 @@ func (d *Dirent) flush() { } // Don't leak a reference. - child.DecRef() + child.DecRef(ctx) } // Check if the child dirent is closed, and mark it as expired if it is. // We must call w.Get() again here, since the child could have been closed // by the calls to flush() and cache.Remove() in the above if-block. if child := w.Get(); child != nil { - child.DecRef() + child.DecRef(ctx) } else { expired[n] = w } @@ -935,7 +935,7 @@ func (d *Dirent) flush() { // Remove expired entries. for n, w := range expired { delete(d.children, n) - w.Drop() + w.Drop(ctx) } } @@ -977,7 +977,7 @@ func (d *Dirent) mount(ctx context.Context, inode *Inode) (newChild *Dirent, err if !ok { panic("mount must mount over an existing dirent") } - weakRef.Drop() + weakRef.Drop(ctx) // Note that even though `d` is now hidden, it still holds a reference // to its parent. @@ -1002,13 +1002,13 @@ func (d *Dirent) unmount(ctx context.Context, replacement *Dirent) error { if !ok { panic("mount must mount over an existing dirent") } - weakRef.Drop() + weakRef.Drop(ctx) // d is not reachable anymore, and hence not mounted anymore. d.mounted = false // Drop mount reference. - d.DecRef() + d.DecRef(ctx) return nil } @@ -1029,7 +1029,7 @@ func (d *Dirent) Remove(ctx context.Context, root *Dirent, name string, dirPath // Child does not exist. return err } - defer child.DecRef() + defer child.DecRef(ctx) // Remove cannot remove directories. if IsDir(child.Inode.StableAttr) { @@ -1055,7 +1055,7 @@ func (d *Dirent) Remove(ctx context.Context, root *Dirent, name string, dirPath atomic.StoreInt32(&child.deleted, 1) if w, ok := d.children[name]; ok { delete(d.children, name) - w.Drop() + w.Drop(ctx) } // Allow the file system to drop extra references on child. @@ -1067,7 +1067,7 @@ func (d *Dirent) Remove(ctx context.Context, root *Dirent, name string, dirPath // inode may have other links. If this was the last link, the events for the // watch removal will be queued by the inode destructor. child.Inode.Watches.MarkUnlinked() - child.Inode.Watches.Unpin(child) + child.Inode.Watches.Unpin(ctx, child) d.Inode.Watches.Notify(name, linux.IN_DELETE, 0) return nil @@ -1100,7 +1100,7 @@ func (d *Dirent) RemoveDirectory(ctx context.Context, root *Dirent, name string) // Child does not exist. return err } - defer child.DecRef() + defer child.DecRef(ctx) // RemoveDirectory can only remove directories. if !IsDir(child.Inode.StableAttr) { @@ -1121,7 +1121,7 @@ func (d *Dirent) RemoveDirectory(ctx context.Context, root *Dirent, name string) atomic.StoreInt32(&child.deleted, 1) if w, ok := d.children[name]; ok { delete(d.children, name) - w.Drop() + w.Drop(ctx) } // Allow the file system to drop extra references on child. @@ -1130,14 +1130,14 @@ func (d *Dirent) RemoveDirectory(ctx context.Context, root *Dirent, name string) // Finally, let inotify know the child is being unlinked. Drop any extra // refs from inotify to this child dirent. child.Inode.Watches.MarkUnlinked() - child.Inode.Watches.Unpin(child) + child.Inode.Watches.Unpin(ctx, child) d.Inode.Watches.Notify(name, linux.IN_ISDIR|linux.IN_DELETE, 0) return nil } // destroy closes this node and all children. -func (d *Dirent) destroy() { +func (d *Dirent) destroy(ctx context.Context) { if d.IsNegative() { // Nothing to tear-down and no parent references to drop, since a negative // Dirent does not take a references on its parent, has no Inode and no children. @@ -1153,19 +1153,19 @@ func (d *Dirent) destroy() { if c.(*Dirent).IsNegative() { // The parent holds both weak and strong refs in the case of // negative dirents. - c.DecRef() + c.DecRef(ctx) } // Drop the reference we just acquired in WeakRef.Get. - c.DecRef() + c.DecRef(ctx) } - w.Drop() + w.Drop(ctx) } d.children = nil allDirents.remove(d) // Drop our reference to the Inode. - d.Inode.DecRef() + d.Inode.DecRef(ctx) // Allow the Dirent to be GC'ed after this point, since the Inode may still // be referenced after the Dirent is destroyed (for instance by filesystem @@ -1175,7 +1175,7 @@ func (d *Dirent) destroy() { // Drop the reference we have on our parent if we took one. renameMu doesn't need to be // held because d can't be reparented without any references to it left. if d.parent != nil { - d.parent.DecRef() + d.parent.DecRef(ctx) } } @@ -1201,14 +1201,14 @@ func (d *Dirent) TryIncRef() bool { // DecRef decreases the Dirent's refcount and drops its reference on its mount. // // DecRef implements RefCounter.DecRef with destructor d.destroy. -func (d *Dirent) DecRef() { +func (d *Dirent) DecRef(ctx context.Context) { if d.Inode != nil { // Keep mount around, since DecRef may destroy d.Inode. msrc := d.Inode.MountSource - d.DecRefWithDestructor(d.destroy) + d.DecRefWithDestructor(ctx, d.destroy) msrc.DecDirentRefs() } else { - d.DecRefWithDestructor(d.destroy) + d.DecRefWithDestructor(ctx, d.destroy) } } @@ -1359,7 +1359,7 @@ func (d *Dirent) MayDelete(ctx context.Context, root *Dirent, name string) error if err != nil { return err } - defer victim.DecRef() + defer victim.DecRef(ctx) return d.mayDelete(ctx, victim) } @@ -1411,7 +1411,7 @@ func Rename(ctx context.Context, root *Dirent, oldParent *Dirent, oldName string if err != nil { return err } - defer renamed.DecRef() + defer renamed.DecRef(ctx) // Check that the renamed dirent is deletable. if err := oldParent.mayDelete(ctx, renamed); err != nil { @@ -1453,13 +1453,13 @@ func Rename(ctx context.Context, root *Dirent, oldParent *Dirent, oldName string // Check that we can delete replaced. if err := newParent.mayDelete(ctx, replaced); err != nil { - replaced.DecRef() + replaced.DecRef(ctx) return err } // Target should not be an ancestor of source. if oldParent.descendantOf(replaced) { - replaced.DecRef() + replaced.DecRef(ctx) // Note that Linux returns EINVAL if the source is an // ancestor of target, but ENOTEMPTY if the target is @@ -1470,7 +1470,7 @@ func Rename(ctx context.Context, root *Dirent, oldParent *Dirent, oldName string // Check that replaced is not a mount point. if replaced.isMountPointLocked() { - replaced.DecRef() + replaced.DecRef(ctx) return syscall.EBUSY } @@ -1478,11 +1478,11 @@ func Rename(ctx context.Context, root *Dirent, oldParent *Dirent, oldName string oldIsDir := IsDir(renamed.Inode.StableAttr) newIsDir := IsDir(replaced.Inode.StableAttr) if !newIsDir && oldIsDir { - replaced.DecRef() + replaced.DecRef(ctx) return syscall.ENOTDIR } if !oldIsDir && newIsDir { - replaced.DecRef() + replaced.DecRef(ctx) return syscall.EISDIR } @@ -1493,13 +1493,13 @@ func Rename(ctx context.Context, root *Dirent, oldParent *Dirent, oldName string // open across renames is currently broken for multiple // reasons, so we flush all references on the replaced node and // its children. - replaced.Inode.Watches.Unpin(replaced) + replaced.Inode.Watches.Unpin(ctx, replaced) replaced.mu.Lock() - replaced.flush() + replaced.flush(ctx) replaced.mu.Unlock() // Done with replaced. - replaced.DecRef() + replaced.DecRef(ctx) } if err := renamed.Inode.Rename(ctx, oldParent, renamed, newParent, newName, replaced != nil); err != nil { @@ -1513,14 +1513,14 @@ func Rename(ctx context.Context, root *Dirent, oldParent *Dirent, oldName string // can't destroy oldParent (and try to retake its lock) because // Rename's caller must be holding a reference. newParent.IncRef() - oldParent.DecRef() + oldParent.DecRef(ctx) } if w, ok := newParent.children[newName]; ok { - w.Drop() + w.Drop(ctx) delete(newParent.children, newName) } if w, ok := oldParent.children[oldName]; ok { - w.Drop() + w.Drop(ctx) delete(oldParent.children, oldName) } @@ -1551,7 +1551,7 @@ func Rename(ctx context.Context, root *Dirent, oldParent *Dirent, oldName string // Same as replaced.flush above. renamed.mu.Lock() - renamed.flush() + renamed.flush(ctx) renamed.mu.Unlock() return nil diff --git a/pkg/sentry/fs/dirent_cache.go b/pkg/sentry/fs/dirent_cache.go index 33de32c69..7d9dd717e 100644 --- a/pkg/sentry/fs/dirent_cache.go +++ b/pkg/sentry/fs/dirent_cache.go @@ -17,6 +17,7 @@ package fs import ( "fmt" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sync" ) @@ -101,7 +102,7 @@ func (c *DirentCache) remove(d *Dirent) { panic(fmt.Sprintf("trying to remove %v, which is not in the dirent cache", d)) } c.list.Remove(d) - d.DecRef() + d.DecRef(context.Background()) c.currentSize-- if c.limit != nil { c.limit.dec() diff --git a/pkg/sentry/fs/dirent_refs_test.go b/pkg/sentry/fs/dirent_refs_test.go index 98d69c6f2..176b894ba 100644 --- a/pkg/sentry/fs/dirent_refs_test.go +++ b/pkg/sentry/fs/dirent_refs_test.go @@ -51,7 +51,7 @@ func TestWalkPositive(t *testing.T) { t.Fatalf("child name = %q has a ref count of %d, want %d", d.name, got, 1) } - d.DecRef() + d.DecRef(ctx) if got := root.ReadRefs(); got != 1 { t.Fatalf("root has a ref count of %d, want %d", got, 1) @@ -61,7 +61,7 @@ func TestWalkPositive(t *testing.T) { t.Fatalf("child name = %q has a ref count of %d, want %d", d.name, got, 0) } - root.flush() + root.flush(ctx) if got := len(root.children); got != 0 { t.Fatalf("root has %d children, want %d", got, 0) @@ -114,7 +114,7 @@ func TestWalkNegative(t *testing.T) { t.Fatalf("child has a ref count of %d, want %d", got, 2) } - child.DecRef() + child.DecRef(ctx) if got := child.(*Dirent).ReadRefs(); got != 1 { t.Fatalf("child has a ref count of %d, want %d", got, 1) @@ -124,7 +124,7 @@ func TestWalkNegative(t *testing.T) { t.Fatalf("root has %d children, want %d", got, 1) } - root.DecRef() + root.DecRef(ctx) if got := root.ReadRefs(); got != 0 { t.Fatalf("root has a ref count of %d, want %d", got, 0) @@ -351,9 +351,9 @@ func TestRemoveExtraRefs(t *testing.T) { t.Fatalf("dirent has a ref count of %d, want %d", got, 1) } - d.DecRef() + d.DecRef(ctx) - test.root.flush() + test.root.flush(ctx) if got := len(test.root.children); got != 0 { t.Errorf("root has %d children, want %d", got, 0) @@ -403,8 +403,8 @@ func TestRenameExtraRefs(t *testing.T) { t.Fatalf("Rename got error %v, want nil", err) } - oldParent.flush() - newParent.flush() + oldParent.flush(ctx) + newParent.flush(ctx) // Expect to have only active references. if got := renamed.ReadRefs(); got != 1 { diff --git a/pkg/sentry/fs/dirent_state.go b/pkg/sentry/fs/dirent_state.go index f623d6c0e..67a35f0b2 100644 --- a/pkg/sentry/fs/dirent_state.go +++ b/pkg/sentry/fs/dirent_state.go @@ -18,6 +18,7 @@ import ( "fmt" "sync/atomic" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" ) @@ -48,7 +49,7 @@ func (d *Dirent) saveChildren() map[string]*Dirent { for name, w := range d.children { if rc := w.Get(); rc != nil { // Drop the reference count obtain in w.Get() - rc.DecRef() + rc.DecRef(context.Background()) cd := rc.(*Dirent) if cd.IsNegative() { diff --git a/pkg/sentry/fs/fdpipe/pipe.go b/pkg/sentry/fs/fdpipe/pipe.go index 9fce177ad..b99199798 100644 --- a/pkg/sentry/fs/fdpipe/pipe.go +++ b/pkg/sentry/fs/fdpipe/pipe.go @@ -115,7 +115,7 @@ func (p *pipeOperations) Readiness(mask waiter.EventMask) (eventMask waiter.Even } // Release implements fs.FileOperations.Release. -func (p *pipeOperations) Release() { +func (p *pipeOperations) Release(context.Context) { fdnotifier.RemoveFD(int32(p.file.FD())) p.file.Close() p.file = nil diff --git a/pkg/sentry/fs/fdpipe/pipe_opener_test.go b/pkg/sentry/fs/fdpipe/pipe_opener_test.go index e556da48a..b9cec4b13 100644 --- a/pkg/sentry/fs/fdpipe/pipe_opener_test.go +++ b/pkg/sentry/fs/fdpipe/pipe_opener_test.go @@ -182,7 +182,7 @@ func TestTryOpen(t *testing.T) { // Cleanup the state of the pipe, and remove the fd from the // fdnotifier. Sadly this needed to maintain the correctness // of other tests because the fdnotifier is global. - pipeOps.Release() + pipeOps.Release(ctx) } continue } @@ -191,7 +191,7 @@ func TestTryOpen(t *testing.T) { } if pipeOps != nil { // Same as above. - pipeOps.Release() + pipeOps.Release(ctx) } } } @@ -279,7 +279,7 @@ func TestPipeOpenUnblocksEventually(t *testing.T) { pipeOps, err := Open(ctx, opener, flags) if pipeOps != nil { // Same as TestTryOpen. - pipeOps.Release() + pipeOps.Release(ctx) } // Check that the partner opened the file successfully. @@ -325,7 +325,7 @@ func TestCopiedReadAheadBuffer(t *testing.T) { ctx := contexttest.Context(t) pipeOps, err := pipeOpenState.TryOpen(ctx, opener, fs.FileFlags{Read: true}) if pipeOps != nil { - pipeOps.Release() + pipeOps.Release(ctx) t.Fatalf("open(%s, %o) got file, want nil", name, syscall.O_RDONLY) } if err != syserror.ErrWouldBlock { @@ -351,7 +351,7 @@ func TestCopiedReadAheadBuffer(t *testing.T) { if pipeOps == nil { t.Fatalf("open(%s, %o) got nil file, want not nil", name, syscall.O_RDONLY) } - defer pipeOps.Release() + defer pipeOps.Release(ctx) if err != nil { t.Fatalf("open(%s, %o) got error %v, want nil", name, syscall.O_RDONLY, err) @@ -471,14 +471,14 @@ func TestPipeHangup(t *testing.T) { f := <-fdchan if f < 0 { t.Errorf("%s: partner routine got fd %d, want > 0", test.desc, f) - pipeOps.Release() + pipeOps.Release(ctx) continue } if test.hangupSelf { // Hangup self and assert that our partner got the expected hangup // error. - pipeOps.Release() + pipeOps.Release(ctx) if test.flags.Read { // Partner is writer. @@ -490,7 +490,7 @@ func TestPipeHangup(t *testing.T) { } else { // Hangup our partner and expect us to get the hangup error. syscall.Close(f) - defer pipeOps.Release() + defer pipeOps.Release(ctx) if test.flags.Read { assertReaderHungup(t, test.desc, pipeOps.(*pipeOperations).file) diff --git a/pkg/sentry/fs/fdpipe/pipe_test.go b/pkg/sentry/fs/fdpipe/pipe_test.go index a0082ecca..1c9e82562 100644 --- a/pkg/sentry/fs/fdpipe/pipe_test.go +++ b/pkg/sentry/fs/fdpipe/pipe_test.go @@ -98,10 +98,11 @@ func TestNewPipe(t *testing.T) { } f := fd.New(gfd) - p, err := newPipeOperations(contexttest.Context(t), nil, test.flags, f, test.readAheadBuffer) + ctx := contexttest.Context(t) + p, err := newPipeOperations(ctx, nil, test.flags, f, test.readAheadBuffer) if p != nil { // This is necessary to remove the fd from the global fd notifier. - defer p.Release() + defer p.Release(ctx) } else { // If there is no p to DecRef on, because newPipeOperations failed, then the // file still needs to be closed. @@ -153,13 +154,14 @@ func TestPipeDestruction(t *testing.T) { syscall.Close(fds[1]) // Test the read end, but it doesn't really matter which. - p, err := newPipeOperations(contexttest.Context(t), nil, fs.FileFlags{Read: true}, f, nil) + ctx := contexttest.Context(t) + p, err := newPipeOperations(ctx, nil, fs.FileFlags{Read: true}, f, nil) if err != nil { f.Close() t.Fatalf("newPipeOperations got error %v, want nil", err) } // Drop our only reference, which should trigger the destructor. - p.Release() + p.Release(ctx) if fdnotifier.HasFD(int32(fds[0])) { t.Fatalf("after DecRef fdnotifier has fd %d, want no longer registered", fds[0]) @@ -282,7 +284,7 @@ func TestPipeRequest(t *testing.T) { if err != nil { t.Fatalf("%s: newPipeOperations got error %v, want nil", test.desc, err) } - defer p.Release() + defer p.Release(ctx) inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{Type: fs.Pipe}) file := fs.NewFile(ctx, fs.NewDirent(ctx, inode, "pipe"), fs.FileFlags{Read: true}, p) @@ -334,7 +336,7 @@ func TestPipeReadAheadBuffer(t *testing.T) { rfile.Close() t.Fatalf("newPipeOperations got error %v, want nil", err) } - defer p.Release() + defer p.Release(ctx) inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{ Type: fs.Pipe, @@ -380,7 +382,7 @@ func TestPipeReadsAccumulate(t *testing.T) { } // Don't forget to remove the fd from the fd notifier. Otherwise other tests will // likely be borked, because it's global :( - defer p.Release() + defer p.Release(ctx) inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{ Type: fs.Pipe, @@ -448,7 +450,7 @@ func TestPipeWritesAccumulate(t *testing.T) { } // Don't forget to remove the fd from the fd notifier. Otherwise other tests // will likely be borked, because it's global :( - defer p.Release() + defer p.Release(ctx) inode := fs.NewMockInode(ctx, fs.NewMockMountSource(nil), fs.StableAttr{ Type: fs.Pipe, diff --git a/pkg/sentry/fs/file.go b/pkg/sentry/fs/file.go index ca41520b4..72ea70fcf 100644 --- a/pkg/sentry/fs/file.go +++ b/pkg/sentry/fs/file.go @@ -142,17 +142,17 @@ func NewFile(ctx context.Context, dirent *Dirent, flags FileFlags, fops FileOper } // DecRef destroys the File when it is no longer referenced. -func (f *File) DecRef() { - f.DecRefWithDestructor(func() { +func (f *File) DecRef(ctx context.Context) { + f.DecRefWithDestructor(ctx, func(context.Context) { // Drop BSD style locks. lockRng := lock.LockRange{Start: 0, End: lock.LockEOF} f.Dirent.Inode.LockCtx.BSD.UnlockRegion(f, lockRng) // Release resources held by the FileOperations. - f.FileOperations.Release() + f.FileOperations.Release(ctx) // Release a reference on the Dirent. - f.Dirent.DecRef() + f.Dirent.DecRef(ctx) // Only unregister if we are currently registered. There is nothing // to register if f.async is nil (this happens when async mode is @@ -460,7 +460,7 @@ func (f *File) UnstableAttr(ctx context.Context) (UnstableAttr, error) { func (f *File) MappedName(ctx context.Context) string { root := RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } name, _ := f.Dirent.FullName(root) return name diff --git a/pkg/sentry/fs/file_operations.go b/pkg/sentry/fs/file_operations.go index f5537411e..305c0f840 100644 --- a/pkg/sentry/fs/file_operations.go +++ b/pkg/sentry/fs/file_operations.go @@ -67,7 +67,7 @@ type SpliceOpts struct { // - File.Flags(): This value may change during the operation. type FileOperations interface { // Release release resources held by FileOperations. - Release() + Release(ctx context.Context) // Waitable defines how this File can be waited on for read and // write readiness. diff --git a/pkg/sentry/fs/file_overlay.go b/pkg/sentry/fs/file_overlay.go index dcc1df38f..9dc58d5ff 100644 --- a/pkg/sentry/fs/file_overlay.go +++ b/pkg/sentry/fs/file_overlay.go @@ -54,7 +54,7 @@ func overlayFile(ctx context.Context, inode *Inode, flags FileFlags) (*File, err // Drop the extra reference on the Dirent. Now there's only one reference // on the dirent, either owned by f (if non-nil), or the Dirent is about // to be destroyed (if GetFile failed). - dirent.DecRef() + dirent.DecRef(ctx) return f, err } @@ -89,12 +89,12 @@ type overlayFileOperations struct { } // Release implements FileOperations.Release. -func (f *overlayFileOperations) Release() { +func (f *overlayFileOperations) Release(ctx context.Context) { if f.upper != nil { - f.upper.DecRef() + f.upper.DecRef(ctx) } if f.lower != nil { - f.lower.DecRef() + f.lower.DecRef(ctx) } } @@ -164,7 +164,7 @@ func (f *overlayFileOperations) Seek(ctx context.Context, file *File, whence See func (f *overlayFileOperations) Readdir(ctx context.Context, file *File, serializer DentrySerializer) (int64, error) { root := RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } dirCtx := &DirCtx{ @@ -497,7 +497,7 @@ func readdirOne(ctx context.Context, d *Dirent) (map[string]DentAttr, error) { if err != nil { return nil, err } - defer dir.DecRef() + defer dir.DecRef(ctx) // Use a stub serializer to read the entries into memory. stubSerializer := &CollectEntriesSerializer{} @@ -521,10 +521,10 @@ type overlayMappingIdentity struct { } // DecRef implements AtomicRefCount.DecRef. -func (omi *overlayMappingIdentity) DecRef() { - omi.AtomicRefCount.DecRefWithDestructor(func() { - omi.overlayFile.DecRef() - omi.id.DecRef() +func (omi *overlayMappingIdentity) DecRef(ctx context.Context) { + omi.AtomicRefCount.DecRefWithDestructor(ctx, func(context.Context) { + omi.overlayFile.DecRef(ctx) + omi.id.DecRef(ctx) }) } @@ -544,7 +544,7 @@ func (omi *overlayMappingIdentity) InodeID() uint64 { func (omi *overlayMappingIdentity) MappedName(ctx context.Context) string { root := RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } name, _ := omi.overlayFile.Dirent.FullName(root) return name diff --git a/pkg/sentry/fs/fsutil/file.go b/pkg/sentry/fs/fsutil/file.go index 08695391c..dc9efa5df 100644 --- a/pkg/sentry/fs/fsutil/file.go +++ b/pkg/sentry/fs/fsutil/file.go @@ -31,7 +31,7 @@ import ( type FileNoopRelease struct{} // Release is a no-op. -func (FileNoopRelease) Release() {} +func (FileNoopRelease) Release(context.Context) {} // SeekWithDirCursor is used to implement fs.FileOperations.Seek. If dirCursor // is not nil and the seek was on a directory, the cursor will be updated. @@ -296,7 +296,7 @@ func (sdfo *StaticDirFileOperations) IterateDir(ctx context.Context, d *fs.Diren func (sdfo *StaticDirFileOperations) Readdir(ctx context.Context, file *fs.File, serializer fs.DentrySerializer) (int64, error) { root := fs.RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } dirCtx := &fs.DirCtx{ Serializer: serializer, diff --git a/pkg/sentry/fs/gofer/file.go b/pkg/sentry/fs/gofer/file.go index b2fcab127..c0bc63a32 100644 --- a/pkg/sentry/fs/gofer/file.go +++ b/pkg/sentry/fs/gofer/file.go @@ -114,7 +114,7 @@ func NewFile(ctx context.Context, dirent *fs.Dirent, name string, flags fs.FileF } // Release implements fs.FileOpeations.Release. -func (f *fileOperations) Release() { +func (f *fileOperations) Release(context.Context) { f.handles.DecRef() } @@ -122,7 +122,7 @@ func (f *fileOperations) Release() { func (f *fileOperations) Readdir(ctx context.Context, file *fs.File, serializer fs.DentrySerializer) (int64, error) { root := fs.RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } dirCtx := &fs.DirCtx{ diff --git a/pkg/sentry/fs/gofer/gofer_test.go b/pkg/sentry/fs/gofer/gofer_test.go index 2df2fe889..326fed954 100644 --- a/pkg/sentry/fs/gofer/gofer_test.go +++ b/pkg/sentry/fs/gofer/gofer_test.go @@ -232,7 +232,7 @@ func TestRevalidation(t *testing.T) { // We must release the dirent, of the test will fail // with a reference leak. This is tracked by p9test. - defer dirent.DecRef() + defer dirent.DecRef(ctx) // Walk again. Depending on the cache policy, we may // get a new dirent. @@ -246,7 +246,7 @@ func TestRevalidation(t *testing.T) { if !test.preModificationWantReload && dirent != newDirent { t.Errorf("Lookup with cachePolicy=%s got new dirent %+v, wanted old dirent %+v", test.cachePolicy, newDirent, dirent) } - newDirent.DecRef() // See above. + newDirent.DecRef(ctx) // See above. // Modify the underlying mocked file's modification // time for the next walk that occurs. @@ -287,7 +287,7 @@ func TestRevalidation(t *testing.T) { if test.postModificationWantUpdatedAttrs && gotModTimeSeconds != nowSeconds { t.Fatalf("Lookup with cachePolicy=%s got new modification time %v, wanted %v", test.cachePolicy, gotModTimeSeconds, nowSeconds) } - newDirent.DecRef() // See above. + newDirent.DecRef(ctx) // See above. // Remove the file from the remote fs, subsequent walks // should now fail to find anything. @@ -303,7 +303,7 @@ func TestRevalidation(t *testing.T) { t.Errorf("Lookup with cachePolicy=%s got new dirent and error %v, wanted old dirent and nil error", test.cachePolicy, err) } if err == nil { - newDirent.DecRef() // See above. + newDirent.DecRef(ctx) // See above. } }) } diff --git a/pkg/sentry/fs/gofer/handles.go b/pkg/sentry/fs/gofer/handles.go index fc14249be..f324dbf26 100644 --- a/pkg/sentry/fs/gofer/handles.go +++ b/pkg/sentry/fs/gofer/handles.go @@ -47,7 +47,8 @@ type handles struct { // DecRef drops a reference on handles. func (h *handles) DecRef() { - h.DecRefWithDestructor(func() { + ctx := context.Background() + h.DecRefWithDestructor(ctx, func(context.Context) { if h.Host != nil { if h.isHostBorrowed { h.Host.Release() @@ -57,7 +58,7 @@ func (h *handles) DecRef() { } } } - if err := h.File.close(context.Background()); err != nil { + if err := h.File.close(ctx); err != nil { log.Warningf("error closing p9 file: %v", err) } }) diff --git a/pkg/sentry/fs/gofer/inode.go b/pkg/sentry/fs/gofer/inode.go index 51d7368a1..3a225fd39 100644 --- a/pkg/sentry/fs/gofer/inode.go +++ b/pkg/sentry/fs/gofer/inode.go @@ -441,8 +441,9 @@ func (i *inodeOperations) Release(ctx context.Context) { // asynchronously. // // We use AsyncWithContext to avoid needing to allocate an extra - // anonymous function on the heap. - fs.AsyncWithContext(ctx, i.fileState.Release) + // anonymous function on the heap. We must use background context + // because the async work cannot happen on the task context. + fs.AsyncWithContext(context.Background(), i.fileState.Release) } // Mappable implements fs.InodeOperations.Mappable. diff --git a/pkg/sentry/fs/gofer/path.go b/pkg/sentry/fs/gofer/path.go index cf9800100..3c66dc3c2 100644 --- a/pkg/sentry/fs/gofer/path.go +++ b/pkg/sentry/fs/gofer/path.go @@ -168,7 +168,7 @@ func (i *inodeOperations) Create(ctx context.Context, dir *fs.Inode, name string // Construct the positive Dirent. d := fs.NewDirent(ctx, fs.NewInode(ctx, iops, dir.MountSource, sattr), name) - defer d.DecRef() + defer d.DecRef(ctx) // Construct the new file, caching the handles if allowed. h := handles{ @@ -371,7 +371,7 @@ func (i *inodeOperations) Remove(ctx context.Context, dir *fs.Inode, name string // Find out if file being deleted is a socket or pipe that needs to be // removed from endpoint map. if d, err := i.Lookup(ctx, dir, name); err == nil { - defer d.DecRef() + defer d.DecRef(ctx) if fs.IsSocket(d.Inode.StableAttr) || fs.IsPipe(d.Inode.StableAttr) { switch iops := d.Inode.InodeOperations.(type) { @@ -392,7 +392,7 @@ func (i *inodeOperations) Remove(ctx context.Context, dir *fs.Inode, name string return err } if key != nil { - i.session().overrides.remove(*key) + i.session().overrides.remove(ctx, *key) } i.touchModificationAndStatusChangeTime(ctx, dir) diff --git a/pkg/sentry/fs/gofer/session.go b/pkg/sentry/fs/gofer/session.go index b5efc86f2..7cf3522ff 100644 --- a/pkg/sentry/fs/gofer/session.go +++ b/pkg/sentry/fs/gofer/session.go @@ -89,10 +89,10 @@ func (e *overrideMaps) addPipe(key device.MultiDeviceKey, d *fs.Dirent, inode *f // remove deletes the key from the maps. // // Precondition: maps must have been locked with 'lock'. -func (e *overrideMaps) remove(key device.MultiDeviceKey) { +func (e *overrideMaps) remove(ctx context.Context, key device.MultiDeviceKey) { endpoint := e.keyMap[key] delete(e.keyMap, key) - endpoint.dirent.DecRef() + endpoint.dirent.DecRef(ctx) } // lock blocks other addition and removal operations from happening while @@ -197,7 +197,7 @@ type session struct { } // Destroy tears down the session. -func (s *session) Destroy() { +func (s *session) Destroy(ctx context.Context) { s.client.Close() } @@ -329,7 +329,7 @@ func Root(ctx context.Context, dev string, filesystem fs.Filesystem, superBlockF s.client, err = p9.NewClient(conn, s.msize, s.version) if err != nil { // Drop our reference on the session, it needs to be torn down. - s.DecRef() + s.DecRef(ctx) return nil, err } @@ -340,7 +340,7 @@ func Root(ctx context.Context, dev string, filesystem fs.Filesystem, superBlockF ctx.UninterruptibleSleepFinish(false) if err != nil { // Same as above. - s.DecRef() + s.DecRef(ctx) return nil, err } @@ -348,7 +348,7 @@ func Root(ctx context.Context, dev string, filesystem fs.Filesystem, superBlockF if err != nil { s.attach.close(ctx) // Same as above, but after we execute the Close request. - s.DecRef() + s.DecRef(ctx) return nil, err } @@ -393,13 +393,13 @@ func (s *session) fillKeyMap(ctx context.Context) error { // fillPathMap populates paths for overrides from dirents in direntMap // before save. -func (s *session) fillPathMap() error { +func (s *session) fillPathMap(ctx context.Context) error { unlock := s.overrides.lock() defer unlock() for _, endpoint := range s.overrides.keyMap { mountRoot := endpoint.dirent.MountRoot() - defer mountRoot.DecRef() + defer mountRoot.DecRef(ctx) dirPath, _ := endpoint.dirent.FullName(mountRoot) if dirPath == "" { return fmt.Errorf("error getting path from dirent") diff --git a/pkg/sentry/fs/gofer/session_state.go b/pkg/sentry/fs/gofer/session_state.go index 2d398b753..48b423dd8 100644 --- a/pkg/sentry/fs/gofer/session_state.go +++ b/pkg/sentry/fs/gofer/session_state.go @@ -26,7 +26,8 @@ import ( // beforeSave is invoked by stateify. func (s *session) beforeSave() { if s.overrides != nil { - if err := s.fillPathMap(); err != nil { + ctx := &dummyClockContext{context.Background()} + if err := s.fillPathMap(ctx); err != nil { panic("failed to save paths to override map before saving" + err.Error()) } } diff --git a/pkg/sentry/fs/gofer/socket.go b/pkg/sentry/fs/gofer/socket.go index 40f2c1cad..8a1c69ac2 100644 --- a/pkg/sentry/fs/gofer/socket.go +++ b/pkg/sentry/fs/gofer/socket.go @@ -134,14 +134,14 @@ func (e *endpoint) UnidirectionalConnect(ctx context.Context) (transport.Connect // We don't need the receiver. c.CloseRecv() - c.Release() + c.Release(ctx) return c, nil } // Release implements transport.BoundEndpoint.Release. -func (e *endpoint) Release() { - e.inode.DecRef() +func (e *endpoint) Release(ctx context.Context) { + e.inode.DecRef(ctx) } // Passcred implements transport.BoundEndpoint.Passcred. diff --git a/pkg/sentry/fs/host/control.go b/pkg/sentry/fs/host/control.go index 39299b7e4..0d8d36afa 100644 --- a/pkg/sentry/fs/host/control.go +++ b/pkg/sentry/fs/host/control.go @@ -57,7 +57,7 @@ func (c *scmRights) Clone() transport.RightsControlMessage { } // Release implements transport.RightsControlMessage.Release. -func (c *scmRights) Release() { +func (c *scmRights) Release(ctx context.Context) { for _, fd := range c.fds { syscall.Close(fd) } diff --git a/pkg/sentry/fs/host/file.go b/pkg/sentry/fs/host/file.go index 3e48b8b2c..86d1a87f0 100644 --- a/pkg/sentry/fs/host/file.go +++ b/pkg/sentry/fs/host/file.go @@ -110,7 +110,7 @@ func newFileFromDonatedFD(ctx context.Context, donated int, saveable, isTTY bool name := fmt.Sprintf("host:[%d]", inode.StableAttr.InodeID) dirent := fs.NewDirent(ctx, inode, name) - defer dirent.DecRef() + defer dirent.DecRef(ctx) if isTTY { return newTTYFile(ctx, dirent, flags, iops), nil @@ -169,7 +169,7 @@ func (f *fileOperations) Readiness(mask waiter.EventMask) waiter.EventMask { func (f *fileOperations) Readdir(ctx context.Context, file *fs.File, serializer fs.DentrySerializer) (int64, error) { root := fs.RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } dirCtx := &fs.DirCtx{ Serializer: serializer, diff --git a/pkg/sentry/fs/host/inode_test.go b/pkg/sentry/fs/host/inode_test.go index c507f57eb..41a23b5da 100644 --- a/pkg/sentry/fs/host/inode_test.go +++ b/pkg/sentry/fs/host/inode_test.go @@ -36,7 +36,7 @@ func TestCloseFD(t *testing.T) { if err != nil { t.Fatalf("Failed to create File: %v", err) } - file.DecRef() + file.DecRef(ctx) s := make([]byte, 10) if c, err := syscall.Read(p[0], s); c != 0 || err != nil { diff --git a/pkg/sentry/fs/host/socket.go b/pkg/sentry/fs/host/socket.go index cfb089e43..a2f3d5918 100644 --- a/pkg/sentry/fs/host/socket.go +++ b/pkg/sentry/fs/host/socket.go @@ -194,7 +194,7 @@ func newSocket(ctx context.Context, orgfd int, saveable bool) (*fs.File, error) } // Send implements transport.ConnectedEndpoint.Send. -func (c *ConnectedEndpoint) Send(data [][]byte, controlMessages transport.ControlMessages, from tcpip.FullAddress) (int64, bool, *syserr.Error) { +func (c *ConnectedEndpoint) Send(ctx context.Context, data [][]byte, controlMessages transport.ControlMessages, from tcpip.FullAddress) (int64, bool, *syserr.Error) { c.mu.RLock() defer c.mu.RUnlock() @@ -271,7 +271,7 @@ func (c *ConnectedEndpoint) EventUpdate() { } // Recv implements transport.Receiver.Recv. -func (c *ConnectedEndpoint) Recv(data [][]byte, creds bool, numRights int, peek bool) (int64, int64, transport.ControlMessages, bool, tcpip.FullAddress, bool, *syserr.Error) { +func (c *ConnectedEndpoint) Recv(ctx context.Context, data [][]byte, creds bool, numRights int, peek bool) (int64, int64, transport.ControlMessages, bool, tcpip.FullAddress, bool, *syserr.Error) { c.mu.RLock() defer c.mu.RUnlock() @@ -318,7 +318,7 @@ func (c *ConnectedEndpoint) Recv(data [][]byte, creds bool, numRights int, peek } // close releases all resources related to the endpoint. -func (c *ConnectedEndpoint) close() { +func (c *ConnectedEndpoint) close(context.Context) { fdnotifier.RemoveFD(int32(c.file.FD())) c.file.Close() c.file = nil @@ -374,8 +374,8 @@ func (c *ConnectedEndpoint) RecvMaxQueueSize() int64 { } // Release implements transport.ConnectedEndpoint.Release and transport.Receiver.Release. -func (c *ConnectedEndpoint) Release() { - c.ref.DecRefWithDestructor(c.close) +func (c *ConnectedEndpoint) Release(ctx context.Context) { + c.ref.DecRefWithDestructor(ctx, c.close) } // CloseUnread implements transport.ConnectedEndpoint.CloseUnread. diff --git a/pkg/sentry/fs/host/socket_test.go b/pkg/sentry/fs/host/socket_test.go index affdbcacb..9d58ea448 100644 --- a/pkg/sentry/fs/host/socket_test.go +++ b/pkg/sentry/fs/host/socket_test.go @@ -67,11 +67,12 @@ func TestSocketIsBlocking(t *testing.T) { if fl&syscall.O_NONBLOCK == syscall.O_NONBLOCK { t.Fatalf("Expected socket %v to be blocking", pair[1]) } - sock, err := newSocket(contexttest.Context(t), pair[0], false) + ctx := contexttest.Context(t) + sock, err := newSocket(ctx, pair[0], false) if err != nil { t.Fatalf("newSocket(%v) failed => %v", pair[0], err) } - defer sock.DecRef() + defer sock.DecRef(ctx) // Test that the socket now is non-blocking. if fl, err = getFl(pair[0]); err != nil { t.Fatalf("getFl: fcntl(%v, GETFL) => %v", pair[0], err) @@ -93,11 +94,12 @@ func TestSocketWritev(t *testing.T) { if err != nil { t.Fatalf("host socket creation failed: %v", err) } - socket, err := newSocket(contexttest.Context(t), pair[0], false) + ctx := contexttest.Context(t) + socket, err := newSocket(ctx, pair[0], false) if err != nil { t.Fatalf("newSocket(%v) => %v", pair[0], err) } - defer socket.DecRef() + defer socket.DecRef(ctx) buf := []byte("hello world\n") n, err := socket.Writev(contexttest.Context(t), usermem.BytesIOSequence(buf)) if err != nil { @@ -115,11 +117,12 @@ func TestSocketWritevLen0(t *testing.T) { if err != nil { t.Fatalf("host socket creation failed: %v", err) } - socket, err := newSocket(contexttest.Context(t), pair[0], false) + ctx := contexttest.Context(t) + socket, err := newSocket(ctx, pair[0], false) if err != nil { t.Fatalf("newSocket(%v) => %v", pair[0], err) } - defer socket.DecRef() + defer socket.DecRef(ctx) n, err := socket.Writev(contexttest.Context(t), usermem.BytesIOSequence(nil)) if err != nil { t.Fatalf("socket writev failed: %v", err) @@ -136,11 +139,12 @@ func TestSocketSendMsgLen0(t *testing.T) { if err != nil { t.Fatalf("host socket creation failed: %v", err) } - sfile, err := newSocket(contexttest.Context(t), pair[0], false) + ctx := contexttest.Context(t) + sfile, err := newSocket(ctx, pair[0], false) if err != nil { t.Fatalf("newSocket(%v) => %v", pair[0], err) } - defer sfile.DecRef() + defer sfile.DecRef(ctx) s := sfile.FileOperations.(socket.Socket) n, terr := s.SendMsg(nil, usermem.BytesIOSequence(nil), []byte{}, 0, false, ktime.Time{}, socket.ControlMessages{}) @@ -158,18 +162,19 @@ func TestListen(t *testing.T) { if err != nil { t.Fatalf("syscall.Socket(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) => %v", err) } - sfile1, err := newSocket(contexttest.Context(t), pair[0], false) + ctx := contexttest.Context(t) + sfile1, err := newSocket(ctx, pair[0], false) if err != nil { t.Fatalf("newSocket(%v) => %v", pair[0], err) } - defer sfile1.DecRef() + defer sfile1.DecRef(ctx) socket1 := sfile1.FileOperations.(socket.Socket) - sfile2, err := newSocket(contexttest.Context(t), pair[1], false) + sfile2, err := newSocket(ctx, pair[1], false) if err != nil { t.Fatalf("newSocket(%v) => %v", pair[1], err) } - defer sfile2.DecRef() + defer sfile2.DecRef(ctx) socket2 := sfile2.FileOperations.(socket.Socket) // Socketpairs can not be listened to. @@ -185,11 +190,11 @@ func TestListen(t *testing.T) { if err != nil { t.Fatalf("syscall.Socket(syscall.AF_UNIX, syscall.SOCK_STREAM, 0) => %v", err) } - sfile3, err := newSocket(contexttest.Context(t), sock, false) + sfile3, err := newSocket(ctx, sock, false) if err != nil { t.Fatalf("newSocket(%v) => %v", sock, err) } - defer sfile3.DecRef() + defer sfile3.DecRef(ctx) socket3 := sfile3.FileOperations.(socket.Socket) // This socket is not bound so we can't listen on it. @@ -237,9 +242,10 @@ func TestRelease(t *testing.T) { } c := &ConnectedEndpoint{queue: &waiter.Queue{}, file: fd.New(f)} want := &ConnectedEndpoint{queue: c.queue} - want.ref.DecRef() + ctx := contexttest.Context(t) + want.ref.DecRef(ctx) fdnotifier.AddFD(int32(c.file.FD()), nil) - c.Release() + c.Release(ctx) if !reflect.DeepEqual(c, want) { t.Errorf("got = %#v, want = %#v", c, want) } diff --git a/pkg/sentry/fs/host/tty.go b/pkg/sentry/fs/host/tty.go index 82a02fcb2..b5229098c 100644 --- a/pkg/sentry/fs/host/tty.go +++ b/pkg/sentry/fs/host/tty.go @@ -113,12 +113,12 @@ func (t *TTYFileOperations) Write(ctx context.Context, file *fs.File, src userme } // Release implements fs.FileOperations.Release. -func (t *TTYFileOperations) Release() { +func (t *TTYFileOperations) Release(ctx context.Context) { t.mu.Lock() t.fgProcessGroup = nil t.mu.Unlock() - t.fileOperations.Release() + t.fileOperations.Release(ctx) } // Ioctl implements fs.FileOperations.Ioctl. diff --git a/pkg/sentry/fs/host/wait_test.go b/pkg/sentry/fs/host/wait_test.go index ce397a5e3..c143f4ce2 100644 --- a/pkg/sentry/fs/host/wait_test.go +++ b/pkg/sentry/fs/host/wait_test.go @@ -39,7 +39,7 @@ func TestWait(t *testing.T) { t.Fatalf("NewFile failed: %v", err) } - defer file.DecRef() + defer file.DecRef(ctx) r := file.Readiness(waiter.EventIn) if r != 0 { diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go index a34fbc946..b79cd9877 100644 --- a/pkg/sentry/fs/inode.go +++ b/pkg/sentry/fs/inode.go @@ -96,13 +96,12 @@ func NewInode(ctx context.Context, iops InodeOperations, msrc *MountSource, satt } // DecRef drops a reference on the Inode. -func (i *Inode) DecRef() { - i.DecRefWithDestructor(i.destroy) +func (i *Inode) DecRef(ctx context.Context) { + i.DecRefWithDestructor(ctx, i.destroy) } // destroy releases the Inode and releases the msrc reference taken. -func (i *Inode) destroy() { - ctx := context.Background() +func (i *Inode) destroy(ctx context.Context) { if err := i.WriteOut(ctx); err != nil { // FIXME(b/65209558): Mark as warning again once noatime is // properly supported. @@ -122,12 +121,12 @@ func (i *Inode) destroy() { i.Watches.targetDestroyed() if i.overlay != nil { - i.overlay.release() + i.overlay.release(ctx) } else { i.InodeOperations.Release(ctx) } - i.MountSource.DecRef() + i.MountSource.DecRef(ctx) } // Mappable calls i.InodeOperations.Mappable. diff --git a/pkg/sentry/fs/inode_inotify.go b/pkg/sentry/fs/inode_inotify.go index efd3c962b..9911a00c2 100644 --- a/pkg/sentry/fs/inode_inotify.go +++ b/pkg/sentry/fs/inode_inotify.go @@ -17,6 +17,7 @@ package fs import ( "fmt" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sync" ) @@ -136,11 +137,11 @@ func (w *Watches) Notify(name string, events, cookie uint32) { } // Unpin unpins dirent from all watches in this set. -func (w *Watches) Unpin(d *Dirent) { +func (w *Watches) Unpin(ctx context.Context, d *Dirent) { w.mu.RLock() defer w.mu.RUnlock() for _, watch := range w.ws { - watch.Unpin(d) + watch.Unpin(ctx, d) } } diff --git a/pkg/sentry/fs/inode_overlay.go b/pkg/sentry/fs/inode_overlay.go index 537c8d257..dc2e353d9 100644 --- a/pkg/sentry/fs/inode_overlay.go +++ b/pkg/sentry/fs/inode_overlay.go @@ -85,7 +85,7 @@ func overlayLookup(ctx context.Context, parent *overlayEntry, inode *Inode, name upperInode = child.Inode upperInode.IncRef() } - child.DecRef() + child.DecRef(ctx) } // Are we done? @@ -108,7 +108,7 @@ func overlayLookup(ctx context.Context, parent *overlayEntry, inode *Inode, name entry, err := newOverlayEntry(ctx, upperInode, nil, false) if err != nil { // Don't leak resources. - upperInode.DecRef() + upperInode.DecRef(ctx) parent.copyMu.RUnlock() return nil, false, err } @@ -129,7 +129,7 @@ func overlayLookup(ctx context.Context, parent *overlayEntry, inode *Inode, name if err != nil && err != syserror.ENOENT { // Don't leak resources. if upperInode != nil { - upperInode.DecRef() + upperInode.DecRef(ctx) } parent.copyMu.RUnlock() return nil, false, err @@ -152,7 +152,7 @@ func overlayLookup(ctx context.Context, parent *overlayEntry, inode *Inode, name } } } - child.DecRef() + child.DecRef(ctx) } } @@ -183,7 +183,7 @@ func overlayLookup(ctx context.Context, parent *overlayEntry, inode *Inode, name // unnecessary because we don't need to copy-up and we will always // operate (e.g. read/write) on the upper Inode. if !IsDir(upperInode.StableAttr) { - lowerInode.DecRef() + lowerInode.DecRef(ctx) lowerInode = nil } } @@ -194,10 +194,10 @@ func overlayLookup(ctx context.Context, parent *overlayEntry, inode *Inode, name // Well, not quite, we failed at the last moment, how depressing. // Be sure not to leak resources. if upperInode != nil { - upperInode.DecRef() + upperInode.DecRef(ctx) } if lowerInode != nil { - lowerInode.DecRef() + lowerInode.DecRef(ctx) } parent.copyMu.RUnlock() return nil, false, err @@ -248,7 +248,7 @@ func overlayCreate(ctx context.Context, o *overlayEntry, parent *Dirent, name st // user) will clobber the real path for the underlying Inode. upperFile.Dirent.Inode.IncRef() upperDirent := NewTransientDirent(upperFile.Dirent.Inode) - upperFile.Dirent.DecRef() + upperFile.Dirent.DecRef(ctx) upperFile.Dirent = upperDirent // Create the overlay inode and dirent. We need this to construct the @@ -259,7 +259,7 @@ func overlayCreate(ctx context.Context, o *overlayEntry, parent *Dirent, name st // The overlay file created below with NewFile will take a reference on // the overlayDirent, and it should be the only thing holding a // reference at the time of creation, so we must drop this reference. - defer overlayDirent.DecRef() + defer overlayDirent.DecRef(ctx) // Create a new overlay file that wraps the upper file. flags.Pread = upperFile.Flags().Pread @@ -399,7 +399,7 @@ func overlayRename(ctx context.Context, o *overlayEntry, oldParent *Dirent, rena if !replaced.IsNegative() && IsDir(replaced.Inode.StableAttr) { children, err := readdirOne(ctx, replaced) if err != nil { - replaced.DecRef() + replaced.DecRef(ctx) return err } @@ -407,12 +407,12 @@ func overlayRename(ctx context.Context, o *overlayEntry, oldParent *Dirent, rena // included among the returned children, so we don't // need to bother checking for them. if len(children) > 0 { - replaced.DecRef() + replaced.DecRef(ctx) return syserror.ENOTEMPTY } } - replaced.DecRef() + replaced.DecRef(ctx) } } @@ -455,12 +455,12 @@ func overlayBind(ctx context.Context, o *overlayEntry, parent *Dirent, name stri // Grab the inode and drop the dirent, we don't need it. inode := d.Inode inode.IncRef() - d.DecRef() + d.DecRef(ctx) // Create a new overlay entry and dirent for the socket. entry, err := newOverlayEntry(ctx, inode, nil, false) if err != nil { - inode.DecRef() + inode.DecRef(ctx) return nil, err } // Use the parent's MountSource, since that corresponds to the overlay, @@ -672,7 +672,7 @@ func overlayGetlink(ctx context.Context, o *overlayEntry) (*Dirent, error) { // ground and claim that jumping around the filesystem like this // is not supported. name, _ := dirent.FullName(nil) - dirent.DecRef() + dirent.DecRef(ctx) // Claim that the path is not accessible. err = syserror.EACCES diff --git a/pkg/sentry/fs/inode_overlay_test.go b/pkg/sentry/fs/inode_overlay_test.go index 389c219d6..aa9851b26 100644 --- a/pkg/sentry/fs/inode_overlay_test.go +++ b/pkg/sentry/fs/inode_overlay_test.go @@ -316,7 +316,7 @@ func TestCacheFlush(t *testing.T) { t.Fatalf("NewMountNamespace failed: %v", err) } root := mns.Root() - defer root.DecRef() + defer root.DecRef(ctx) ctx = &rootContext{ Context: ctx, @@ -345,7 +345,7 @@ func TestCacheFlush(t *testing.T) { } // Drop the file reference. - file.DecRef() + file.DecRef(ctx) // Dirent should have 2 refs left. if got, want := dirent.ReadRefs(), 2; int(got) != want { @@ -361,7 +361,7 @@ func TestCacheFlush(t *testing.T) { } // Drop our ref. - dirent.DecRef() + dirent.DecRef(ctx) // We should be back to zero refs. if got, want := dirent.ReadRefs(), 0; int(got) != want { @@ -398,7 +398,7 @@ func (d *dir) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags if err != nil { return nil, err } - defer file.DecRef() + defer file.DecRef(ctx) // Wrap the file's FileOperations in a dirFile. fops := &dirFile{ FileOperations: file.FileOperations, diff --git a/pkg/sentry/fs/inotify.go b/pkg/sentry/fs/inotify.go index e3a715c1f..c5c07d564 100644 --- a/pkg/sentry/fs/inotify.go +++ b/pkg/sentry/fs/inotify.go @@ -80,7 +80,7 @@ func NewInotify(ctx context.Context) *Inotify { // Release implements FileOperations.Release. Release removes all watches and // frees all resources for an inotify instance. -func (i *Inotify) Release() { +func (i *Inotify) Release(ctx context.Context) { // We need to hold i.mu to avoid a race with concurrent calls to // Inotify.targetDestroyed from Watches. There's no risk of Watches // accessing this Inotify after the destructor ends, because we remove all @@ -93,7 +93,7 @@ func (i *Inotify) Release() { // the owner's destructor. w.target.Watches.Remove(w.ID()) // Don't leak any references to the target, held by pins in the watch. - w.destroy() + w.destroy(ctx) } } @@ -321,7 +321,7 @@ func (i *Inotify) AddWatch(target *Dirent, mask uint32) int32 { // // RmWatch looks up an inotify watch for the given 'wd' and configures the // target dirent to stop sending events to this inotify instance. -func (i *Inotify) RmWatch(wd int32) error { +func (i *Inotify) RmWatch(ctx context.Context, wd int32) error { i.mu.Lock() // Find the watch we were asked to removed. @@ -346,7 +346,7 @@ func (i *Inotify) RmWatch(wd int32) error { i.queueEvent(newEvent(watch.wd, "", linux.IN_IGNORED, 0)) // Remove all pins. - watch.destroy() + watch.destroy(ctx) return nil } diff --git a/pkg/sentry/fs/inotify_watch.go b/pkg/sentry/fs/inotify_watch.go index 900cba3ca..605423d22 100644 --- a/pkg/sentry/fs/inotify_watch.go +++ b/pkg/sentry/fs/inotify_watch.go @@ -18,6 +18,7 @@ import ( "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sync" ) @@ -105,12 +106,12 @@ func (w *Watch) Pin(d *Dirent) { // Unpin drops any extra refs held on dirent due to a previous Pin // call. Calling Unpin multiple times for the same dirent, or on a dirent // without a corresponding Pin call is a no-op. -func (w *Watch) Unpin(d *Dirent) { +func (w *Watch) Unpin(ctx context.Context, d *Dirent) { w.mu.Lock() defer w.mu.Unlock() if w.pins[d] { delete(w.pins, d) - d.DecRef() + d.DecRef(ctx) } } @@ -125,11 +126,11 @@ func (w *Watch) TargetDestroyed() { // this watch. Destroy does not cause any new events to be generated. The caller // is responsible for ensuring there are no outstanding references to this // watch. -func (w *Watch) destroy() { +func (w *Watch) destroy(ctx context.Context) { w.mu.Lock() defer w.mu.Unlock() for d := range w.pins { - d.DecRef() + d.DecRef(ctx) } w.pins = nil } diff --git a/pkg/sentry/fs/mount.go b/pkg/sentry/fs/mount.go index 37bae6810..ee69b10e8 100644 --- a/pkg/sentry/fs/mount.go +++ b/pkg/sentry/fs/mount.go @@ -51,7 +51,7 @@ type MountSourceOperations interface { DirentOperations // Destroy destroys the MountSource. - Destroy() + Destroy(ctx context.Context) // Below are MountSourceOperations that do not conform to Linux. @@ -165,16 +165,16 @@ func (msrc *MountSource) DecDirentRefs() { } } -func (msrc *MountSource) destroy() { +func (msrc *MountSource) destroy(ctx context.Context) { if c := msrc.DirentRefs(); c != 0 { panic(fmt.Sprintf("MountSource with non-zero direntRefs is being destroyed: %d", c)) } - msrc.MountSourceOperations.Destroy() + msrc.MountSourceOperations.Destroy(ctx) } // DecRef drops a reference on the MountSource. -func (msrc *MountSource) DecRef() { - msrc.DecRefWithDestructor(msrc.destroy) +func (msrc *MountSource) DecRef(ctx context.Context) { + msrc.DecRefWithDestructor(ctx, msrc.destroy) } // FlushDirentRefs drops all references held by the MountSource on Dirents. @@ -264,7 +264,7 @@ func (*SimpleMountSourceOperations) ResetInodeMappings() {} func (*SimpleMountSourceOperations) SaveInodeMapping(*Inode, string) {} // Destroy implements MountSourceOperations.Destroy. -func (*SimpleMountSourceOperations) Destroy() {} +func (*SimpleMountSourceOperations) Destroy(context.Context) {} // Info defines attributes of a filesystem. type Info struct { diff --git a/pkg/sentry/fs/mount_overlay.go b/pkg/sentry/fs/mount_overlay.go index 78e35b1e6..7badc75d6 100644 --- a/pkg/sentry/fs/mount_overlay.go +++ b/pkg/sentry/fs/mount_overlay.go @@ -115,9 +115,9 @@ func (o *overlayMountSourceOperations) SaveInodeMapping(inode *Inode, path strin } // Destroy drops references on the upper and lower MountSource. -func (o *overlayMountSourceOperations) Destroy() { - o.upper.DecRef() - o.lower.DecRef() +func (o *overlayMountSourceOperations) Destroy(ctx context.Context) { + o.upper.DecRef(ctx) + o.lower.DecRef(ctx) } // type overlayFilesystem is the filesystem for overlay mounts. diff --git a/pkg/sentry/fs/mount_test.go b/pkg/sentry/fs/mount_test.go index a3d10770b..6c296f5d0 100644 --- a/pkg/sentry/fs/mount_test.go +++ b/pkg/sentry/fs/mount_test.go @@ -18,6 +18,7 @@ import ( "fmt" "testing" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/contexttest" ) @@ -32,13 +33,13 @@ func cacheReallyContains(cache *DirentCache, d *Dirent) bool { return false } -func mountPathsAre(root *Dirent, got []*Mount, want ...string) error { +func mountPathsAre(ctx context.Context, root *Dirent, got []*Mount, want ...string) error { gotPaths := make(map[string]struct{}, len(got)) gotStr := make([]string, len(got)) for i, g := range got { if groot := g.Root(); groot != nil { name, _ := groot.FullName(root) - groot.DecRef() + groot.DecRef(ctx) gotStr[i] = name gotPaths[name] = struct{}{} } @@ -69,7 +70,7 @@ func TestMountSourceOnlyCachedOnce(t *testing.T) { t.Fatalf("NewMountNamespace failed: %v", err) } rootDirent := mm.Root() - defer rootDirent.DecRef() + defer rootDirent.DecRef(ctx) // Get a child of the root which we will mount over. Note that the // MockInodeOperations causes Walk to always succeed. @@ -125,7 +126,7 @@ func TestAllMountsUnder(t *testing.T) { t.Fatalf("NewMountNamespace failed: %v", err) } rootDirent := mm.Root() - defer rootDirent.DecRef() + defer rootDirent.DecRef(ctx) // Add mounts at the following paths: paths := []string{ @@ -150,14 +151,14 @@ func TestAllMountsUnder(t *testing.T) { if err := mm.Mount(ctx, d, submountInode); err != nil { t.Fatalf("could not mount at %q: %v", p, err) } - d.DecRef() + d.DecRef(ctx) } // mm root should contain all submounts (and does not include the root mount). rootMnt := mm.FindMount(rootDirent) submounts := mm.AllMountsUnder(rootMnt) allPaths := append(paths, "/") - if err := mountPathsAre(rootDirent, submounts, allPaths...); err != nil { + if err := mountPathsAre(ctx, rootDirent, submounts, allPaths...); err != nil { t.Error(err) } @@ -181,9 +182,9 @@ func TestAllMountsUnder(t *testing.T) { if err != nil { t.Fatalf("could not find path %q in mount manager: %v", "/foo", err) } - defer d.DecRef() + defer d.DecRef(ctx) submounts = mm.AllMountsUnder(mm.FindMount(d)) - if err := mountPathsAre(rootDirent, submounts, "/foo", "/foo/bar", "/foo/qux", "/foo/bar/baz"); err != nil { + if err := mountPathsAre(ctx, rootDirent, submounts, "/foo", "/foo/bar", "/foo/qux", "/foo/bar/baz"); err != nil { t.Error(err) } @@ -193,9 +194,9 @@ func TestAllMountsUnder(t *testing.T) { if err != nil { t.Fatalf("could not find path %q in mount manager: %v", "/waldo", err) } - defer waldo.DecRef() + defer waldo.DecRef(ctx) submounts = mm.AllMountsUnder(mm.FindMount(waldo)) - if err := mountPathsAre(rootDirent, submounts, "/waldo"); err != nil { + if err := mountPathsAre(ctx, rootDirent, submounts, "/waldo"); err != nil { t.Error(err) } } @@ -212,7 +213,7 @@ func TestUnmount(t *testing.T) { t.Fatalf("NewMountNamespace failed: %v", err) } rootDirent := mm.Root() - defer rootDirent.DecRef() + defer rootDirent.DecRef(ctx) // Add mounts at the following paths: paths := []string{ @@ -240,7 +241,7 @@ func TestUnmount(t *testing.T) { if err := mm.Mount(ctx, d, submountInode); err != nil { t.Fatalf("could not mount at %q: %v", p, err) } - d.DecRef() + d.DecRef(ctx) } allPaths := make([]string, len(paths)+1) @@ -259,13 +260,13 @@ func TestUnmount(t *testing.T) { if err := mm.Unmount(ctx, d, false); err != nil { t.Fatalf("could not unmount at %q: %v", p, err) } - d.DecRef() + d.DecRef(ctx) // Remove the path that has been unmounted and the check that the remaining // mounts are still there. allPaths = allPaths[:len(allPaths)-1] submounts := mm.AllMountsUnder(rootMnt) - if err := mountPathsAre(rootDirent, submounts, allPaths...); err != nil { + if err := mountPathsAre(ctx, rootDirent, submounts, allPaths...); err != nil { t.Error(err) } } diff --git a/pkg/sentry/fs/mounts.go b/pkg/sentry/fs/mounts.go index 3f2bd0e87..d741c4339 100644 --- a/pkg/sentry/fs/mounts.go +++ b/pkg/sentry/fs/mounts.go @@ -234,7 +234,7 @@ func (mns *MountNamespace) flushMountSourceRefsLocked() { // After destroy is called, the MountNamespace may continue to be referenced (for // example via /proc/mounts), but should free all resources and shouldn't have // Find* methods called. -func (mns *MountNamespace) destroy() { +func (mns *MountNamespace) destroy(ctx context.Context) { mns.mu.Lock() defer mns.mu.Unlock() @@ -247,13 +247,13 @@ func (mns *MountNamespace) destroy() { for _, mp := range mns.mounts { // Drop the mount reference on all mounted dirents. for ; mp != nil; mp = mp.previous { - mp.root.DecRef() + mp.root.DecRef(ctx) } } mns.mounts = nil // Drop reference on the root. - mns.root.DecRef() + mns.root.DecRef(ctx) // Ensure that root cannot be accessed via this MountNamespace any // more. @@ -265,8 +265,8 @@ func (mns *MountNamespace) destroy() { } // DecRef implements RefCounter.DecRef with destructor mns.destroy. -func (mns *MountNamespace) DecRef() { - mns.DecRefWithDestructor(mns.destroy) +func (mns *MountNamespace) DecRef(ctx context.Context) { + mns.DecRefWithDestructor(ctx, mns.destroy) } // withMountLocked prevents further walks to `node`, because `node` is about to @@ -312,7 +312,7 @@ func (mns *MountNamespace) Mount(ctx context.Context, mountPoint *Dirent, inode if err != nil { return err } - defer replacement.DecRef() + defer replacement.DecRef(ctx) // Set the mount's root dirent and id. parentMnt := mns.findMountLocked(mountPoint) @@ -394,7 +394,7 @@ func (mns *MountNamespace) Unmount(ctx context.Context, node *Dirent, detachOnly panic(fmt.Sprintf("Last mount in the chain must be a undo mount: %+v", prev)) } // Drop mount reference taken at the end of MountNamespace.Mount. - prev.root.DecRef() + prev.root.DecRef(ctx) } else { mns.mounts[prev.root] = prev } @@ -496,11 +496,11 @@ func (mns *MountNamespace) FindLink(ctx context.Context, root, wd *Dirent, path // non-directory root is hopeless. if current != root { if !IsDir(current.Inode.StableAttr) { - current.DecRef() // Drop reference from above. + current.DecRef(ctx) // Drop reference from above. return nil, syserror.ENOTDIR } if err := current.Inode.CheckPermission(ctx, PermMask{Execute: true}); err != nil { - current.DecRef() // Drop reference from above. + current.DecRef(ctx) // Drop reference from above. return nil, err } } @@ -511,12 +511,12 @@ func (mns *MountNamespace) FindLink(ctx context.Context, root, wd *Dirent, path // Allow failed walks to cache the dirent, because no // children will acquire a reference at the end. current.maybeExtendReference() - current.DecRef() + current.DecRef(ctx) return nil, err } // Drop old reference. - current.DecRef() + current.DecRef(ctx) if remainder != "" { // Ensure it's resolved, unless it's the last level. @@ -570,11 +570,11 @@ func (mns *MountNamespace) resolve(ctx context.Context, root, node *Dirent, rema case nil: // Make sure we didn't exhaust the traversal budget. if *remainingTraversals == 0 { - target.DecRef() + target.DecRef(ctx) return nil, syscall.ELOOP } - node.DecRef() // Drop the original reference. + node.DecRef(ctx) // Drop the original reference. return target, nil case syscall.ENOLINK: @@ -582,7 +582,7 @@ func (mns *MountNamespace) resolve(ctx context.Context, root, node *Dirent, rema return node, nil case ErrResolveViaReadlink: - defer node.DecRef() // See above. + defer node.DecRef(ctx) // See above. // First, check if we should traverse. if *remainingTraversals == 0 { @@ -608,7 +608,7 @@ func (mns *MountNamespace) resolve(ctx context.Context, root, node *Dirent, rema return d, err default: - node.DecRef() // Drop for err; see above. + node.DecRef(ctx) // Drop for err; see above. // Propagate the error. return nil, err diff --git a/pkg/sentry/fs/mounts_test.go b/pkg/sentry/fs/mounts_test.go index a69b41468..975d6cbc9 100644 --- a/pkg/sentry/fs/mounts_test.go +++ b/pkg/sentry/fs/mounts_test.go @@ -51,7 +51,7 @@ func TestFindLink(t *testing.T) { } root := mm.Root() - defer root.DecRef() + defer root.DecRef(ctx) foo, err := root.Walk(ctx, root, "foo") if err != nil { t.Fatalf("Error walking to foo: %v", err) diff --git a/pkg/sentry/fs/overlay.go b/pkg/sentry/fs/overlay.go index a8ae7d81d..35013a21b 100644 --- a/pkg/sentry/fs/overlay.go +++ b/pkg/sentry/fs/overlay.go @@ -107,7 +107,7 @@ func NewOverlayRoot(ctx context.Context, upper *Inode, lower *Inode, flags Mount msrc := newOverlayMountSource(ctx, upper.MountSource, lower.MountSource, flags) overlay, err := newOverlayEntry(ctx, upper, lower, true) if err != nil { - msrc.DecRef() + msrc.DecRef(ctx) return nil, err } @@ -130,7 +130,7 @@ func NewOverlayRootFile(ctx context.Context, upperMS *MountSource, lower *Inode, msrc := newOverlayMountSource(ctx, upperMS, lower.MountSource, flags) overlay, err := newOverlayEntry(ctx, nil, lower, true) if err != nil { - msrc.DecRef() + msrc.DecRef(ctx) return nil, err } return newOverlayInode(ctx, overlay, msrc), nil @@ -230,16 +230,16 @@ func newOverlayEntry(ctx context.Context, upper *Inode, lower *Inode, lowerExist }, nil } -func (o *overlayEntry) release() { +func (o *overlayEntry) release(ctx context.Context) { // We drop a reference on upper and lower file system Inodes // rather than releasing them, because in-memory filesystems // may hold an extra reference to these Inodes so that they // stay in memory. if o.upper != nil { - o.upper.DecRef() + o.upper.DecRef(ctx) } if o.lower != nil { - o.lower.DecRef() + o.lower.DecRef(ctx) } } diff --git a/pkg/sentry/fs/proc/fds.go b/pkg/sentry/fs/proc/fds.go index 35972e23c..45523adf8 100644 --- a/pkg/sentry/fs/proc/fds.go +++ b/pkg/sentry/fs/proc/fds.go @@ -56,11 +56,11 @@ func walkDescriptors(t *kernel.Task, p string, toInode func(*fs.File, kernel.FDF // readDescriptors reads fds in the task starting at offset, and calls the // toDentAttr callback for each to get a DentAttr, which it then emits. This is // a helper for implementing fs.InodeOperations.Readdir. -func readDescriptors(t *kernel.Task, c *fs.DirCtx, offset int64, toDentAttr func(int) fs.DentAttr) (int64, error) { +func readDescriptors(ctx context.Context, t *kernel.Task, c *fs.DirCtx, offset int64, toDentAttr func(int) fs.DentAttr) (int64, error) { var fds []int32 t.WithMuLocked(func(t *kernel.Task) { if fdTable := t.FDTable(); fdTable != nil { - fds = fdTable.GetFDs() + fds = fdTable.GetFDs(ctx) } }) @@ -116,7 +116,7 @@ func (f *fd) GetFile(context.Context, *fs.Dirent, fs.FileFlags) (*fs.File, error func (f *fd) Readlink(ctx context.Context, _ *fs.Inode) (string, error) { root := fs.RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } n, _ := f.file.Dirent.FullName(root) return n, nil @@ -135,13 +135,7 @@ func (f *fd) Truncate(context.Context, *fs.Inode, int64) error { func (f *fd) Release(ctx context.Context) { f.Symlink.Release(ctx) - f.file.DecRef() -} - -// Close releases the reference on the file. -func (f *fd) Close() error { - f.file.DecRef() - return nil + f.file.DecRef(ctx) } // fdDir is an InodeOperations for /proc/TID/fd. @@ -227,7 +221,7 @@ func (f *fdDirFile) Readdir(ctx context.Context, file *fs.File, ser fs.DentrySer if f.isInfoFile { typ = fs.Symlink } - return readDescriptors(f.t, dirCtx, file.Offset(), func(fd int) fs.DentAttr { + return readDescriptors(ctx, f.t, dirCtx, file.Offset(), func(fd int) fs.DentAttr { return fs.GenericDentAttr(typ, device.ProcDevice) }) } @@ -261,7 +255,7 @@ func (fdid *fdInfoDir) Lookup(ctx context.Context, dir *fs.Inode, p string) (*fs // locks, and other data. For now we only have flags. // See https://www.kernel.org/doc/Documentation/filesystems/proc.txt flags := file.Flags().ToLinux() | fdFlags.ToLinuxFileFlags() - file.DecRef() + file.DecRef(ctx) contents := []byte(fmt.Sprintf("flags:\t0%o\n", flags)) return newStaticProcInode(ctx, dir.MountSource, contents) }) diff --git a/pkg/sentry/fs/proc/mounts.go b/pkg/sentry/fs/proc/mounts.go index 1fc9c703c..6a63c47b3 100644 --- a/pkg/sentry/fs/proc/mounts.go +++ b/pkg/sentry/fs/proc/mounts.go @@ -47,7 +47,7 @@ func forEachMount(t *kernel.Task, fn func(string, *fs.Mount)) { // The task has been destroyed. Nothing to show here. return } - defer rootDir.DecRef() + defer rootDir.DecRef(t) mnt := t.MountNamespace().FindMount(rootDir) if mnt == nil { @@ -64,7 +64,7 @@ func forEachMount(t *kernel.Task, fn func(string, *fs.Mount)) { continue // No longer valid. } mountPath, desc := mroot.FullName(rootDir) - mroot.DecRef() + mroot.DecRef(t) if !desc { // MountSources that are not descendants of the chroot jail are ignored. continue @@ -97,7 +97,7 @@ func (mif *mountInfoFile) ReadSeqFileData(ctx context.Context, handle seqfile.Se if mroot == nil { return // No longer valid. } - defer mroot.DecRef() + defer mroot.DecRef(ctx) // Format: // 36 35 98:0 /mnt1 /mnt2 rw,noatime master:1 - ext3 /dev/root rw,errors=continue @@ -216,7 +216,7 @@ func (mf *mountsFile) ReadSeqFileData(ctx context.Context, handle seqfile.SeqHan if root == nil { return // No longer valid. } - defer root.DecRef() + defer root.DecRef(ctx) flags := root.Inode.MountSource.Flags opts := "rw" diff --git a/pkg/sentry/fs/proc/net.go b/pkg/sentry/fs/proc/net.go index bd18177d4..83a43aa26 100644 --- a/pkg/sentry/fs/proc/net.go +++ b/pkg/sentry/fs/proc/net.go @@ -419,7 +419,7 @@ func (n *netUnix) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]s } sfile := s.(*fs.File) if family, _, _ := sfile.FileOperations.(socket.Socket).Type(); family != linux.AF_UNIX { - s.DecRef() + s.DecRef(ctx) // Not a unix socket. continue } @@ -479,7 +479,7 @@ func (n *netUnix) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]s } fmt.Fprintf(&buf, "\n") - s.DecRef() + s.DecRef(ctx) } data := []seqfile.SeqData{ @@ -574,7 +574,7 @@ func commonReadSeqFileDataTCP(ctx context.Context, n seqfile.SeqHandle, k *kerne panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) } if family, stype, _ := sops.Type(); !(family == fa && stype == linux.SOCK_STREAM) { - s.DecRef() + s.DecRef(ctx) // Not tcp4 sockets. continue } @@ -664,7 +664,7 @@ func commonReadSeqFileDataTCP(ctx context.Context, n seqfile.SeqHandle, k *kerne fmt.Fprintf(&buf, "\n") - s.DecRef() + s.DecRef(ctx) } data := []seqfile.SeqData{ @@ -752,7 +752,7 @@ func (n *netUDP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) } if family, stype, _ := sops.Type(); family != linux.AF_INET || stype != linux.SOCK_DGRAM { - s.DecRef() + s.DecRef(ctx) // Not udp4 socket. continue } @@ -822,7 +822,7 @@ func (n *netUDP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se fmt.Fprintf(&buf, "\n") - s.DecRef() + s.DecRef(ctx) } data := []seqfile.SeqData{ diff --git a/pkg/sentry/fs/proc/proc.go b/pkg/sentry/fs/proc/proc.go index c659224a7..77e0e1d26 100644 --- a/pkg/sentry/fs/proc/proc.go +++ b/pkg/sentry/fs/proc/proc.go @@ -213,7 +213,7 @@ func (rpf *rootProcFile) Readdir(ctx context.Context, file *fs.File, ser fs.Dent // Add dot and dotdot. root := fs.RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } dot, dotdot := file.Dirent.GetDotAttrs(root) names = append(names, ".", "..") diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index 4bbe90198..9cf7f2a62 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -185,7 +185,7 @@ func (f *subtasksFile) Readdir(ctx context.Context, file *fs.File, ser fs.Dentry // Serialize "." and "..". root := fs.RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } dot, dotdot := file.Dirent.GetDotAttrs(root) if err := dirCtx.DirEmit(".", dot); err != nil { @@ -295,7 +295,7 @@ func (e *exe) Readlink(ctx context.Context, inode *fs.Inode) (string, error) { if err != nil { return "", err } - defer exec.DecRef() + defer exec.DecRef(ctx) return exec.PathnameWithDeleted(ctx), nil } diff --git a/pkg/sentry/fs/ramfs/dir.go b/pkg/sentry/fs/ramfs/dir.go index bfa304552..f4fcddecb 100644 --- a/pkg/sentry/fs/ramfs/dir.go +++ b/pkg/sentry/fs/ramfs/dir.go @@ -219,7 +219,7 @@ func (d *Dir) Remove(ctx context.Context, _ *fs.Inode, name string) error { } // Remove our reference on the inode. - inode.DecRef() + inode.DecRef(ctx) return nil } @@ -250,7 +250,7 @@ func (d *Dir) RemoveDirectory(ctx context.Context, _ *fs.Inode, name string) err } // Remove our reference on the inode. - inode.DecRef() + inode.DecRef(ctx) return nil } @@ -326,7 +326,7 @@ func (d *Dir) Create(ctx context.Context, dir *fs.Inode, name string, flags fs.F // Create the Dirent and corresponding file. created := fs.NewDirent(ctx, inode, name) - defer created.DecRef() + defer created.DecRef(ctx) return created.Inode.GetFile(ctx, created, flags) } @@ -412,11 +412,11 @@ func (*Dir) Rename(ctx context.Context, inode *fs.Inode, oldParent *fs.Inode, ol } // Release implements fs.InodeOperation.Release. -func (d *Dir) Release(_ context.Context) { +func (d *Dir) Release(ctx context.Context) { // Drop references on all children. d.mu.Lock() for _, i := range d.children { - i.DecRef() + i.DecRef(ctx) } d.mu.Unlock() } @@ -456,7 +456,7 @@ func (dfo *dirFileOperations) IterateDir(ctx context.Context, d *fs.Dirent, dirC func (dfo *dirFileOperations) Readdir(ctx context.Context, file *fs.File, serializer fs.DentrySerializer) (int64, error) { root := fs.RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } dirCtx := &fs.DirCtx{ Serializer: serializer, @@ -473,13 +473,13 @@ func hasChildren(ctx context.Context, inode *fs.Inode) (bool, error) { // dropped when that dirent is destroyed. inode.IncRef() d := fs.NewTransientDirent(inode) - defer d.DecRef() + defer d.DecRef(ctx) file, err := inode.GetFile(ctx, d, fs.FileFlags{Read: true}) if err != nil { return false, err } - defer file.DecRef() + defer file.DecRef(ctx) ser := &fs.CollectEntriesSerializer{} if err := file.Readdir(ctx, ser); err != nil { @@ -530,7 +530,7 @@ func Rename(ctx context.Context, oldParent fs.InodeOperations, oldName string, n if err != nil { return err } - inode.DecRef() + inode.DecRef(ctx) } // Be careful, we may have already grabbed this mutex above. diff --git a/pkg/sentry/fs/ramfs/tree_test.go b/pkg/sentry/fs/ramfs/tree_test.go index a6ed8b2c5..3e0d1e07e 100644 --- a/pkg/sentry/fs/ramfs/tree_test.go +++ b/pkg/sentry/fs/ramfs/tree_test.go @@ -67,7 +67,7 @@ func TestMakeDirectoryTree(t *testing.T) { continue } root := mm.Root() - defer mm.DecRef() + defer mm.DecRef(ctx) for _, p := range test.subdirs { maxTraversals := uint(0) diff --git a/pkg/sentry/fs/timerfd/timerfd.go b/pkg/sentry/fs/timerfd/timerfd.go index 88c344089..f362ca9b6 100644 --- a/pkg/sentry/fs/timerfd/timerfd.go +++ b/pkg/sentry/fs/timerfd/timerfd.go @@ -55,7 +55,7 @@ type TimerOperations struct { func NewFile(ctx context.Context, c ktime.Clock) *fs.File { dirent := fs.NewDirent(ctx, anon.NewInode(ctx), "anon_inode:[timerfd]") // Release the initial dirent reference after NewFile takes a reference. - defer dirent.DecRef() + defer dirent.DecRef(ctx) tops := &TimerOperations{} tops.timer = ktime.NewTimer(c, tops) // Timerfds reject writes, but the Write flag must be set in order to @@ -65,7 +65,7 @@ func NewFile(ctx context.Context, c ktime.Clock) *fs.File { } // Release implements fs.FileOperations.Release. -func (t *TimerOperations) Release() { +func (t *TimerOperations) Release(context.Context) { t.timer.Destroy() } diff --git a/pkg/sentry/fs/tmpfs/file_test.go b/pkg/sentry/fs/tmpfs/file_test.go index aaba35502..d4d613ea9 100644 --- a/pkg/sentry/fs/tmpfs/file_test.go +++ b/pkg/sentry/fs/tmpfs/file_test.go @@ -46,7 +46,7 @@ func newFile(ctx context.Context) *fs.File { func TestGrow(t *testing.T) { ctx := contexttest.Context(t) f := newFile(ctx) - defer f.DecRef() + defer f.DecRef(ctx) abuf := bytes.Repeat([]byte{'a'}, 68) n, err := f.Pwritev(ctx, usermem.BytesIOSequence(abuf), 0) diff --git a/pkg/sentry/fs/tty/dir.go b/pkg/sentry/fs/tty/dir.go index 108654827..463f6189e 100644 --- a/pkg/sentry/fs/tty/dir.go +++ b/pkg/sentry/fs/tty/dir.go @@ -132,7 +132,7 @@ func (d *dirInodeOperations) Release(ctx context.Context) { d.mu.Lock() defer d.mu.Unlock() - d.master.DecRef() + d.master.DecRef(ctx) if len(d.slaves) != 0 { panic(fmt.Sprintf("devpts directory still contains active terminals: %+v", d)) } @@ -263,7 +263,7 @@ func (d *dirInodeOperations) allocateTerminal(ctx context.Context) (*Terminal, e } // masterClose is called when the master end of t is closed. -func (d *dirInodeOperations) masterClose(t *Terminal) { +func (d *dirInodeOperations) masterClose(ctx context.Context, t *Terminal) { d.mu.Lock() defer d.mu.Unlock() @@ -277,7 +277,7 @@ func (d *dirInodeOperations) masterClose(t *Terminal) { panic(fmt.Sprintf("Terminal %+v doesn't exist in %+v?", t, d)) } - s.DecRef() + s.DecRef(ctx) delete(d.slaves, t.n) d.dentryMap.Remove(strconv.FormatUint(uint64(t.n), 10)) } @@ -322,7 +322,7 @@ func (df *dirFileOperations) IterateDir(ctx context.Context, d *fs.Dirent, dirCt func (df *dirFileOperations) Readdir(ctx context.Context, file *fs.File, serializer fs.DentrySerializer) (int64, error) { root := fs.RootFromContext(ctx) if root != nil { - defer root.DecRef() + defer root.DecRef(ctx) } dirCtx := &fs.DirCtx{ Serializer: serializer, diff --git a/pkg/sentry/fs/tty/fs.go b/pkg/sentry/fs/tty/fs.go index 8fe05ebe5..2d4d44bf3 100644 --- a/pkg/sentry/fs/tty/fs.go +++ b/pkg/sentry/fs/tty/fs.go @@ -108,4 +108,4 @@ func (superOperations) ResetInodeMappings() {} func (superOperations) SaveInodeMapping(*fs.Inode, string) {} // Destroy implements MountSourceOperations.Destroy. -func (superOperations) Destroy() {} +func (superOperations) Destroy(context.Context) {} diff --git a/pkg/sentry/fs/tty/master.go b/pkg/sentry/fs/tty/master.go index fe07fa929..e00746017 100644 --- a/pkg/sentry/fs/tty/master.go +++ b/pkg/sentry/fs/tty/master.go @@ -75,7 +75,7 @@ func newMasterInode(ctx context.Context, d *dirInodeOperations, owner fs.FileOwn } // Release implements fs.InodeOperations.Release. -func (mi *masterInodeOperations) Release(ctx context.Context) { +func (mi *masterInodeOperations) Release(context.Context) { } // Truncate implements fs.InodeOperations.Truncate. @@ -120,9 +120,9 @@ type masterFileOperations struct { var _ fs.FileOperations = (*masterFileOperations)(nil) // Release implements fs.FileOperations.Release. -func (mf *masterFileOperations) Release() { - mf.d.masterClose(mf.t) - mf.t.DecRef() +func (mf *masterFileOperations) Release(ctx context.Context) { + mf.d.masterClose(ctx, mf.t) + mf.t.DecRef(ctx) } // EventRegister implements waiter.Waitable.EventRegister. diff --git a/pkg/sentry/fs/tty/slave.go b/pkg/sentry/fs/tty/slave.go index 9871f6fc6..7c7292687 100644 --- a/pkg/sentry/fs/tty/slave.go +++ b/pkg/sentry/fs/tty/slave.go @@ -71,7 +71,7 @@ func newSlaveInode(ctx context.Context, d *dirInodeOperations, t *Terminal, owne // Release implements fs.InodeOperations.Release. func (si *slaveInodeOperations) Release(ctx context.Context) { - si.t.DecRef() + si.t.DecRef(ctx) } // Truncate implements fs.InodeOperations.Truncate. @@ -106,7 +106,7 @@ type slaveFileOperations struct { var _ fs.FileOperations = (*slaveFileOperations)(nil) // Release implements fs.FileOperations.Release. -func (sf *slaveFileOperations) Release() { +func (sf *slaveFileOperations) Release(context.Context) { } // EventRegister implements waiter.Waitable.EventRegister. diff --git a/pkg/sentry/fs/user/path.go b/pkg/sentry/fs/user/path.go index 397e96045..2f5a43b84 100644 --- a/pkg/sentry/fs/user/path.go +++ b/pkg/sentry/fs/user/path.go @@ -82,7 +82,7 @@ func resolve(ctx context.Context, mns *fs.MountNamespace, paths []string, name s // Caller has no root. Don't bother traversing anything. return "", syserror.ENOENT } - defer root.DecRef() + defer root.DecRef(ctx) for _, p := range paths { if !path.IsAbs(p) { // Relative paths aren't safe, no one should be using them. @@ -100,7 +100,7 @@ func resolve(ctx context.Context, mns *fs.MountNamespace, paths []string, name s if err != nil { return "", err } - defer d.DecRef() + defer d.DecRef(ctx) // Check that it is a regular file. if !fs.IsRegular(d.Inode.StableAttr) { @@ -121,7 +121,7 @@ func resolve(ctx context.Context, mns *fs.MountNamespace, paths []string, name s func resolveVFS2(ctx context.Context, creds *auth.Credentials, mns *vfs.MountNamespace, paths []string, name string) (string, error) { root := mns.Root() - defer root.DecRef() + defer root.DecRef(ctx) for _, p := range paths { if !path.IsAbs(p) { // Relative paths aren't safe, no one should be using them. @@ -148,7 +148,7 @@ func resolveVFS2(ctx context.Context, creds *auth.Credentials, mns *vfs.MountNam if err != nil { return "", err } - dentry.DecRef() + dentry.DecRef(ctx) return binPath, nil } diff --git a/pkg/sentry/fs/user/user.go b/pkg/sentry/fs/user/user.go index f4d525523..936fd3932 100644 --- a/pkg/sentry/fs/user/user.go +++ b/pkg/sentry/fs/user/user.go @@ -62,7 +62,7 @@ func getExecUserHome(ctx context.Context, rootMns *fs.MountNamespace, uid auth.K // doesn't exist we will return the default home directory. return defaultHome, nil } - defer dirent.DecRef() + defer dirent.DecRef(ctx) // Check read permissions on the file. if err := dirent.Inode.CheckPermission(ctx, fs.PermMask{Read: true}); err != nil { @@ -81,7 +81,7 @@ func getExecUserHome(ctx context.Context, rootMns *fs.MountNamespace, uid auth.K if err != nil { return "", err } - defer f.DecRef() + defer f.DecRef(ctx) r := &fileReader{ Ctx: ctx, @@ -105,7 +105,7 @@ func getExecUserHomeVFS2(ctx context.Context, mns *vfs.MountNamespace, uid auth. const defaultHome = "/" root := mns.Root() - defer root.DecRef() + defer root.DecRef(ctx) creds := auth.CredentialsFromContext(ctx) @@ -123,7 +123,7 @@ func getExecUserHomeVFS2(ctx context.Context, mns *vfs.MountNamespace, uid auth. if err != nil { return defaultHome, nil } - defer fd.DecRef() + defer fd.DecRef(ctx) r := &fileReaderVFS2{ ctx: ctx, diff --git a/pkg/sentry/fs/user/user_test.go b/pkg/sentry/fs/user/user_test.go index 7d8e9ac7c..12b786224 100644 --- a/pkg/sentry/fs/user/user_test.go +++ b/pkg/sentry/fs/user/user_test.go @@ -39,7 +39,7 @@ func createEtcPasswd(ctx context.Context, root *fs.Dirent, contents string, mode if err != nil { return err } - defer etc.DecRef() + defer etc.DecRef(ctx) switch mode.FileType() { case 0: // Don't create anything. @@ -49,7 +49,7 @@ func createEtcPasswd(ctx context.Context, root *fs.Dirent, contents string, mode if err != nil { return err } - defer passwd.DecRef() + defer passwd.DecRef(ctx) if _, err := passwd.Writev(ctx, usermem.BytesIOSequence([]byte(contents))); err != nil { return err } @@ -110,9 +110,9 @@ func TestGetExecUserHome(t *testing.T) { if err != nil { t.Fatalf("NewMountNamespace failed: %v", err) } - defer mns.DecRef() + defer mns.DecRef(ctx) root := mns.Root() - defer root.DecRef() + defer root.DecRef(ctx) ctx = fs.WithRoot(ctx, root) if err := createEtcPasswd(ctx, root, tc.passwdContents, tc.passwdMode); err != nil { diff --git a/pkg/sentry/fsbridge/bridge.go b/pkg/sentry/fsbridge/bridge.go index 8e7590721..7e61209ee 100644 --- a/pkg/sentry/fsbridge/bridge.go +++ b/pkg/sentry/fsbridge/bridge.go @@ -44,7 +44,7 @@ type File interface { IncRef() // DecRef decrements reference. - DecRef() + DecRef(ctx context.Context) } // Lookup provides a common interface to open files. diff --git a/pkg/sentry/fsbridge/fs.go b/pkg/sentry/fsbridge/fs.go index 093ce1fb3..9785fd62a 100644 --- a/pkg/sentry/fsbridge/fs.go +++ b/pkg/sentry/fsbridge/fs.go @@ -49,7 +49,7 @@ func (f *fsFile) PathnameWithDeleted(ctx context.Context) string { // global there. return "" } - defer root.DecRef() + defer root.DecRef(ctx) name, _ := f.file.Dirent.FullName(root) return name @@ -87,8 +87,8 @@ func (f *fsFile) IncRef() { } // DecRef implements File. -func (f *fsFile) DecRef() { - f.file.DecRef() +func (f *fsFile) DecRef(ctx context.Context) { + f.file.DecRef(ctx) } // fsLookup implements Lookup interface using fs.File. @@ -124,7 +124,7 @@ func (l *fsLookup) OpenPath(ctx context.Context, path string, opts vfs.OpenOptio if err != nil { return nil, err } - defer d.DecRef() + defer d.DecRef(ctx) if !resolveFinal && fs.IsSymlink(d.Inode.StableAttr) { return nil, syserror.ELOOP diff --git a/pkg/sentry/fsbridge/vfs.go b/pkg/sentry/fsbridge/vfs.go index 89168220a..323506d33 100644 --- a/pkg/sentry/fsbridge/vfs.go +++ b/pkg/sentry/fsbridge/vfs.go @@ -43,7 +43,7 @@ func NewVFSFile(file *vfs.FileDescription) File { // PathnameWithDeleted implements File. func (f *VFSFile) PathnameWithDeleted(ctx context.Context) string { root := vfs.RootFromContext(ctx) - defer root.DecRef() + defer root.DecRef(ctx) vfsObj := f.file.VirtualDentry().Mount().Filesystem().VirtualFilesystem() name, _ := vfsObj.PathnameWithDeleted(ctx, root, f.file.VirtualDentry()) @@ -86,8 +86,8 @@ func (f *VFSFile) IncRef() { } // DecRef implements File. -func (f *VFSFile) DecRef() { - f.file.DecRef() +func (f *VFSFile) DecRef(ctx context.Context) { + f.file.DecRef(ctx) } // FileDescription returns the FileDescription represented by f. It does not diff --git a/pkg/sentry/fsimpl/devpts/devpts.go b/pkg/sentry/fsimpl/devpts/devpts.go index e6fda2b4f..7169e91af 100644 --- a/pkg/sentry/fsimpl/devpts/devpts.go +++ b/pkg/sentry/fsimpl/devpts/devpts.go @@ -103,9 +103,9 @@ func (fstype FilesystemType) newFilesystem(vfsObj *vfs.VirtualFilesystem, creds } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) - fs.Filesystem.Release() + fs.Filesystem.Release(ctx) } // rootInode is the root directory inode for the devpts mounts. diff --git a/pkg/sentry/fsimpl/devpts/master.go b/pkg/sentry/fsimpl/devpts/master.go index 1081fff52..3bb397f71 100644 --- a/pkg/sentry/fsimpl/devpts/master.go +++ b/pkg/sentry/fsimpl/devpts/master.go @@ -60,7 +60,7 @@ func (mi *masterInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vf } fd.LockFD.Init(&mi.locks) if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { - mi.DecRef() + mi.DecRef(ctx) return nil, err } return &fd.vfsfd, nil @@ -98,9 +98,9 @@ type masterFileDescription struct { var _ vfs.FileDescriptionImpl = (*masterFileDescription)(nil) // Release implements vfs.FileDescriptionImpl.Release. -func (mfd *masterFileDescription) Release() { +func (mfd *masterFileDescription) Release(ctx context.Context) { mfd.inode.root.masterClose(mfd.t) - mfd.inode.DecRef() + mfd.inode.DecRef(ctx) } // EventRegister implements waiter.Waitable.EventRegister. diff --git a/pkg/sentry/fsimpl/devpts/slave.go b/pkg/sentry/fsimpl/devpts/slave.go index a91cae3ef..32e4e1908 100644 --- a/pkg/sentry/fsimpl/devpts/slave.go +++ b/pkg/sentry/fsimpl/devpts/slave.go @@ -56,7 +56,7 @@ func (si *slaveInode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs } fd.LockFD.Init(&si.locks) if err := fd.vfsfd.Init(fd, opts.Flags, rp.Mount(), vfsd, &vfs.FileDescriptionOptions{}); err != nil { - si.DecRef() + si.DecRef(ctx) return nil, err } return &fd.vfsfd, nil @@ -103,8 +103,8 @@ type slaveFileDescription struct { var _ vfs.FileDescriptionImpl = (*slaveFileDescription)(nil) // Release implements fs.FileOperations.Release. -func (sfd *slaveFileDescription) Release() { - sfd.inode.DecRef() +func (sfd *slaveFileDescription) Release(ctx context.Context) { + sfd.inode.DecRef(ctx) } // EventRegister implements waiter.Waitable.EventRegister. diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go index d0e06cdc0..2ed5fa8a9 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs.go @@ -92,9 +92,9 @@ func NewAccessor(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *auth } // Release must be called when a is no longer in use. -func (a *Accessor) Release() { - a.root.DecRef() - a.mntns.DecRef() +func (a *Accessor) Release(ctx context.Context) { + a.root.DecRef(ctx) + a.mntns.DecRef(ctx) } // accessorContext implements context.Context by extending an existing diff --git a/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go b/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go index b6d52c015..747867cca 100644 --- a/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go +++ b/pkg/sentry/fsimpl/devtmpfs/devtmpfs_test.go @@ -30,7 +30,7 @@ func TestDevtmpfs(t *testing.T) { creds := auth.CredentialsFromContext(ctx) vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { t.Fatalf("VFS init: %v", err) } // Register tmpfs just so that we can have a root filesystem that isn't @@ -48,9 +48,9 @@ func TestDevtmpfs(t *testing.T) { if err != nil { t.Fatalf("failed to create tmpfs root mount: %v", err) } - defer mntns.DecRef() + defer mntns.DecRef(ctx) root := mntns.Root() - defer root.DecRef() + defer root.DecRef(ctx) devpop := vfs.PathOperation{ Root: root, Start: root, @@ -69,7 +69,7 @@ func TestDevtmpfs(t *testing.T) { if err != nil { t.Fatalf("failed to create devtmpfs.Accessor: %v", err) } - defer a.Release() + defer a.Release(ctx) // Create "userspace-initialized" files using a devtmpfs.Accessor. if err := a.UserspaceInit(ctx); err != nil { diff --git a/pkg/sentry/fsimpl/eventfd/eventfd.go b/pkg/sentry/fsimpl/eventfd/eventfd.go index d12d78b84..812171fa3 100644 --- a/pkg/sentry/fsimpl/eventfd/eventfd.go +++ b/pkg/sentry/fsimpl/eventfd/eventfd.go @@ -59,9 +59,9 @@ type EventFileDescription struct { var _ vfs.FileDescriptionImpl = (*EventFileDescription)(nil) // New creates a new event fd. -func New(vfsObj *vfs.VirtualFilesystem, initVal uint64, semMode bool, flags uint32) (*vfs.FileDescription, error) { +func New(ctx context.Context, vfsObj *vfs.VirtualFilesystem, initVal uint64, semMode bool, flags uint32) (*vfs.FileDescription, error) { vd := vfsObj.NewAnonVirtualDentry("[eventfd]") - defer vd.DecRef() + defer vd.DecRef(ctx) efd := &EventFileDescription{ val: initVal, semMode: semMode, @@ -107,7 +107,7 @@ func (efd *EventFileDescription) HostFD() (int, error) { } // Release implements FileDescriptionImpl.Release() -func (efd *EventFileDescription) Release() { +func (efd *EventFileDescription) Release(context.Context) { efd.mu.Lock() defer efd.mu.Unlock() if efd.hostfd >= 0 { diff --git a/pkg/sentry/fsimpl/eventfd/eventfd_test.go b/pkg/sentry/fsimpl/eventfd/eventfd_test.go index 20e3adffc..49916fa81 100644 --- a/pkg/sentry/fsimpl/eventfd/eventfd_test.go +++ b/pkg/sentry/fsimpl/eventfd/eventfd_test.go @@ -36,16 +36,16 @@ func TestEventFD(t *testing.T) { for _, initVal := range initVals { ctx := contexttest.Context(t) vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { t.Fatalf("VFS init: %v", err) } // Make a new eventfd that is writable. - eventfd, err := New(vfsObj, initVal, false, linux.O_RDWR) + eventfd, err := New(ctx, vfsObj, initVal, false, linux.O_RDWR) if err != nil { t.Fatalf("New() failed: %v", err) } - defer eventfd.DecRef() + defer eventfd.DecRef(ctx) // Register a callback for a write event. w, ch := waiter.NewChannelEntry(nil) @@ -74,16 +74,16 @@ func TestEventFD(t *testing.T) { func TestEventFDStat(t *testing.T) { ctx := contexttest.Context(t) vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { t.Fatalf("VFS init: %v", err) } // Make a new eventfd that is writable. - eventfd, err := New(vfsObj, 0, false, linux.O_RDWR) + eventfd, err := New(ctx, vfsObj, 0, false, linux.O_RDWR) if err != nil { t.Fatalf("New() failed: %v", err) } - defer eventfd.DecRef() + defer eventfd.DecRef(ctx) statx, err := eventfd.Stat(ctx, vfs.StatOptions{ Mask: linux.STATX_BASIC_STATS, diff --git a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go index 89caee3df..8f7d5a9bb 100644 --- a/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go +++ b/pkg/sentry/fsimpl/ext/benchmark/benchmark_test.go @@ -53,7 +53,7 @@ func setUp(b *testing.B, imagePath string) (context.Context, *vfs.VirtualFilesys // Create VFS. vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { return nil, nil, nil, nil, err } vfsObj.MustRegisterFilesystemType("extfs", ext.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ @@ -68,7 +68,7 @@ func setUp(b *testing.B, imagePath string) (context.Context, *vfs.VirtualFilesys root := mntns.Root() tearDown := func() { - root.DecRef() + root.DecRef(ctx) if err := f.Close(); err != nil { b.Fatalf("tearDown failed: %v", err) @@ -169,7 +169,7 @@ func BenchmarkVFS2ExtfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to walk to mount point: %v", err) } - defer mountPoint.DecRef() + defer mountPoint.DecRef(ctx) // Create extfs submount. mountTearDown := mount(b, fmt.Sprintf("/tmp/image-%d.ext4", depth), vfsfs, &pop) diff --git a/pkg/sentry/fsimpl/ext/dentry.go b/pkg/sentry/fsimpl/ext/dentry.go index 55902322a..7a1b4219f 100644 --- a/pkg/sentry/fsimpl/ext/dentry.go +++ b/pkg/sentry/fsimpl/ext/dentry.go @@ -15,6 +15,7 @@ package ext import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/vfs" ) @@ -55,7 +56,7 @@ func (d *dentry) TryIncRef() bool { } // DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef() { +func (d *dentry) DecRef(ctx context.Context) { // FIXME(b/134676337): filesystem.mu may not be locked as required by // inode.decRef(). d.inode.decRef() @@ -64,7 +65,7 @@ func (d *dentry) DecRef() { // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. // // TODO(b/134676337): Implement inotify. -func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) {} +func (d *dentry) InotifyWithParent(ctx context.Context, events, cookie uint32, et vfs.EventType) {} // Watches implements vfs.DentryImpl.Watches. // @@ -76,4 +77,4 @@ func (d *dentry) Watches() *vfs.Watches { // OnZeroWatches implements vfs.Dentry.OnZeroWatches. // // TODO(b/134676337): Implement inotify. -func (d *dentry) OnZeroWatches() {} +func (d *dentry) OnZeroWatches(context.Context) {} diff --git a/pkg/sentry/fsimpl/ext/directory.go b/pkg/sentry/fsimpl/ext/directory.go index 357512c7e..0fc01668d 100644 --- a/pkg/sentry/fsimpl/ext/directory.go +++ b/pkg/sentry/fsimpl/ext/directory.go @@ -142,7 +142,7 @@ type directoryFD struct { var _ vfs.FileDescriptionImpl = (*directoryFD)(nil) // Release implements vfs.FileDescriptionImpl.Release. -func (fd *directoryFD) Release() { +func (fd *directoryFD) Release(ctx context.Context) { if fd.iter == nil { return } diff --git a/pkg/sentry/fsimpl/ext/ext.go b/pkg/sentry/fsimpl/ext/ext.go index dac6effbf..08ffc2834 100644 --- a/pkg/sentry/fsimpl/ext/ext.go +++ b/pkg/sentry/fsimpl/ext/ext.go @@ -123,32 +123,32 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt fs.vfsfs.Init(vfsObj, &fsType, &fs) fs.sb, err = readSuperBlock(dev) if err != nil { - fs.vfsfs.DecRef() + fs.vfsfs.DecRef(ctx) return nil, nil, err } if fs.sb.Magic() != linux.EXT_SUPER_MAGIC { // mount(2) specifies that EINVAL should be returned if the superblock is // invalid. - fs.vfsfs.DecRef() + fs.vfsfs.DecRef(ctx) return nil, nil, syserror.EINVAL } // Refuse to mount if the filesystem is incompatible. if !isCompatible(fs.sb) { - fs.vfsfs.DecRef() + fs.vfsfs.DecRef(ctx) return nil, nil, syserror.EINVAL } fs.bgs, err = readBlockGroups(dev, fs.sb) if err != nil { - fs.vfsfs.DecRef() + fs.vfsfs.DecRef(ctx) return nil, nil, err } rootInode, err := fs.getOrCreateInodeLocked(disklayout.RootDirInode) if err != nil { - fs.vfsfs.DecRef() + fs.vfsfs.DecRef(ctx) return nil, nil, err } rootInode.incRef() diff --git a/pkg/sentry/fsimpl/ext/ext_test.go b/pkg/sentry/fsimpl/ext/ext_test.go index 64e9a579f..2dbaee287 100644 --- a/pkg/sentry/fsimpl/ext/ext_test.go +++ b/pkg/sentry/fsimpl/ext/ext_test.go @@ -65,7 +65,7 @@ func setUp(t *testing.T, imagePath string) (context.Context, *vfs.VirtualFilesys // Create VFS. vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { t.Fatalf("VFS init: %v", err) } vfsObj.MustRegisterFilesystemType("extfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ @@ -80,7 +80,7 @@ func setUp(t *testing.T, imagePath string) (context.Context, *vfs.VirtualFilesys root := mntns.Root() tearDown := func() { - root.DecRef() + root.DecRef(ctx) if err := f.Close(); err != nil { t.Fatalf("tearDown failed: %v", err) diff --git a/pkg/sentry/fsimpl/ext/filesystem.go b/pkg/sentry/fsimpl/ext/filesystem.go index 557963e03..c714ddf73 100644 --- a/pkg/sentry/fsimpl/ext/filesystem.go +++ b/pkg/sentry/fsimpl/ext/filesystem.go @@ -84,7 +84,7 @@ var _ vfs.FilesystemImpl = (*filesystem)(nil) // - filesystem.mu must be locked (for writing if write param is true). // - !rp.Done(). // - inode == vfsd.Impl().(*Dentry).inode. -func stepLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, inode *inode, write bool) (*vfs.Dentry, *inode, error) { +func stepLocked(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, inode *inode, write bool) (*vfs.Dentry, *inode, error) { if !inode.isDir() { return nil, nil, syserror.ENOTDIR } @@ -100,7 +100,7 @@ func stepLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, inode *inode, write boo } d := vfsd.Impl().(*dentry) if name == ".." { - isRoot, err := rp.CheckRoot(vfsd) + isRoot, err := rp.CheckRoot(ctx, vfsd) if err != nil { return nil, nil, err } @@ -108,7 +108,7 @@ func stepLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, inode *inode, write boo rp.Advance() return vfsd, inode, nil } - if err := rp.CheckMount(&d.parent.vfsd); err != nil { + if err := rp.CheckMount(ctx, &d.parent.vfsd); err != nil { return nil, nil, err } rp.Advance() @@ -143,7 +143,7 @@ func stepLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, inode *inode, write boo child.name = name dir.childCache[name] = child } - if err := rp.CheckMount(&child.vfsd); err != nil { + if err := rp.CheckMount(ctx, &child.vfsd); err != nil { return nil, nil, err } if child.inode.isSymlink() && rp.ShouldFollowSymlink() { @@ -167,12 +167,12 @@ func stepLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, inode *inode, write boo // // Preconditions: // - filesystem.mu must be locked (for writing if write param is true). -func walkLocked(rp *vfs.ResolvingPath, write bool) (*vfs.Dentry, *inode, error) { +func walkLocked(ctx context.Context, rp *vfs.ResolvingPath, write bool) (*vfs.Dentry, *inode, error) { vfsd := rp.Start() inode := vfsd.Impl().(*dentry).inode for !rp.Done() { var err error - vfsd, inode, err = stepLocked(rp, vfsd, inode, write) + vfsd, inode, err = stepLocked(ctx, rp, vfsd, inode, write) if err != nil { return nil, nil, err } @@ -196,12 +196,12 @@ func walkLocked(rp *vfs.ResolvingPath, write bool) (*vfs.Dentry, *inode, error) // Preconditions: // - filesystem.mu must be locked (for writing if write param is true). // - !rp.Done(). -func walkParentLocked(rp *vfs.ResolvingPath, write bool) (*vfs.Dentry, *inode, error) { +func walkParentLocked(ctx context.Context, rp *vfs.ResolvingPath, write bool) (*vfs.Dentry, *inode, error) { vfsd := rp.Start() inode := vfsd.Impl().(*dentry).inode for !rp.Final() { var err error - vfsd, inode, err = stepLocked(rp, vfsd, inode, write) + vfsd, inode, err = stepLocked(ctx, rp, vfsd, inode, write) if err != nil { return nil, nil, err } @@ -216,7 +216,7 @@ func walkParentLocked(rp *vfs.ResolvingPath, write bool) (*vfs.Dentry, *inode, e // the rp till the parent of the last component which should be an existing // directory. If parent is false then resolves rp entirely. Attemps to resolve // the path as far as it can with a read lock and upgrades the lock if needed. -func (fs *filesystem) walk(rp *vfs.ResolvingPath, parent bool) (*vfs.Dentry, *inode, error) { +func (fs *filesystem) walk(ctx context.Context, rp *vfs.ResolvingPath, parent bool) (*vfs.Dentry, *inode, error) { var ( vfsd *vfs.Dentry inode *inode @@ -227,9 +227,9 @@ func (fs *filesystem) walk(rp *vfs.ResolvingPath, parent bool) (*vfs.Dentry, *in // of disk. This reduces congestion (allows concurrent walks). fs.mu.RLock() if parent { - vfsd, inode, err = walkParentLocked(rp, false) + vfsd, inode, err = walkParentLocked(ctx, rp, false) } else { - vfsd, inode, err = walkLocked(rp, false) + vfsd, inode, err = walkLocked(ctx, rp, false) } fs.mu.RUnlock() @@ -238,9 +238,9 @@ func (fs *filesystem) walk(rp *vfs.ResolvingPath, parent bool) (*vfs.Dentry, *in // walk is fine as this is a read only filesystem. fs.mu.Lock() if parent { - vfsd, inode, err = walkParentLocked(rp, true) + vfsd, inode, err = walkParentLocked(ctx, rp, true) } else { - vfsd, inode, err = walkLocked(rp, true) + vfsd, inode, err = walkLocked(ctx, rp, true) } fs.mu.Unlock() } @@ -283,7 +283,7 @@ func (fs *filesystem) statTo(stat *linux.Statfs) { // AccessAt implements vfs.Filesystem.Impl.AccessAt. func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { - _, inode, err := fs.walk(rp, false) + _, inode, err := fs.walk(ctx, rp, false) if err != nil { return err } @@ -292,7 +292,7 @@ func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds // GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { - vfsd, inode, err := fs.walk(rp, false) + vfsd, inode, err := fs.walk(ctx, rp, false) if err != nil { return nil, err } @@ -312,7 +312,7 @@ func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op // GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { - vfsd, inode, err := fs.walk(rp, true) + vfsd, inode, err := fs.walk(ctx, rp, true) if err != nil { return nil, err } @@ -322,7 +322,7 @@ func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPa // OpenAt implements vfs.FilesystemImpl.OpenAt. func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.OpenOptions) (*vfs.FileDescription, error) { - vfsd, inode, err := fs.walk(rp, false) + vfsd, inode, err := fs.walk(ctx, rp, false) if err != nil { return nil, err } @@ -336,7 +336,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf // ReadlinkAt implements vfs.FilesystemImpl.ReadlinkAt. func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { - _, inode, err := fs.walk(rp, false) + _, inode, err := fs.walk(ctx, rp, false) if err != nil { return "", err } @@ -349,7 +349,7 @@ func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (st // StatAt implements vfs.FilesystemImpl.StatAt. func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { - _, inode, err := fs.walk(rp, false) + _, inode, err := fs.walk(ctx, rp, false) if err != nil { return linux.Statx{}, err } @@ -360,7 +360,7 @@ func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf // StatFSAt implements vfs.FilesystemImpl.StatFSAt. func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { - if _, _, err := fs.walk(rp, false); err != nil { + if _, _, err := fs.walk(ctx, rp, false); err != nil { return linux.Statfs{}, err } @@ -370,7 +370,7 @@ func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.vfsfs.VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) } @@ -390,7 +390,7 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. return syserror.EEXIST } - if _, _, err := fs.walk(rp, true); err != nil { + if _, _, err := fs.walk(ctx, rp, true); err != nil { return err } @@ -403,7 +403,7 @@ func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v return syserror.EEXIST } - if _, _, err := fs.walk(rp, true); err != nil { + if _, _, err := fs.walk(ctx, rp, true); err != nil { return err } @@ -416,7 +416,7 @@ func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v return syserror.EEXIST } - _, _, err := fs.walk(rp, true) + _, _, err := fs.walk(ctx, rp, true) if err != nil { return err } @@ -430,7 +430,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa return syserror.ENOENT } - _, _, err := fs.walk(rp, false) + _, _, err := fs.walk(ctx, rp, false) if err != nil { return err } @@ -440,7 +440,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa // RmdirAt implements vfs.FilesystemImpl.RmdirAt. func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { - _, inode, err := fs.walk(rp, false) + _, inode, err := fs.walk(ctx, rp, false) if err != nil { return err } @@ -454,7 +454,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error // SetStatAt implements vfs.FilesystemImpl.SetStatAt. func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { - _, _, err := fs.walk(rp, false) + _, _, err := fs.walk(ctx, rp, false) if err != nil { return err } @@ -468,7 +468,7 @@ func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ return syserror.EEXIST } - _, _, err := fs.walk(rp, true) + _, _, err := fs.walk(ctx, rp, true) if err != nil { return err } @@ -478,7 +478,7 @@ func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ // UnlinkAt implements vfs.FilesystemImpl.UnlinkAt. func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { - _, inode, err := fs.walk(rp, false) + _, inode, err := fs.walk(ctx, rp, false) if err != nil { return err } @@ -492,7 +492,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error // BoundEndpointAt implements FilesystemImpl.BoundEndpointAt. func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { - _, inode, err := fs.walk(rp, false) + _, inode, err := fs.walk(ctx, rp, false) if err != nil { return nil, err } @@ -506,7 +506,7 @@ func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath // ListxattrAt implements vfs.FilesystemImpl.ListxattrAt. func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) { - _, _, err := fs.walk(rp, false) + _, _, err := fs.walk(ctx, rp, false) if err != nil { return nil, err } @@ -515,7 +515,7 @@ func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, si // GetxattrAt implements vfs.FilesystemImpl.GetxattrAt. func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) { - _, _, err := fs.walk(rp, false) + _, _, err := fs.walk(ctx, rp, false) if err != nil { return "", err } @@ -524,7 +524,7 @@ func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt // SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { - _, _, err := fs.walk(rp, false) + _, _, err := fs.walk(ctx, rp, false) if err != nil { return err } @@ -533,7 +533,7 @@ func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt // RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { - _, _, err := fs.walk(rp, false) + _, _, err := fs.walk(ctx, rp, false) if err != nil { return err } diff --git a/pkg/sentry/fsimpl/ext/regular_file.go b/pkg/sentry/fsimpl/ext/regular_file.go index 66d14bb95..e73e740d6 100644 --- a/pkg/sentry/fsimpl/ext/regular_file.go +++ b/pkg/sentry/fsimpl/ext/regular_file.go @@ -79,7 +79,7 @@ type regularFileFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *regularFileFD) Release() {} +func (fd *regularFileFD) Release(context.Context) {} // PRead implements vfs.FileDescriptionImpl.PRead. func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { diff --git a/pkg/sentry/fsimpl/ext/symlink.go b/pkg/sentry/fsimpl/ext/symlink.go index 62efd4095..2fd0d1fa8 100644 --- a/pkg/sentry/fsimpl/ext/symlink.go +++ b/pkg/sentry/fsimpl/ext/symlink.go @@ -73,7 +73,7 @@ type symlinkFD struct { var _ vfs.FileDescriptionImpl = (*symlinkFD)(nil) // Release implements vfs.FileDescriptionImpl.Release. -func (fd *symlinkFD) Release() {} +func (fd *symlinkFD) Release(context.Context) {} // PRead implements vfs.FileDescriptionImpl.PRead. func (fd *symlinkFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { diff --git a/pkg/sentry/fsimpl/fuse/dev.go b/pkg/sentry/fsimpl/fuse/dev.go index 2225076bc..e522ff9a0 100644 --- a/pkg/sentry/fsimpl/fuse/dev.go +++ b/pkg/sentry/fsimpl/fuse/dev.go @@ -99,7 +99,7 @@ type DeviceFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *DeviceFD) Release() { +func (fd *DeviceFD) Release(context.Context) { fd.fs.conn.connected = false } diff --git a/pkg/sentry/fsimpl/fuse/dev_test.go b/pkg/sentry/fsimpl/fuse/dev_test.go index 84c222ad6..1ffe7ccd2 100644 --- a/pkg/sentry/fsimpl/fuse/dev_test.go +++ b/pkg/sentry/fsimpl/fuse/dev_test.go @@ -356,12 +356,12 @@ func newTestConnection(system *testutil.System, k *kernel.Kernel, maxActiveReque vfsObj := &vfs.VirtualFilesystem{} fuseDev := &DeviceFD{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(system.Ctx); err != nil { return nil, nil, err } vd := vfsObj.NewAnonVirtualDentry("genCountFD") - defer vd.DecRef() + defer vd.DecRef(system.Ctx) if err := fuseDev.vfsfd.Init(fuseDev, linux.O_RDWR|linux.O_CREAT, vd.Mount(), vd.Dentry(), &vfs.FileDescriptionOptions{}); err != nil { return nil, nil, err } diff --git a/pkg/sentry/fsimpl/fuse/fusefs.go b/pkg/sentry/fsimpl/fuse/fusefs.go index 200a93bbf..a1405f7c3 100644 --- a/pkg/sentry/fsimpl/fuse/fusefs.go +++ b/pkg/sentry/fsimpl/fuse/fusefs.go @@ -191,9 +191,9 @@ func NewFUSEFilesystem(ctx context.Context, devMinor uint32, opts *filesystemOpt } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) - fs.Filesystem.Release() + fs.Filesystem.Release(ctx) } // inode implements kernfs.Inode. diff --git a/pkg/sentry/fsimpl/gofer/directory.go b/pkg/sentry/fsimpl/gofer/directory.go index 8c7c8e1b3..1679066ba 100644 --- a/pkg/sentry/fsimpl/gofer/directory.go +++ b/pkg/sentry/fsimpl/gofer/directory.go @@ -122,7 +122,7 @@ type directoryFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *directoryFD) Release() { +func (fd *directoryFD) Release(context.Context) { } // IterDirents implements vfs.FileDescriptionImpl.IterDirents. @@ -139,7 +139,7 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba fd.dirents = ds } - d.InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + d.InotifyWithParent(ctx, linux.IN_ACCESS, 0, vfs.PathEvent) if d.cachedMetadataAuthoritative() { d.touchAtime(fd.vfsfd.Mount()) } diff --git a/pkg/sentry/fsimpl/gofer/filesystem.go b/pkg/sentry/fsimpl/gofer/filesystem.go index 00e3c99cd..e6af37d0d 100644 --- a/pkg/sentry/fsimpl/gofer/filesystem.go +++ b/pkg/sentry/fsimpl/gofer/filesystem.go @@ -55,7 +55,7 @@ func (fs *filesystem) Sync(ctx context.Context) error { // Sync regular files. for _, d := range ds { err := d.syncSharedHandle(ctx) - d.DecRef() + d.DecRef(ctx) if err != nil && retErr == nil { retErr = err } @@ -65,7 +65,7 @@ func (fs *filesystem) Sync(ctx context.Context) error { // handles (so they won't be synced by the above). for _, sffd := range sffds { err := sffd.Sync(ctx) - sffd.vfsfd.DecRef() + sffd.vfsfd.DecRef(ctx) if err != nil && retErr == nil { retErr = err } @@ -133,7 +133,7 @@ afterSymlink: return d, nil } if name == ".." { - if isRoot, err := rp.CheckRoot(&d.vfsd); err != nil { + if isRoot, err := rp.CheckRoot(ctx, &d.vfsd); err != nil { return nil, err } else if isRoot || d.parent == nil { rp.Advance() @@ -146,7 +146,7 @@ afterSymlink: // // Call rp.CheckMount() before updating d.parent's metadata, since if // we traverse to another mount then d.parent's metadata is irrelevant. - if err := rp.CheckMount(&d.parent.vfsd); err != nil { + if err := rp.CheckMount(ctx, &d.parent.vfsd); err != nil { return nil, err } if d != d.parent && !d.cachedMetadataAuthoritative() { @@ -164,7 +164,7 @@ afterSymlink: if child == nil { return nil, syserror.ENOENT } - if err := rp.CheckMount(&child.vfsd); err != nil { + if err := rp.CheckMount(ctx, &child.vfsd); err != nil { return nil, err } if child.isSymlink() && mayFollowSymlinks && rp.ShouldFollowSymlink() { @@ -239,7 +239,7 @@ func (fs *filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir // has 0 references, drop it). Wait to update parent.children until we // know what to replace the existing dentry with (i.e. one of the // returns below), to avoid a redundant map access. - vfsObj.InvalidateDentry(&child.vfsd) + vfsObj.InvalidateDentry(ctx, &child.vfsd) if child.isSynthetic() { // Normally we don't mark invalidated dentries as deleted since // they may still exist (but at a different path), and also for @@ -332,7 +332,7 @@ func (fs *filesystem) resolveLocked(ctx context.Context, rp *vfs.ResolvingPath, func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, createInRemoteDir func(parent *dentry, name string) error, createInSyntheticDir func(parent *dentry, name string) error) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) start := rp.Start().Impl().(*dentry) if !start.cachedMetadataAuthoritative() { // Get updated metadata for start as required by @@ -384,7 +384,7 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir if dir { ev |= linux.IN_ISDIR } - parent.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) + parent.watches.Notify(ctx, name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) return nil } if fs.opts.interop == InteropModeShared { @@ -405,7 +405,7 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir if dir { ev |= linux.IN_ISDIR } - parent.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) + parent.watches.Notify(ctx, name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) return nil } if child := parent.children[name]; child != nil { @@ -426,7 +426,7 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir if dir { ev |= linux.IN_ISDIR } - parent.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) + parent.watches.Notify(ctx, name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) return nil } @@ -434,7 +434,7 @@ func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) start := rp.Start().Impl().(*dentry) if !start.cachedMetadataAuthoritative() { // Get updated metadata for start as required by @@ -470,7 +470,7 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b } vfsObj := rp.VirtualFilesystem() mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) parent.dirMu.Lock() defer parent.dirMu.Unlock() @@ -600,17 +600,17 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b // Generate inotify events for rmdir or unlink. if dir { - parent.watches.Notify(name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent, true /* unlinked */) + parent.watches.Notify(ctx, name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent, true /* unlinked */) } else { var cw *vfs.Watches if child != nil { cw = &child.watches } - vfs.InotifyRemoveChild(cw, &parent.watches, name) + vfs.InotifyRemoveChild(ctx, cw, &parent.watches, name) } if child != nil { - vfsObj.CommitDeleteDentry(&child.vfsd) + vfsObj.CommitDeleteDentry(ctx, &child.vfsd) child.setDeleted() if child.isSynthetic() { parent.syntheticChildren-- @@ -637,7 +637,7 @@ func (fs *filesystem) unlinkAt(ctx context.Context, rp *vfs.ResolvingPath, dir b // but dentry slices are allocated lazily, and it's much easier to say "defer // fs.renameMuRUnlockAndCheckCaching(&ds)" than "defer func() { // fs.renameMuRUnlockAndCheckCaching(ds) }()" to work around this. -func (fs *filesystem) renameMuRUnlockAndCheckCaching(ds **[]*dentry) { +func (fs *filesystem) renameMuRUnlockAndCheckCaching(ctx context.Context, ds **[]*dentry) { fs.renameMu.RUnlock() if *ds == nil { return @@ -645,20 +645,20 @@ func (fs *filesystem) renameMuRUnlockAndCheckCaching(ds **[]*dentry) { if len(**ds) != 0 { fs.renameMu.Lock() for _, d := range **ds { - d.checkCachingLocked() + d.checkCachingLocked(ctx) } fs.renameMu.Unlock() } putDentrySlice(*ds) } -func (fs *filesystem) renameMuUnlockAndCheckCaching(ds **[]*dentry) { +func (fs *filesystem) renameMuUnlockAndCheckCaching(ctx context.Context, ds **[]*dentry) { if *ds == nil { fs.renameMu.Unlock() return } for _, d := range **ds { - d.checkCachingLocked() + d.checkCachingLocked(ctx) } fs.renameMu.Unlock() putDentrySlice(*ds) @@ -668,7 +668,7 @@ func (fs *filesystem) renameMuUnlockAndCheckCaching(ds **[]*dentry) { func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return err @@ -680,7 +680,7 @@ func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return nil, err @@ -701,7 +701,7 @@ func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) start := rp.Start().Impl().(*dentry) if !start.cachedMetadataAuthoritative() { // Get updated metadata for start as required by @@ -812,7 +812,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) start := rp.Start().Impl().(*dentry) if !start.cachedMetadataAuthoritative() { @@ -1126,7 +1126,7 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving } childVFSFD = &fd.vfsfd } - d.watches.Notify(name, linux.IN_CREATE, 0, vfs.PathEvent, false /* unlinked */) + d.watches.Notify(ctx, name, linux.IN_CREATE, 0, vfs.PathEvent, false /* unlinked */) return childVFSFD, nil } @@ -1134,7 +1134,7 @@ func (d *dentry) createAndOpenChildLocked(ctx context.Context, rp *vfs.Resolving func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return "", err @@ -1154,7 +1154,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa var ds *[]*dentry fs.renameMu.Lock() - defer fs.renameMuUnlockAndCheckCaching(&ds) + defer fs.renameMuUnlockAndCheckCaching(ctx, &ds) newParent, err := fs.walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry), &ds) if err != nil { return err @@ -1244,7 +1244,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa return nil } mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) if err := vfsObj.PrepareRenameDentry(mntns, &renamed.vfsd, replacedVFSD); err != nil { return err } @@ -1269,7 +1269,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } // Update the dentry tree. - vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, replacedVFSD) + vfsObj.CommitRenameReplaceDentry(ctx, &renamed.vfsd, replacedVFSD) if replaced != nil { replaced.setDeleted() if replaced.isSynthetic() { @@ -1331,17 +1331,17 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts fs.renameMu.RLock() d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } if err := d.setStat(ctx, rp.Credentials(), &opts, rp.Mount()); err != nil { - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { - d.InotifyWithParent(ev, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, ev, 0, vfs.InodeEvent) } return nil } @@ -1350,7 +1350,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return linux.Statx{}, err @@ -1367,7 +1367,7 @@ func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return linux.Statfs{}, err @@ -1417,7 +1417,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return nil, err @@ -1443,7 +1443,7 @@ func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return nil, err @@ -1455,7 +1455,7 @@ func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, si func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckCaching(&ds) + defer fs.renameMuRUnlockAndCheckCaching(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return "", err @@ -1469,16 +1469,16 @@ func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt fs.renameMu.RLock() d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } if err := d.setxattr(ctx, rp.Credentials(), &opts); err != nil { - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) - d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } @@ -1488,16 +1488,16 @@ func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, fs.renameMu.RLock() d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } if err := d.removexattr(ctx, rp.Credentials(), name); err != nil { - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) return err } - fs.renameMuRUnlockAndCheckCaching(&ds) + fs.renameMuRUnlockAndCheckCaching(ctx, &ds) - d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go index e20de84b5..2e5575d8d 100644 --- a/pkg/sentry/fsimpl/gofer/gofer.go +++ b/pkg/sentry/fsimpl/gofer/gofer.go @@ -482,7 +482,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt root, err := fs.newDentry(ctx, attachFile, qid, attrMask, &attr) if err != nil { attachFile.close(ctx) - fs.vfsfs.DecRef() + fs.vfsfs.DecRef(ctx) return nil, nil, err } // Set the root's reference count to 2. One reference is returned to the @@ -495,8 +495,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { - ctx := context.Background() +func (fs *filesystem) Release(ctx context.Context) { mf := fs.mfp.MemoryFile() fs.syncMu.Lock() @@ -1089,10 +1088,10 @@ func (d *dentry) TryIncRef() bool { } // DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef() { +func (d *dentry) DecRef(ctx context.Context) { if refs := atomic.AddInt64(&d.refs, -1); refs == 0 { d.fs.renameMu.Lock() - d.checkCachingLocked() + d.checkCachingLocked(ctx) d.fs.renameMu.Unlock() } else if refs < 0 { panic("gofer.dentry.DecRef() called without holding a reference") @@ -1109,7 +1108,7 @@ func (d *dentry) decRefLocked() { } // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. -func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) { +func (d *dentry) InotifyWithParent(ctx context.Context, events, cookie uint32, et vfs.EventType) { if d.isDir() { events |= linux.IN_ISDIR } @@ -1117,9 +1116,9 @@ func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) { d.fs.renameMu.RLock() // The ordering below is important, Linux always notifies the parent first. if d.parent != nil { - d.parent.watches.Notify(d.name, events, cookie, et, d.isDeleted()) + d.parent.watches.Notify(ctx, d.name, events, cookie, et, d.isDeleted()) } - d.watches.Notify("", events, cookie, et, d.isDeleted()) + d.watches.Notify(ctx, "", events, cookie, et, d.isDeleted()) d.fs.renameMu.RUnlock() } @@ -1131,10 +1130,10 @@ func (d *dentry) Watches() *vfs.Watches { // OnZeroWatches implements vfs.DentryImpl.OnZeroWatches. // // If no watches are left on this dentry and it has no references, cache it. -func (d *dentry) OnZeroWatches() { +func (d *dentry) OnZeroWatches(ctx context.Context) { if atomic.LoadInt64(&d.refs) == 0 { d.fs.renameMu.Lock() - d.checkCachingLocked() + d.checkCachingLocked(ctx) d.fs.renameMu.Unlock() } } @@ -1149,7 +1148,7 @@ func (d *dentry) OnZeroWatches() { // do nothing. // // Preconditions: d.fs.renameMu must be locked for writing. -func (d *dentry) checkCachingLocked() { +func (d *dentry) checkCachingLocked(ctx context.Context) { // Dentries with a non-zero reference count must be retained. (The only way // to obtain a reference on a dentry with zero references is via path // resolution, which requires renameMu, so if d.refs is zero then it will @@ -1171,14 +1170,14 @@ func (d *dentry) checkCachingLocked() { // reachable by path resolution and should be dropped immediately. if d.vfsd.IsDead() { if d.isDeleted() { - d.watches.HandleDeletion() + d.watches.HandleDeletion(ctx) } if d.cached { d.fs.cachedDentries.Remove(d) d.fs.cachedDentriesLen-- d.cached = false } - d.destroyLocked() + d.destroyLocked(ctx) return } // If d still has inotify watches and it is not deleted or invalidated, we @@ -1213,7 +1212,7 @@ func (d *dentry) checkCachingLocked() { if !victim.vfsd.IsDead() { // Note that victim can't be a mount point (in any mount // namespace), since VFS holds references on mount points. - d.fs.vfsfs.VirtualFilesystem().InvalidateDentry(&victim.vfsd) + d.fs.vfsfs.VirtualFilesystem().InvalidateDentry(ctx, &victim.vfsd) delete(victim.parent.children, victim.name) // We're only deleting the dentry, not the file it // represents, so we don't need to update @@ -1221,7 +1220,7 @@ func (d *dentry) checkCachingLocked() { } victim.parent.dirMu.Unlock() } - victim.destroyLocked() + victim.destroyLocked(ctx) } // Whether or not victim was destroyed, we brought fs.cachedDentriesLen // back down to fs.opts.maxCachedDentries, so we don't loop. @@ -1233,7 +1232,7 @@ func (d *dentry) checkCachingLocked() { // // Preconditions: d.fs.renameMu must be locked for writing. d.refs == 0. d is // not a child dentry. -func (d *dentry) destroyLocked() { +func (d *dentry) destroyLocked(ctx context.Context) { switch atomic.LoadInt64(&d.refs) { case 0: // Mark the dentry destroyed. @@ -1244,7 +1243,6 @@ func (d *dentry) destroyLocked() { panic("dentry.destroyLocked() called with references on the dentry") } - ctx := context.Background() d.handleMu.Lock() if !d.handle.file.isNil() { mf := d.fs.mfp.MemoryFile() @@ -1276,7 +1274,7 @@ func (d *dentry) destroyLocked() { // d.fs.renameMu. if d.parent != nil { if refs := atomic.AddInt64(&d.parent.refs, -1); refs == 0 { - d.parent.checkCachingLocked() + d.parent.checkCachingLocked(ctx) } else if refs < 0 { panic("gofer.dentry.DecRef() called without holding a reference") } @@ -1514,7 +1512,7 @@ func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) return err } if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { - fd.dentry().InotifyWithParent(ev, 0, vfs.InodeEvent) + fd.dentry().InotifyWithParent(ctx, ev, 0, vfs.InodeEvent) } return nil } @@ -1535,7 +1533,7 @@ func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOption if err := d.setxattr(ctx, auth.CredentialsFromContext(ctx), &opts); err != nil { return err } - d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } @@ -1545,7 +1543,7 @@ func (fd *fileDescription) Removexattr(ctx context.Context, name string) error { if err := d.removexattr(ctx, auth.CredentialsFromContext(ctx), name); err != nil { return err } - d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } diff --git a/pkg/sentry/fsimpl/gofer/gofer_test.go b/pkg/sentry/fsimpl/gofer/gofer_test.go index adff39490..56d80bcf8 100644 --- a/pkg/sentry/fsimpl/gofer/gofer_test.go +++ b/pkg/sentry/fsimpl/gofer/gofer_test.go @@ -50,7 +50,7 @@ func TestDestroyIdempotent(t *testing.T) { } parent.cacheNewChildLocked(child, "child") - child.checkCachingLocked() + child.checkCachingLocked(ctx) if got := atomic.LoadInt64(&child.refs); got != -1 { t.Fatalf("child.refs=%d, want: -1", got) } @@ -58,6 +58,6 @@ func TestDestroyIdempotent(t *testing.T) { if got := atomic.LoadInt64(&parent.refs); got != -1 { t.Fatalf("parent.refs=%d, want: -1", got) } - child.checkCachingLocked() - child.checkCachingLocked() + child.checkCachingLocked(ctx) + child.checkCachingLocked(ctx) } diff --git a/pkg/sentry/fsimpl/gofer/regular_file.go b/pkg/sentry/fsimpl/gofer/regular_file.go index 09f142cfc..420e8efe2 100644 --- a/pkg/sentry/fsimpl/gofer/regular_file.go +++ b/pkg/sentry/fsimpl/gofer/regular_file.go @@ -48,7 +48,7 @@ type regularFileFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *regularFileFD) Release() { +func (fd *regularFileFD) Release(context.Context) { } // OnClose implements vfs.FileDescriptionImpl.OnClose. diff --git a/pkg/sentry/fsimpl/gofer/socket.go b/pkg/sentry/fsimpl/gofer/socket.go index d6dbe9092..85d2bee72 100644 --- a/pkg/sentry/fsimpl/gofer/socket.go +++ b/pkg/sentry/fsimpl/gofer/socket.go @@ -108,7 +108,7 @@ func (e *endpoint) UnidirectionalConnect(ctx context.Context) (transport.Connect // We don't need the receiver. c.CloseRecv() - c.Release() + c.Release(ctx) return c, nil } @@ -136,8 +136,8 @@ func (e *endpoint) newConnectedEndpoint(ctx context.Context, flags p9.ConnectFla } // Release implements transport.BoundEndpoint.Release. -func (e *endpoint) Release() { - e.dentry.DecRef() +func (e *endpoint) Release(ctx context.Context) { + e.dentry.DecRef(ctx) } // Passcred implements transport.BoundEndpoint.Passcred. diff --git a/pkg/sentry/fsimpl/gofer/special_file.go b/pkg/sentry/fsimpl/gofer/special_file.go index 811528982..fc269ef2b 100644 --- a/pkg/sentry/fsimpl/gofer/special_file.go +++ b/pkg/sentry/fsimpl/gofer/special_file.go @@ -80,11 +80,11 @@ func newSpecialFileFD(h handle, mnt *vfs.Mount, d *dentry, locks *vfs.FileLocks, } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *specialFileFD) Release() { +func (fd *specialFileFD) Release(ctx context.Context) { if fd.haveQueue { fdnotifier.RemoveFD(fd.handle.fd) } - fd.handle.close(context.Background()) + fd.handle.close(ctx) fs := fd.vfsfd.Mount().Filesystem().Impl().(*filesystem) fs.syncMu.Lock() delete(fs.specialFileFDs, fd) diff --git a/pkg/sentry/fsimpl/host/control.go b/pkg/sentry/fsimpl/host/control.go index b9082a20f..0135e4428 100644 --- a/pkg/sentry/fsimpl/host/control.go +++ b/pkg/sentry/fsimpl/host/control.go @@ -58,7 +58,7 @@ func (c *scmRights) Clone() transport.RightsControlMessage { } // Release implements transport.RightsControlMessage.Release. -func (c *scmRights) Release() { +func (c *scmRights) Release(ctx context.Context) { for _, fd := range c.fds { syscall.Close(fd) } diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index c894f2ca0..bf922c566 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -117,7 +117,7 @@ func NewFD(ctx context.Context, mnt *vfs.Mount, hostFD int, opts *NewFDOptions) d.Init(i) // i.open will take a reference on d. - defer d.DecRef() + defer d.DecRef(ctx) // For simplicity, fileDescription.offset is set to 0. Technically, we // should only set to 0 on files that are not seekable (sockets, pipes, @@ -168,9 +168,9 @@ type filesystem struct { devMinor uint32 } -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) - fs.Filesystem.Release() + fs.Filesystem.Release(ctx) } func (fs *filesystem) PrependPath(ctx context.Context, vfsroot, vd vfs.VirtualDentry, b *fspath.Builder) error { @@ -431,12 +431,12 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre } // DecRef implements kernfs.Inode. -func (i *inode) DecRef() { - i.AtomicRefCount.DecRefWithDestructor(i.Destroy) +func (i *inode) DecRef(ctx context.Context) { + i.AtomicRefCount.DecRefWithDestructor(ctx, i.Destroy) } // Destroy implements kernfs.Inode. -func (i *inode) Destroy() { +func (i *inode) Destroy(context.Context) { if i.wouldBlock { fdnotifier.RemoveFD(int32(i.hostFD)) } @@ -542,7 +542,7 @@ func (f *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux } // Release implements vfs.FileDescriptionImpl. -func (f *fileDescription) Release() { +func (f *fileDescription) Release(context.Context) { // noop } diff --git a/pkg/sentry/fsimpl/host/socket.go b/pkg/sentry/fsimpl/host/socket.go index fd16bd92d..4979dd0a9 100644 --- a/pkg/sentry/fsimpl/host/socket.go +++ b/pkg/sentry/fsimpl/host/socket.go @@ -139,7 +139,7 @@ func NewConnectedEndpoint(ctx context.Context, hostFD int, addr string, saveable } // Send implements transport.ConnectedEndpoint.Send. -func (c *ConnectedEndpoint) Send(data [][]byte, controlMessages transport.ControlMessages, from tcpip.FullAddress) (int64, bool, *syserr.Error) { +func (c *ConnectedEndpoint) Send(ctx context.Context, data [][]byte, controlMessages transport.ControlMessages, from tcpip.FullAddress) (int64, bool, *syserr.Error) { c.mu.RLock() defer c.mu.RUnlock() @@ -216,7 +216,7 @@ func (c *ConnectedEndpoint) EventUpdate() { } // Recv implements transport.Receiver.Recv. -func (c *ConnectedEndpoint) Recv(data [][]byte, creds bool, numRights int, peek bool) (int64, int64, transport.ControlMessages, bool, tcpip.FullAddress, bool, *syserr.Error) { +func (c *ConnectedEndpoint) Recv(ctx context.Context, data [][]byte, creds bool, numRights int, peek bool) (int64, int64, transport.ControlMessages, bool, tcpip.FullAddress, bool, *syserr.Error) { c.mu.RLock() defer c.mu.RUnlock() @@ -317,8 +317,8 @@ func (c *ConnectedEndpoint) destroyLocked() { // Release implements transport.ConnectedEndpoint.Release and // transport.Receiver.Release. -func (c *ConnectedEndpoint) Release() { - c.ref.DecRefWithDestructor(func() { +func (c *ConnectedEndpoint) Release(ctx context.Context) { + c.ref.DecRefWithDestructor(ctx, func(context.Context) { c.mu.Lock() c.destroyLocked() c.mu.Unlock() @@ -347,8 +347,8 @@ func (e *SCMConnectedEndpoint) Init() error { // Release implements transport.ConnectedEndpoint.Release and // transport.Receiver.Release. -func (e *SCMConnectedEndpoint) Release() { - e.ref.DecRefWithDestructor(func() { +func (e *SCMConnectedEndpoint) Release(ctx context.Context) { + e.ref.DecRefWithDestructor(ctx, func(context.Context) { e.mu.Lock() if err := syscall.Close(e.fd); err != nil { log.Warningf("Failed to close host fd %d: %v", err) diff --git a/pkg/sentry/fsimpl/host/tty.go b/pkg/sentry/fsimpl/host/tty.go index 4ee9270cc..d372c60cb 100644 --- a/pkg/sentry/fsimpl/host/tty.go +++ b/pkg/sentry/fsimpl/host/tty.go @@ -67,12 +67,12 @@ func (t *TTYFileDescription) ForegroundProcessGroup() *kernel.ProcessGroup { } // Release implements fs.FileOperations.Release. -func (t *TTYFileDescription) Release() { +func (t *TTYFileDescription) Release(ctx context.Context) { t.mu.Lock() t.fgProcessGroup = nil t.mu.Unlock() - t.fileDescription.Release() + t.fileDescription.Release(ctx) } // PRead implements vfs.FileDescriptionImpl. diff --git a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go index c6c4472e7..12adf727a 100644 --- a/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go +++ b/pkg/sentry/fsimpl/kernfs/dynamic_bytes_file.go @@ -122,7 +122,7 @@ func (fd *DynamicBytesFD) PWrite(ctx context.Context, src usermem.IOSequence, of } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *DynamicBytesFD) Release() {} +func (fd *DynamicBytesFD) Release(context.Context) {} // Stat implements vfs.FileDescriptionImpl.Stat. func (fd *DynamicBytesFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { diff --git a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go index 1d37ccb98..fcee6200a 100644 --- a/pkg/sentry/fsimpl/kernfs/fd_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/fd_impl_util.go @@ -113,7 +113,7 @@ func (fd *GenericDirectoryFD) PWrite(ctx context.Context, src usermem.IOSequence } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *GenericDirectoryFD) Release() {} +func (fd *GenericDirectoryFD) Release(context.Context) {} func (fd *GenericDirectoryFD) filesystem() *vfs.Filesystem { return fd.vfsfd.VirtualDentry().Mount().Filesystem() diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 61a36cff9..d7edb6342 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -56,13 +56,13 @@ afterSymlink: return vfsd, nil } if name == ".." { - if isRoot, err := rp.CheckRoot(vfsd); err != nil { + if isRoot, err := rp.CheckRoot(ctx, vfsd); err != nil { return nil, err } else if isRoot || d.parent == nil { rp.Advance() return vfsd, nil } - if err := rp.CheckMount(&d.parent.vfsd); err != nil { + if err := rp.CheckMount(ctx, &d.parent.vfsd); err != nil { return nil, err } rp.Advance() @@ -77,7 +77,7 @@ afterSymlink: if err != nil { return nil, err } - if err := rp.CheckMount(&next.vfsd); err != nil { + if err := rp.CheckMount(ctx, &next.vfsd); err != nil { return nil, err } // Resolve any symlink at current path component. @@ -88,7 +88,7 @@ afterSymlink: } if targetVD.Ok() { err := rp.HandleJump(targetVD) - targetVD.DecRef() + targetVD.DecRef(ctx) if err != nil { return nil, err } @@ -116,7 +116,7 @@ func (fs *Filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir // Cached dentry exists, revalidate. if !child.inode.Valid(ctx) { delete(parent.children, name) - vfsObj.InvalidateDentry(&child.vfsd) + vfsObj.InvalidateDentry(ctx, &child.vfsd) fs.deferDecRef(&child.vfsd) // Reference from Lookup. child = nil } @@ -234,7 +234,7 @@ func checkDeleteLocked(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Den } // Release implements vfs.FilesystemImpl.Release. -func (fs *Filesystem) Release() { +func (fs *Filesystem) Release(context.Context) { } // Sync implements vfs.FilesystemImpl.Sync. @@ -246,7 +246,7 @@ func (fs *Filesystem) Sync(ctx context.Context) error { // AccessAt implements vfs.Filesystem.Impl.AccessAt. func (fs *Filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { fs.mu.RLock() - defer fs.processDeferredDecRefs() + defer fs.processDeferredDecRefs(ctx) defer fs.mu.RUnlock() _, inode, err := fs.walkExistingLocked(ctx, rp) @@ -259,7 +259,7 @@ func (fs *Filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds // GetDentryAt implements vfs.FilesystemImpl.GetDentryAt. func (fs *Filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { fs.mu.RLock() - defer fs.processDeferredDecRefs() + defer fs.processDeferredDecRefs(ctx) defer fs.mu.RUnlock() vfsd, inode, err := fs.walkExistingLocked(ctx, rp) if err != nil { @@ -282,7 +282,7 @@ func (fs *Filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op // GetParentDentryAt implements vfs.FilesystemImpl.GetParentDentryAt. func (fs *Filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { fs.mu.RLock() - defer fs.processDeferredDecRefs() + defer fs.processDeferredDecRefs(ctx) defer fs.mu.RUnlock() vfsd, _, err := fs.walkParentDirLocked(ctx, rp) if err != nil { @@ -300,7 +300,7 @@ func (fs *Filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. fs.mu.Lock() defer fs.mu.Unlock() parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) - fs.processDeferredDecRefsLocked() + fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } @@ -337,7 +337,7 @@ func (fs *Filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v fs.mu.Lock() defer fs.mu.Unlock() parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) - fs.processDeferredDecRefsLocked() + fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } @@ -365,7 +365,7 @@ func (fs *Filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v fs.mu.Lock() defer fs.mu.Unlock() parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) - fs.processDeferredDecRefsLocked() + fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } @@ -397,7 +397,7 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf // Do not create new file. if opts.Flags&linux.O_CREAT == 0 { fs.mu.RLock() - defer fs.processDeferredDecRefs() + defer fs.processDeferredDecRefs(ctx) defer fs.mu.RUnlock() vfsd, inode, err := fs.walkExistingLocked(ctx, rp) if err != nil { @@ -429,7 +429,7 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf } afterTrailingSymlink: parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) - fs.processDeferredDecRefsLocked() + fs.processDeferredDecRefsLocked(ctx) if err != nil { return nil, err } @@ -483,7 +483,7 @@ afterTrailingSymlink: } if targetVD.Ok() { err := rp.HandleJump(targetVD) - targetVD.DecRef() + targetVD.DecRef(ctx) if err != nil { return nil, err } @@ -507,7 +507,7 @@ func (fs *Filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (st fs.mu.RLock() d, inode, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return "", err } @@ -526,7 +526,7 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa noReplace := opts.Flags&linux.RENAME_NOREPLACE != 0 fs.mu.Lock() - defer fs.processDeferredDecRefsLocked() + defer fs.processDeferredDecRefsLocked(ctx) defer fs.mu.Unlock() // Resolve the destination directory first to verify that it's on this @@ -584,7 +584,7 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) virtfs := rp.VirtualFilesystem() // We can't deadlock here due to lock ordering because we're protected from @@ -615,7 +615,7 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa dstDir.children = make(map[string]*Dentry) } dstDir.children[pc] = src - virtfs.CommitRenameReplaceDentry(srcVFSD, replaced) + virtfs.CommitRenameReplaceDentry(ctx, srcVFSD, replaced) return nil } @@ -624,7 +624,7 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error fs.mu.Lock() defer fs.mu.Unlock() vfsd, inode, err := fs.walkExistingLocked(ctx, rp) - fs.processDeferredDecRefsLocked() + fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } @@ -648,7 +648,7 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error defer parentDentry.dirMu.Unlock() mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) if err := virtfs.PrepareDeleteDentry(mntns, vfsd); err != nil { return err } @@ -656,7 +656,7 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error virtfs.AbortDeleteDentry(vfsd) return err } - virtfs.CommitDeleteDentry(vfsd) + virtfs.CommitDeleteDentry(ctx, vfsd) return nil } @@ -665,7 +665,7 @@ func (fs *Filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts fs.mu.RLock() _, inode, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return err } @@ -680,7 +680,7 @@ func (fs *Filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf fs.mu.RLock() _, inode, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return linux.Statx{}, err } @@ -692,7 +692,7 @@ func (fs *Filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu fs.mu.RLock() _, _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return linux.Statfs{}, err } @@ -708,7 +708,7 @@ func (fs *Filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ fs.mu.Lock() defer fs.mu.Unlock() parentVFSD, parentInode, err := fs.walkParentDirLocked(ctx, rp) - fs.processDeferredDecRefsLocked() + fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } @@ -733,7 +733,7 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error fs.mu.Lock() defer fs.mu.Unlock() vfsd, _, err := fs.walkExistingLocked(ctx, rp) - fs.processDeferredDecRefsLocked() + fs.processDeferredDecRefsLocked(ctx) if err != nil { return err } @@ -753,7 +753,7 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error parentDentry.dirMu.Lock() defer parentDentry.dirMu.Unlock() mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) if err := virtfs.PrepareDeleteDentry(mntns, vfsd); err != nil { return err } @@ -761,7 +761,7 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error virtfs.AbortDeleteDentry(vfsd) return err } - virtfs.CommitDeleteDentry(vfsd) + virtfs.CommitDeleteDentry(ctx, vfsd) return nil } @@ -770,7 +770,7 @@ func (fs *Filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath fs.mu.RLock() _, inode, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return nil, err } @@ -785,7 +785,7 @@ func (fs *Filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, si fs.mu.RLock() _, _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return nil, err } @@ -798,7 +798,7 @@ func (fs *Filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt fs.mu.RLock() _, _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return "", err } @@ -811,7 +811,7 @@ func (fs *Filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt fs.mu.RLock() _, _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return err } @@ -824,7 +824,7 @@ func (fs *Filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, fs.mu.RLock() _, _, err := fs.walkExistingLocked(ctx, rp) fs.mu.RUnlock() - fs.processDeferredDecRefs() + fs.processDeferredDecRefs(ctx) if err != nil { return err } diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 579e627f0..c3efcf3ec 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -40,7 +40,7 @@ func (InodeNoopRefCount) IncRef() { } // DecRef implements Inode.DecRef. -func (InodeNoopRefCount) DecRef() { +func (InodeNoopRefCount) DecRef(context.Context) { } // TryIncRef implements Inode.TryIncRef. @@ -49,7 +49,7 @@ func (InodeNoopRefCount) TryIncRef() bool { } // Destroy implements Inode.Destroy. -func (InodeNoopRefCount) Destroy() { +func (InodeNoopRefCount) Destroy(context.Context) { } // InodeDirectoryNoNewChildren partially implements the Inode interface. @@ -366,12 +366,12 @@ func (o *OrderedChildren) Init(opts OrderedChildrenOptions) { } // DecRef implements Inode.DecRef. -func (o *OrderedChildren) DecRef() { - o.AtomicRefCount.DecRefWithDestructor(o.Destroy) +func (o *OrderedChildren) DecRef(ctx context.Context) { + o.AtomicRefCount.DecRefWithDestructor(ctx, o.Destroy) } // Destroy cleans up resources referenced by this OrderedChildren. -func (o *OrderedChildren) Destroy() { +func (o *OrderedChildren) Destroy(context.Context) { o.mu.Lock() defer o.mu.Unlock() o.order.Reset() diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 46f207664..080118841 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -116,17 +116,17 @@ func (fs *Filesystem) deferDecRef(d *vfs.Dentry) { // processDeferredDecRefs calls vfs.Dentry.DecRef on all dentries in the // droppedDentries list. See comment on Filesystem.mu. -func (fs *Filesystem) processDeferredDecRefs() { +func (fs *Filesystem) processDeferredDecRefs(ctx context.Context) { fs.mu.Lock() - fs.processDeferredDecRefsLocked() + fs.processDeferredDecRefsLocked(ctx) fs.mu.Unlock() } // Precondition: fs.mu must be held for writing. -func (fs *Filesystem) processDeferredDecRefsLocked() { +func (fs *Filesystem) processDeferredDecRefsLocked(ctx context.Context) { fs.droppedDentriesMu.Lock() for _, d := range fs.droppedDentries { - d.DecRef() + d.DecRef(ctx) } fs.droppedDentries = fs.droppedDentries[:0] // Keep slice memory for reuse. fs.droppedDentriesMu.Unlock() @@ -212,16 +212,16 @@ func (d *Dentry) isSymlink() bool { } // DecRef implements vfs.DentryImpl.DecRef. -func (d *Dentry) DecRef() { - d.AtomicRefCount.DecRefWithDestructor(d.destroy) +func (d *Dentry) DecRef(ctx context.Context) { + d.AtomicRefCount.DecRefWithDestructor(ctx, d.destroy) } // Precondition: Dentry must be removed from VFS' dentry cache. -func (d *Dentry) destroy() { - d.inode.DecRef() // IncRef from Init. +func (d *Dentry) destroy(ctx context.Context) { + d.inode.DecRef(ctx) // IncRef from Init. d.inode = nil if d.parent != nil { - d.parent.DecRef() // IncRef from Dentry.InsertChild. + d.parent.DecRef(ctx) // IncRef from Dentry.InsertChild. } } @@ -230,7 +230,7 @@ func (d *Dentry) destroy() { // Although Linux technically supports inotify on pseudo filesystems (inotify // is implemented at the vfs layer), it is not particularly useful. It is left // unimplemented until someone actually needs it. -func (d *Dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) {} +func (d *Dentry) InotifyWithParent(ctx context.Context, events, cookie uint32, et vfs.EventType) {} // Watches implements vfs.DentryImpl.Watches. func (d *Dentry) Watches() *vfs.Watches { @@ -238,7 +238,7 @@ func (d *Dentry) Watches() *vfs.Watches { } // OnZeroWatches implements vfs.Dentry.OnZeroWatches. -func (d *Dentry) OnZeroWatches() {} +func (d *Dentry) OnZeroWatches(context.Context) {} // InsertChild inserts child into the vfs dentry cache with the given name under // this dentry. This does not update the directory inode, so calling this on @@ -326,12 +326,12 @@ type Inode interface { type inodeRefs interface { IncRef() - DecRef() + DecRef(ctx context.Context) TryIncRef() bool // Destroy is called when the inode reaches zero references. Destroy release // all resources (references) on objects referenced by the inode, including // any child dentries. - Destroy() + Destroy(ctx context.Context) } type inodeMetadata interface { diff --git a/pkg/sentry/fsimpl/kernfs/kernfs_test.go b/pkg/sentry/fsimpl/kernfs/kernfs_test.go index dc407eb1d..c5d5afedf 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs_test.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs_test.go @@ -46,7 +46,7 @@ func newTestSystem(t *testing.T, rootFn RootDentryFn) *testutil.System { ctx := contexttest.Context(t) creds := auth.CredentialsFromContext(ctx) v := &vfs.VirtualFilesystem{} - if err := v.Init(); err != nil { + if err := v.Init(ctx); err != nil { t.Fatalf("VFS init: %v", err) } v.MustRegisterFilesystemType("testfs", &fsType{rootFn: rootFn}, &vfs.RegisterFilesystemTypeOptions{ @@ -163,7 +163,7 @@ func (d *dir) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (* dir := d.fs.newDir(creds, opts.Mode, nil) dirVFSD := dir.VFSDentry() if err := d.OrderedChildren.Insert(name, dirVFSD); err != nil { - dir.DecRef() + dir.DecRef(ctx) return nil, err } d.IncLinks(1) @@ -175,7 +175,7 @@ func (d *dir) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (* f := d.fs.newFile(creds, "") fVFSD := f.VFSDentry() if err := d.OrderedChildren.Insert(name, fVFSD); err != nil { - f.DecRef() + f.DecRef(ctx) return nil, err } return fVFSD, nil @@ -213,7 +213,7 @@ func TestBasic(t *testing.T) { }) }) defer sys.Destroy() - sys.GetDentryOrDie(sys.PathOpAtRoot("file1")).DecRef() + sys.GetDentryOrDie(sys.PathOpAtRoot("file1")).DecRef(sys.Ctx) } func TestMkdirGetDentry(t *testing.T) { @@ -228,7 +228,7 @@ func TestMkdirGetDentry(t *testing.T) { if err := sys.VFS.MkdirAt(sys.Ctx, sys.Creds, pop, &vfs.MkdirOptions{Mode: 0755}); err != nil { t.Fatalf("MkdirAt for PathOperation %+v failed: %v", pop, err) } - sys.GetDentryOrDie(pop).DecRef() + sys.GetDentryOrDie(pop).DecRef(sys.Ctx) } func TestReadStaticFile(t *testing.T) { @@ -246,7 +246,7 @@ func TestReadStaticFile(t *testing.T) { if err != nil { t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) } - defer fd.DecRef() + defer fd.DecRef(sys.Ctx) content, err := sys.ReadToEnd(fd) if err != nil { @@ -273,7 +273,7 @@ func TestCreateNewFileInStaticDir(t *testing.T) { } // Close the file. The file should persist. - fd.DecRef() + fd.DecRef(sys.Ctx) fd, err = sys.VFS.OpenAt(sys.Ctx, sys.Creds, pop, &vfs.OpenOptions{ Flags: linux.O_RDONLY, @@ -281,7 +281,7 @@ func TestCreateNewFileInStaticDir(t *testing.T) { if err != nil { t.Fatalf("OpenAt(pop:%+v) = %+v failed: %v", pop, fd, err) } - fd.DecRef() + fd.DecRef(sys.Ctx) } func TestDirFDReadWrite(t *testing.T) { @@ -297,7 +297,7 @@ func TestDirFDReadWrite(t *testing.T) { if err != nil { t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) } - defer fd.DecRef() + defer fd.DecRef(sys.Ctx) // Read/Write should fail for directory FDs. if _, err := fd.Read(sys.Ctx, usermem.BytesIOSequence([]byte{}), vfs.ReadOptions{}); err != syserror.EISDIR { diff --git a/pkg/sentry/fsimpl/overlay/copy_up.go b/pkg/sentry/fsimpl/overlay/copy_up.go index 8f8dcfafe..b3d19ff82 100644 --- a/pkg/sentry/fsimpl/overlay/copy_up.go +++ b/pkg/sentry/fsimpl/overlay/copy_up.go @@ -98,7 +98,7 @@ func (d *dentry) copyUpLocked(ctx context.Context) error { if err != nil { return err } - defer oldFD.DecRef() + defer oldFD.DecRef(ctx) newFD, err := vfsObj.OpenAt(ctx, d.fs.creds, &newpop, &vfs.OpenOptions{ Flags: linux.O_WRONLY | linux.O_CREAT | linux.O_EXCL, Mode: linux.FileMode(d.mode &^ linux.S_IFMT), @@ -106,7 +106,7 @@ func (d *dentry) copyUpLocked(ctx context.Context) error { if err != nil { return err } - defer newFD.DecRef() + defer newFD.DecRef(ctx) bufIOSeq := usermem.BytesIOSequence(make([]byte, 32*1024)) // arbitrary buffer size for { readN, readErr := oldFD.Read(ctx, bufIOSeq, vfs.ReadOptions{}) @@ -241,13 +241,13 @@ func (d *dentry) copyUpLocked(ctx context.Context) error { Mask: linux.STATX_INO, }) if err != nil { - d.upperVD.DecRef() + d.upperVD.DecRef(ctx) d.upperVD = vfs.VirtualDentry{} cleanupUndoCopyUp() return err } if upperStat.Mask&linux.STATX_INO == 0 { - d.upperVD.DecRef() + d.upperVD.DecRef(ctx) d.upperVD = vfs.VirtualDentry{} cleanupUndoCopyUp() return syserror.EREMOTE diff --git a/pkg/sentry/fsimpl/overlay/directory.go b/pkg/sentry/fsimpl/overlay/directory.go index f5c2462a5..fccb94105 100644 --- a/pkg/sentry/fsimpl/overlay/directory.go +++ b/pkg/sentry/fsimpl/overlay/directory.go @@ -46,7 +46,7 @@ func (d *dentry) collectWhiteoutsForRmdirLocked(ctx context.Context) (map[string readdirErr = err return false } - defer layerFD.DecRef() + defer layerFD.DecRef(ctx) // Reuse slice allocated for maybeWhiteouts from a previous layer to // reduce allocations. @@ -108,7 +108,7 @@ type directoryFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *directoryFD) Release() { +func (fd *directoryFD) Release(ctx context.Context) { } // IterDirents implements vfs.FileDescriptionImpl.IterDirents. @@ -177,7 +177,7 @@ func (d *dentry) getDirents(ctx context.Context) ([]vfs.Dirent, error) { readdirErr = err return false } - defer layerFD.DecRef() + defer layerFD.DecRef(ctx) // Reuse slice allocated for maybeWhiteouts from a previous layer to // reduce allocations. @@ -282,6 +282,6 @@ func (fd *directoryFD) Sync(ctx context.Context) error { return err } err = upperFD.Sync(ctx) - upperFD.DecRef() + upperFD.DecRef(ctx) return err } diff --git a/pkg/sentry/fsimpl/overlay/filesystem.go b/pkg/sentry/fsimpl/overlay/filesystem.go index 6b705e955..986b36ead 100644 --- a/pkg/sentry/fsimpl/overlay/filesystem.go +++ b/pkg/sentry/fsimpl/overlay/filesystem.go @@ -77,7 +77,7 @@ func putDentrySlice(ds *[]*dentry) { // but dentry slices are allocated lazily, and it's much easier to say "defer // fs.renameMuRUnlockAndCheckDrop(&ds)" than "defer func() { // fs.renameMuRUnlockAndCheckDrop(ds) }()" to work around this. -func (fs *filesystem) renameMuRUnlockAndCheckDrop(ds **[]*dentry) { +func (fs *filesystem) renameMuRUnlockAndCheckDrop(ctx context.Context, ds **[]*dentry) { fs.renameMu.RUnlock() if *ds == nil { return @@ -85,20 +85,20 @@ func (fs *filesystem) renameMuRUnlockAndCheckDrop(ds **[]*dentry) { if len(**ds) != 0 { fs.renameMu.Lock() for _, d := range **ds { - d.checkDropLocked() + d.checkDropLocked(ctx) } fs.renameMu.Unlock() } putDentrySlice(*ds) } -func (fs *filesystem) renameMuUnlockAndCheckDrop(ds **[]*dentry) { +func (fs *filesystem) renameMuUnlockAndCheckDrop(ctx context.Context, ds **[]*dentry) { if *ds == nil { fs.renameMu.Unlock() return } for _, d := range **ds { - d.checkDropLocked() + d.checkDropLocked(ctx) } fs.renameMu.Unlock() putDentrySlice(*ds) @@ -126,13 +126,13 @@ afterSymlink: return d, nil } if name == ".." { - if isRoot, err := rp.CheckRoot(&d.vfsd); err != nil { + if isRoot, err := rp.CheckRoot(ctx, &d.vfsd); err != nil { return nil, err } else if isRoot || d.parent == nil { rp.Advance() return d, nil } - if err := rp.CheckMount(&d.parent.vfsd); err != nil { + if err := rp.CheckMount(ctx, &d.parent.vfsd); err != nil { return nil, err } rp.Advance() @@ -142,7 +142,7 @@ afterSymlink: if err != nil { return nil, err } - if err := rp.CheckMount(&child.vfsd); err != nil { + if err := rp.CheckMount(ctx, &child.vfsd); err != nil { return nil, err } if child.isSymlink() && mayFollowSymlinks && rp.ShouldFollowSymlink() { @@ -272,11 +272,11 @@ func (fs *filesystem) lookupLocked(ctx context.Context, parent *dentry, name str }) if lookupErr != nil { - child.destroyLocked() + child.destroyLocked(ctx) return nil, lookupErr } if !existsOnAnyLayer { - child.destroyLocked() + child.destroyLocked(ctx) return nil, syserror.ENOENT } @@ -430,7 +430,7 @@ func (fs *filesystem) resolveLocked(ctx context.Context, rp *vfs.ResolvingPath, func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, create func(parent *dentry, name string, haveUpperWhiteout bool) error) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) start := rp.Start().Impl().(*dentry) parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) if err != nil { @@ -501,7 +501,7 @@ func (fs *filesystem) cleanupRecreateWhiteout(ctx context.Context, vfsObj *vfs.V func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return err @@ -513,7 +513,7 @@ func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return nil, err @@ -532,7 +532,7 @@ func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return nil, err @@ -553,7 +553,7 @@ func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) start := rp.Start().Impl().(*dentry) d, err := fs.walkParentDirLocked(ctx, rp, start, &ds) if err != nil { @@ -720,7 +720,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) start := rp.Start().Impl().(*dentry) if rp.Done() { @@ -825,7 +825,7 @@ func (d *dentry) openLocked(ctx context.Context, rp *vfs.ResolvingPath, opts *vf fd.LockFD.Init(&d.locks) layerFDOpts := layerFD.Options() if err := fd.vfsfd.Init(fd, layerFlags, mnt, &d.vfsd, &layerFDOpts); err != nil { - layerFD.DecRef() + layerFD.DecRef(ctx) return nil, err } return &fd.vfsfd, nil @@ -920,7 +920,7 @@ func (fs *filesystem) createAndOpenLocked(ctx context.Context, rp *vfs.Resolving fd.LockFD.Init(&child.locks) upperFDOpts := upperFD.Options() if err := fd.vfsfd.Init(fd, upperFlags, mnt, &child.vfsd, &upperFDOpts); err != nil { - upperFD.DecRef() + upperFD.DecRef(ctx) // Don't bother with cleanup; the file was created successfully, we // just can't open it anymore for some reason. return nil, err @@ -932,7 +932,7 @@ func (fs *filesystem) createAndOpenLocked(ctx context.Context, rp *vfs.Resolving func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return "", err @@ -952,7 +952,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa var ds *[]*dentry fs.renameMu.Lock() - defer fs.renameMuUnlockAndCheckDrop(&ds) + defer fs.renameMuUnlockAndCheckDrop(ctx, &ds) newParent, err := fs.walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry), &ds) if err != nil { return err @@ -979,7 +979,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) start := rp.Start().Impl().(*dentry) parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) if err != nil { @@ -1001,7 +1001,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error } vfsObj := rp.VirtualFilesystem() mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) parent.dirMu.Lock() defer parent.dirMu.Unlock() @@ -1086,7 +1086,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error return err } - vfsObj.CommitDeleteDentry(&child.vfsd) + vfsObj.CommitDeleteDentry(ctx, &child.vfsd) delete(parent.children, name) ds = appendDentry(ds, child) parent.dirents = nil @@ -1097,7 +1097,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return err @@ -1132,7 +1132,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) d, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return linux.Statx{}, err @@ -1160,7 +1160,7 @@ func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) _, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return linux.Statfs{}, err @@ -1211,7 +1211,7 @@ func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) start := rp.Start().Impl().(*dentry) parent, err := fs.walkParentDirLocked(ctx, rp, start, &ds) if err != nil { @@ -1233,7 +1233,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error } vfsObj := rp.VirtualFilesystem() mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) parent.dirMu.Lock() defer parent.dirMu.Unlock() @@ -1298,7 +1298,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error } if child != nil { - vfsObj.CommitDeleteDentry(&child.vfsd) + vfsObj.CommitDeleteDentry(ctx, &child.vfsd) delete(parent.children, name) ds = appendDentry(ds, child) } @@ -1310,7 +1310,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) _, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return nil, err @@ -1324,7 +1324,7 @@ func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, si func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) _, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return "", err @@ -1336,7 +1336,7 @@ func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) _, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return err @@ -1348,7 +1348,7 @@ func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { var ds *[]*dentry fs.renameMu.RLock() - defer fs.renameMuRUnlockAndCheckDrop(&ds) + defer fs.renameMuRUnlockAndCheckDrop(ctx, &ds) _, err := fs.resolveLocked(ctx, rp, &ds) if err != nil { return err diff --git a/pkg/sentry/fsimpl/overlay/non_directory.go b/pkg/sentry/fsimpl/overlay/non_directory.go index c0749e711..d3060a481 100644 --- a/pkg/sentry/fsimpl/overlay/non_directory.go +++ b/pkg/sentry/fsimpl/overlay/non_directory.go @@ -81,11 +81,11 @@ func (fd *nonDirectoryFD) currentFDLocked(ctx context.Context) (*vfs.FileDescrip oldOff, oldOffErr := fd.cachedFD.Seek(ctx, 0, linux.SEEK_CUR) if oldOffErr == nil { if _, err := upperFD.Seek(ctx, oldOff, linux.SEEK_SET); err != nil { - upperFD.DecRef() + upperFD.DecRef(ctx) return nil, err } } - fd.cachedFD.DecRef() + fd.cachedFD.DecRef(ctx) fd.copiedUp = true fd.cachedFD = upperFD fd.cachedFlags = statusFlags @@ -99,8 +99,8 @@ func (fd *nonDirectoryFD) currentFDLocked(ctx context.Context) (*vfs.FileDescrip } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *nonDirectoryFD) Release() { - fd.cachedFD.DecRef() +func (fd *nonDirectoryFD) Release(ctx context.Context) { + fd.cachedFD.DecRef(ctx) fd.cachedFD = nil } @@ -138,7 +138,7 @@ func (fd *nonDirectoryFD) Stat(ctx context.Context, opts vfs.StatOptions) (linux Mask: layerMask, Sync: opts.Sync, }) - wrappedFD.DecRef() + wrappedFD.DecRef(ctx) if err != nil { return linux.Statx{}, err } @@ -187,7 +187,7 @@ func (fd *nonDirectoryFD) PRead(ctx context.Context, dst usermem.IOSequence, off if err != nil { return 0, err } - defer wrappedFD.DecRef() + defer wrappedFD.DecRef(ctx) return wrappedFD.PRead(ctx, dst, offset, opts) } @@ -209,7 +209,7 @@ func (fd *nonDirectoryFD) PWrite(ctx context.Context, src usermem.IOSequence, of if err != nil { return 0, err } - defer wrappedFD.DecRef() + defer wrappedFD.DecRef(ctx) return wrappedFD.PWrite(ctx, src, offset, opts) } @@ -250,7 +250,7 @@ func (fd *nonDirectoryFD) Sync(ctx context.Context) error { return err } wrappedFD.IncRef() - defer wrappedFD.DecRef() + defer wrappedFD.DecRef(ctx) fd.mu.Unlock() return wrappedFD.Sync(ctx) } @@ -261,6 +261,6 @@ func (fd *nonDirectoryFD) ConfigureMMap(ctx context.Context, opts *memmap.MMapOp if err != nil { return err } - defer wrappedFD.DecRef() + defer wrappedFD.DecRef(ctx) return wrappedFD.ConfigureMMap(ctx, opts) } diff --git a/pkg/sentry/fsimpl/overlay/overlay.go b/pkg/sentry/fsimpl/overlay/overlay.go index e720d4825..75cc006bf 100644 --- a/pkg/sentry/fsimpl/overlay/overlay.go +++ b/pkg/sentry/fsimpl/overlay/overlay.go @@ -123,7 +123,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt // filesystem with any number of lower layers. } else { vfsroot := vfs.RootFromContext(ctx) - defer vfsroot.DecRef() + defer vfsroot.DecRef(ctx) upperPathname, ok := mopts["upperdir"] if ok { delete(mopts, "upperdir") @@ -147,13 +147,13 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt ctx.Warningf("overlay.FilesystemType.GetFilesystem: failed to resolve upperdir %q: %v", upperPathname, err) return nil, nil, err } - defer upperRoot.DecRef() + defer upperRoot.DecRef(ctx) privateUpperRoot, err := clonePrivateMount(vfsObj, upperRoot, false /* forceReadOnly */) if err != nil { ctx.Warningf("overlay.FilesystemType.GetFilesystem: failed to make private bind mount of upperdir %q: %v", upperPathname, err) return nil, nil, err } - defer privateUpperRoot.DecRef() + defer privateUpperRoot.DecRef(ctx) fsopts.UpperRoot = privateUpperRoot } lowerPathnamesStr, ok := mopts["lowerdir"] @@ -190,13 +190,13 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt ctx.Warningf("overlay.FilesystemType.GetFilesystem: failed to resolve lowerdir %q: %v", lowerPathname, err) return nil, nil, err } - defer lowerRoot.DecRef() + defer lowerRoot.DecRef(ctx) privateLowerRoot, err := clonePrivateMount(vfsObj, lowerRoot, true /* forceReadOnly */) if err != nil { ctx.Warningf("overlay.FilesystemType.GetFilesystem: failed to make private bind mount of lowerdir %q: %v", lowerPathname, err) return nil, nil, err } - defer privateLowerRoot.DecRef() + defer privateLowerRoot.DecRef(ctx) fsopts.LowerRoots = append(fsopts.LowerRoots, privateLowerRoot) } } @@ -264,19 +264,19 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt Mask: rootStatMask, }) if err != nil { - root.destroyLocked() - fs.vfsfs.DecRef() + root.destroyLocked(ctx) + fs.vfsfs.DecRef(ctx) return nil, nil, err } if rootStat.Mask&rootStatMask != rootStatMask { - root.destroyLocked() - fs.vfsfs.DecRef() + root.destroyLocked(ctx) + fs.vfsfs.DecRef(ctx) return nil, nil, syserror.EREMOTE } if isWhiteout(&rootStat) { ctx.Warningf("overlay.FilesystemType.GetFilesystem: filesystem root is a whiteout") - root.destroyLocked() - fs.vfsfs.DecRef() + root.destroyLocked(ctx) + fs.vfsfs.DecRef(ctx) return nil, nil, syserror.EINVAL } root.mode = uint32(rootStat.Mode) @@ -319,17 +319,17 @@ func clonePrivateMount(vfsObj *vfs.VirtualFilesystem, vd vfs.VirtualDentry, forc } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { vfsObj := fs.vfsfs.VirtualFilesystem() vfsObj.PutAnonBlockDevMinor(fs.dirDevMinor) for _, lowerDevMinor := range fs.lowerDevMinors { vfsObj.PutAnonBlockDevMinor(lowerDevMinor) } if fs.opts.UpperRoot.Ok() { - fs.opts.UpperRoot.DecRef() + fs.opts.UpperRoot.DecRef(ctx) } for _, lowerRoot := range fs.opts.LowerRoots { - lowerRoot.DecRef() + lowerRoot.DecRef(ctx) } } @@ -452,10 +452,10 @@ func (d *dentry) TryIncRef() bool { } // DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef() { +func (d *dentry) DecRef(ctx context.Context) { if refs := atomic.AddInt64(&d.refs, -1); refs == 0 { d.fs.renameMu.Lock() - d.checkDropLocked() + d.checkDropLocked(ctx) d.fs.renameMu.Unlock() } else if refs < 0 { panic("overlay.dentry.DecRef() called without holding a reference") @@ -466,7 +466,7 @@ func (d *dentry) DecRef() { // becomes deleted. // // Preconditions: d.fs.renameMu must be locked for writing. -func (d *dentry) checkDropLocked() { +func (d *dentry) checkDropLocked(ctx context.Context) { // Dentries with a positive reference count must be retained. (The only way // to obtain a reference on a dentry with zero references is via path // resolution, which requires renameMu, so if d.refs is zero then it will @@ -476,14 +476,14 @@ func (d *dentry) checkDropLocked() { return } // Refs is still zero; destroy it. - d.destroyLocked() + d.destroyLocked(ctx) return } // destroyLocked destroys the dentry. // // Preconditions: d.fs.renameMu must be locked for writing. d.refs == 0. -func (d *dentry) destroyLocked() { +func (d *dentry) destroyLocked(ctx context.Context) { switch atomic.LoadInt64(&d.refs) { case 0: // Mark the dentry destroyed. @@ -495,10 +495,10 @@ func (d *dentry) destroyLocked() { } if d.upperVD.Ok() { - d.upperVD.DecRef() + d.upperVD.DecRef(ctx) } for _, lowerVD := range d.lowerVDs { - lowerVD.DecRef() + lowerVD.DecRef(ctx) } if d.parent != nil { @@ -510,7 +510,7 @@ func (d *dentry) destroyLocked() { // Drop the reference held by d on its parent without recursively // locking d.fs.renameMu. if refs := atomic.AddInt64(&d.parent.refs, -1); refs == 0 { - d.parent.checkDropLocked() + d.parent.checkDropLocked(ctx) } else if refs < 0 { panic("overlay.dentry.DecRef() called without holding a reference") } @@ -518,7 +518,7 @@ func (d *dentry) destroyLocked() { } // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. -func (d *dentry) InotifyWithParent(events uint32, cookie uint32, et vfs.EventType) { +func (d *dentry) InotifyWithParent(ctx context.Context, events uint32, cookie uint32, et vfs.EventType) { // TODO(gvisor.dev/issue/1479): Implement inotify. } @@ -531,7 +531,7 @@ func (d *dentry) Watches() *vfs.Watches { // OnZeroWatches implements vfs.DentryImpl.OnZeroWatches. // // TODO(gvisor.dev/issue/1479): Implement inotify. -func (d *dentry) OnZeroWatches() {} +func (d *dentry) OnZeroWatches(context.Context) {} // iterLayers invokes yield on each layer comprising d, from top to bottom. If // any call to yield returns false, iterLayer stops iteration. diff --git a/pkg/sentry/fsimpl/pipefs/pipefs.go b/pkg/sentry/fsimpl/pipefs/pipefs.go index 811f80a5f..2ca793db9 100644 --- a/pkg/sentry/fsimpl/pipefs/pipefs.go +++ b/pkg/sentry/fsimpl/pipefs/pipefs.go @@ -63,9 +63,9 @@ func NewFilesystem(vfsObj *vfs.VirtualFilesystem) (*vfs.Filesystem, error) { } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) - fs.Filesystem.Release() + fs.Filesystem.Release(ctx) } // PrependPath implements vfs.FilesystemImpl.PrependPath. @@ -160,6 +160,6 @@ func NewConnectedPipeFDs(ctx context.Context, mnt *vfs.Mount, flags uint32) (*vf inode := newInode(ctx, fs) var d kernfs.Dentry d.Init(inode) - defer d.DecRef() + defer d.DecRef(ctx) return inode.pipe.ReaderWriterPair(mnt, d.VFSDentry(), flags) } diff --git a/pkg/sentry/fsimpl/proc/filesystem.go b/pkg/sentry/fsimpl/proc/filesystem.go index 609210253..2463d51cd 100644 --- a/pkg/sentry/fsimpl/proc/filesystem.go +++ b/pkg/sentry/fsimpl/proc/filesystem.go @@ -77,9 +77,9 @@ func (ft FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualF } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) - fs.Filesystem.Release() + fs.Filesystem.Release(ctx) } // dynamicInode is an overfitted interface for common Inodes with diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index fea29e5f0..f0d3f7f5e 100644 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -43,12 +43,12 @@ func getTaskFD(t *kernel.Task, fd int32) (*vfs.FileDescription, kernel.FDFlags) return file, flags } -func taskFDExists(t *kernel.Task, fd int32) bool { +func taskFDExists(ctx context.Context, t *kernel.Task, fd int32) bool { file, _ := getTaskFD(t, fd) if file == nil { return false } - file.DecRef() + file.DecRef(ctx) return true } @@ -68,7 +68,7 @@ func (i *fdDir) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, off var fds []int32 i.task.WithMuLocked(func(t *kernel.Task) { if fdTable := t.FDTable(); fdTable != nil { - fds = fdTable.GetFDs() + fds = fdTable.GetFDs(ctx) } }) @@ -135,7 +135,7 @@ func (i *fdDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, erro return nil, syserror.ENOENT } fd := int32(fdInt) - if !taskFDExists(i.task, fd) { + if !taskFDExists(ctx, i.task, fd) { return nil, syserror.ENOENT } taskDentry := i.fs.newFDSymlink(i.task, fd, i.fs.NextIno()) @@ -204,9 +204,9 @@ func (s *fdSymlink) Readlink(ctx context.Context) (string, error) { if file == nil { return "", syserror.ENOENT } - defer file.DecRef() + defer file.DecRef(ctx) root := vfs.RootFromContext(ctx) - defer root.DecRef() + defer root.DecRef(ctx) return s.task.Kernel().VFS().PathnameWithDeleted(ctx, root, file.VirtualDentry()) } @@ -215,7 +215,7 @@ func (s *fdSymlink) Getlink(ctx context.Context, mnt *vfs.Mount) (vfs.VirtualDen if file == nil { return vfs.VirtualDentry{}, "", syserror.ENOENT } - defer file.DecRef() + defer file.DecRef(ctx) vd := file.VirtualDentry() vd.IncRef() return vd, "", nil @@ -258,7 +258,7 @@ func (i *fdInfoDirInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, return nil, syserror.ENOENT } fd := int32(fdInt) - if !taskFDExists(i.task, fd) { + if !taskFDExists(ctx, i.task, fd) { return nil, syserror.ENOENT } data := &fdInfoData{ @@ -297,7 +297,7 @@ func (d *fdInfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { if file == nil { return syserror.ENOENT } - defer file.DecRef() + defer file.DecRef(ctx) // TODO(b/121266871): Include pos, locks, and other data. For now we only // have flags. // See https://www.kernel.org/doc/Documentation/filesystems/proc.txt diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 859b7d727..830b78949 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -677,7 +677,7 @@ func (s *exeSymlink) Readlink(ctx context.Context) (string, error) { if err != nil { return "", err } - defer exec.DecRef() + defer exec.DecRef(ctx) return exec.PathnameWithDeleted(ctx), nil } @@ -692,7 +692,7 @@ func (s *exeSymlink) Getlink(ctx context.Context, _ *vfs.Mount) (vfs.VirtualDent if err != nil { return vfs.VirtualDentry{}, "", err } - defer exec.DecRef() + defer exec.DecRef(ctx) vd := exec.(*fsbridge.VFSFile).FileDescription().VirtualDentry() vd.IncRef() @@ -748,7 +748,7 @@ func (i *mountInfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { // Root has been destroyed. Don't try to read mounts. return nil } - defer rootDir.DecRef() + defer rootDir.DecRef(ctx) i.task.Kernel().VFS().GenerateProcMountInfo(ctx, rootDir, buf) return nil } @@ -779,7 +779,7 @@ func (i *mountsData) Generate(ctx context.Context, buf *bytes.Buffer) error { // Root has been destroyed. Don't try to read mounts. return nil } - defer rootDir.DecRef() + defer rootDir.DecRef(ctx) i.task.Kernel().VFS().GenerateProcMounts(ctx, rootDir, buf) return nil } @@ -825,7 +825,7 @@ func (s *namespaceSymlink) Getlink(ctx context.Context, mnt *vfs.Mount) (vfs.Vir dentry.Init(&namespaceInode{}) vd := vfs.MakeVirtualDentry(mnt, dentry.VFSDentry()) vd.IncRef() - dentry.DecRef() + dentry.DecRef(ctx) return vd, "", nil } @@ -887,8 +887,8 @@ func (fd *namespaceFD) SetStat(ctx context.Context, opts vfs.SetStatOptions) err } // Release implements FileDescriptionImpl. -func (fd *namespaceFD) Release() { - fd.inode.DecRef() +func (fd *namespaceFD) Release(ctx context.Context) { + fd.inode.DecRef(ctx) } // LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX. diff --git a/pkg/sentry/fsimpl/proc/task_net.go b/pkg/sentry/fsimpl/proc/task_net.go index 6bde27376..a4c884bf9 100644 --- a/pkg/sentry/fsimpl/proc/task_net.go +++ b/pkg/sentry/fsimpl/proc/task_net.go @@ -212,7 +212,7 @@ func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error { continue } if family, _, _ := s.Impl().(socket.SocketVFS2).Type(); family != linux.AF_UNIX { - s.DecRef() + s.DecRef(ctx) // Not a unix socket. continue } @@ -281,7 +281,7 @@ func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error { } fmt.Fprintf(buf, "\n") - s.DecRef() + s.DecRef(ctx) } return nil } @@ -359,7 +359,7 @@ func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, panic(fmt.Sprintf("Found non-socket file in socket table: %+v", s)) } if fa, stype, _ := sops.Type(); !(family == fa && stype == linux.SOCK_STREAM) { - s.DecRef() + s.DecRef(ctx) // Not tcp4 sockets. continue } @@ -455,7 +455,7 @@ func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, fmt.Fprintf(buf, "\n") - s.DecRef() + s.DecRef(ctx) } return nil @@ -524,7 +524,7 @@ func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error { panic(fmt.Sprintf("Found non-socket file in socket table: %+v", s)) } if family, stype, _ := sops.Type(); family != linux.AF_INET || stype != linux.SOCK_DGRAM { - s.DecRef() + s.DecRef(ctx) // Not udp4 socket. continue } @@ -600,7 +600,7 @@ func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error { fmt.Fprintf(buf, "\n") - s.DecRef() + s.DecRef(ctx) } return nil } diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index 19abb5034..3c9297dee 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -218,7 +218,7 @@ func TestTasks(t *testing.T) { if err != nil { t.Fatalf("vfsfs.OpenAt(%q) failed: %v", path, err) } - defer fd.DecRef() + defer fd.DecRef(s.Ctx) buf := make([]byte, 1) bufIOSeq := usermem.BytesIOSequence(buf) if _, err := fd.Read(s.Ctx, bufIOSeq, vfs.ReadOptions{}); err != syserror.EISDIR { @@ -336,7 +336,7 @@ func TestTasksOffset(t *testing.T) { if err != nil { t.Fatalf("vfsfs.OpenAt(/) failed: %v", err) } - defer fd.DecRef() + defer fd.DecRef(s.Ctx) if _, err := fd.Seek(s.Ctx, tc.offset, linux.SEEK_SET); err != nil { t.Fatalf("Seek(%d, SEEK_SET): %v", tc.offset, err) } @@ -441,7 +441,7 @@ func iterateDir(ctx context.Context, t *testing.T, s *testutil.System, fd *vfs.F t.Errorf("vfsfs.OpenAt(%v) failed: %v", absPath, err) continue } - defer child.DecRef() + defer child.DecRef(ctx) stat, err := child.Stat(ctx, vfs.StatOptions{}) if err != nil { t.Errorf("Stat(%v) failed: %v", absPath, err) @@ -476,7 +476,7 @@ func TestTree(t *testing.T) { if err != nil { t.Fatalf("failed to create test file: %v", err) } - defer file.DecRef() + defer file.DecRef(s.Ctx) var tasks []*kernel.Task for i := 0; i < 5; i++ { @@ -501,5 +501,5 @@ func TestTree(t *testing.T) { t.Fatalf("vfsfs.OpenAt(/proc) failed: %v", err) } iterateDir(ctx, t, s, fd) - fd.DecRef() + fd.DecRef(ctx) } diff --git a/pkg/sentry/fsimpl/signalfd/signalfd.go b/pkg/sentry/fsimpl/signalfd/signalfd.go index 242ba9b5d..6297e1df4 100644 --- a/pkg/sentry/fsimpl/signalfd/signalfd.go +++ b/pkg/sentry/fsimpl/signalfd/signalfd.go @@ -54,7 +54,7 @@ var _ vfs.FileDescriptionImpl = (*SignalFileDescription)(nil) // New creates a new signal fd. func New(vfsObj *vfs.VirtualFilesystem, target *kernel.Task, mask linux.SignalSet, flags uint32) (*vfs.FileDescription, error) { vd := vfsObj.NewAnonVirtualDentry("[signalfd]") - defer vd.DecRef() + defer vd.DecRef(target) sfd := &SignalFileDescription{ target: target, mask: mask, @@ -133,4 +133,4 @@ func (sfd *SignalFileDescription) EventUnregister(entry *waiter.Entry) { } // Release implements FileDescriptionImpl.Release() -func (sfd *SignalFileDescription) Release() {} +func (sfd *SignalFileDescription) Release(context.Context) {} diff --git a/pkg/sentry/fsimpl/sockfs/sockfs.go b/pkg/sentry/fsimpl/sockfs/sockfs.go index ee0828a15..c61818ff6 100644 --- a/pkg/sentry/fsimpl/sockfs/sockfs.go +++ b/pkg/sentry/fsimpl/sockfs/sockfs.go @@ -67,9 +67,9 @@ func NewFilesystem(vfsObj *vfs.VirtualFilesystem) (*vfs.Filesystem, error) { } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) - fs.Filesystem.Release() + fs.Filesystem.Release(ctx) } // PrependPath implements vfs.FilesystemImpl.PrependPath. diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go index 01ce30a4d..0401726b6 100644 --- a/pkg/sentry/fsimpl/sys/sys.go +++ b/pkg/sentry/fsimpl/sys/sys.go @@ -30,6 +30,7 @@ import ( // Name is the default filesystem name. const Name = "sysfs" +const defaultSysDirMode = linux.FileMode(0755) // FilesystemType implements vfs.FilesystemType. type FilesystemType struct{} @@ -57,9 +58,6 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt devMinor: devMinor, } fs.VFSFilesystem().Init(vfsObj, &fsType, fs) - k := kernel.KernelFromContext(ctx) - maxCPUCores := k.ApplicationCores() - defaultSysDirMode := linux.FileMode(0755) root := fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ "block": fs.newDir(creds, defaultSysDirMode, nil), @@ -70,11 +68,7 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt "dev": fs.newDir(creds, defaultSysDirMode, nil), "devices": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ "system": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ - "cpu": fs.newDir(creds, defaultSysDirMode, map[string]*kernfs.Dentry{ - "online": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), - "possible": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), - "present": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), - }), + "cpu": cpuDir(ctx, fs, creds), }), }), "firmware": fs.newDir(creds, defaultSysDirMode, nil), @@ -86,10 +80,24 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt return fs.VFSFilesystem(), root.VFSDentry(), nil } +func cpuDir(ctx context.Context, fs *filesystem, creds *auth.Credentials) *kernfs.Dentry { + k := kernel.KernelFromContext(ctx) + maxCPUCores := k.ApplicationCores() + children := map[string]*kernfs.Dentry{ + "online": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), + "possible": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), + "present": fs.newCPUFile(creds, maxCPUCores, linux.FileMode(0444)), + } + for i := uint(0); i < maxCPUCores; i++ { + children[fmt.Sprintf("cpu%d", i)] = fs.newDir(creds, linux.FileMode(0555), nil) + } + return fs.newDir(creds, defaultSysDirMode, children) +} + // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) - fs.Filesystem.Release() + fs.Filesystem.Release(ctx) } // dir implements kernfs.Inode. diff --git a/pkg/sentry/fsimpl/sys/sys_test.go b/pkg/sentry/fsimpl/sys/sys_test.go index 242d5fd12..9fd38b295 100644 --- a/pkg/sentry/fsimpl/sys/sys_test.go +++ b/pkg/sentry/fsimpl/sys/sys_test.go @@ -59,7 +59,7 @@ func TestReadCPUFile(t *testing.T) { if err != nil { t.Fatalf("OpenAt(pop:%+v) = %+v failed: %v", pop, fd, err) } - defer fd.DecRef() + defer fd.DecRef(s.Ctx) content, err := s.ReadToEnd(fd) if err != nil { t.Fatalf("Read failed: %v", err) diff --git a/pkg/sentry/fsimpl/testutil/kernel.go b/pkg/sentry/fsimpl/testutil/kernel.go index e743e8114..1813269e0 100644 --- a/pkg/sentry/fsimpl/testutil/kernel.go +++ b/pkg/sentry/fsimpl/testutil/kernel.go @@ -122,12 +122,16 @@ func Boot() (*kernel.Kernel, error) { // CreateTask creates a new bare bones task for tests. func CreateTask(ctx context.Context, name string, tc *kernel.ThreadGroup, mntns *vfs.MountNamespace, root, cwd vfs.VirtualDentry) (*kernel.Task, error) { k := kernel.KernelFromContext(ctx) + if k == nil { + return nil, fmt.Errorf("cannot find kernel from context") + } + exe, err := newFakeExecutable(ctx, k.VFS(), auth.CredentialsFromContext(ctx), root) if err != nil { return nil, err } m := mm.NewMemoryManager(k, k, k.SleepForAddressSpaceActivation) - m.SetExecutable(fsbridge.NewVFSFile(exe)) + m.SetExecutable(ctx, fsbridge.NewVFSFile(exe)) config := &kernel.TaskConfig{ Kernel: k, diff --git a/pkg/sentry/fsimpl/testutil/testutil.go b/pkg/sentry/fsimpl/testutil/testutil.go index 0556af877..568132121 100644 --- a/pkg/sentry/fsimpl/testutil/testutil.go +++ b/pkg/sentry/fsimpl/testutil/testutil.go @@ -97,8 +97,8 @@ func (s *System) WithTemporaryContext(ctx context.Context) *System { // Destroy release resources associated with a test system. func (s *System) Destroy() { - s.Root.DecRef() - s.MntNs.DecRef() // Reference on MntNs passed to NewSystem. + s.Root.DecRef(s.Ctx) + s.MntNs.DecRef(s.Ctx) // Reference on MntNs passed to NewSystem. } // ReadToEnd reads the contents of fd until EOF to a string. @@ -149,7 +149,7 @@ func (s *System) ListDirents(pop *vfs.PathOperation) *DirentCollector { if err != nil { s.t.Fatalf("OpenAt for PathOperation %+v failed: %v", pop, err) } - defer fd.DecRef() + defer fd.DecRef(s.Ctx) collector := &DirentCollector{} if err := fd.IterDirents(s.Ctx, collector); err != nil { diff --git a/pkg/sentry/fsimpl/timerfd/timerfd.go b/pkg/sentry/fsimpl/timerfd/timerfd.go index 2dc90d484..86beaa0a8 100644 --- a/pkg/sentry/fsimpl/timerfd/timerfd.go +++ b/pkg/sentry/fsimpl/timerfd/timerfd.go @@ -47,9 +47,9 @@ var _ vfs.FileDescriptionImpl = (*TimerFileDescription)(nil) var _ ktime.TimerListener = (*TimerFileDescription)(nil) // New returns a new timer fd. -func New(vfsObj *vfs.VirtualFilesystem, clock ktime.Clock, flags uint32) (*vfs.FileDescription, error) { +func New(ctx context.Context, vfsObj *vfs.VirtualFilesystem, clock ktime.Clock, flags uint32) (*vfs.FileDescription, error) { vd := vfsObj.NewAnonVirtualDentry("[timerfd]") - defer vd.DecRef() + defer vd.DecRef(ctx) tfd := &TimerFileDescription{} tfd.timer = ktime.NewTimer(clock, tfd) if err := tfd.vfsfd.Init(tfd, flags, vd.Mount(), vd.Dentry(), &vfs.FileDescriptionOptions{ @@ -129,7 +129,7 @@ func (tfd *TimerFileDescription) ResumeTimer() { } // Release implements FileDescriptionImpl.Release() -func (tfd *TimerFileDescription) Release() { +func (tfd *TimerFileDescription) Release(context.Context) { tfd.timer.Destroy() } diff --git a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go index 2fb5c4d84..d263147c2 100644 --- a/pkg/sentry/fsimpl/tmpfs/benchmark_test.go +++ b/pkg/sentry/fsimpl/tmpfs/benchmark_test.go @@ -83,7 +83,7 @@ func fileOpOn(ctx context.Context, mntns *fs.MountNamespace, root, wd *fs.Dirent } err = fn(root, d) - d.DecRef() + d.DecRef(ctx) return err } @@ -105,17 +105,17 @@ func BenchmarkVFS1TmpfsStat(b *testing.B) { if err != nil { b.Fatalf("failed to create mount namespace: %v", err) } - defer mntns.DecRef() + defer mntns.DecRef(ctx) var filePathBuilder strings.Builder filePathBuilder.WriteByte('/') // Create nested directories with given depth. root := mntns.Root() - defer root.DecRef() + defer root.DecRef(ctx) d := root d.IncRef() - defer d.DecRef() + defer d.DecRef(ctx) for i := depth; i > 0; i-- { name := fmt.Sprintf("%d", i) if err := d.Inode.CreateDirectory(ctx, d, name, fs.FilePermsFromMode(0755)); err != nil { @@ -125,7 +125,7 @@ func BenchmarkVFS1TmpfsStat(b *testing.B) { if err != nil { b.Fatalf("failed to walk to directory %q: %v", name, err) } - d.DecRef() + d.DecRef(ctx) d = next filePathBuilder.WriteString(name) filePathBuilder.WriteByte('/') @@ -136,7 +136,7 @@ func BenchmarkVFS1TmpfsStat(b *testing.B) { if err != nil { b.Fatalf("failed to create file %q: %v", filename, err) } - file.DecRef() + file.DecRef(ctx) filePathBuilder.WriteString(filename) filePath := filePathBuilder.String() @@ -176,7 +176,7 @@ func BenchmarkVFS2TmpfsStat(b *testing.B) { // Create VFS. vfsObj := vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { b.Fatalf("VFS init: %v", err) } vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ @@ -186,14 +186,14 @@ func BenchmarkVFS2TmpfsStat(b *testing.B) { if err != nil { b.Fatalf("failed to create tmpfs root mount: %v", err) } - defer mntns.DecRef() + defer mntns.DecRef(ctx) var filePathBuilder strings.Builder filePathBuilder.WriteByte('/') // Create nested directories with given depth. root := mntns.Root() - defer root.DecRef() + defer root.DecRef(ctx) vd := root vd.IncRef() for i := depth; i > 0; i-- { @@ -212,7 +212,7 @@ func BenchmarkVFS2TmpfsStat(b *testing.B) { if err != nil { b.Fatalf("failed to walk to directory %q: %v", name, err) } - vd.DecRef() + vd.DecRef(ctx) vd = nextVD filePathBuilder.WriteString(name) filePathBuilder.WriteByte('/') @@ -228,12 +228,12 @@ func BenchmarkVFS2TmpfsStat(b *testing.B) { Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, Mode: 0644, }) - vd.DecRef() + vd.DecRef(ctx) vd = vfs.VirtualDentry{} if err != nil { b.Fatalf("failed to create file %q: %v", filename, err) } - defer fd.DecRef() + defer fd.DecRef(ctx) filePathBuilder.WriteString(filename) filePath := filePathBuilder.String() @@ -278,14 +278,14 @@ func BenchmarkVFS1TmpfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to create mount namespace: %v", err) } - defer mntns.DecRef() + defer mntns.DecRef(ctx) var filePathBuilder strings.Builder filePathBuilder.WriteByte('/') // Create and mount the submount. root := mntns.Root() - defer root.DecRef() + defer root.DecRef(ctx) if err := root.Inode.CreateDirectory(ctx, root, mountPointName, fs.FilePermsFromMode(0755)); err != nil { b.Fatalf("failed to create mount point: %v", err) } @@ -293,7 +293,7 @@ func BenchmarkVFS1TmpfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to walk to mount point: %v", err) } - defer mountPoint.DecRef() + defer mountPoint.DecRef(ctx) submountInode, err := tmpfsFS.Mount(ctx, "tmpfs", fs.MountSourceFlags{}, "", nil) if err != nil { b.Fatalf("failed to create tmpfs submount: %v", err) @@ -309,7 +309,7 @@ func BenchmarkVFS1TmpfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to walk to mount root: %v", err) } - defer d.DecRef() + defer d.DecRef(ctx) for i := depth; i > 0; i-- { name := fmt.Sprintf("%d", i) if err := d.Inode.CreateDirectory(ctx, d, name, fs.FilePermsFromMode(0755)); err != nil { @@ -319,7 +319,7 @@ func BenchmarkVFS1TmpfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to walk to directory %q: %v", name, err) } - d.DecRef() + d.DecRef(ctx) d = next filePathBuilder.WriteString(name) filePathBuilder.WriteByte('/') @@ -330,7 +330,7 @@ func BenchmarkVFS1TmpfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to create file %q: %v", filename, err) } - file.DecRef() + file.DecRef(ctx) filePathBuilder.WriteString(filename) filePath := filePathBuilder.String() @@ -370,7 +370,7 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) { // Create VFS. vfsObj := vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { b.Fatalf("VFS init: %v", err) } vfsObj.MustRegisterFilesystemType("tmpfs", tmpfs.FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ @@ -380,14 +380,14 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to create tmpfs root mount: %v", err) } - defer mntns.DecRef() + defer mntns.DecRef(ctx) var filePathBuilder strings.Builder filePathBuilder.WriteByte('/') // Create the mount point. root := mntns.Root() - defer root.DecRef() + defer root.DecRef(ctx) pop := vfs.PathOperation{ Root: root, Start: root, @@ -403,7 +403,7 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to walk to mount point: %v", err) } - defer mountPoint.DecRef() + defer mountPoint.DecRef(ctx) // Create and mount the submount. if err := vfsObj.MountAt(ctx, creds, "", &pop, "tmpfs", &vfs.MountOptions{}); err != nil { b.Fatalf("failed to mount tmpfs submount: %v", err) @@ -432,7 +432,7 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) { if err != nil { b.Fatalf("failed to walk to directory %q: %v", name, err) } - vd.DecRef() + vd.DecRef(ctx) vd = nextVD filePathBuilder.WriteString(name) filePathBuilder.WriteByte('/') @@ -448,11 +448,11 @@ func BenchmarkVFS2TmpfsMountStat(b *testing.B) { Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, Mode: 0644, }) - vd.DecRef() + vd.DecRef(ctx) if err != nil { b.Fatalf("failed to create file %q: %v", filename, err) } - fd.DecRef() + fd.DecRef(ctx) filePathBuilder.WriteString(filename) filePath := filePathBuilder.String() diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go index 0a1ad4765..78b4fc5be 100644 --- a/pkg/sentry/fsimpl/tmpfs/directory.go +++ b/pkg/sentry/fsimpl/tmpfs/directory.go @@ -95,7 +95,7 @@ type directoryFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *directoryFD) Release() { +func (fd *directoryFD) Release(ctx context.Context) { if fd.iter != nil { dir := fd.inode().impl.(*directory) dir.iterMu.Lock() @@ -110,7 +110,7 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba fs := fd.filesystem() dir := fd.inode().impl.(*directory) - defer fd.dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + defer fd.dentry().InotifyWithParent(ctx, linux.IN_ACCESS, 0, vfs.PathEvent) // fs.mu is required to read d.parent and dentry.name. fs.mu.RLock() diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index ef210a69b..fb77f95cc 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -40,7 +40,7 @@ func (fs *filesystem) Sync(ctx context.Context) error { // stepLocked is loosely analogous to fs/namei.c:walk_component(). // // Preconditions: filesystem.mu must be locked. !rp.Done(). -func stepLocked(rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { +func stepLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry) (*dentry, error) { dir, ok := d.inode.impl.(*directory) if !ok { return nil, syserror.ENOTDIR @@ -55,13 +55,13 @@ afterSymlink: return d, nil } if name == ".." { - if isRoot, err := rp.CheckRoot(&d.vfsd); err != nil { + if isRoot, err := rp.CheckRoot(ctx, &d.vfsd); err != nil { return nil, err } else if isRoot || d.parent == nil { rp.Advance() return d, nil } - if err := rp.CheckMount(&d.parent.vfsd); err != nil { + if err := rp.CheckMount(ctx, &d.parent.vfsd); err != nil { return nil, err } rp.Advance() @@ -74,7 +74,7 @@ afterSymlink: if !ok { return nil, syserror.ENOENT } - if err := rp.CheckMount(&child.vfsd); err != nil { + if err := rp.CheckMount(ctx, &child.vfsd); err != nil { return nil, err } if symlink, ok := child.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { @@ -98,9 +98,9 @@ afterSymlink: // fs/namei.c:path_parentat(). // // Preconditions: filesystem.mu must be locked. !rp.Done(). -func walkParentDirLocked(rp *vfs.ResolvingPath, d *dentry) (*directory, error) { +func walkParentDirLocked(ctx context.Context, rp *vfs.ResolvingPath, d *dentry) (*directory, error) { for !rp.Final() { - next, err := stepLocked(rp, d) + next, err := stepLocked(ctx, rp, d) if err != nil { return nil, err } @@ -118,10 +118,10 @@ func walkParentDirLocked(rp *vfs.ResolvingPath, d *dentry) (*directory, error) { // resolveLocked is loosely analogous to Linux's fs/namei.c:path_lookupat(). // // Preconditions: filesystem.mu must be locked. -func resolveLocked(rp *vfs.ResolvingPath) (*dentry, error) { +func resolveLocked(ctx context.Context, rp *vfs.ResolvingPath) (*dentry, error) { d := rp.Start().Impl().(*dentry) for !rp.Done() { - next, err := stepLocked(rp, d) + next, err := stepLocked(ctx, rp, d) if err != nil { return nil, err } @@ -141,10 +141,10 @@ func resolveLocked(rp *vfs.ResolvingPath) (*dentry, error) { // // Preconditions: !rp.Done(). For the final path component in rp, // !rp.ShouldFollowSymlink(). -func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(parentDir *directory, name string) error) error { +func (fs *filesystem) doCreateAt(ctx context.Context, rp *vfs.ResolvingPath, dir bool, create func(parentDir *directory, name string) error) error { fs.mu.Lock() defer fs.mu.Unlock() - parentDir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + parentDir, err := walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry)) if err != nil { return err } @@ -182,7 +182,7 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa if dir { ev |= linux.IN_ISDIR } - parentDir.inode.watches.Notify(name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) + parentDir.inode.watches.Notify(ctx, name, uint32(ev), 0, vfs.InodeEvent, false /* unlinked */) parentDir.inode.touchCMtime() return nil } @@ -191,7 +191,7 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds *auth.Credentials, ats vfs.AccessTypes) error { fs.mu.RLock() defer fs.mu.RUnlock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { return err } @@ -202,7 +202,7 @@ func (fs *filesystem) AccessAt(ctx context.Context, rp *vfs.ResolvingPath, creds func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetDentryOptions) (*vfs.Dentry, error) { fs.mu.RLock() defer fs.mu.RUnlock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { return nil, err } @@ -222,7 +222,7 @@ func (fs *filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPath) (*vfs.Dentry, error) { fs.mu.RLock() defer fs.mu.RUnlock() - dir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + dir, err := walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry)) if err != nil { return nil, err } @@ -232,7 +232,7 @@ func (fs *filesystem) GetParentDentryAt(ctx context.Context, rp *vfs.ResolvingPa // LinkAt implements vfs.FilesystemImpl.LinkAt. func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.VirtualDentry) error { - return fs.doCreateAt(rp, false /* dir */, func(parentDir *directory, name string) error { + return fs.doCreateAt(ctx, rp, false /* dir */, func(parentDir *directory, name string) error { if rp.Mount() != vd.Mount() { return syserror.EXDEV } @@ -251,7 +251,7 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. return syserror.EMLINK } i.incLinksLocked() - i.watches.Notify("", linux.IN_ATTRIB, 0, vfs.InodeEvent, false /* unlinked */) + i.watches.Notify(ctx, "", linux.IN_ATTRIB, 0, vfs.InodeEvent, false /* unlinked */) parentDir.insertChildLocked(fs.newDentry(i), name) return nil }) @@ -259,7 +259,7 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. // MkdirAt implements vfs.FilesystemImpl.MkdirAt. func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MkdirOptions) error { - return fs.doCreateAt(rp, true /* dir */, func(parentDir *directory, name string) error { + return fs.doCreateAt(ctx, rp, true /* dir */, func(parentDir *directory, name string) error { creds := rp.Credentials() if parentDir.inode.nlink == maxLinks { return syserror.EMLINK @@ -273,7 +273,7 @@ func (fs *filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v // MknodAt implements vfs.FilesystemImpl.MknodAt. func (fs *filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.MknodOptions) error { - return fs.doCreateAt(rp, false /* dir */, func(parentDir *directory, name string) error { + return fs.doCreateAt(ctx, rp, false /* dir */, func(parentDir *directory, name string) error { creds := rp.Credentials() var childInode *inode switch opts.Mode.FileType() { @@ -308,7 +308,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if opts.Flags&linux.O_CREAT == 0 { fs.mu.RLock() defer fs.mu.RUnlock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { return nil, err } @@ -330,7 +330,7 @@ func (fs *filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf return start.open(ctx, rp, &opts, false /* afterCreate */) } afterTrailingSymlink: - parentDir, err := walkParentDirLocked(rp, start) + parentDir, err := walkParentDirLocked(ctx, rp, start) if err != nil { return nil, err } @@ -368,7 +368,7 @@ afterTrailingSymlink: if err != nil { return nil, err } - parentDir.inode.watches.Notify(name, linux.IN_CREATE, 0, vfs.PathEvent, false /* unlinked */) + parentDir.inode.watches.Notify(ctx, name, linux.IN_CREATE, 0, vfs.PathEvent, false /* unlinked */) parentDir.inode.touchCMtime() return fd, nil } @@ -376,7 +376,7 @@ afterTrailingSymlink: return nil, syserror.EEXIST } // Is the file mounted over? - if err := rp.CheckMount(&child.vfsd); err != nil { + if err := rp.CheckMount(ctx, &child.vfsd); err != nil { return nil, err } // Do we need to resolve a trailing symlink? @@ -445,7 +445,7 @@ func (d *dentry) open(ctx context.Context, rp *vfs.ResolvingPath, opts *vfs.Open func (fs *filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (string, error) { fs.mu.RLock() defer fs.mu.RUnlock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { return "", err } @@ -467,7 +467,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa // Resolve newParent first to verify that it's on this Mount. fs.mu.Lock() defer fs.mu.Unlock() - newParentDir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + newParentDir, err := walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry)) if err != nil { return err } @@ -555,7 +555,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa } vfsObj := rp.VirtualFilesystem() mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) var replacedVFSD *vfs.Dentry if replaced != nil { replacedVFSD = &replaced.vfsd @@ -566,17 +566,17 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa if replaced != nil { newParentDir.removeChildLocked(replaced) if replaced.inode.isDir() { - newParentDir.inode.decLinksLocked() // from replaced's ".." + newParentDir.inode.decLinksLocked(ctx) // from replaced's ".." } - replaced.inode.decLinksLocked() + replaced.inode.decLinksLocked(ctx) } oldParentDir.removeChildLocked(renamed) newParentDir.insertChildLocked(renamed, newName) - vfsObj.CommitRenameReplaceDentry(&renamed.vfsd, replacedVFSD) + vfsObj.CommitRenameReplaceDentry(ctx, &renamed.vfsd, replacedVFSD) oldParentDir.inode.touchCMtime() if oldParentDir != newParentDir { if renamed.inode.isDir() { - oldParentDir.inode.decLinksLocked() + oldParentDir.inode.decLinksLocked(ctx) newParentDir.inode.incLinksLocked() } newParentDir.inode.touchCMtime() @@ -591,7 +591,7 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { fs.mu.Lock() defer fs.mu.Unlock() - parentDir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + parentDir, err := walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry)) if err != nil { return err } @@ -626,17 +626,17 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error defer mnt.EndWrite() vfsObj := rp.VirtualFilesystem() mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil { return err } parentDir.removeChildLocked(child) - parentDir.inode.watches.Notify(name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent, true /* unlinked */) + parentDir.inode.watches.Notify(ctx, name, linux.IN_DELETE|linux.IN_ISDIR, 0, vfs.InodeEvent, true /* unlinked */) // Remove links for child, child/., and child/.. - child.inode.decLinksLocked() - child.inode.decLinksLocked() - parentDir.inode.decLinksLocked() - vfsObj.CommitDeleteDentry(&child.vfsd) + child.inode.decLinksLocked(ctx) + child.inode.decLinksLocked(ctx) + parentDir.inode.decLinksLocked(ctx) + vfsObj.CommitDeleteDentry(ctx, &child.vfsd) parentDir.inode.touchCMtime() return nil } @@ -644,7 +644,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error // SetStatAt implements vfs.FilesystemImpl.SetStatAt. func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { fs.mu.RLock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { fs.mu.RUnlock() return err @@ -656,7 +656,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts fs.mu.RUnlock() if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { - d.InotifyWithParent(ev, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, ev, 0, vfs.InodeEvent) } return nil } @@ -665,7 +665,7 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.StatOptions) (linux.Statx, error) { fs.mu.RLock() defer fs.mu.RUnlock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { return linux.Statx{}, err } @@ -678,7 +678,7 @@ func (fs *filesystem) StatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linux.Statfs, error) { fs.mu.RLock() defer fs.mu.RUnlock() - if _, err := resolveLocked(rp); err != nil { + if _, err := resolveLocked(ctx, rp); err != nil { return linux.Statfs{}, err } statfs := linux.Statfs{ @@ -695,7 +695,7 @@ func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu // SymlinkAt implements vfs.FilesystemImpl.SymlinkAt. func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, target string) error { - return fs.doCreateAt(rp, false /* dir */, func(parentDir *directory, name string) error { + return fs.doCreateAt(ctx, rp, false /* dir */, func(parentDir *directory, name string) error { creds := rp.Credentials() child := fs.newDentry(fs.newSymlink(creds.EffectiveKUID, creds.EffectiveKGID, 0777, target)) parentDir.insertChildLocked(child, name) @@ -707,7 +707,7 @@ func (fs *filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { fs.mu.Lock() defer fs.mu.Unlock() - parentDir, err := walkParentDirLocked(rp, rp.Start().Impl().(*dentry)) + parentDir, err := walkParentDirLocked(ctx, rp, rp.Start().Impl().(*dentry)) if err != nil { return err } @@ -738,7 +738,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error defer mnt.EndWrite() vfsObj := rp.VirtualFilesystem() mntns := vfs.MountNamespaceFromContext(ctx) - defer mntns.DecRef() + defer mntns.DecRef(ctx) if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil { return err } @@ -746,11 +746,11 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error // Generate inotify events. Note that this must take place before the link // count of the child is decremented, or else the watches may be dropped // before these events are added. - vfs.InotifyRemoveChild(&child.inode.watches, &parentDir.inode.watches, name) + vfs.InotifyRemoveChild(ctx, &child.inode.watches, &parentDir.inode.watches, name) parentDir.removeChildLocked(child) - child.inode.decLinksLocked() - vfsObj.CommitDeleteDentry(&child.vfsd) + child.inode.decLinksLocked(ctx) + vfsObj.CommitDeleteDentry(ctx, &child.vfsd) parentDir.inode.touchCMtime() return nil } @@ -759,7 +759,7 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.BoundEndpointOptions) (transport.BoundEndpoint, error) { fs.mu.RLock() defer fs.mu.RUnlock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { return nil, err } @@ -778,7 +778,7 @@ func (fs *filesystem) BoundEndpointAt(ctx context.Context, rp *vfs.ResolvingPath func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, size uint64) ([]string, error) { fs.mu.RLock() defer fs.mu.RUnlock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { return nil, err } @@ -789,7 +789,7 @@ func (fs *filesystem) ListxattrAt(ctx context.Context, rp *vfs.ResolvingPath, si func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.GetxattrOptions) (string, error) { fs.mu.RLock() defer fs.mu.RUnlock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { return "", err } @@ -799,7 +799,7 @@ func (fs *filesystem) GetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt // SetxattrAt implements vfs.FilesystemImpl.SetxattrAt. func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetxattrOptions) error { fs.mu.RLock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { fs.mu.RUnlock() return err @@ -810,14 +810,14 @@ func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt } fs.mu.RUnlock() - d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } // RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt. func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, name string) error { fs.mu.RLock() - d, err := resolveLocked(rp) + d, err := resolveLocked(ctx, rp) if err != nil { fs.mu.RUnlock() return err @@ -828,7 +828,7 @@ func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath, } fs.mu.RUnlock() - d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } diff --git a/pkg/sentry/fsimpl/tmpfs/pipe_test.go b/pkg/sentry/fsimpl/tmpfs/pipe_test.go index 1614f2c39..ec2701d8b 100644 --- a/pkg/sentry/fsimpl/tmpfs/pipe_test.go +++ b/pkg/sentry/fsimpl/tmpfs/pipe_test.go @@ -32,7 +32,7 @@ const fileName = "mypipe" func TestSeparateFDs(t *testing.T) { ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() + defer root.DecRef(ctx) // Open the read side. This is done in a concurrently because opening // One end the pipe blocks until the other end is opened. @@ -55,13 +55,13 @@ func TestSeparateFDs(t *testing.T) { if err != nil { t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) } - defer wfd.DecRef() + defer wfd.DecRef(ctx) rfd, ok := <-rfdchan if !ok { t.Fatalf("failed to open pipe for reading %q", fileName) } - defer rfd.DecRef() + defer rfd.DecRef(ctx) const msg = "vamos azul" checkEmpty(ctx, t, rfd) @@ -71,7 +71,7 @@ func TestSeparateFDs(t *testing.T) { func TestNonblockingRead(t *testing.T) { ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() + defer root.DecRef(ctx) // Open the read side as nonblocking. pop := vfs.PathOperation{ @@ -85,7 +85,7 @@ func TestNonblockingRead(t *testing.T) { if err != nil { t.Fatalf("failed to open pipe for reading %q: %v", fileName, err) } - defer rfd.DecRef() + defer rfd.DecRef(ctx) // Open the write side. openOpts = vfs.OpenOptions{Flags: linux.O_WRONLY} @@ -93,7 +93,7 @@ func TestNonblockingRead(t *testing.T) { if err != nil { t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) } - defer wfd.DecRef() + defer wfd.DecRef(ctx) const msg = "geh blau" checkEmpty(ctx, t, rfd) @@ -103,7 +103,7 @@ func TestNonblockingRead(t *testing.T) { func TestNonblockingWriteError(t *testing.T) { ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() + defer root.DecRef(ctx) // Open the write side as nonblocking, which should return ENXIO. pop := vfs.PathOperation{ @@ -121,7 +121,7 @@ func TestNonblockingWriteError(t *testing.T) { func TestSingleFD(t *testing.T) { ctx, creds, vfsObj, root := setup(t) - defer root.DecRef() + defer root.DecRef(ctx) // Open the pipe as readable and writable. pop := vfs.PathOperation{ @@ -135,7 +135,7 @@ func TestSingleFD(t *testing.T) { if err != nil { t.Fatalf("failed to open pipe for writing %q: %v", fileName, err) } - defer fd.DecRef() + defer fd.DecRef(ctx) const msg = "forza blu" checkEmpty(ctx, t, fd) @@ -152,7 +152,7 @@ func setup(t *testing.T) (context.Context, *auth.Credentials, *vfs.VirtualFilesy // Create VFS. vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { t.Fatalf("VFS init: %v", err) } vfsObj.MustRegisterFilesystemType("tmpfs", FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index abbaa5d60..0710b65db 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -270,7 +270,7 @@ type regularFileFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *regularFileFD) Release() { +func (fd *regularFileFD) Release(context.Context) { // noop } diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 2545d88e9..68e615e8b 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -185,7 +185,7 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt case linux.S_IFDIR: root = &fs.newDirectory(rootKUID, rootKGID, rootMode).dentry default: - fs.vfsfs.DecRef() + fs.vfsfs.DecRef(ctx) return nil, nil, fmt.Errorf("invalid tmpfs root file type: %#o", rootFileType) } return &fs.vfsfs, &root.vfsd, nil @@ -197,7 +197,7 @@ func NewFilesystem(ctx context.Context, vfsObj *vfs.VirtualFilesystem, creds *au } // Release implements vfs.FilesystemImpl.Release. -func (fs *filesystem) Release() { +func (fs *filesystem) Release(ctx context.Context) { fs.vfsfs.VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) } @@ -249,12 +249,12 @@ func (d *dentry) TryIncRef() bool { } // DecRef implements vfs.DentryImpl.DecRef. -func (d *dentry) DecRef() { - d.inode.decRef() +func (d *dentry) DecRef(ctx context.Context) { + d.inode.decRef(ctx) } // InotifyWithParent implements vfs.DentryImpl.InotifyWithParent. -func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) { +func (d *dentry) InotifyWithParent(ctx context.Context, events, cookie uint32, et vfs.EventType) { if d.inode.isDir() { events |= linux.IN_ISDIR } @@ -266,9 +266,9 @@ func (d *dentry) InotifyWithParent(events, cookie uint32, et vfs.EventType) { d.inode.fs.mu.RLock() // The ordering below is important, Linux always notifies the parent first. if d.parent != nil { - d.parent.inode.watches.Notify(d.name, events, cookie, et, deleted) + d.parent.inode.watches.Notify(ctx, d.name, events, cookie, et, deleted) } - d.inode.watches.Notify("", events, cookie, et, deleted) + d.inode.watches.Notify(ctx, "", events, cookie, et, deleted) d.inode.fs.mu.RUnlock() } @@ -278,7 +278,7 @@ func (d *dentry) Watches() *vfs.Watches { } // OnZeroWatches implements vfs.Dentry.OnZeroWatches. -func (d *dentry) OnZeroWatches() {} +func (d *dentry) OnZeroWatches(context.Context) {} // inode represents a filesystem object. type inode struct { @@ -359,12 +359,12 @@ func (i *inode) incLinksLocked() { // remove a reference on i as well. // // Preconditions: filesystem.mu must be locked for writing. i.nlink != 0. -func (i *inode) decLinksLocked() { +func (i *inode) decLinksLocked(ctx context.Context) { if i.nlink == 0 { panic("tmpfs.inode.decLinksLocked() called with no existing links") } if atomic.AddUint32(&i.nlink, ^uint32(0)) == 0 { - i.decRef() + i.decRef(ctx) } } @@ -386,9 +386,9 @@ func (i *inode) tryIncRef() bool { } } -func (i *inode) decRef() { +func (i *inode) decRef(ctx context.Context) { if refs := atomic.AddInt64(&i.refs, -1); refs == 0 { - i.watches.HandleDeletion() + i.watches.HandleDeletion(ctx) if regFile, ok := i.impl.(*regularFile); ok { // Release memory used by regFile to store data. Since regFile is // no longer usable, we don't need to grab any locks or update any @@ -701,7 +701,7 @@ func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) } if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 { - d.InotifyWithParent(ev, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, ev, 0, vfs.InodeEvent) } return nil } @@ -724,7 +724,7 @@ func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOption } // Generate inotify events. - d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } @@ -736,13 +736,13 @@ func (fd *fileDescription) Removexattr(ctx context.Context, name string) error { } // Generate inotify events. - d.InotifyWithParent(linux.IN_ATTRIB, 0, vfs.InodeEvent) + d.InotifyWithParent(ctx, linux.IN_ATTRIB, 0, vfs.InodeEvent) return nil } // NewMemfd creates a new tmpfs regular file and file description that can back // an anonymous fd created by memfd_create. -func NewMemfd(mount *vfs.Mount, creds *auth.Credentials, allowSeals bool, name string) (*vfs.FileDescription, error) { +func NewMemfd(ctx context.Context, creds *auth.Credentials, mount *vfs.Mount, allowSeals bool, name string) (*vfs.FileDescription, error) { fs, ok := mount.Filesystem().Impl().(*filesystem) if !ok { panic("NewMemfd() called with non-tmpfs mount") @@ -757,7 +757,7 @@ func NewMemfd(mount *vfs.Mount, creds *auth.Credentials, allowSeals bool, name s } d := fs.newDentry(inode) - defer d.DecRef() + defer d.DecRef(ctx) d.name = name // Per Linux, mm/shmem.c:__shmem_file_setup(), memfd files are set up with diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go b/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go index a240fb276..6f3e3ae6f 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs_test.go @@ -34,7 +34,7 @@ func newTmpfsRoot(ctx context.Context) (*vfs.VirtualFilesystem, vfs.VirtualDentr creds := auth.CredentialsFromContext(ctx) vfsObj := &vfs.VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("VFS init: %v", err) } @@ -47,8 +47,8 @@ func newTmpfsRoot(ctx context.Context) (*vfs.VirtualFilesystem, vfs.VirtualDentr } root := mntns.Root() return vfsObj, root, func() { - root.DecRef() - mntns.DecRef() + root.DecRef(ctx) + mntns.DecRef(ctx) }, nil } diff --git a/pkg/sentry/kernel/abstract_socket_namespace.go b/pkg/sentry/kernel/abstract_socket_namespace.go index 920fe4329..52ed5cea2 100644 --- a/pkg/sentry/kernel/abstract_socket_namespace.go +++ b/pkg/sentry/kernel/abstract_socket_namespace.go @@ -17,6 +17,7 @@ package kernel import ( "syscall" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sync" @@ -31,7 +32,7 @@ type abstractEndpoint struct { } // WeakRefGone implements refs.WeakRefUser.WeakRefGone. -func (e *abstractEndpoint) WeakRefGone() { +func (e *abstractEndpoint) WeakRefGone(context.Context) { e.ns.mu.Lock() if e.ns.endpoints[e.name].ep == e.ep { delete(e.ns.endpoints, e.name) @@ -64,9 +65,9 @@ type boundEndpoint struct { } // Release implements transport.BoundEndpoint.Release. -func (e *boundEndpoint) Release() { - e.rc.DecRef() - e.BoundEndpoint.Release() +func (e *boundEndpoint) Release(ctx context.Context) { + e.rc.DecRef(ctx) + e.BoundEndpoint.Release(ctx) } // BoundEndpoint retrieves the endpoint bound to the given name. The return @@ -93,13 +94,13 @@ func (a *AbstractSocketNamespace) BoundEndpoint(name string) transport.BoundEndp // // When the last reference managed by rc is dropped, ep may be removed from the // namespace. -func (a *AbstractSocketNamespace) Bind(name string, ep transport.BoundEndpoint, rc refs.RefCounter) error { +func (a *AbstractSocketNamespace) Bind(ctx context.Context, name string, ep transport.BoundEndpoint, rc refs.RefCounter) error { a.mu.Lock() defer a.mu.Unlock() if ep, ok := a.endpoints[name]; ok { if rc := ep.wr.Get(); rc != nil { - rc.DecRef() + rc.DecRef(ctx) return syscall.EADDRINUSE } } diff --git a/pkg/sentry/kernel/epoll/epoll.go b/pkg/sentry/kernel/epoll/epoll.go index 4c0f1e41f..15519f0df 100644 --- a/pkg/sentry/kernel/epoll/epoll.go +++ b/pkg/sentry/kernel/epoll/epoll.go @@ -76,8 +76,8 @@ type pollEntry struct { // WeakRefGone implements refs.WeakRefUser.WeakRefGone. // weakReferenceGone is called when the file in the weak reference is destroyed. // The poll entry is removed in response to this. -func (p *pollEntry) WeakRefGone() { - p.epoll.RemoveEntry(p.id) +func (p *pollEntry) WeakRefGone(ctx context.Context) { + p.epoll.RemoveEntry(ctx, p.id) } // EventPoll holds all the state associated with an event poll object, that is, @@ -144,14 +144,14 @@ func NewEventPoll(ctx context.Context) *fs.File { // name matches fs/eventpoll.c:epoll_create1. dirent := fs.NewDirent(ctx, anon.NewInode(ctx), fmt.Sprintf("anon_inode:[eventpoll]")) // Release the initial dirent reference after NewFile takes a reference. - defer dirent.DecRef() + defer dirent.DecRef(ctx) return fs.NewFile(ctx, dirent, fs.FileFlags{}, &EventPoll{ files: make(map[FileIdentifier]*pollEntry), }) } // Release implements fs.FileOperations.Release. -func (e *EventPoll) Release() { +func (e *EventPoll) Release(ctx context.Context) { // We need to take the lock now because files may be attempting to // remove entries in parallel if they get destroyed. e.mu.Lock() @@ -160,7 +160,7 @@ func (e *EventPoll) Release() { // Go through all entries and clean up. for _, entry := range e.files { entry.id.File.EventUnregister(&entry.waiter) - entry.file.Drop() + entry.file.Drop(ctx) } e.files = nil } @@ -423,7 +423,7 @@ func (e *EventPoll) UpdateEntry(id FileIdentifier, flags EntryFlags, mask waiter } // RemoveEntry a files from the collection of observed files. -func (e *EventPoll) RemoveEntry(id FileIdentifier) error { +func (e *EventPoll) RemoveEntry(ctx context.Context, id FileIdentifier) error { e.mu.Lock() defer e.mu.Unlock() @@ -445,7 +445,7 @@ func (e *EventPoll) RemoveEntry(id FileIdentifier) error { // Remove file from map, and drop weak reference. delete(e.files, id) - entry.file.Drop() + entry.file.Drop(ctx) return nil } diff --git a/pkg/sentry/kernel/epoll/epoll_test.go b/pkg/sentry/kernel/epoll/epoll_test.go index 22630e9c5..55b505593 100644 --- a/pkg/sentry/kernel/epoll/epoll_test.go +++ b/pkg/sentry/kernel/epoll/epoll_test.go @@ -26,7 +26,8 @@ func TestFileDestroyed(t *testing.T) { f := filetest.NewTestFile(t) id := FileIdentifier{f, 12} - efile := NewEventPoll(contexttest.Context(t)) + ctx := contexttest.Context(t) + efile := NewEventPoll(ctx) e := efile.FileOperations.(*EventPoll) if err := e.AddEntry(id, 0, waiter.EventIn, [2]int32{}); err != nil { t.Fatalf("addEntry failed: %v", err) @@ -44,7 +45,7 @@ func TestFileDestroyed(t *testing.T) { } // Destroy the file. Check that we get no more events. - f.DecRef() + f.DecRef(ctx) evt = e.ReadEvents(1) if len(evt) != 0 { diff --git a/pkg/sentry/kernel/eventfd/eventfd.go b/pkg/sentry/kernel/eventfd/eventfd.go index 87951adeb..bbf568dfc 100644 --- a/pkg/sentry/kernel/eventfd/eventfd.go +++ b/pkg/sentry/kernel/eventfd/eventfd.go @@ -70,7 +70,7 @@ func New(ctx context.Context, initVal uint64, semMode bool) *fs.File { // name matches fs/eventfd.c:eventfd_file_create. dirent := fs.NewDirent(ctx, anon.NewInode(ctx), "anon_inode:[eventfd]") // Release the initial dirent reference after NewFile takes a reference. - defer dirent.DecRef() + defer dirent.DecRef(ctx) return fs.NewFile(ctx, dirent, fs.FileFlags{Read: true, Write: true}, &EventOperations{ val: initVal, semMode: semMode, @@ -106,7 +106,7 @@ func (e *EventOperations) HostFD() (int, error) { } // Release implements fs.FileOperations.Release. -func (e *EventOperations) Release() { +func (e *EventOperations) Release(context.Context) { e.mu.Lock() defer e.mu.Unlock() if e.hostfd >= 0 { diff --git a/pkg/sentry/kernel/fd_table.go b/pkg/sentry/kernel/fd_table.go index 4b7d234a4..ce53af69b 100644 --- a/pkg/sentry/kernel/fd_table.go +++ b/pkg/sentry/kernel/fd_table.go @@ -98,7 +98,7 @@ type FDTable struct { func (f *FDTable) saveDescriptorTable() map[int32]descriptor { m := make(map[int32]descriptor) - f.forEach(func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { + f.forEach(context.Background(), func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { m[fd] = descriptor{ file: file, fileVFS2: fileVFS2, @@ -109,6 +109,7 @@ func (f *FDTable) saveDescriptorTable() map[int32]descriptor { } func (f *FDTable) loadDescriptorTable(m map[int32]descriptor) { + ctx := context.Background() f.init() // Initialize table. for fd, d := range m { f.setAll(fd, d.file, d.fileVFS2, d.flags) @@ -118,9 +119,9 @@ func (f *FDTable) loadDescriptorTable(m map[int32]descriptor) { // reference taken by set above. switch { case d.file != nil: - d.file.DecRef() + d.file.DecRef(ctx) case d.fileVFS2 != nil: - d.fileVFS2.DecRef() + d.fileVFS2.DecRef(ctx) } } } @@ -144,14 +145,15 @@ func (f *FDTable) drop(file *fs.File) { d.InotifyEvent(ev, 0) // Drop the table reference. - file.DecRef() + file.DecRef(context.Background()) } // dropVFS2 drops the table reference. func (f *FDTable) dropVFS2(file *vfs.FileDescription) { // Release any POSIX lock possibly held by the FDTable. Range {0, 0} means the // entire file. - err := file.UnlockPOSIX(context.Background(), f, 0, 0, linux.SEEK_SET) + ctx := context.Background() + err := file.UnlockPOSIX(ctx, f, 0, 0, linux.SEEK_SET) if err != nil && err != syserror.ENOLCK { panic(fmt.Sprintf("UnlockPOSIX failed: %v", err)) } @@ -161,10 +163,10 @@ func (f *FDTable) dropVFS2(file *vfs.FileDescription) { if file.IsWritable() { ev = linux.IN_CLOSE_WRITE } - file.Dentry().InotifyWithParent(ev, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(ctx, ev, 0, vfs.PathEvent) // Drop the table's reference. - file.DecRef() + file.DecRef(ctx) } // NewFDTable allocates a new FDTable that may be used by tasks in k. @@ -175,15 +177,15 @@ func (k *Kernel) NewFDTable() *FDTable { } // destroy removes all of the file descriptors from the map. -func (f *FDTable) destroy() { - f.RemoveIf(func(*fs.File, *vfs.FileDescription, FDFlags) bool { +func (f *FDTable) destroy(ctx context.Context) { + f.RemoveIf(ctx, func(*fs.File, *vfs.FileDescription, FDFlags) bool { return true }) } // DecRef implements RefCounter.DecRef with destructor f.destroy. -func (f *FDTable) DecRef() { - f.DecRefWithDestructor(f.destroy) +func (f *FDTable) DecRef(ctx context.Context) { + f.DecRefWithDestructor(ctx, f.destroy) } // Size returns the number of file descriptor slots currently allocated. @@ -195,7 +197,7 @@ func (f *FDTable) Size() int { // forEach iterates over all non-nil files in sorted order. // // It is the caller's responsibility to acquire an appropriate lock. -func (f *FDTable) forEach(fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags)) { +func (f *FDTable) forEach(ctx context.Context, fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags)) { // retries tracks the number of failed TryIncRef attempts for the same FD. retries := 0 fd := int32(0) @@ -214,7 +216,7 @@ func (f *FDTable) forEach(fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDes continue // Race caught. } fn(fd, file, nil, flags) - file.DecRef() + file.DecRef(ctx) case fileVFS2 != nil: if !fileVFS2.TryIncRef() { retries++ @@ -224,7 +226,7 @@ func (f *FDTable) forEach(fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDes continue // Race caught. } fn(fd, nil, fileVFS2, flags) - fileVFS2.DecRef() + fileVFS2.DecRef(ctx) } retries = 0 fd++ @@ -234,7 +236,8 @@ func (f *FDTable) forEach(fn func(fd int32, file *fs.File, fileVFS2 *vfs.FileDes // String is a stringer for FDTable. func (f *FDTable) String() string { var buf strings.Builder - f.forEach(func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { + ctx := context.Background() + f.forEach(ctx, func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { switch { case file != nil: n, _ := file.Dirent.FullName(nil /* root */) @@ -242,7 +245,7 @@ func (f *FDTable) String() string { case fileVFS2 != nil: vfsObj := fileVFS2.Mount().Filesystem().VirtualFilesystem() - name, err := vfsObj.PathnameWithDeleted(context.Background(), vfs.VirtualDentry{}, fileVFS2.VirtualDentry()) + name, err := vfsObj.PathnameWithDeleted(ctx, vfs.VirtualDentry{}, fileVFS2.VirtualDentry()) if err != nil { fmt.Fprintf(&buf, "<err: %v>\n", err) return @@ -541,9 +544,9 @@ func (f *FDTable) GetVFS2(fd int32) (*vfs.FileDescription, FDFlags) { // // Precondition: The caller must be running on the task goroutine, or Task.mu // must be locked. -func (f *FDTable) GetFDs() []int32 { +func (f *FDTable) GetFDs(ctx context.Context) []int32 { fds := make([]int32, 0, int(atomic.LoadInt32(&f.used))) - f.forEach(func(fd int32, _ *fs.File, _ *vfs.FileDescription, _ FDFlags) { + f.forEach(ctx, func(fd int32, _ *fs.File, _ *vfs.FileDescription, _ FDFlags) { fds = append(fds, fd) }) return fds @@ -552,9 +555,9 @@ func (f *FDTable) GetFDs() []int32 { // GetRefs returns a stable slice of references to all files and bumps the // reference count on each. The caller must use DecRef on each reference when // they're done using the slice. -func (f *FDTable) GetRefs() []*fs.File { +func (f *FDTable) GetRefs(ctx context.Context) []*fs.File { files := make([]*fs.File, 0, f.Size()) - f.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { + f.forEach(ctx, func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { file.IncRef() // Acquire a reference for caller. files = append(files, file) }) @@ -564,9 +567,9 @@ func (f *FDTable) GetRefs() []*fs.File { // GetRefsVFS2 returns a stable slice of references to all files and bumps the // reference count on each. The caller must use DecRef on each reference when // they're done using the slice. -func (f *FDTable) GetRefsVFS2() []*vfs.FileDescription { +func (f *FDTable) GetRefsVFS2(ctx context.Context) []*vfs.FileDescription { files := make([]*vfs.FileDescription, 0, f.Size()) - f.forEach(func(_ int32, _ *fs.File, file *vfs.FileDescription, _ FDFlags) { + f.forEach(ctx, func(_ int32, _ *fs.File, file *vfs.FileDescription, _ FDFlags) { file.IncRef() // Acquire a reference for caller. files = append(files, file) }) @@ -574,10 +577,10 @@ func (f *FDTable) GetRefsVFS2() []*vfs.FileDescription { } // Fork returns an independent FDTable. -func (f *FDTable) Fork() *FDTable { +func (f *FDTable) Fork(ctx context.Context) *FDTable { clone := f.k.NewFDTable() - f.forEach(func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { + f.forEach(ctx, func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { // The set function here will acquire an appropriate table // reference for the clone. We don't need anything else. switch { @@ -622,11 +625,11 @@ func (f *FDTable) Remove(fd int32) (*fs.File, *vfs.FileDescription) { } // RemoveIf removes all FDs where cond is true. -func (f *FDTable) RemoveIf(cond func(*fs.File, *vfs.FileDescription, FDFlags) bool) { +func (f *FDTable) RemoveIf(ctx context.Context, cond func(*fs.File, *vfs.FileDescription, FDFlags) bool) { f.mu.Lock() defer f.mu.Unlock() - f.forEach(func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { + f.forEach(ctx, func(fd int32, file *fs.File, fileVFS2 *vfs.FileDescription, flags FDFlags) { if cond(file, fileVFS2, flags) { f.set(fd, nil, FDFlags{}) // Clear from table. // Update current available position. diff --git a/pkg/sentry/kernel/fd_table_test.go b/pkg/sentry/kernel/fd_table_test.go index 29f95a2c4..e3f30ba2a 100644 --- a/pkg/sentry/kernel/fd_table_test.go +++ b/pkg/sentry/kernel/fd_table_test.go @@ -154,7 +154,7 @@ func TestFDTable(t *testing.T) { if ref == nil { t.Fatalf("fdTable.Remove(1) for an existing FD: failed, want success") } - ref.DecRef() + ref.DecRef(ctx) if ref, _ := fdTable.Remove(1); ref != nil { t.Fatalf("r.Remove(1) for a removed FD: got success, want failure") @@ -191,7 +191,7 @@ func BenchmarkFDLookupAndDecRef(b *testing.B) { b.StartTimer() // Benchmark. for i := 0; i < b.N; i++ { tf, _ := fdTable.Get(fds[i%len(fds)]) - tf.DecRef() + tf.DecRef(ctx) } }) } @@ -219,7 +219,7 @@ func BenchmarkFDLookupAndDecRefConcurrent(b *testing.B) { defer wg.Done() for i := 0; i < each; i++ { tf, _ := fdTable.Get(fds[i%len(fds)]) - tf.DecRef() + tf.DecRef(ctx) } }() } diff --git a/pkg/sentry/kernel/fs_context.go b/pkg/sentry/kernel/fs_context.go index 47f78df9a..8f2d36d5a 100644 --- a/pkg/sentry/kernel/fs_context.go +++ b/pkg/sentry/kernel/fs_context.go @@ -17,6 +17,7 @@ package kernel import ( "fmt" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/vfs" @@ -89,28 +90,28 @@ func NewFSContextVFS2(root, cwd vfs.VirtualDentry, umask uint) *FSContext { // Note that there may still be calls to WorkingDirectory() or RootDirectory() // (that return nil). This is because valid references may still be held via // proc files or other mechanisms. -func (f *FSContext) destroy() { +func (f *FSContext) destroy(ctx context.Context) { // Hold f.mu so that we don't race with RootDirectory() and // WorkingDirectory(). f.mu.Lock() defer f.mu.Unlock() if VFS2Enabled { - f.rootVFS2.DecRef() + f.rootVFS2.DecRef(ctx) f.rootVFS2 = vfs.VirtualDentry{} - f.cwdVFS2.DecRef() + f.cwdVFS2.DecRef(ctx) f.cwdVFS2 = vfs.VirtualDentry{} } else { - f.root.DecRef() + f.root.DecRef(ctx) f.root = nil - f.cwd.DecRef() + f.cwd.DecRef(ctx) f.cwd = nil } } // DecRef implements RefCounter.DecRef with destructor f.destroy. -func (f *FSContext) DecRef() { - f.DecRefWithDestructor(f.destroy) +func (f *FSContext) DecRef(ctx context.Context) { + f.DecRefWithDestructor(ctx, f.destroy) } // Fork forks this FSContext. @@ -165,7 +166,7 @@ func (f *FSContext) WorkingDirectoryVFS2() vfs.VirtualDentry { // This will take an extra reference on the Dirent. // // This is not a valid call after destroy. -func (f *FSContext) SetWorkingDirectory(d *fs.Dirent) { +func (f *FSContext) SetWorkingDirectory(ctx context.Context, d *fs.Dirent) { if d == nil { panic("FSContext.SetWorkingDirectory called with nil dirent") } @@ -180,21 +181,21 @@ func (f *FSContext) SetWorkingDirectory(d *fs.Dirent) { old := f.cwd f.cwd = d d.IncRef() - old.DecRef() + old.DecRef(ctx) } // SetWorkingDirectoryVFS2 sets the current working directory. // This will take an extra reference on the VirtualDentry. // // This is not a valid call after destroy. -func (f *FSContext) SetWorkingDirectoryVFS2(d vfs.VirtualDentry) { +func (f *FSContext) SetWorkingDirectoryVFS2(ctx context.Context, d vfs.VirtualDentry) { f.mu.Lock() defer f.mu.Unlock() old := f.cwdVFS2 f.cwdVFS2 = d d.IncRef() - old.DecRef() + old.DecRef(ctx) } // RootDirectory returns the current filesystem root. @@ -226,7 +227,7 @@ func (f *FSContext) RootDirectoryVFS2() vfs.VirtualDentry { // This will take an extra reference on the Dirent. // // This is not a valid call after free. -func (f *FSContext) SetRootDirectory(d *fs.Dirent) { +func (f *FSContext) SetRootDirectory(ctx context.Context, d *fs.Dirent) { if d == nil { panic("FSContext.SetRootDirectory called with nil dirent") } @@ -241,13 +242,13 @@ func (f *FSContext) SetRootDirectory(d *fs.Dirent) { old := f.root f.root = d d.IncRef() - old.DecRef() + old.DecRef(ctx) } // SetRootDirectoryVFS2 sets the root directory. It takes a reference on vd. // // This is not a valid call after free. -func (f *FSContext) SetRootDirectoryVFS2(vd vfs.VirtualDentry) { +func (f *FSContext) SetRootDirectoryVFS2(ctx context.Context, vd vfs.VirtualDentry) { if !vd.Ok() { panic("FSContext.SetRootDirectoryVFS2 called with zero-value VirtualDentry") } @@ -263,7 +264,7 @@ func (f *FSContext) SetRootDirectoryVFS2(vd vfs.VirtualDentry) { vd.IncRef() f.rootVFS2 = vd f.mu.Unlock() - old.DecRef() + old.DecRef(ctx) } // Umask returns the current umask. diff --git a/pkg/sentry/kernel/futex/BUILD b/pkg/sentry/kernel/futex/BUILD index c5021f2db..daa2dae76 100644 --- a/pkg/sentry/kernel/futex/BUILD +++ b/pkg/sentry/kernel/futex/BUILD @@ -51,6 +51,7 @@ go_test( srcs = ["futex_test.go"], library = ":futex", deps = [ + "//pkg/context", "//pkg/sync", "//pkg/usermem", ], diff --git a/pkg/sentry/kernel/futex/futex.go b/pkg/sentry/kernel/futex/futex.go index bcc1b29a8..e4dcc4d40 100644 --- a/pkg/sentry/kernel/futex/futex.go +++ b/pkg/sentry/kernel/futex/futex.go @@ -19,6 +19,7 @@ package futex import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" @@ -66,9 +67,9 @@ type Key struct { Offset uint64 } -func (k *Key) release() { +func (k *Key) release(t Target) { if k.MappingIdentity != nil { - k.MappingIdentity.DecRef() + k.MappingIdentity.DecRef(t) } k.Mappable = nil k.MappingIdentity = nil @@ -94,6 +95,8 @@ func (k *Key) matches(k2 *Key) bool { // Target abstracts memory accesses and keys. type Target interface { + context.Context + // SwapUint32 gives access to usermem.IO.SwapUint32. SwapUint32(addr usermem.Addr, new uint32) (uint32, error) @@ -296,7 +299,7 @@ func (b *bucket) wakeWaiterLocked(w *Waiter) { // bucket "to". // // Preconditions: b and to must be locked. -func (b *bucket) requeueLocked(to *bucket, key, nkey *Key, n int) int { +func (b *bucket) requeueLocked(t Target, to *bucket, key, nkey *Key, n int) int { done := 0 for w := b.waiters.Front(); done < n && w != nil; { if !w.key.matches(key) { @@ -308,7 +311,7 @@ func (b *bucket) requeueLocked(to *bucket, key, nkey *Key, n int) int { requeued := w w = w.Next() // Next iteration. b.waiters.Remove(requeued) - requeued.key.release() + requeued.key.release(t) requeued.key = nkey.clone() to.waiters.PushBack(requeued) requeued.bucket.Store(to) @@ -456,7 +459,7 @@ func (m *Manager) Wake(t Target, addr usermem.Addr, private bool, bitmask uint32 r := b.wakeLocked(&k, bitmask, n) b.mu.Unlock() - k.release() + k.release(t) return r, nil } @@ -465,12 +468,12 @@ func (m *Manager) doRequeue(t Target, addr, naddr usermem.Addr, private bool, ch if err != nil { return 0, err } - defer k1.release() + defer k1.release(t) k2, err := getKey(t, naddr, private) if err != nil { return 0, err } - defer k2.release() + defer k2.release(t) b1, b2 := m.lockBuckets(&k1, &k2) defer b1.mu.Unlock() @@ -488,7 +491,7 @@ func (m *Manager) doRequeue(t Target, addr, naddr usermem.Addr, private bool, ch done := b1.wakeLocked(&k1, ^uint32(0), nwake) // Requeue the number required. - b1.requeueLocked(b2, &k1, &k2, nreq) + b1.requeueLocked(t, b2, &k1, &k2, nreq) return done, nil } @@ -515,12 +518,12 @@ func (m *Manager) WakeOp(t Target, addr1, addr2 usermem.Addr, private bool, nwak if err != nil { return 0, err } - defer k1.release() + defer k1.release(t) k2, err := getKey(t, addr2, private) if err != nil { return 0, err } - defer k2.release() + defer k2.release(t) b1, b2 := m.lockBuckets(&k1, &k2) defer b1.mu.Unlock() @@ -571,7 +574,7 @@ func (m *Manager) WaitPrepare(w *Waiter, t Target, addr usermem.Addr, private bo // Perform our atomic check. if err := check(t, addr, val); err != nil { b.mu.Unlock() - w.key.release() + w.key.release(t) return err } @@ -585,7 +588,7 @@ func (m *Manager) WaitPrepare(w *Waiter, t Target, addr usermem.Addr, private bo // WaitComplete must be called when a Waiter previously added by WaitPrepare is // no longer eligible to be woken. -func (m *Manager) WaitComplete(w *Waiter) { +func (m *Manager) WaitComplete(w *Waiter, t Target) { // Remove w from the bucket it's in. for { b := w.bucket.Load() @@ -617,7 +620,7 @@ func (m *Manager) WaitComplete(w *Waiter) { } // Release references held by the waiter. - w.key.release() + w.key.release(t) } // LockPI attempts to lock the futex following the Priority-inheritance futex @@ -648,13 +651,13 @@ func (m *Manager) LockPI(w *Waiter, t Target, addr usermem.Addr, tid uint32, pri success, err := m.lockPILocked(w, t, addr, tid, b, try) if err != nil { - w.key.release() + w.key.release(t) b.mu.Unlock() return false, err } if success || try { // Release waiter if it's not going to be a wait. - w.key.release() + w.key.release(t) } b.mu.Unlock() return success, nil @@ -730,7 +733,7 @@ func (m *Manager) UnlockPI(t Target, addr usermem.Addr, tid uint32, private bool err = m.unlockPILocked(t, addr, tid, b, &k) - k.release() + k.release(t) b.mu.Unlock() return err } diff --git a/pkg/sentry/kernel/futex/futex_test.go b/pkg/sentry/kernel/futex/futex_test.go index 7c5c7665b..d0128c548 100644 --- a/pkg/sentry/kernel/futex/futex_test.go +++ b/pkg/sentry/kernel/futex/futex_test.go @@ -22,6 +22,7 @@ import ( "testing" "unsafe" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/usermem" ) @@ -29,28 +30,33 @@ import ( // testData implements the Target interface, and allows us to // treat the address passed for futex operations as an index in // a byte slice for testing simplicity. -type testData []byte +type testData struct { + context.Context + data []byte +} const sizeofInt32 = 4 func newTestData(size uint) testData { - return make([]byte, size) + return testData{ + data: make([]byte, size), + } } func (t testData) SwapUint32(addr usermem.Addr, new uint32) (uint32, error) { - val := atomic.SwapUint32((*uint32)(unsafe.Pointer(&t[addr])), new) + val := atomic.SwapUint32((*uint32)(unsafe.Pointer(&t.data[addr])), new) return val, nil } func (t testData) CompareAndSwapUint32(addr usermem.Addr, old, new uint32) (uint32, error) { - if atomic.CompareAndSwapUint32((*uint32)(unsafe.Pointer(&t[addr])), old, new) { + if atomic.CompareAndSwapUint32((*uint32)(unsafe.Pointer(&t.data[addr])), old, new) { return old, nil } - return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t[addr]))), nil + return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t.data[addr]))), nil } func (t testData) LoadUint32(addr usermem.Addr) (uint32, error) { - return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t[addr]))), nil + return atomic.LoadUint32((*uint32)(unsafe.Pointer(&t.data[addr]))), nil } func (t testData) GetSharedKey(addr usermem.Addr) (Key, error) { @@ -83,7 +89,7 @@ func TestFutexWake(t *testing.T) { // Start waiting for wakeup. w := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w) + defer m.WaitComplete(w, d) // Perform a wakeup. if n, err := m.Wake(d, 0, private, ^uint32(0), 1); err != nil || n != 1 { @@ -106,7 +112,7 @@ func TestFutexWakeBitmask(t *testing.T) { // Start waiting for wakeup. w := newPreparedTestWaiter(t, m, d, 0, private, 0, 0x0000ffff) - defer m.WaitComplete(w) + defer m.WaitComplete(w, d) // Perform a wakeup using the wrong bitmask. if n, err := m.Wake(d, 0, private, 0xffff0000, 1); err != nil || n != 0 { @@ -141,7 +147,7 @@ func TestFutexWakeTwo(t *testing.T) { var ws [3]*Waiter for i := range ws { ws[i] = newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(ws[i]) + defer m.WaitComplete(ws[i], d) } // Perform two wakeups. @@ -174,9 +180,9 @@ func TestFutexWakeUnrelated(t *testing.T) { // Start two waiters waiting for wakeup on different addresses. w1 := newPreparedTestWaiter(t, m, d, 0*sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, 1*sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Perform two wakeups on the second address. if n, err := m.Wake(d, 1*sizeofInt32, private, ^uint32(0), 2); err != nil || n != 1 { @@ -216,9 +222,9 @@ func TestWakeOpFirstNonEmpty(t *testing.T) { // Add two waiters on address 0. w1 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Perform 10 wakeups on address 0. if n, err := m.WakeOp(d, 0, sizeofInt32, private, 10, 0, 0); err != nil || n != 2 { @@ -244,9 +250,9 @@ func TestWakeOpSecondNonEmpty(t *testing.T) { // Add two waiters on address sizeofInt32. w1 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Perform 10 wakeups on address sizeofInt32 (contingent on // d.Op(0), which should succeed). @@ -273,9 +279,9 @@ func TestWakeOpSecondNonEmptyFailingOp(t *testing.T) { // Add two waiters on address sizeofInt32. w1 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Perform 10 wakeups on address sizeofInt32 (contingent on // d.Op(1), which should fail). @@ -302,15 +308,15 @@ func TestWakeOpAllNonEmpty(t *testing.T) { // Add two waiters on address 0. w1 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Add two waiters on address sizeofInt32. w3 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w3) + defer m.WaitComplete(w3, d) w4 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w4) + defer m.WaitComplete(w4, d) // Perform 10 wakeups on address 0 (unconditionally), and 10 // wakeups on address sizeofInt32 (contingent on d.Op(0), which @@ -344,15 +350,15 @@ func TestWakeOpAllNonEmptyFailingOp(t *testing.T) { // Add two waiters on address 0. w1 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w1) + defer m.WaitComplete(w1, d) w2 := newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(w2) + defer m.WaitComplete(w2, d) // Add two waiters on address sizeofInt32. w3 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w3) + defer m.WaitComplete(w3, d) w4 := newPreparedTestWaiter(t, m, d, sizeofInt32, private, 0, ^uint32(0)) - defer m.WaitComplete(w4) + defer m.WaitComplete(w4, d) // Perform 10 wakeups on address 0 (unconditionally), and 10 // wakeups on address sizeofInt32 (contingent on d.Op(1), which @@ -388,7 +394,7 @@ func TestWakeOpSameAddress(t *testing.T) { var ws [4]*Waiter for i := range ws { ws[i] = newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(ws[i]) + defer m.WaitComplete(ws[i], d) } // Perform 1 wakeup on address 0 (unconditionally), and 1 wakeup @@ -422,7 +428,7 @@ func TestWakeOpSameAddressFailingOp(t *testing.T) { var ws [4]*Waiter for i := range ws { ws[i] = newPreparedTestWaiter(t, m, d, 0, private, 0, ^uint32(0)) - defer m.WaitComplete(ws[i]) + defer m.WaitComplete(ws[i], d) } // Perform 1 wakeup on address 0 (unconditionally), and 1 wakeup @@ -472,7 +478,7 @@ func (t *testMutex) Lock() { for { // Attempt to grab the lock. if atomic.CompareAndSwapUint32( - (*uint32)(unsafe.Pointer(&t.d[t.a])), + (*uint32)(unsafe.Pointer(&t.d.data[t.a])), testMutexUnlocked, testMutexLocked) { // Lock held. @@ -490,7 +496,7 @@ func (t *testMutex) Lock() { panic("WaitPrepare returned unexpected error: " + err.Error()) } <-w.C - t.m.WaitComplete(w) + t.m.WaitComplete(w, t.d) } } @@ -498,7 +504,7 @@ func (t *testMutex) Lock() { // This will notify any waiters via the futex manager. func (t *testMutex) Unlock() { // Unlock. - atomic.StoreUint32((*uint32)(unsafe.Pointer(&t.d[t.a])), testMutexUnlocked) + atomic.StoreUint32((*uint32)(unsafe.Pointer(&t.d.data[t.a])), testMutexUnlocked) // Notify all waiters. t.m.Wake(t.d, t.a, true, ^uint32(0), math.MaxInt32) diff --git a/pkg/sentry/kernel/kernel.go b/pkg/sentry/kernel/kernel.go index 15dae0f5b..1028d13c6 100644 --- a/pkg/sentry/kernel/kernel.go +++ b/pkg/sentry/kernel/kernel.go @@ -376,7 +376,8 @@ func (k *Kernel) Init(args InitKernelArgs) error { k.netlinkPorts = port.New() if VFS2Enabled { - if err := k.vfs.Init(); err != nil { + ctx := k.SupervisorContext() + if err := k.vfs.Init(ctx); err != nil { return fmt.Errorf("failed to initialize VFS: %v", err) } @@ -384,19 +385,19 @@ func (k *Kernel) Init(args InitKernelArgs) error { if err != nil { return fmt.Errorf("failed to create pipefs filesystem: %v", err) } - defer pipeFilesystem.DecRef() + defer pipeFilesystem.DecRef(ctx) pipeMount, err := k.vfs.NewDisconnectedMount(pipeFilesystem, nil, &vfs.MountOptions{}) if err != nil { return fmt.Errorf("failed to create pipefs mount: %v", err) } k.pipeMount = pipeMount - tmpfsFilesystem, tmpfsRoot, err := tmpfs.NewFilesystem(k.SupervisorContext(), &k.vfs, auth.NewRootCredentials(k.rootUserNamespace)) + tmpfsFilesystem, tmpfsRoot, err := tmpfs.NewFilesystem(ctx, &k.vfs, auth.NewRootCredentials(k.rootUserNamespace)) if err != nil { return fmt.Errorf("failed to create tmpfs filesystem: %v", err) } - defer tmpfsFilesystem.DecRef() - defer tmpfsRoot.DecRef() + defer tmpfsFilesystem.DecRef(ctx) + defer tmpfsRoot.DecRef(ctx) shmMount, err := k.vfs.NewDisconnectedMount(tmpfsFilesystem, tmpfsRoot, &vfs.MountOptions{}) if err != nil { return fmt.Errorf("failed to create tmpfs mount: %v", err) @@ -407,7 +408,7 @@ func (k *Kernel) Init(args InitKernelArgs) error { if err != nil { return fmt.Errorf("failed to create sockfs filesystem: %v", err) } - defer socketFilesystem.DecRef() + defer socketFilesystem.DecRef(ctx) socketMount, err := k.vfs.NewDisconnectedMount(socketFilesystem, nil, &vfs.MountOptions{}) if err != nil { return fmt.Errorf("failed to create sockfs mount: %v", err) @@ -430,8 +431,8 @@ func (k *Kernel) SaveTo(w wire.Writer) error { defer k.extMu.Unlock() // Stop time. - k.pauseTimeLocked() - defer k.resumeTimeLocked() + k.pauseTimeLocked(ctx) + defer k.resumeTimeLocked(ctx) // Evict all evictable MemoryFile allocations. k.mf.StartEvictions() @@ -447,12 +448,12 @@ func (k *Kernel) SaveTo(w wire.Writer) error { // Remove all epoll waiter objects from underlying wait queues. // NOTE: for programs to resume execution in future snapshot scenarios, // we will need to re-establish these waiter objects after saving. - k.tasks.unregisterEpollWaiters() + k.tasks.unregisterEpollWaiters(ctx) // Clear the dirent cache before saving because Dirents must be Loaded in a // particular order (parents before children), and Loading dirents from a cache // breaks that order. - if err := k.flushMountSourceRefs(); err != nil { + if err := k.flushMountSourceRefs(ctx); err != nil { return err } @@ -505,7 +506,7 @@ func (k *Kernel) SaveTo(w wire.Writer) error { // flushMountSourceRefs flushes the MountSources for all mounted filesystems // and open FDs. -func (k *Kernel) flushMountSourceRefs() error { +func (k *Kernel) flushMountSourceRefs(ctx context.Context) error { // Flush all mount sources for currently mounted filesystems in each task. flushed := make(map[*fs.MountNamespace]struct{}) k.tasks.mu.RLock() @@ -521,7 +522,7 @@ func (k *Kernel) flushMountSourceRefs() error { // There may be some open FDs whose filesystems have been unmounted. We // must flush those as well. - return k.tasks.forEachFDPaused(func(file *fs.File, _ *vfs.FileDescription) error { + return k.tasks.forEachFDPaused(ctx, func(file *fs.File, _ *vfs.FileDescription) error { file.Dirent.Inode.MountSource.FlushDirentRefs() return nil }) @@ -531,7 +532,7 @@ func (k *Kernel) flushMountSourceRefs() error { // each task. // // Precondition: Must be called with the kernel paused. -func (ts *TaskSet) forEachFDPaused(f func(*fs.File, *vfs.FileDescription) error) (err error) { +func (ts *TaskSet) forEachFDPaused(ctx context.Context, f func(*fs.File, *vfs.FileDescription) error) (err error) { // TODO(gvisor.dev/issue/1663): Add save support for VFS2. if VFS2Enabled { return nil @@ -544,7 +545,7 @@ func (ts *TaskSet) forEachFDPaused(f func(*fs.File, *vfs.FileDescription) error) if t.fdTable == nil { continue } - t.fdTable.forEach(func(_ int32, file *fs.File, fileVFS2 *vfs.FileDescription, _ FDFlags) { + t.fdTable.forEach(ctx, func(_ int32, file *fs.File, fileVFS2 *vfs.FileDescription, _ FDFlags) { if lastErr := f(file, fileVFS2); lastErr != nil && err == nil { err = lastErr } @@ -555,7 +556,7 @@ func (ts *TaskSet) forEachFDPaused(f func(*fs.File, *vfs.FileDescription) error) func (ts *TaskSet) flushWritesToFiles(ctx context.Context) error { // TODO(gvisor.dev/issue/1663): Add save support for VFS2. - return ts.forEachFDPaused(func(file *fs.File, _ *vfs.FileDescription) error { + return ts.forEachFDPaused(ctx, func(file *fs.File, _ *vfs.FileDescription) error { if flags := file.Flags(); !flags.Write { return nil } @@ -602,7 +603,7 @@ func (k *Kernel) invalidateUnsavableMappings(ctx context.Context) error { return nil } -func (ts *TaskSet) unregisterEpollWaiters() { +func (ts *TaskSet) unregisterEpollWaiters(ctx context.Context) { // TODO(gvisor.dev/issue/1663): Add save support for VFS2. if VFS2Enabled { return @@ -623,7 +624,7 @@ func (ts *TaskSet) unregisterEpollWaiters() { if _, ok := processed[t.fdTable]; ok { continue } - t.fdTable.forEach(func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { + t.fdTable.forEach(ctx, func(_ int32, file *fs.File, _ *vfs.FileDescription, _ FDFlags) { if e, ok := file.FileOperations.(*epoll.EventPoll); ok { e.UnregisterEpollWaiters() } @@ -900,7 +901,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, root := args.MountNamespaceVFS2.Root() // The call to newFSContext below will take a reference on root, so we // don't need to hold this one. - defer root.DecRef() + defer root.DecRef(ctx) // Grab the working directory. wd := root // Default. @@ -918,7 +919,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, if err != nil { return nil, 0, fmt.Errorf("failed to find initial working directory %q: %v", args.WorkingDirectory, err) } - defer wd.DecRef() + defer wd.DecRef(ctx) } opener = fsbridge.NewVFSLookup(mntnsVFS2, root, wd) fsContext = NewFSContextVFS2(root, wd, args.Umask) @@ -933,7 +934,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, root := mntns.Root() // The call to newFSContext below will take a reference on root, so we // don't need to hold this one. - defer root.DecRef() + defer root.DecRef(ctx) // Grab the working directory. remainingTraversals := args.MaxSymlinkTraversals @@ -944,7 +945,7 @@ func (k *Kernel) CreateProcess(args CreateProcessArgs) (*ThreadGroup, ThreadID, if err != nil { return nil, 0, fmt.Errorf("failed to find initial working directory %q: %v", args.WorkingDirectory, err) } - defer wd.DecRef() + defer wd.DecRef(ctx) } opener = fsbridge.NewFSLookup(mntns, root, wd) fsContext = newFSContext(root, wd, args.Umask) @@ -1054,7 +1055,7 @@ func (k *Kernel) Start() error { // If k was created by LoadKernelFrom, timers were stopped during // Kernel.SaveTo and need to be resumed. If k was created by NewKernel, // this is a no-op. - k.resumeTimeLocked() + k.resumeTimeLocked(k.SupervisorContext()) // Start task goroutines. k.tasks.mu.RLock() defer k.tasks.mu.RUnlock() @@ -1068,7 +1069,7 @@ func (k *Kernel) Start() error { // // Preconditions: Any task goroutines running in k must be stopped. k.extMu // must be locked. -func (k *Kernel) pauseTimeLocked() { +func (k *Kernel) pauseTimeLocked(ctx context.Context) { // k.cpuClockTicker may be nil since Kernel.SaveTo() may be called before // Kernel.Start(). if k.cpuClockTicker != nil { @@ -1090,7 +1091,7 @@ func (k *Kernel) pauseTimeLocked() { // This means we'll iterate FDTables shared by multiple tasks repeatedly, // but ktime.Timer.Pause is idempotent so this is harmless. if t.fdTable != nil { - t.fdTable.forEach(func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) { + t.fdTable.forEach(ctx, func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) { if VFS2Enabled { if tfd, ok := fd.Impl().(*timerfd.TimerFileDescription); ok { tfd.PauseTimer() @@ -1112,7 +1113,7 @@ func (k *Kernel) pauseTimeLocked() { // // Preconditions: Any task goroutines running in k must be stopped. k.extMu // must be locked. -func (k *Kernel) resumeTimeLocked() { +func (k *Kernel) resumeTimeLocked(ctx context.Context) { if k.cpuClockTicker != nil { k.cpuClockTicker.Resume() } @@ -1126,7 +1127,7 @@ func (k *Kernel) resumeTimeLocked() { } } if t.fdTable != nil { - t.fdTable.forEach(func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) { + t.fdTable.forEach(ctx, func(_ int32, file *fs.File, fd *vfs.FileDescription, _ FDFlags) { if VFS2Enabled { if tfd, ok := fd.Impl().(*timerfd.TimerFileDescription); ok { tfd.ResumeTimer() @@ -1262,6 +1263,13 @@ func (k *Kernel) Pause() { k.tasks.aioGoroutines.Wait() } +// ReceiveTaskStates receives full states for all tasks. +func (k *Kernel) ReceiveTaskStates() { + k.extMu.Lock() + k.tasks.PullFullState() + k.extMu.Unlock() +} + // Unpause ends the effect of a previous call to Pause. If Unpause is called // without a matching preceding call to Pause, Unpause may panic. func (k *Kernel) Unpause() { @@ -1511,7 +1519,7 @@ type SocketEntry struct { } // WeakRefGone implements refs.WeakRefUser.WeakRefGone. -func (s *SocketEntry) WeakRefGone() { +func (s *SocketEntry) WeakRefGone(context.Context) { s.k.extMu.Lock() s.k.sockets.Remove(s) s.k.extMu.Unlock() @@ -1600,7 +1608,7 @@ func (ctx supervisorContext) Value(key interface{}) interface{} { return vfs.VirtualDentry{} } mntns := ctx.k.GlobalInit().Leader().MountNamespaceVFS2() - defer mntns.DecRef() + defer mntns.DecRef(ctx) // Root() takes a reference on the root dirent for us. return mntns.Root() case vfs.CtxMountNamespace: diff --git a/pkg/sentry/kernel/pipe/node.go b/pkg/sentry/kernel/pipe/node.go index 4b688c627..6497dc4ba 100644 --- a/pkg/sentry/kernel/pipe/node.go +++ b/pkg/sentry/kernel/pipe/node.go @@ -93,7 +93,7 @@ func (i *inodeOperations) GetFile(ctx context.Context, d *fs.Dirent, flags fs.Fi if i.p.isNamed && !flags.NonBlocking && !i.p.HasWriters() { if !waitFor(&i.mu, &i.wWakeup, ctx) { - r.DecRef() + r.DecRef(ctx) return nil, syserror.ErrInterrupted } } @@ -111,12 +111,12 @@ func (i *inodeOperations) GetFile(ctx context.Context, d *fs.Dirent, flags fs.Fi // On a nonblocking, write-only open, the open fails with ENXIO if the // read side isn't open yet. if flags.NonBlocking { - w.DecRef() + w.DecRef(ctx) return nil, syserror.ENXIO } if !waitFor(&i.mu, &i.rWakeup, ctx) { - w.DecRef() + w.DecRef(ctx) return nil, syserror.ErrInterrupted } } diff --git a/pkg/sentry/kernel/pipe/node_test.go b/pkg/sentry/kernel/pipe/node_test.go index ab75a87ff..ce0db5583 100644 --- a/pkg/sentry/kernel/pipe/node_test.go +++ b/pkg/sentry/kernel/pipe/node_test.go @@ -167,7 +167,7 @@ func TestClosedReaderBlocksWriteOpen(t *testing.T) { f := NewInodeOperations(ctx, perms, newNamedPipe(t)) rFile, _ := testOpenOrDie(ctx, t, f, fs.FileFlags{Read: true, NonBlocking: true}, nil) - rFile.DecRef() + rFile.DecRef(ctx) wDone := make(chan struct{}) // This open for write should block because the reader is now gone. diff --git a/pkg/sentry/kernel/pipe/pipe.go b/pkg/sentry/kernel/pipe/pipe.go index 79645d7d2..297e8f28f 100644 --- a/pkg/sentry/kernel/pipe/pipe.go +++ b/pkg/sentry/kernel/pipe/pipe.go @@ -152,7 +152,7 @@ func NewConnectedPipe(ctx context.Context, sizeBytes, atomicIOBytes int64) (*fs. d := fs.NewDirent(ctx, fs.NewInode(ctx, iops, ms, sattr), fmt.Sprintf("pipe:[%d]", ino)) // The p.Open calls below will each take a reference on the Dirent. We // must drop the one we already have. - defer d.DecRef() + defer d.DecRef(ctx) return p.Open(ctx, d, fs.FileFlags{Read: true}), p.Open(ctx, d, fs.FileFlags{Write: true}) } diff --git a/pkg/sentry/kernel/pipe/pipe_test.go b/pkg/sentry/kernel/pipe/pipe_test.go index bda739dbe..fe97e9800 100644 --- a/pkg/sentry/kernel/pipe/pipe_test.go +++ b/pkg/sentry/kernel/pipe/pipe_test.go @@ -27,8 +27,8 @@ import ( func TestPipeRW(t *testing.T) { ctx := contexttest.Context(t) r, w := NewConnectedPipe(ctx, 65536, 4096) - defer r.DecRef() - defer w.DecRef() + defer r.DecRef(ctx) + defer w.DecRef(ctx) msg := []byte("here's some bytes") wantN := int64(len(msg)) @@ -47,8 +47,8 @@ func TestPipeRW(t *testing.T) { func TestPipeReadBlock(t *testing.T) { ctx := contexttest.Context(t) r, w := NewConnectedPipe(ctx, 65536, 4096) - defer r.DecRef() - defer w.DecRef() + defer r.DecRef(ctx) + defer w.DecRef(ctx) n, err := r.Readv(ctx, usermem.BytesIOSequence(make([]byte, 1))) if n != 0 || err != syserror.ErrWouldBlock { @@ -62,8 +62,8 @@ func TestPipeWriteBlock(t *testing.T) { ctx := contexttest.Context(t) r, w := NewConnectedPipe(ctx, capacity, atomicIOBytes) - defer r.DecRef() - defer w.DecRef() + defer r.DecRef(ctx) + defer w.DecRef(ctx) msg := make([]byte, capacity+1) n, err := w.Writev(ctx, usermem.BytesIOSequence(msg)) @@ -77,8 +77,8 @@ func TestPipeWriteUntilEnd(t *testing.T) { ctx := contexttest.Context(t) r, w := NewConnectedPipe(ctx, atomicIOBytes, atomicIOBytes) - defer r.DecRef() - defer w.DecRef() + defer r.DecRef(ctx) + defer w.DecRef(ctx) msg := []byte("here's some bytes") diff --git a/pkg/sentry/kernel/pipe/pipe_util.go b/pkg/sentry/kernel/pipe/pipe_util.go index aacf28da2..6d58b682f 100644 --- a/pkg/sentry/kernel/pipe/pipe_util.go +++ b/pkg/sentry/kernel/pipe/pipe_util.go @@ -33,7 +33,7 @@ import ( // the old fs architecture. // Release cleans up the pipe's state. -func (p *Pipe) Release() { +func (p *Pipe) Release(context.Context) { p.rClose() p.wClose() diff --git a/pkg/sentry/kernel/pipe/reader.go b/pkg/sentry/kernel/pipe/reader.go index 7724b4452..ac18785c0 100644 --- a/pkg/sentry/kernel/pipe/reader.go +++ b/pkg/sentry/kernel/pipe/reader.go @@ -15,6 +15,7 @@ package pipe import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/waiter" ) @@ -29,7 +30,7 @@ type Reader struct { // Release implements fs.FileOperations.Release. // // This overrides ReaderWriter.Release. -func (r *Reader) Release() { +func (r *Reader) Release(context.Context) { r.Pipe.rClose() // Wake up writers. diff --git a/pkg/sentry/kernel/pipe/vfs.go b/pkg/sentry/kernel/pipe/vfs.go index 45d4c5fc1..28f998e45 100644 --- a/pkg/sentry/kernel/pipe/vfs.go +++ b/pkg/sentry/kernel/pipe/vfs.go @@ -101,7 +101,7 @@ func (vp *VFSPipe) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, s // If this pipe is being opened as blocking and there's no // writer, we have to wait for a writer to open the other end. if vp.pipe.isNamed && statusFlags&linux.O_NONBLOCK == 0 && !vp.pipe.HasWriters() && !waitFor(&vp.mu, &vp.wWakeup, ctx) { - fd.DecRef() + fd.DecRef(ctx) return nil, syserror.EINTR } @@ -112,12 +112,12 @@ func (vp *VFSPipe) Open(ctx context.Context, mnt *vfs.Mount, vfsd *vfs.Dentry, s // Non-blocking, write-only opens fail with ENXIO when the read // side isn't open yet. if statusFlags&linux.O_NONBLOCK != 0 { - fd.DecRef() + fd.DecRef(ctx) return nil, syserror.ENXIO } // Wait for a reader to open the other end. if !waitFor(&vp.mu, &vp.rWakeup, ctx) { - fd.DecRef() + fd.DecRef(ctx) return nil, syserror.EINTR } } @@ -169,7 +169,7 @@ type VFSPipeFD struct { } // Release implements vfs.FileDescriptionImpl.Release. -func (fd *VFSPipeFD) Release() { +func (fd *VFSPipeFD) Release(context.Context) { var event waiter.EventMask if fd.vfsfd.IsReadable() { fd.pipe.rClose() diff --git a/pkg/sentry/kernel/pipe/writer.go b/pkg/sentry/kernel/pipe/writer.go index 5bc6aa931..ef4b70ca3 100644 --- a/pkg/sentry/kernel/pipe/writer.go +++ b/pkg/sentry/kernel/pipe/writer.go @@ -15,6 +15,7 @@ package pipe import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/waiter" ) @@ -29,7 +30,7 @@ type Writer struct { // Release implements fs.FileOperations.Release. // // This overrides ReaderWriter.Release. -func (w *Writer) Release() { +func (w *Writer) Release(context.Context) { w.Pipe.wClose() // Wake up readers. diff --git a/pkg/sentry/kernel/ptrace.go b/pkg/sentry/kernel/ptrace.go index e23e796ef..6c03d9041 100644 --- a/pkg/sentry/kernel/ptrace.go +++ b/pkg/sentry/kernel/ptrace.go @@ -1018,6 +1018,9 @@ func (t *Task) Ptrace(req int64, pid ThreadID, addr, data usermem.Addr) error { if err != nil { return err } + + t.p.PullFullState(t.MemoryManager().AddressSpace(), t.Arch()) + ar := ars.Head() n, err := target.Arch().PtraceGetRegSet(uintptr(addr), &usermem.IOReadWriter{ Ctx: t, @@ -1044,10 +1047,14 @@ func (t *Task) Ptrace(req int64, pid ThreadID, addr, data usermem.Addr) error { if err != nil { return err } + + mm := t.MemoryManager() + t.p.PullFullState(mm.AddressSpace(), t.Arch()) + ar := ars.Head() n, err := target.Arch().PtraceSetRegSet(uintptr(addr), &usermem.IOReadWriter{ Ctx: t, - IO: t.MemoryManager(), + IO: mm, Addr: ar.Start, Opts: usermem.IOOpts{ AddressSpaceActive: true, @@ -1056,6 +1063,7 @@ func (t *Task) Ptrace(req int64, pid ThreadID, addr, data usermem.Addr) error { if err != nil { return err } + t.p.FloatingPointStateChanged() ar.End -= usermem.Addr(n) return t.CopyOutIovecs(data, usermem.AddrRangeSeqOf(ar)) diff --git a/pkg/sentry/kernel/sessions.go b/pkg/sentry/kernel/sessions.go index 0e19286de..5c4c622c2 100644 --- a/pkg/sentry/kernel/sessions.go +++ b/pkg/sentry/kernel/sessions.go @@ -16,6 +16,7 @@ package kernel import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/syserror" @@ -70,7 +71,7 @@ func (s *Session) incRef() { // // Precondition: callers must hold TaskSet.mu for writing. func (s *Session) decRef() { - s.refs.DecRefWithDestructor(func() { + s.refs.DecRefWithDestructor(nil, func(context.Context) { // Remove translations from the leader. for ns := s.leader.pidns; ns != nil; ns = ns.parent { id := ns.sids[s] @@ -162,7 +163,7 @@ func (pg *ProcessGroup) decRefWithParent(parentPG *ProcessGroup) { } alive := true - pg.refs.DecRefWithDestructor(func() { + pg.refs.DecRefWithDestructor(nil, func(context.Context) { alive = false // don't bother with handleOrphan. // Remove translations from the originator. diff --git a/pkg/sentry/kernel/shm/shm.go b/pkg/sentry/kernel/shm/shm.go index 55b4c2cdb..13ec7afe0 100644 --- a/pkg/sentry/kernel/shm/shm.go +++ b/pkg/sentry/kernel/shm/shm.go @@ -431,8 +431,8 @@ func (s *Shm) InodeID() uint64 { // DecRef overrides refs.RefCount.DecRef with a destructor. // // Precondition: Caller must not hold s.mu. -func (s *Shm) DecRef() { - s.DecRefWithDestructor(s.destroy) +func (s *Shm) DecRef(ctx context.Context) { + s.DecRefWithDestructor(ctx, s.destroy) } // Msync implements memmap.MappingIdentity.Msync. Msync is a no-op for shm @@ -642,7 +642,7 @@ func (s *Shm) Set(ctx context.Context, ds *linux.ShmidDS) error { return nil } -func (s *Shm) destroy() { +func (s *Shm) destroy(context.Context) { s.mfp.MemoryFile().DecRef(s.fr) s.registry.remove(s) } @@ -651,7 +651,7 @@ func (s *Shm) destroy() { // destroyed once it has no references. MarkDestroyed may be called multiple // times, and is safe to call after a segment has already been destroyed. See // shmctl(IPC_RMID). -func (s *Shm) MarkDestroyed() { +func (s *Shm) MarkDestroyed(ctx context.Context) { s.registry.dissociateKey(s) s.mu.Lock() @@ -663,7 +663,7 @@ func (s *Shm) MarkDestroyed() { // // N.B. This cannot be the final DecRef, as the caller also // holds a reference. - s.DecRef() + s.DecRef(ctx) return } } diff --git a/pkg/sentry/kernel/signalfd/signalfd.go b/pkg/sentry/kernel/signalfd/signalfd.go index 8243bb93e..b07e1c1bd 100644 --- a/pkg/sentry/kernel/signalfd/signalfd.go +++ b/pkg/sentry/kernel/signalfd/signalfd.go @@ -76,7 +76,7 @@ func New(ctx context.Context, mask linux.SignalSet) (*fs.File, error) { } // Release implements fs.FileOperations.Release. -func (s *SignalOperations) Release() {} +func (s *SignalOperations) Release(context.Context) {} // Mask returns the signal mask. func (s *SignalOperations) Mask() linux.SignalSet { diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go index c4db05bd8..5aee699e7 100644 --- a/pkg/sentry/kernel/task.go +++ b/pkg/sentry/kernel/task.go @@ -730,17 +730,17 @@ func (t *Task) SyscallRestartBlock() SyscallRestartBlock { func (t *Task) IsChrooted() bool { if VFS2Enabled { realRoot := t.mountNamespaceVFS2.Root() - defer realRoot.DecRef() + defer realRoot.DecRef(t) root := t.fsContext.RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) return root != realRoot } realRoot := t.tg.mounts.Root() - defer realRoot.DecRef() + defer realRoot.DecRef(t) root := t.fsContext.RootDirectory() if root != nil { - defer root.DecRef() + defer root.DecRef(t) } return root != realRoot } diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go index e1ecca99e..9d7a9128f 100644 --- a/pkg/sentry/kernel/task_clone.go +++ b/pkg/sentry/kernel/task_clone.go @@ -161,6 +161,10 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { return 0, nil, syserror.EINVAL } + // Pull task registers and FPU state, a cloned task will inherit the + // state of the current task. + t.p.PullFullState(t.MemoryManager().AddressSpace(), t.Arch()) + // "If CLONE_NEWUSER is specified along with other CLONE_NEW* flags in a // single clone(2) or unshare(2) call, the user namespace is guaranteed to // be created first, giving the child (clone(2)) or caller (unshare(2)) @@ -237,7 +241,7 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { var fdTable *FDTable if opts.NewFiles { - fdTable = t.fdTable.Fork() + fdTable = t.fdTable.Fork(t) } else { fdTable = t.fdTable fdTable.IncRef() @@ -294,7 +298,7 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { nt, err := t.tg.pidns.owner.NewTask(cfg) if err != nil { if opts.NewThreadGroup { - tg.release() + tg.release(t) } return 0, nil, err } @@ -510,7 +514,7 @@ func (t *Task) Unshare(opts *SharingOptions) error { var oldFDTable *FDTable if opts.NewFiles { oldFDTable = t.fdTable - t.fdTable = oldFDTable.Fork() + t.fdTable = oldFDTable.Fork(t) } var oldFSContext *FSContext if opts.NewFSContext { @@ -519,10 +523,10 @@ func (t *Task) Unshare(opts *SharingOptions) error { } t.mu.Unlock() if oldFDTable != nil { - oldFDTable.DecRef() + oldFDTable.DecRef(t) } if oldFSContext != nil { - oldFSContext.DecRef() + oldFSContext.DecRef(t) } return nil } diff --git a/pkg/sentry/kernel/task_exec.go b/pkg/sentry/kernel/task_exec.go index 7803b98d0..47c28b8ff 100644 --- a/pkg/sentry/kernel/task_exec.go +++ b/pkg/sentry/kernel/task_exec.go @@ -199,11 +199,11 @@ func (r *runSyscallAfterExecStop) execute(t *Task) taskRunState { t.tg.pidns.owner.mu.Unlock() oldFDTable := t.fdTable - t.fdTable = t.fdTable.Fork() - oldFDTable.DecRef() + t.fdTable = t.fdTable.Fork(t) + oldFDTable.DecRef(t) // Remove FDs with the CloseOnExec flag set. - t.fdTable.RemoveIf(func(_ *fs.File, _ *vfs.FileDescription, flags FDFlags) bool { + t.fdTable.RemoveIf(t, func(_ *fs.File, _ *vfs.FileDescription, flags FDFlags) bool { return flags.CloseOnExec }) diff --git a/pkg/sentry/kernel/task_exit.go b/pkg/sentry/kernel/task_exit.go index 231ac548a..c165d6cb1 100644 --- a/pkg/sentry/kernel/task_exit.go +++ b/pkg/sentry/kernel/task_exit.go @@ -269,12 +269,12 @@ func (*runExitMain) execute(t *Task) taskRunState { // Releasing the MM unblocks a blocked CLONE_VFORK parent. t.unstopVforkParent() - t.fsContext.DecRef() - t.fdTable.DecRef() + t.fsContext.DecRef(t) + t.fdTable.DecRef(t) t.mu.Lock() if t.mountNamespaceVFS2 != nil { - t.mountNamespaceVFS2.DecRef() + t.mountNamespaceVFS2.DecRef(t) t.mountNamespaceVFS2 = nil } t.mu.Unlock() @@ -282,7 +282,7 @@ func (*runExitMain) execute(t *Task) taskRunState { // If this is the last task to exit from the thread group, release the // thread group's resources. if lastExiter { - t.tg.release() + t.tg.release(t) } // Detach tracees. diff --git a/pkg/sentry/kernel/task_log.go b/pkg/sentry/kernel/task_log.go index eeccaa197..d23cea802 100644 --- a/pkg/sentry/kernel/task_log.go +++ b/pkg/sentry/kernel/task_log.go @@ -27,6 +27,9 @@ const ( // maxStackDebugBytes is the maximum number of user stack bytes that may be // printed by debugDumpStack. maxStackDebugBytes = 1024 + // maxCodeDebugBytes is the maximum number of user code bytes that may be + // printed by debugDumpCode. + maxCodeDebugBytes = 128 ) // Infof logs an formatted info message by calling log.Infof. @@ -61,6 +64,7 @@ func (t *Task) IsLogging(level log.Level) bool { func (t *Task) DebugDumpState() { t.debugDumpRegisters() t.debugDumpStack() + t.debugDumpCode() if mm := t.MemoryManager(); mm != nil { t.Debugf("Mappings:\n%s", mm) } @@ -128,6 +132,45 @@ func (t *Task) debugDumpStack() { } } +// debugDumpCode logs user code contents at log level debug. +// +// Preconditions: The caller must be running on the task goroutine. +func (t *Task) debugDumpCode() { + if !t.IsLogging(log.Debug) { + return + } + m := t.MemoryManager() + if m == nil { + t.Debugf("Memory manager for task is gone, skipping application code dump.") + return + } + t.Debugf("Code:") + // Print code on both sides of the instruction register. + start := usermem.Addr(t.Arch().IP()) - maxCodeDebugBytes/2 + // Round addr down to a 16-byte boundary. + start &= ^usermem.Addr(15) + // Print 16 bytes per line, one byte at a time. + for offset := uint64(0); offset < maxCodeDebugBytes; offset += 16 { + addr, ok := start.AddLength(offset) + if !ok { + break + } + var data [16]byte + n, err := m.CopyIn(t, addr, data[:], usermem.IOOpts{ + IgnorePermissions: true, + }) + // Print as much of the line as we can, even if an error was + // encountered. + if n > 0 { + t.Debugf("%x: % x", addr, data[:n]) + } + if err != nil { + t.Debugf("Error reading stack at address %x: %v", addr+usermem.Addr(n), err) + break + } + } +} + // trace definitions. // // Note that all region names are prefixed by ':' in order to ensure that they @@ -203,6 +246,6 @@ func (t *Task) traceExecEvent(tc *TaskContext) { trace.Logf(t.traceContext, traceCategory, "exec: << unknown >>") return } - defer file.DecRef() + defer file.DecRef(t) trace.Logf(t.traceContext, traceCategory, "exec: %s", file.PathnameWithDeleted(t)) } diff --git a/pkg/sentry/kernel/task_run.go b/pkg/sentry/kernel/task_run.go index 7d4f44caf..abaf29216 100644 --- a/pkg/sentry/kernel/task_run.go +++ b/pkg/sentry/kernel/task_run.go @@ -260,7 +260,7 @@ func (app *runApp) execute(t *Task) taskRunState { region := trace.StartRegion(t.traceContext, runRegion) t.accountTaskGoroutineEnter(TaskGoroutineRunningApp) - info, at, err := t.p.Switch(t.MemoryManager().AddressSpace(), t.Arch(), t.rseqCPU) + info, at, err := t.p.Switch(t, t.MemoryManager(), t.Arch(), t.rseqCPU) t.accountTaskGoroutineLeave(TaskGoroutineRunningApp) region.End() diff --git a/pkg/sentry/kernel/task_signals.go b/pkg/sentry/kernel/task_signals.go index 79766cafe..2180fd27d 100644 --- a/pkg/sentry/kernel/task_signals.go +++ b/pkg/sentry/kernel/task_signals.go @@ -255,10 +255,11 @@ func (t *Task) deliverSignalToHandler(info *arch.SignalInfo, act arch.SignalAct) } } + mm := t.MemoryManager() // Set up the signal handler. If we have a saved signal mask, the signal // handler should run with the current mask, but sigreturn should restore // the saved one. - st := &arch.Stack{t.Arch(), t.MemoryManager(), sp} + st := &arch.Stack{t.Arch(), mm, sp} mask := t.signalMask if t.haveSavedSignalMask { mask = t.savedSignalMask @@ -273,12 +274,13 @@ func (t *Task) deliverSignalToHandler(info *arch.SignalInfo, act arch.SignalAct) // Please see the linux code as reference: // linux/arch/arm64/kernel/signal.c:setup_return() if act.Flags&linux.SA_RESTORER == 0 { - act.Restorer = t.MemoryManager().VDSOSigReturn() + act.Restorer = mm.VDSOSigReturn() } if err := t.Arch().SignalSetup(st, &act, info, &alt, mask); err != nil { return err } + t.p.FloatingPointStateChanged() t.haveSavedSignalMask = false // Add our signal mask. @@ -310,6 +312,7 @@ func (t *Task) SignalReturn(rt bool) (*SyscallControl, error) { // Restore our signal mask. SIGKILL and SIGSTOP should not be blocked. t.SetSignalMask(sigset &^ UnblockableSignals) + t.p.FloatingPointStateChanged() return ctrlResume, nil } @@ -636,6 +639,7 @@ func (t *Task) SetSavedSignalMask(mask linux.SignalSet) { // SignalStack returns the task-private signal stack. func (t *Task) SignalStack() arch.SignalStack { + t.p.PullFullState(t.MemoryManager().AddressSpace(), t.Arch()) alt := t.signalStack if t.onSignalStack(alt) { alt.Flags |= arch.SignalStackFlagOnStack @@ -1050,6 +1054,8 @@ func (*runInterrupt) execute(t *Task) taskRunState { // Are there signals pending? if info := t.dequeueSignalLocked(t.signalMask); info != nil { + t.p.PullFullState(t.MemoryManager().AddressSpace(), t.Arch()) + if linux.SignalSetOf(linux.Signal(info.Signo))&StopSignals != 0 { // Indicate that we've dequeued a stop signal before unlocking the // signal mutex; initiateGroupStop will check for races with diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go index 8485fb4b6..64c1e120a 100644 --- a/pkg/sentry/kernel/task_start.go +++ b/pkg/sentry/kernel/task_start.go @@ -102,10 +102,10 @@ func (ts *TaskSet) NewTask(cfg *TaskConfig) (*Task, error) { t, err := ts.newTask(cfg) if err != nil { cfg.TaskContext.release() - cfg.FSContext.DecRef() - cfg.FDTable.DecRef() + cfg.FSContext.DecRef(t) + cfg.FDTable.DecRef(t) if cfg.MountNamespaceVFS2 != nil { - cfg.MountNamespaceVFS2.DecRef() + cfg.MountNamespaceVFS2.DecRef(t) } return nil, err } diff --git a/pkg/sentry/kernel/task_stop.go b/pkg/sentry/kernel/task_stop.go index 10c6e455c..296735d32 100644 --- a/pkg/sentry/kernel/task_stop.go +++ b/pkg/sentry/kernel/task_stop.go @@ -205,6 +205,22 @@ func (ts *TaskSet) BeginExternalStop() { } } +// PullFullState receives full states for all tasks. +func (ts *TaskSet) PullFullState() { + ts.mu.Lock() + defer ts.mu.Unlock() + if ts.Root == nil { + return + } + for t := range ts.Root.tids { + t.Activate() + if mm := t.MemoryManager(); mm != nil { + t.p.PullFullState(t.MemoryManager().AddressSpace(), t.Arch()) + } + t.Deactivate() + } +} + // EndExternalStop indicates the end of an external stop started by a previous // call to TaskSet.BeginExternalStop. EndExternalStop does not wait for task // goroutines to resume. diff --git a/pkg/sentry/kernel/thread_group.go b/pkg/sentry/kernel/thread_group.go index 4dfd2c990..0b34c0099 100644 --- a/pkg/sentry/kernel/thread_group.go +++ b/pkg/sentry/kernel/thread_group.go @@ -308,7 +308,7 @@ func (tg *ThreadGroup) Limits() *limits.LimitSet { } // release releases the thread group's resources. -func (tg *ThreadGroup) release() { +func (tg *ThreadGroup) release(t *Task) { // Timers must be destroyed without holding the TaskSet or signal mutexes // since timers send signals with Timer.mu locked. tg.itimerRealTimer.Destroy() @@ -325,7 +325,7 @@ func (tg *ThreadGroup) release() { it.DestroyTimer() } if tg.mounts != nil { - tg.mounts.DecRef() + tg.mounts.DecRef(t) } } diff --git a/pkg/sentry/loader/elf.go b/pkg/sentry/loader/elf.go index ddeaff3db..20dd1cc21 100644 --- a/pkg/sentry/loader/elf.go +++ b/pkg/sentry/loader/elf.go @@ -281,7 +281,7 @@ func mapSegment(ctx context.Context, m *mm.MemoryManager, f fsbridge.File, phdr } defer func() { if mopts.MappingIdentity != nil { - mopts.MappingIdentity.DecRef() + mopts.MappingIdentity.DecRef(ctx) } }() if err := f.ConfigureMMap(ctx, &mopts); err != nil { @@ -663,7 +663,7 @@ func loadELF(ctx context.Context, args LoadArgs) (loadedELF, arch.Context, error ctx.Infof("Error opening interpreter %s: %v", bin.interpreter, err) return loadedELF{}, nil, err } - defer intFile.DecRef() + defer intFile.DecRef(ctx) interp, err = loadInterpreterELF(ctx, args.MemoryManager, intFile, bin) if err != nil { diff --git a/pkg/sentry/loader/loader.go b/pkg/sentry/loader/loader.go index 986c7fb4d..8d6802ea3 100644 --- a/pkg/sentry/loader/loader.go +++ b/pkg/sentry/loader/loader.go @@ -154,7 +154,7 @@ func loadExecutable(ctx context.Context, args LoadArgs) (loadedELF, arch.Context return loadedELF{}, nil, nil, nil, err } // Ensure file is release in case the code loops or errors out. - defer args.File.DecRef() + defer args.File.DecRef(ctx) } else { if err := checkIsRegularFile(ctx, args.File, args.Filename); err != nil { return loadedELF{}, nil, nil, nil, err @@ -223,7 +223,7 @@ func Load(ctx context.Context, args LoadArgs, extraAuxv []arch.AuxEntry, vdso *V if err != nil { return 0, nil, "", syserr.NewDynamic(fmt.Sprintf("failed to load %s: %v", args.Filename, err), syserr.FromError(err).ToLinux()) } - defer file.DecRef() + defer file.DecRef(ctx) // Load the VDSO. vdsoAddr, err := loadVDSO(ctx, args.MemoryManager, vdso, loaded) @@ -292,7 +292,7 @@ func Load(ctx context.Context, args LoadArgs, extraAuxv []arch.AuxEntry, vdso *V m.SetEnvvStart(sl.EnvvStart) m.SetEnvvEnd(sl.EnvvEnd) m.SetAuxv(auxv) - m.SetExecutable(file) + m.SetExecutable(ctx, file) symbolValue, err := getSymbolValueFromVDSO("rt_sigreturn") if err != nil { diff --git a/pkg/sentry/memmap/memmap.go b/pkg/sentry/memmap/memmap.go index c188f6c29..65d83096f 100644 --- a/pkg/sentry/memmap/memmap.go +++ b/pkg/sentry/memmap/memmap.go @@ -238,7 +238,7 @@ type MappingIdentity interface { IncRef() // DecRef decrements the MappingIdentity's reference count. - DecRef() + DecRef(ctx context.Context) // MappedName returns the application-visible name shown in // /proc/[pid]/maps. @@ -360,6 +360,13 @@ type MMapOpts struct { // // TODO(jamieliu): Replace entirely with MappingIdentity? Hint string + + // Force means to skip validation checks of Addr and Length. It can be + // used to create special mappings below mm.layout.MinAddr and + // mm.layout.MaxAddr. It has to be used with caution. + // + // If Force is true, Unmap and Fixed must be true. + Force bool } // File represents a host file that may be mapped into an platform.AddressSpace. diff --git a/pkg/sentry/mm/aio_context.go b/pkg/sentry/mm/aio_context.go index 1999ec706..16fea53c4 100644 --- a/pkg/sentry/mm/aio_context.go +++ b/pkg/sentry/mm/aio_context.go @@ -258,8 +258,8 @@ func newAIOMappable(mfp pgalloc.MemoryFileProvider) (*aioMappable, error) { } // DecRef implements refs.RefCounter.DecRef. -func (m *aioMappable) DecRef() { - m.AtomicRefCount.DecRefWithDestructor(func() { +func (m *aioMappable) DecRef(ctx context.Context) { + m.AtomicRefCount.DecRefWithDestructor(ctx, func(context.Context) { m.mfp.MemoryFile().DecRef(m.fr) }) } @@ -367,7 +367,7 @@ func (mm *MemoryManager) NewAIOContext(ctx context.Context, events uint32) (uint if err != nil { return 0, err } - defer m.DecRef() + defer m.DecRef(ctx) addr, err := mm.MMap(ctx, memmap.MMapOpts{ Length: aioRingBufferSize, MappingIdentity: m, diff --git a/pkg/sentry/mm/lifecycle.go b/pkg/sentry/mm/lifecycle.go index aac56679b..09dbc06a4 100644 --- a/pkg/sentry/mm/lifecycle.go +++ b/pkg/sentry/mm/lifecycle.go @@ -57,6 +57,8 @@ func (mm *MemoryManager) SetMmapLayout(ac arch.Context, r *limits.LimitSet) (arc // Fork creates a copy of mm with 1 user, as for Linux syscalls fork() or // clone() (without CLONE_VM). func (mm *MemoryManager) Fork(ctx context.Context) (*MemoryManager, error) { + mm.AddressSpace().PreFork() + defer mm.AddressSpace().PostFork() mm.metadataMu.Lock() defer mm.metadataMu.Unlock() mm.mappingMu.RLock() @@ -258,7 +260,7 @@ func (mm *MemoryManager) DecUsers(ctx context.Context) { mm.executable = nil mm.metadataMu.Unlock() if exe != nil { - exe.DecRef() + exe.DecRef(ctx) } mm.activeMu.Lock() diff --git a/pkg/sentry/mm/metadata.go b/pkg/sentry/mm/metadata.go index 28e5057f7..0cfd60f6c 100644 --- a/pkg/sentry/mm/metadata.go +++ b/pkg/sentry/mm/metadata.go @@ -15,6 +15,7 @@ package mm import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/usermem" @@ -147,7 +148,7 @@ func (mm *MemoryManager) Executable() fsbridge.File { // SetExecutable sets the executable. // // This takes a reference on d. -func (mm *MemoryManager) SetExecutable(file fsbridge.File) { +func (mm *MemoryManager) SetExecutable(ctx context.Context, file fsbridge.File) { mm.metadataMu.Lock() // Grab a new reference. @@ -164,7 +165,7 @@ func (mm *MemoryManager) SetExecutable(file fsbridge.File) { // Do this without holding the lock, since it may wind up doing some // I/O to sync the dirent, etc. if orig != nil { - orig.DecRef() + orig.DecRef(ctx) } } diff --git a/pkg/sentry/mm/special_mappable.go b/pkg/sentry/mm/special_mappable.go index 0e142fb11..4cdb52eb6 100644 --- a/pkg/sentry/mm/special_mappable.go +++ b/pkg/sentry/mm/special_mappable.go @@ -50,8 +50,8 @@ func NewSpecialMappable(name string, mfp pgalloc.MemoryFileProvider, fr memmap.F } // DecRef implements refs.RefCounter.DecRef. -func (m *SpecialMappable) DecRef() { - m.AtomicRefCount.DecRefWithDestructor(func() { +func (m *SpecialMappable) DecRef(ctx context.Context) { + m.AtomicRefCount.DecRefWithDestructor(ctx, func(context.Context) { m.mfp.MemoryFile().DecRef(m.fr) }) } diff --git a/pkg/sentry/mm/syscalls.go b/pkg/sentry/mm/syscalls.go index 3f496aa9f..e74d4e1c1 100644 --- a/pkg/sentry/mm/syscalls.go +++ b/pkg/sentry/mm/syscalls.go @@ -101,7 +101,7 @@ func (mm *MemoryManager) MMap(ctx context.Context, opts memmap.MMapOpts) (userme if err != nil { return 0, err } - defer m.DecRef() + defer m.DecRef(ctx) opts.MappingIdentity = m opts.Mappable = m } @@ -1191,7 +1191,7 @@ func (mm *MemoryManager) MSync(ctx context.Context, addr usermem.Addr, length ui mr := vseg.mappableRangeOf(vseg.Range().Intersect(ar)) mm.mappingMu.RUnlock() err := id.Msync(ctx, mr) - id.DecRef() + id.DecRef(ctx) if err != nil { return err } diff --git a/pkg/sentry/mm/vma.go b/pkg/sentry/mm/vma.go index 16d8207e9..c4e1989ed 100644 --- a/pkg/sentry/mm/vma.go +++ b/pkg/sentry/mm/vma.go @@ -42,7 +42,12 @@ func (mm *MemoryManager) createVMALocked(ctx context.Context, opts memmap.MMapOp Map32Bit: opts.Map32Bit, }) if err != nil { - return vmaIterator{}, usermem.AddrRange{}, err + // Can't force without opts.Unmap and opts.Fixed. + if opts.Force && opts.Unmap && opts.Fixed { + addr = opts.Addr + } else { + return vmaIterator{}, usermem.AddrRange{}, err + } } ar, _ := addr.ToRange(opts.Length) @@ -377,7 +382,7 @@ func (mm *MemoryManager) removeVMAsLocked(ctx context.Context, ar usermem.AddrRa vma.mappable.RemoveMapping(ctx, mm, vmaAR, vma.off, vma.canWriteMappableLocked()) } if vma.id != nil { - vma.id.DecRef() + vma.id.DecRef(ctx) } mm.usageAS -= uint64(vmaAR.Length()) if vma.isPrivateDataLocked() { @@ -446,7 +451,7 @@ func (vmaSetFunctions) Merge(ar1 usermem.AddrRange, vma1 vma, ar2 usermem.AddrRa } if vma2.id != nil { - vma2.id.DecRef() + vma2.id.DecRef(context.Background()) } return vma1, true } diff --git a/pkg/sentry/pgalloc/pgalloc.go b/pkg/sentry/pgalloc/pgalloc.go index 3243d7214..46d3be58c 100644 --- a/pkg/sentry/pgalloc/pgalloc.go +++ b/pkg/sentry/pgalloc/pgalloc.go @@ -409,7 +409,6 @@ func (f *MemoryFile) Allocate(length uint64, kind usage.MemoryKind) (memmap.File f.mappingsMu.Unlock() } - // Mark selected pages as in use. if f.opts.ManualZeroing { if err := f.forEachMappingSlice(fr, func(bs []byte) { for i := range bs { @@ -419,6 +418,7 @@ func (f *MemoryFile) Allocate(length uint64, kind usage.MemoryKind) (memmap.File return memmap.FileRange{}, err } } + // Mark selected pages as in use. if !f.usage.Add(fr, usageInfo{ kind: kind, refs: 1, diff --git a/pkg/sentry/platform/kvm/BUILD b/pkg/sentry/platform/kvm/BUILD index b5d27a72a..3970dd81d 100644 --- a/pkg/sentry/platform/kvm/BUILD +++ b/pkg/sentry/platform/kvm/BUILD @@ -41,6 +41,7 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/atomicbitops", + "//pkg/context", "//pkg/cpuid", "//pkg/log", "//pkg/procid", diff --git a/pkg/sentry/platform/kvm/address_space.go b/pkg/sentry/platform/kvm/address_space.go index 98a3e539d..af5c5e191 100644 --- a/pkg/sentry/platform/kvm/address_space.go +++ b/pkg/sentry/platform/kvm/address_space.go @@ -248,3 +248,9 @@ func (as *addressSpace) Release() { // Drop all cached machine references. as.machine.dropPageTables(as.pageTables) } + +// PreFork implements platform.AddressSpace.PreFork. +func (as *addressSpace) PreFork() {} + +// PostFork implements platform.AddressSpace.PostFork. +func (as *addressSpace) PostFork() {} diff --git a/pkg/sentry/platform/kvm/context.go b/pkg/sentry/platform/kvm/context.go index 6507121ea..eb92721fb 100644 --- a/pkg/sentry/platform/kvm/context.go +++ b/pkg/sentry/platform/kvm/context.go @@ -15,6 +15,7 @@ package kvm import ( + pkgcontext "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/interrupt" @@ -37,7 +38,8 @@ type context struct { } // Switch runs the provided context in the given address space. -func (c *context) Switch(as platform.AddressSpace, ac arch.Context, _ int32) (*arch.SignalInfo, usermem.AccessType, error) { +func (c *context) Switch(ctx pkgcontext.Context, mm platform.MemoryManager, ac arch.Context, _ int32) (*arch.SignalInfo, usermem.AccessType, error) { + as := mm.AddressSpace() localAS := as.(*addressSpace) // Grab a vCPU. @@ -88,3 +90,9 @@ func (c *context) Interrupt() { // Release implements platform.Context.Release(). func (c *context) Release() {} + +// FloatingPointStateChanged implements platform.Context.FloatingPointStateChanged. +func (c *context) FloatingPointStateChanged() {} + +// PullFullState implements platform.Context.PullFullState. +func (c *context) PullFullState(as platform.AddressSpace, ac arch.Context) {} diff --git a/pkg/sentry/platform/kvm/kvm_const_arm64.go b/pkg/sentry/platform/kvm/kvm_const_arm64.go index 6f0539c29..fdc599477 100644 --- a/pkg/sentry/platform/kvm/kvm_const_arm64.go +++ b/pkg/sentry/platform/kvm/kvm_const_arm64.go @@ -116,6 +116,17 @@ const ( // Arm64: Exception Syndrome Register EL1. const ( + _ESR_ELx_EC_SHIFT = 26 + _ESR_ELx_EC_MASK = 0x3F << _ESR_ELx_EC_SHIFT + + _ESR_ELx_EC_IMP_DEF = 0x1f + _ESR_ELx_EC_IABT_LOW = 0x20 + _ESR_ELx_EC_IABT_CUR = 0x21 + _ESR_ELx_EC_PC_ALIGN = 0x22 + + _ESR_ELx_CM = 1 << 8 + _ESR_ELx_WNR = 1 << 6 + _ESR_ELx_FSC = 0x3F _ESR_SEGV_MAPERR_L0 = 0x4 diff --git a/pkg/sentry/platform/kvm/machine_arm64.go b/pkg/sentry/platform/kvm/machine_arm64.go index f3bf973de..9db171af9 100644 --- a/pkg/sentry/platform/kvm/machine_arm64.go +++ b/pkg/sentry/platform/kvm/machine_arm64.go @@ -125,71 +125,59 @@ func nonCanonical(addr uint64, signal int32, info *arch.SignalInfo) (usermem.Acc return usermem.NoAccess, platform.ErrContextSignal } +// isInstructionAbort returns true if it is an instruction abort. +// +//go:nosplit +func isInstructionAbort(code uint64) bool { + value := (code & _ESR_ELx_EC_MASK) >> _ESR_ELx_EC_SHIFT + return value == _ESR_ELx_EC_IABT_LOW +} + +// isWriteFault returns whether it is a write fault. +// +//go:nosplit +func isWriteFault(code uint64) bool { + if isInstructionAbort(code) { + return false + } + + return (code & _ESR_ELx_WNR) != 0 +} + // fault generates an appropriate fault return. // //go:nosplit func (c *vCPU) fault(signal int32, info *arch.SignalInfo) (usermem.AccessType, error) { + bluepill(c) // Probably no-op, but may not be. faultAddr := c.GetFaultAddr() code, user := c.ErrorCode() + if !user { + // The last fault serviced by this CPU was not a user + // fault, so we can't reliably trust the faultAddr or + // the code provided here. We need to re-execute. + return usermem.NoAccess, platform.ErrContextInterrupt + } + // Reset the pointed SignalInfo. *info = arch.SignalInfo{Signo: signal} info.SetAddr(uint64(faultAddr)) - read := true - write := false - execute := true - ret := code & _ESR_ELx_FSC switch ret { case _ESR_SEGV_MAPERR_L0, _ESR_SEGV_MAPERR_L1, _ESR_SEGV_MAPERR_L2, _ESR_SEGV_MAPERR_L3: info.Code = 1 //SEGV_MAPERR - read = false - write = true - execute = false case _ESR_SEGV_ACCERR_L1, _ESR_SEGV_ACCERR_L2, _ESR_SEGV_ACCERR_L3, _ESR_SEGV_PEMERR_L1, _ESR_SEGV_PEMERR_L2, _ESR_SEGV_PEMERR_L3: info.Code = 2 // SEGV_ACCERR. - read = true - write = false - execute = false default: info.Code = 2 } - if !user { - read = true - write = false - execute = true - - } accessType := usermem.AccessType{ - Read: read, - Write: write, - Execute: execute, + Read: !isWriteFault(uint64(code)), + Write: isWriteFault(uint64(code)), + Execute: isInstructionAbort(uint64(code)), } return accessType, platform.ErrContextSignal } - -// retryInGuest runs the given function in guest mode. -// -// If the function does not complete in guest mode (due to execution of a -// system call due to a GC stall, for example), then it will be retried. The -// given function must be idempotent as a result of the retry mechanism. -func (m *machine) retryInGuest(fn func()) { - c := m.Get() - defer m.Put(c) - for { - c.ClearErrorCode() // See below. - bluepill(c) // Force guest mode. - fn() // Execute the given function. - _, user := c.ErrorCode() - if user { - // If user is set, then we haven't bailed back to host - // mode via a kernel exception or system call. We - // consider the full function to have executed in guest - // mode and we can return. - break - } - } -} diff --git a/pkg/sentry/platform/kvm/testutil/testutil_arm64.go b/pkg/sentry/platform/kvm/testutil/testutil_arm64.go index ca902c8c1..4dad877ba 100644 --- a/pkg/sentry/platform/kvm/testutil/testutil_arm64.go +++ b/pkg/sentry/platform/kvm/testutil/testutil_arm64.go @@ -56,5 +56,9 @@ func CheckTestRegs(regs *arch.Registers, full bool) (err error) { err = addRegisterMismatch(err, fmt.Sprintf("R%d", i), regs.Regs[i], need) } } + // Check tls. + if need := ^uint64(11); regs.TPIDR_EL0 != need { + err = addRegisterMismatch(err, "tpdir_el0", regs.TPIDR_EL0, need) + } return } diff --git a/pkg/sentry/platform/kvm/testutil/testutil_arm64.s b/pkg/sentry/platform/kvm/testutil/testutil_arm64.s index 07658144e..6caf7282d 100644 --- a/pkg/sentry/platform/kvm/testutil/testutil_arm64.s +++ b/pkg/sentry/platform/kvm/testutil/testutil_arm64.s @@ -52,6 +52,8 @@ start: TEXT ·FloatingPointWorks(SB),NOSPLIT,$0-8 NO_LOCAL_POINTERS + // gc will touch fpsimd, so we should test it. + // such as in <runtime.deductSweepCredit>. FMOVD $(9.9), F0 MOVD $SYS_GETPID, R8 // getpid SVC @@ -102,11 +104,15 @@ isNaN: TEXT ·TwiddleRegsSyscall(SB),NOSPLIT,$0 TWIDDLE_REGS() + MSR R10, TPIDR_EL0 + // Trapped in el0_svc. SVC RET // never reached TEXT ·TwiddleRegsFault(SB),NOSPLIT,$0 - TWIDDLE_REGS() - // Branch to Register branches unconditionally to an address in <Rn>. - JMP (R4) // <=> br x4, must fault - RET // never reached + TWIDDLE_REGS() + MSR R10, TPIDR_EL0 + // Trapped in el0_ia. + // Branch to Register branches unconditionally to an address in <Rn>. + JMP (R6) // <=> br x6, must fault + RET // never reached diff --git a/pkg/sentry/platform/platform.go b/pkg/sentry/platform/platform.go index 4b13eec30..3f99afdd1 100644 --- a/pkg/sentry/platform/platform.go +++ b/pkg/sentry/platform/platform.go @@ -22,6 +22,7 @@ import ( "os" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/seccomp" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/memmap" @@ -114,6 +115,17 @@ func (NoCPUPreemptionDetection) PreemptAllCPUs() error { panic("This platform does not support CPU preemption detection") } +// MemoryManager represents an abstraction above the platform address space +// which manages memory mappings and their contents. +type MemoryManager interface { + //usermem.IO provides access to the contents of a virtual memory space. + usermem.IO + // MMap establishes a memory mapping. + MMap(ctx context.Context, opts memmap.MMapOpts) (usermem.Addr, error) + // AddressSpace returns the AddressSpace bound to mm. + AddressSpace() AddressSpace +} + // Context represents the execution context for a single thread. type Context interface { // Switch resumes execution of the thread specified by the arch.Context @@ -143,7 +155,30 @@ type Context interface { // concurrent call to Switch(). // // - ErrContextCPUPreempted: See the definition of that error for details. - Switch(as AddressSpace, ac arch.Context, cpu int32) (*arch.SignalInfo, usermem.AccessType, error) + Switch(ctx context.Context, mm MemoryManager, ac arch.Context, cpu int32) (*arch.SignalInfo, usermem.AccessType, error) + + // PullFullState() pulls a full state of the application thread. + // + // A platform can support lazy loading/restoring of a thread state + // which includes registers and a floating point state. + // + // For example, when the Sentry handles a system call, it may have only + // syscall arguments without other registers and a floating point + // state. And in this case, if the Sentry will need to construct a + // signal frame to call a signal handler, it will need to call + // PullFullState() to load all registers and FPU state. + // + // Preconditions: The caller must be running on the task goroutine. + PullFullState(as AddressSpace, ac arch.Context) + + // FloatingPointStateChanged forces restoring a full state of the application thread. + // + // A platform can support lazy loading/restoring of a thread state. + // This means that if the Sentry has not changed a thread state, + // the platform may not restore it. + // + // Preconditions: The caller must be running on the task goroutine. + FloatingPointStateChanged() // Interrupt interrupts a concurrent call to Switch(), causing it to return // ErrContextInterrupt. @@ -218,6 +253,13 @@ type AddressSpace interface { // must be acquired via platform.NewAddressSpace(). Release() + // PreFork() is called before creating a copy of AddressSpace. This + // guarantees that this address space will be in a consistent state. + PreFork() + + // PostFork() is called after creating a copy of AddressSpace. + PostFork() + // AddressSpaceIO methods are supported iff the associated platform's // Platform.SupportsAddressSpaceIO() == true. AddressSpaces for which this // does not hold may panic if AddressSpaceIO methods are invoked. diff --git a/pkg/sentry/platform/ptrace/BUILD b/pkg/sentry/platform/ptrace/BUILD index 29fd23cc3..e04165fbf 100644 --- a/pkg/sentry/platform/ptrace/BUILD +++ b/pkg/sentry/platform/ptrace/BUILD @@ -24,6 +24,7 @@ go_library( visibility = ["//:sandbox"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/log", "//pkg/procid", "//pkg/safecopy", diff --git a/pkg/sentry/platform/ptrace/ptrace.go b/pkg/sentry/platform/ptrace/ptrace.go index 08d055e05..45ff2bcc3 100644 --- a/pkg/sentry/platform/ptrace/ptrace.go +++ b/pkg/sentry/platform/ptrace/ptrace.go @@ -48,6 +48,7 @@ import ( "os" "gvisor.dev/gvisor/pkg/abi/linux" + pkgcontext "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/platform" "gvisor.dev/gvisor/pkg/sentry/platform/interrupt" @@ -95,7 +96,8 @@ type context struct { } // Switch runs the provided context in the given address space. -func (c *context) Switch(as platform.AddressSpace, ac arch.Context, cpu int32) (*arch.SignalInfo, usermem.AccessType, error) { +func (c *context) Switch(ctx pkgcontext.Context, mm platform.MemoryManager, ac arch.Context, cpu int32) (*arch.SignalInfo, usermem.AccessType, error) { + as := mm.AddressSpace() s := as.(*subprocess) isSyscall := s.switchToApp(c, ac) @@ -180,6 +182,12 @@ func (c *context) Interrupt() { // Release implements platform.Context.Release(). func (c *context) Release() {} +// FloatingPointStateChanged implements platform.Context.FloatingPointStateChanged. +func (c *context) FloatingPointStateChanged() {} + +// PullFullState implements platform.Context.PullFullState. +func (c *context) PullFullState(as platform.AddressSpace, ac arch.Context) {} + // PTrace represents a collection of ptrace subprocesses. type PTrace struct { platform.MMapMinAddr diff --git a/pkg/sentry/platform/ptrace/subprocess.go b/pkg/sentry/platform/ptrace/subprocess.go index c990f3454..e1d54d8a2 100644 --- a/pkg/sentry/platform/ptrace/subprocess.go +++ b/pkg/sentry/platform/ptrace/subprocess.go @@ -662,3 +662,9 @@ func (s *subprocess) Unmap(addr usermem.Addr, length uint64) { panic(fmt.Sprintf("munmap(%x, %x)) failed: %v", addr, length, err)) } } + +// PreFork implements platform.AddressSpace.PreFork. +func (s *subprocess) PreFork() {} + +// PostFork implements platform.AddressSpace.PostFork. +func (s *subprocess) PostFork() {} diff --git a/pkg/sentry/platform/ring0/kernel_arm64.go b/pkg/sentry/platform/ring0/kernel_arm64.go index fca3a5478..d483ff03c 100644 --- a/pkg/sentry/platform/ring0/kernel_arm64.go +++ b/pkg/sentry/platform/ring0/kernel_arm64.go @@ -59,11 +59,13 @@ func (c *CPU) SwitchToUser(switchOpts SwitchOpts) (vector Vector) { regs.Pstate &= ^uint64(UserFlagsClear) regs.Pstate |= UserFlagsSet + LoadFloatingPoint(switchOpts.FloatingPointState) SetTLS(regs.TPIDR_EL0) kernelExitToEl0() regs.TPIDR_EL0 = GetTLS() + SaveFloatingPoint(switchOpts.FloatingPointState) vector = c.vecCode diff --git a/pkg/sentry/platform/ring0/lib_arm64.go b/pkg/sentry/platform/ring0/lib_arm64.go index a6345010d..00e52c8af 100644 --- a/pkg/sentry/platform/ring0/lib_arm64.go +++ b/pkg/sentry/platform/ring0/lib_arm64.go @@ -38,6 +38,12 @@ func SaveVRegs(*byte) // LoadVRegs loads V0-V31 registers. func LoadVRegs(*byte) +// LoadFloatingPoint loads floating point state. +func LoadFloatingPoint(*byte) + +// SaveFloatingPoint saves floating point state. +func SaveFloatingPoint(*byte) + // GetTLS returns the value of TPIDR_EL0 register. func GetTLS() (value uint64) diff --git a/pkg/sentry/platform/ring0/lib_arm64.s b/pkg/sentry/platform/ring0/lib_arm64.s index b63e14b41..86bfbe46f 100644 --- a/pkg/sentry/platform/ring0/lib_arm64.s +++ b/pkg/sentry/platform/ring0/lib_arm64.s @@ -129,3 +129,89 @@ TEXT ·LoadVRegs(SB),NOSPLIT,$0-8 ISB $15 RET + +TEXT ·LoadFloatingPoint(SB),NOSPLIT,$0-8 + MOVD addr+0(FP), R0 + + MOVD 0(R0), R1 + MOVD R1, FPSR + MOVD 8(R0), R1 + MOVD R1, NZCV + + FMOVD 16*1(R0), F0 + FMOVD 16*2(R0), F1 + FMOVD 16*3(R0), F2 + FMOVD 16*4(R0), F3 + FMOVD 16*5(R0), F4 + FMOVD 16*6(R0), F5 + FMOVD 16*7(R0), F6 + FMOVD 16*8(R0), F7 + FMOVD 16*9(R0), F8 + FMOVD 16*10(R0), F9 + FMOVD 16*11(R0), F10 + FMOVD 16*12(R0), F11 + FMOVD 16*13(R0), F12 + FMOVD 16*14(R0), F13 + FMOVD 16*15(R0), F14 + FMOVD 16*16(R0), F15 + FMOVD 16*17(R0), F16 + FMOVD 16*18(R0), F17 + FMOVD 16*19(R0), F18 + FMOVD 16*20(R0), F19 + FMOVD 16*21(R0), F20 + FMOVD 16*22(R0), F21 + FMOVD 16*23(R0), F22 + FMOVD 16*24(R0), F23 + FMOVD 16*25(R0), F24 + FMOVD 16*26(R0), F25 + FMOVD 16*27(R0), F26 + FMOVD 16*28(R0), F27 + FMOVD 16*29(R0), F28 + FMOVD 16*30(R0), F29 + FMOVD 16*31(R0), F30 + FMOVD 16*32(R0), F31 + + RET + +TEXT ·SaveFloatingPoint(SB),NOSPLIT,$0-8 + MOVD addr+0(FP), R0 + + MOVD FPSR, R1 + MOVD R1, 0(R0) + MOVD NZCV, R1 + MOVD R1, 8(R0) + + FMOVD F0, 16*1(R0) + FMOVD F1, 16*2(R0) + FMOVD F2, 16*3(R0) + FMOVD F3, 16*4(R0) + FMOVD F4, 16*5(R0) + FMOVD F5, 16*6(R0) + FMOVD F6, 16*7(R0) + FMOVD F7, 16*8(R0) + FMOVD F8, 16*9(R0) + FMOVD F9, 16*10(R0) + FMOVD F10, 16*11(R0) + FMOVD F11, 16*12(R0) + FMOVD F12, 16*13(R0) + FMOVD F13, 16*14(R0) + FMOVD F14, 16*15(R0) + FMOVD F15, 16*16(R0) + FMOVD F16, 16*17(R0) + FMOVD F17, 16*18(R0) + FMOVD F18, 16*19(R0) + FMOVD F19, 16*20(R0) + FMOVD F20, 16*21(R0) + FMOVD F21, 16*22(R0) + FMOVD F22, 16*23(R0) + FMOVD F23, 16*24(R0) + FMOVD F24, 16*25(R0) + FMOVD F25, 16*26(R0) + FMOVD F26, 16*27(R0) + FMOVD F27, 16*28(R0) + FMOVD F28, 16*29(R0) + FMOVD F29, 16*30(R0) + FMOVD F30, 16*31(R0) + FMOVD F31, 16*32(R0) + + RET diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index 8b439a078..70ccf77a7 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -68,7 +68,7 @@ func NewSCMRights(t *kernel.Task, fds []int32) (SCMRights, error) { for _, fd := range fds { file := t.GetFile(fd) if file == nil { - files.Release() + files.Release(t) return nil, syserror.EBADF } files = append(files, file) @@ -100,9 +100,9 @@ func (fs *RightsFiles) Clone() transport.RightsControlMessage { } // Release implements transport.RightsControlMessage.Release. -func (fs *RightsFiles) Release() { +func (fs *RightsFiles) Release(ctx context.Context) { for _, f := range *fs { - f.DecRef() + f.DecRef(ctx) } *fs = nil } @@ -115,7 +115,7 @@ func rightsFDs(t *kernel.Task, rights SCMRights, cloexec bool, max int) ([]int32 fd, err := t.NewFDFrom(0, files[0], kernel.FDFlags{ CloseOnExec: cloexec, }) - files[0].DecRef() + files[0].DecRef(t) files = files[1:] if err != nil { t.Warningf("Error inserting FD: %v", err) diff --git a/pkg/sentry/socket/control/control_vfs2.go b/pkg/sentry/socket/control/control_vfs2.go index fd08179be..d9621968c 100644 --- a/pkg/sentry/socket/control/control_vfs2.go +++ b/pkg/sentry/socket/control/control_vfs2.go @@ -46,7 +46,7 @@ func NewSCMRightsVFS2(t *kernel.Task, fds []int32) (SCMRightsVFS2, error) { for _, fd := range fds { file := t.GetFileVFS2(fd) if file == nil { - files.Release() + files.Release(t) return nil, syserror.EBADF } files = append(files, file) @@ -78,9 +78,9 @@ func (fs *RightsFilesVFS2) Clone() transport.RightsControlMessage { } // Release implements transport.RightsControlMessage.Release. -func (fs *RightsFilesVFS2) Release() { +func (fs *RightsFilesVFS2) Release(ctx context.Context) { for _, f := range *fs { - f.DecRef() + f.DecRef(ctx) } *fs = nil } @@ -93,7 +93,7 @@ func rightsFDsVFS2(t *kernel.Task, rights SCMRightsVFS2, cloexec bool, max int) fd, err := t.NewFDFromVFS2(0, files[0], kernel.FDFlags{ CloseOnExec: cloexec, }) - files[0].DecRef() + files[0].DecRef(t) files = files[1:] if err != nil { t.Warningf("Error inserting FD: %v", err) diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index 532a1ea5d..242e6bf76 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -100,12 +100,12 @@ func newSocketFile(ctx context.Context, family int, stype linux.SockType, protoc return nil, syserr.FromError(err) } dirent := socket.NewDirent(ctx, socketDevice) - defer dirent.DecRef() + defer dirent.DecRef(ctx) return fs.NewFile(ctx, dirent, fs.FileFlags{NonBlocking: nonblock, Read: true, Write: true, NonSeekable: true}, s), nil } // Release implements fs.FileOperations.Release. -func (s *socketOpsCommon) Release() { +func (s *socketOpsCommon) Release(context.Context) { fdnotifier.RemoveFD(int32(s.fd)) syscall.Close(s.fd) } @@ -269,7 +269,7 @@ func (s *socketOpsCommon) Accept(t *kernel.Task, peerRequested bool, flags int, syscall.Close(fd) return 0, nil, 0, err } - defer f.DecRef() + defer f.DecRef(t) kfd, kerr = t.NewFDFromVFS2(0, f, kernel.FDFlags{ CloseOnExec: flags&syscall.SOCK_CLOEXEC != 0, @@ -281,7 +281,7 @@ func (s *socketOpsCommon) Accept(t *kernel.Task, peerRequested bool, flags int, syscall.Close(fd) return 0, nil, 0, err } - defer f.DecRef() + defer f.DecRef(t) kfd, kerr = t.NewFDFrom(0, f, kernel.FDFlags{ CloseOnExec: flags&syscall.SOCK_CLOEXEC != 0, diff --git a/pkg/sentry/socket/netlink/provider.go b/pkg/sentry/socket/netlink/provider.go index 0d45e5053..31e374833 100644 --- a/pkg/sentry/socket/netlink/provider.go +++ b/pkg/sentry/socket/netlink/provider.go @@ -97,7 +97,7 @@ func (*socketProvider) Socket(t *kernel.Task, stype linux.SockType, protocol int } d := socket.NewDirent(t, netlinkSocketDevice) - defer d.DecRef() + defer d.DecRef(t) return fs.NewFile(t, d, fs.FileFlags{Read: true, Write: true, NonSeekable: true}, s), nil } diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go index 98ca7add0..68a9b9a96 100644 --- a/pkg/sentry/socket/netlink/socket.go +++ b/pkg/sentry/socket/netlink/socket.go @@ -140,14 +140,14 @@ func NewSocket(t *kernel.Task, skType linux.SockType, protocol Protocol) (*Socke // Bind the endpoint for good measure so we can connect to it. The // bound address will never be exposed. if err := ep.Bind(tcpip.FullAddress{Addr: "dummy"}, nil); err != nil { - ep.Close() + ep.Close(t) return nil, err } // Create a connection from which the kernel can write messages. connection, err := ep.(transport.BoundEndpoint).UnidirectionalConnect(t) if err != nil { - ep.Close() + ep.Close(t) return nil, err } @@ -164,9 +164,9 @@ func NewSocket(t *kernel.Task, skType linux.SockType, protocol Protocol) (*Socke } // Release implements fs.FileOperations.Release. -func (s *socketOpsCommon) Release() { - s.connection.Release() - s.ep.Close() +func (s *socketOpsCommon) Release(ctx context.Context) { + s.connection.Release(ctx) + s.ep.Close(ctx) if s.bound { s.ports.Release(s.protocol.Protocol(), s.portID) @@ -621,7 +621,7 @@ func (s *socketOpsCommon) sendResponse(ctx context.Context, ms *MessageSet) *sys if len(bufs) > 0 { // RecvMsg never receives the address, so we don't need to send // one. - _, notify, err := s.connection.Send(bufs, cms, tcpip.FullAddress{}) + _, notify, err := s.connection.Send(ctx, bufs, cms, tcpip.FullAddress{}) // If the buffer is full, we simply drop messages, just like // Linux. if err != nil && err != syserr.ErrWouldBlock { @@ -648,7 +648,7 @@ func (s *socketOpsCommon) sendResponse(ctx context.Context, ms *MessageSet) *sys // Add the dump_done_errno payload. m.Put(int64(0)) - _, notify, err := s.connection.Send([][]byte{m.Finalize()}, cms, tcpip.FullAddress{}) + _, notify, err := s.connection.Send(ctx, [][]byte{m.Finalize()}, cms, tcpip.FullAddress{}) if err != nil && err != syserr.ErrWouldBlock { return err } diff --git a/pkg/sentry/socket/netlink/socket_vfs2.go b/pkg/sentry/socket/netlink/socket_vfs2.go index dbcd8b49a..a38d25da9 100644 --- a/pkg/sentry/socket/netlink/socket_vfs2.go +++ b/pkg/sentry/socket/netlink/socket_vfs2.go @@ -57,14 +57,14 @@ func NewVFS2(t *kernel.Task, skType linux.SockType, protocol Protocol) (*SocketV // Bind the endpoint for good measure so we can connect to it. The // bound address will never be exposed. if err := ep.Bind(tcpip.FullAddress{Addr: "dummy"}, nil); err != nil { - ep.Close() + ep.Close(t) return nil, err } // Create a connection from which the kernel can write messages. connection, err := ep.(transport.BoundEndpoint).UnidirectionalConnect(t) if err != nil { - ep.Close() + ep.Close(t) return nil, err } diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index f86e6cd7a..e4846bc0b 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -330,7 +330,7 @@ func New(t *kernel.Task, family int, skType linux.SockType, protocol int, queue } dirent := socket.NewDirent(t, netstackDevice) - defer dirent.DecRef() + defer dirent.DecRef(t) return fs.NewFile(t, dirent, fs.FileFlags{Read: true, Write: true, NonSeekable: true}, &SocketOperations{ socketOpsCommon: socketOpsCommon{ Queue: queue, @@ -479,7 +479,7 @@ func (s *socketOpsCommon) fetchReadView() *syserr.Error { } // Release implements fs.FileOperations.Release. -func (s *socketOpsCommon) Release() { +func (s *socketOpsCommon) Release(context.Context) { s.Endpoint.Close() } @@ -854,7 +854,7 @@ func (s *SocketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, if err != nil { return 0, nil, 0, err } - defer ns.DecRef() + defer ns.DecRef(t) if flags&linux.SOCK_NONBLOCK != 0 { flags := ns.Flags() @@ -1490,6 +1490,10 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (marsha vP := primitive.Int32(boolToInt32(v)) return &vP, nil + case linux.SO_ORIGINAL_DST: + // TODO(gvisor.dev/issue/170): ip6tables. + return nil, syserr.ErrInvalidArgument + default: emitUnimplementedEventIPv6(t, name) } @@ -1600,6 +1604,19 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in vP := primitive.Int32(boolToInt32(v)) return &vP, nil + case linux.SO_ORIGINAL_DST: + if outLen < int(binary.Size(linux.SockAddrInet{})) { + return nil, syserr.ErrInvalidArgument + } + + var v tcpip.OriginalDestinationOption + if err := ep.GetSockOpt(&v); err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + a, _ := ConvertAddress(linux.AF_INET, tcpip.FullAddress(v)) + return a.(*linux.SockAddrInet), nil + default: emitUnimplementedEventIP(t, name) } diff --git a/pkg/sentry/socket/netstack/netstack_vfs2.go b/pkg/sentry/socket/netstack/netstack_vfs2.go index a9025b0ec..3335e7430 100644 --- a/pkg/sentry/socket/netstack/netstack_vfs2.go +++ b/pkg/sentry/socket/netstack/netstack_vfs2.go @@ -169,7 +169,7 @@ func (s *SocketVFS2) Accept(t *kernel.Task, peerRequested bool, flags int, block if err != nil { return 0, nil, 0, err } - defer ns.DecRef() + defer ns.DecRef(t) if err := ns.SetStatusFlags(t, t.Credentials(), uint32(flags&linux.SOCK_NONBLOCK)); err != nil { return 0, nil, 0, syserr.FromError(err) diff --git a/pkg/sentry/socket/socket.go b/pkg/sentry/socket/socket.go index d112757fb..04b259d27 100644 --- a/pkg/sentry/socket/socket.go +++ b/pkg/sentry/socket/socket.go @@ -46,8 +46,8 @@ type ControlMessages struct { } // Release releases Unix domain socket credentials and rights. -func (c *ControlMessages) Release() { - c.Unix.Release() +func (c *ControlMessages) Release(ctx context.Context) { + c.Unix.Release(ctx) } // Socket is an interface combining fs.FileOperations and SocketOps, diff --git a/pkg/sentry/socket/unix/transport/connectioned.go b/pkg/sentry/socket/unix/transport/connectioned.go index a1e49cc57..c67b602f0 100644 --- a/pkg/sentry/socket/unix/transport/connectioned.go +++ b/pkg/sentry/socket/unix/transport/connectioned.go @@ -211,7 +211,7 @@ func (e *connectionedEndpoint) Listening() bool { // The socket will be a fresh state after a call to close and may be reused. // That is, close may be used to "unbind" or "disconnect" the socket in error // paths. -func (e *connectionedEndpoint) Close() { +func (e *connectionedEndpoint) Close(ctx context.Context) { e.Lock() var c ConnectedEndpoint var r Receiver @@ -233,7 +233,7 @@ func (e *connectionedEndpoint) Close() { case e.Listening(): close(e.acceptedChan) for n := range e.acceptedChan { - n.Close() + n.Close(ctx) } e.acceptedChan = nil e.path = "" @@ -241,11 +241,11 @@ func (e *connectionedEndpoint) Close() { e.Unlock() if c != nil { c.CloseNotify() - c.Release() + c.Release(ctx) } if r != nil { r.CloseNotify() - r.Release() + r.Release(ctx) } } @@ -340,7 +340,7 @@ func (e *connectionedEndpoint) BidirectionalConnect(ctx context.Context, ce Conn return nil default: // Busy; return ECONNREFUSED per spec. - ne.Close() + ne.Close(ctx) e.Unlock() ce.Unlock() return syserr.ErrConnectionRefused diff --git a/pkg/sentry/socket/unix/transport/connectionless.go b/pkg/sentry/socket/unix/transport/connectionless.go index 4b06d63ac..70ee8f9b8 100644 --- a/pkg/sentry/socket/unix/transport/connectionless.go +++ b/pkg/sentry/socket/unix/transport/connectionless.go @@ -54,10 +54,10 @@ func (e *connectionlessEndpoint) isBound() bool { // Close puts the endpoint in a closed state and frees all resources associated // with it. -func (e *connectionlessEndpoint) Close() { +func (e *connectionlessEndpoint) Close(ctx context.Context) { e.Lock() if e.connected != nil { - e.connected.Release() + e.connected.Release(ctx) e.connected = nil } @@ -71,7 +71,7 @@ func (e *connectionlessEndpoint) Close() { e.Unlock() r.CloseNotify() - r.Release() + r.Release(ctx) } // BidirectionalConnect implements BoundEndpoint.BidirectionalConnect. @@ -108,10 +108,10 @@ func (e *connectionlessEndpoint) SendMsg(ctx context.Context, data [][]byte, c C if err != nil { return 0, syserr.ErrInvalidEndpointState } - defer connected.Release() + defer connected.Release(ctx) e.Lock() - n, notify, err := connected.Send(data, c, tcpip.FullAddress{Addr: tcpip.Address(e.path)}) + n, notify, err := connected.Send(ctx, data, c, tcpip.FullAddress{Addr: tcpip.Address(e.path)}) e.Unlock() if notify { @@ -135,7 +135,7 @@ func (e *connectionlessEndpoint) Connect(ctx context.Context, server BoundEndpoi e.Lock() if e.connected != nil { - e.connected.Release() + e.connected.Release(ctx) } e.connected = connected e.Unlock() diff --git a/pkg/sentry/socket/unix/transport/queue.go b/pkg/sentry/socket/unix/transport/queue.go index d8f3ad63d..ef6043e19 100644 --- a/pkg/sentry/socket/unix/transport/queue.go +++ b/pkg/sentry/socket/unix/transport/queue.go @@ -15,6 +15,7 @@ package transport import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" @@ -57,10 +58,10 @@ func (q *queue) Close() { // Both the read and write queues must be notified after resetting: // q.ReaderQueue.Notify(waiter.EventIn) // q.WriterQueue.Notify(waiter.EventOut) -func (q *queue) Reset() { +func (q *queue) Reset(ctx context.Context) { q.mu.Lock() for cur := q.dataList.Front(); cur != nil; cur = cur.Next() { - cur.Release() + cur.Release(ctx) } q.dataList.Reset() q.used = 0 @@ -68,8 +69,8 @@ func (q *queue) Reset() { } // DecRef implements RefCounter.DecRef with destructor q.Reset. -func (q *queue) DecRef() { - q.DecRefWithDestructor(q.Reset) +func (q *queue) DecRef(ctx context.Context) { + q.DecRefWithDestructor(ctx, q.Reset) // We don't need to notify after resetting because no one cares about // this queue after all references have been dropped. } @@ -111,7 +112,7 @@ func (q *queue) IsWritable() bool { // // If notify is true, ReaderQueue.Notify must be called: // q.ReaderQueue.Notify(waiter.EventIn) -func (q *queue) Enqueue(data [][]byte, c ControlMessages, from tcpip.FullAddress, discardEmpty bool, truncate bool) (l int64, notify bool, err *syserr.Error) { +func (q *queue) Enqueue(ctx context.Context, data [][]byte, c ControlMessages, from tcpip.FullAddress, discardEmpty bool, truncate bool) (l int64, notify bool, err *syserr.Error) { q.mu.Lock() if q.closed { @@ -124,7 +125,7 @@ func (q *queue) Enqueue(data [][]byte, c ControlMessages, from tcpip.FullAddress } if discardEmpty && l == 0 { q.mu.Unlock() - c.Release() + c.Release(ctx) return 0, false, nil } diff --git a/pkg/sentry/socket/unix/transport/unix.go b/pkg/sentry/socket/unix/transport/unix.go index 2f1b127df..475d7177e 100644 --- a/pkg/sentry/socket/unix/transport/unix.go +++ b/pkg/sentry/socket/unix/transport/unix.go @@ -37,7 +37,7 @@ type RightsControlMessage interface { Clone() RightsControlMessage // Release releases any resources owned by the RightsControlMessage. - Release() + Release(ctx context.Context) } // A CredentialsControlMessage is a control message containing Unix credentials. @@ -74,9 +74,9 @@ func (c *ControlMessages) Clone() ControlMessages { } // Release releases both the credentials and the rights. -func (c *ControlMessages) Release() { +func (c *ControlMessages) Release(ctx context.Context) { if c.Rights != nil { - c.Rights.Release() + c.Rights.Release(ctx) } *c = ControlMessages{} } @@ -90,7 +90,7 @@ type Endpoint interface { // Close puts the endpoint in a closed state and frees all resources // associated with it. - Close() + Close(ctx context.Context) // RecvMsg reads data and a control message from the endpoint. This method // does not block if there is no data pending. @@ -252,7 +252,7 @@ type BoundEndpoint interface { // Release releases any resources held by the BoundEndpoint. It must be // called before dropping all references to a BoundEndpoint returned by a // function. - Release() + Release(ctx context.Context) } // message represents a message passed over a Unix domain socket. @@ -281,8 +281,8 @@ func (m *message) Length() int64 { } // Release releases any resources held by the message. -func (m *message) Release() { - m.Control.Release() +func (m *message) Release(ctx context.Context) { + m.Control.Release(ctx) } // Peek returns a copy of the message. @@ -304,7 +304,7 @@ type Receiver interface { // See Endpoint.RecvMsg for documentation on shared arguments. // // notify indicates if RecvNotify should be called. - Recv(data [][]byte, creds bool, numRights int, peek bool) (recvLen, msgLen int64, cm ControlMessages, CMTruncated bool, source tcpip.FullAddress, notify bool, err *syserr.Error) + Recv(ctx context.Context, data [][]byte, creds bool, numRights int, peek bool) (recvLen, msgLen int64, cm ControlMessages, CMTruncated bool, source tcpip.FullAddress, notify bool, err *syserr.Error) // RecvNotify notifies the Receiver of a successful Recv. This must not be // called while holding any endpoint locks. @@ -333,7 +333,7 @@ type Receiver interface { // Release releases any resources owned by the Receiver. It should be // called before droping all references to a Receiver. - Release() + Release(ctx context.Context) } // queueReceiver implements Receiver for datagram sockets. @@ -344,7 +344,7 @@ type queueReceiver struct { } // Recv implements Receiver.Recv. -func (q *queueReceiver) Recv(data [][]byte, creds bool, numRights int, peek bool) (int64, int64, ControlMessages, bool, tcpip.FullAddress, bool, *syserr.Error) { +func (q *queueReceiver) Recv(ctx context.Context, data [][]byte, creds bool, numRights int, peek bool) (int64, int64, ControlMessages, bool, tcpip.FullAddress, bool, *syserr.Error) { var m *message var notify bool var err *syserr.Error @@ -398,8 +398,8 @@ func (q *queueReceiver) RecvMaxQueueSize() int64 { } // Release implements Receiver.Release. -func (q *queueReceiver) Release() { - q.readQueue.DecRef() +func (q *queueReceiver) Release(ctx context.Context) { + q.readQueue.DecRef(ctx) } // streamQueueReceiver implements Receiver for stream sockets. @@ -456,7 +456,7 @@ func (q *streamQueueReceiver) RecvMaxQueueSize() int64 { } // Recv implements Receiver.Recv. -func (q *streamQueueReceiver) Recv(data [][]byte, wantCreds bool, numRights int, peek bool) (int64, int64, ControlMessages, bool, tcpip.FullAddress, bool, *syserr.Error) { +func (q *streamQueueReceiver) Recv(ctx context.Context, data [][]byte, wantCreds bool, numRights int, peek bool) (int64, int64, ControlMessages, bool, tcpip.FullAddress, bool, *syserr.Error) { q.mu.Lock() defer q.mu.Unlock() @@ -502,7 +502,7 @@ func (q *streamQueueReceiver) Recv(data [][]byte, wantCreds bool, numRights int, var cmTruncated bool if c.Rights != nil && numRights == 0 { - c.Rights.Release() + c.Rights.Release(ctx) c.Rights = nil cmTruncated = true } @@ -557,7 +557,7 @@ func (q *streamQueueReceiver) Recv(data [][]byte, wantCreds bool, numRights int, // Consume rights. if numRights == 0 { cmTruncated = true - q.control.Rights.Release() + q.control.Rights.Release(ctx) } else { c.Rights = q.control.Rights haveRights = true @@ -582,7 +582,7 @@ type ConnectedEndpoint interface { // // syserr.ErrWouldBlock can be returned along with a partial write if // the caller should block to send the rest of the data. - Send(data [][]byte, c ControlMessages, from tcpip.FullAddress) (n int64, notify bool, err *syserr.Error) + Send(ctx context.Context, data [][]byte, c ControlMessages, from tcpip.FullAddress) (n int64, notify bool, err *syserr.Error) // SendNotify notifies the ConnectedEndpoint of a successful Send. This // must not be called while holding any endpoint locks. @@ -616,7 +616,7 @@ type ConnectedEndpoint interface { // Release releases any resources owned by the ConnectedEndpoint. It should // be called before droping all references to a ConnectedEndpoint. - Release() + Release(ctx context.Context) // CloseUnread sets the fact that this end is closed with unread data to // the peer socket. @@ -654,7 +654,7 @@ func (e *connectedEndpoint) GetLocalAddress() (tcpip.FullAddress, *tcpip.Error) } // Send implements ConnectedEndpoint.Send. -func (e *connectedEndpoint) Send(data [][]byte, c ControlMessages, from tcpip.FullAddress) (int64, bool, *syserr.Error) { +func (e *connectedEndpoint) Send(ctx context.Context, data [][]byte, c ControlMessages, from tcpip.FullAddress) (int64, bool, *syserr.Error) { discardEmpty := false truncate := false if e.endpoint.Type() == linux.SOCK_STREAM { @@ -669,7 +669,7 @@ func (e *connectedEndpoint) Send(data [][]byte, c ControlMessages, from tcpip.Fu truncate = true } - return e.writeQueue.Enqueue(data, c, from, discardEmpty, truncate) + return e.writeQueue.Enqueue(ctx, data, c, from, discardEmpty, truncate) } // SendNotify implements ConnectedEndpoint.SendNotify. @@ -707,8 +707,8 @@ func (e *connectedEndpoint) SendMaxQueueSize() int64 { } // Release implements ConnectedEndpoint.Release. -func (e *connectedEndpoint) Release() { - e.writeQueue.DecRef() +func (e *connectedEndpoint) Release(ctx context.Context) { + e.writeQueue.DecRef(ctx) } // CloseUnread implements ConnectedEndpoint.CloseUnread. @@ -798,7 +798,7 @@ func (e *baseEndpoint) RecvMsg(ctx context.Context, data [][]byte, creds bool, n return 0, 0, ControlMessages{}, false, syserr.ErrNotConnected } - recvLen, msgLen, cms, cmt, a, notify, err := e.receiver.Recv(data, creds, numRights, peek) + recvLen, msgLen, cms, cmt, a, notify, err := e.receiver.Recv(ctx, data, creds, numRights, peek) e.Unlock() if err != nil { return 0, 0, ControlMessages{}, false, err @@ -827,7 +827,7 @@ func (e *baseEndpoint) SendMsg(ctx context.Context, data [][]byte, c ControlMess return 0, syserr.ErrAlreadyConnected } - n, notify, err := e.connected.Send(data, c, tcpip.FullAddress{Addr: tcpip.Address(e.path)}) + n, notify, err := e.connected.Send(ctx, data, c, tcpip.FullAddress{Addr: tcpip.Address(e.path)}) e.Unlock() if notify { @@ -1001,6 +1001,6 @@ func (e *baseEndpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { } // Release implements BoundEndpoint.Release. -func (*baseEndpoint) Release() { +func (*baseEndpoint) Release(context.Context) { // Binding a baseEndpoint doesn't take a reference. } diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 0482d33cf..2b8454edb 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -62,7 +62,7 @@ type SocketOperations struct { // New creates a new unix socket. func New(ctx context.Context, endpoint transport.Endpoint, stype linux.SockType) *fs.File { dirent := socket.NewDirent(ctx, unixSocketDevice) - defer dirent.DecRef() + defer dirent.DecRef(ctx) return NewWithDirent(ctx, dirent, endpoint, stype, fs.FileFlags{Read: true, Write: true, NonSeekable: true}) } @@ -97,17 +97,17 @@ type socketOpsCommon struct { } // DecRef implements RefCounter.DecRef. -func (s *socketOpsCommon) DecRef() { - s.DecRefWithDestructor(func() { - s.ep.Close() +func (s *socketOpsCommon) DecRef(ctx context.Context) { + s.DecRefWithDestructor(ctx, func(context.Context) { + s.ep.Close(ctx) }) } // Release implemements fs.FileOperations.Release. -func (s *socketOpsCommon) Release() { +func (s *socketOpsCommon) Release(ctx context.Context) { // Release only decrements a reference on s because s may be referenced in // the abstract socket namespace. - s.DecRef() + s.DecRef(ctx) } func (s *socketOpsCommon) isPacket() bool { @@ -234,7 +234,7 @@ func (s *SocketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, } ns := New(t, ep, s.stype) - defer ns.DecRef() + defer ns.DecRef(t) if flags&linux.SOCK_NONBLOCK != 0 { flags := ns.Flags() @@ -284,7 +284,7 @@ func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { if t.IsNetworkNamespaced() { return syserr.ErrInvalidEndpointState } - if err := t.AbstractSockets().Bind(p[1:], bep, s); err != nil { + if err := t.AbstractSockets().Bind(t, p[1:], bep, s); err != nil { // syserr.ErrPortInUse corresponds to EADDRINUSE. return syserr.ErrPortInUse } @@ -294,7 +294,7 @@ func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { var name string cwd := t.FSContext().WorkingDirectory() - defer cwd.DecRef() + defer cwd.DecRef(t) // Is there no slash at all? if !strings.Contains(p, "/") { @@ -302,7 +302,7 @@ func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { name = p } else { root := t.FSContext().RootDirectory() - defer root.DecRef() + defer root.DecRef(t) // Find the last path component, we know that something follows // that final slash, otherwise extractPath() would have failed. lastSlash := strings.LastIndex(p, "/") @@ -318,7 +318,7 @@ func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { // No path available. return syserr.ErrNoSuchFile } - defer d.DecRef() + defer d.DecRef(t) name = p[lastSlash+1:] } @@ -332,7 +332,7 @@ func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { if err != nil { return syserr.ErrPortInUse } - childDir.DecRef() + childDir.DecRef(t) } return nil @@ -378,9 +378,9 @@ func extractEndpoint(t *kernel.Task, sockaddr []byte) (transport.BoundEndpoint, FollowFinalSymlink: true, } ep, e := t.Kernel().VFS().BoundEndpointAt(t, t.Credentials(), &pop, &vfs.BoundEndpointOptions{path}) - root.DecRef() + root.DecRef(t) if relPath { - start.DecRef() + start.DecRef(t) } if e != nil { return nil, syserr.FromError(e) @@ -393,15 +393,15 @@ func extractEndpoint(t *kernel.Task, sockaddr []byte) (transport.BoundEndpoint, cwd := t.FSContext().WorkingDirectory() remainingTraversals := uint(fs.DefaultTraversalLimit) d, e := t.MountNamespace().FindInode(t, root, cwd, path, &remainingTraversals) - cwd.DecRef() - root.DecRef() + cwd.DecRef(t) + root.DecRef(t) if e != nil { return nil, syserr.FromError(e) } // Extract the endpoint if one is there. ep := d.Inode.BoundEndpoint(path) - d.DecRef() + d.DecRef(t) if ep == nil { // No socket! return nil, syserr.ErrConnectionRefused @@ -415,7 +415,7 @@ func (s *socketOpsCommon) Connect(t *kernel.Task, sockaddr []byte, blocking bool if err != nil { return err } - defer ep.Release() + defer ep.Release(t) // Connect the server endpoint. err = s.ep.Connect(t, ep) @@ -473,7 +473,7 @@ func (s *socketOpsCommon) SendMsg(t *kernel.Task, src usermem.IOSequence, to []b if err != nil { return 0, err } - defer ep.Release() + defer ep.Release(t) w.To = ep if ep.Passcred() && w.Control.Credentials == nil { diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index 05c16fcfe..dfa25241a 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -136,7 +136,7 @@ func (s *SocketVFS2) Accept(t *kernel.Task, peerRequested bool, flags int, block if err != nil { return 0, nil, 0, err } - defer ns.DecRef() + defer ns.DecRef(t) if flags&linux.SOCK_NONBLOCK != 0 { ns.SetStatusFlags(t, t.Credentials(), linux.SOCK_NONBLOCK) @@ -183,19 +183,19 @@ func (s *SocketVFS2) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { if t.IsNetworkNamespaced() { return syserr.ErrInvalidEndpointState } - if err := t.AbstractSockets().Bind(p[1:], bep, s); err != nil { + if err := t.AbstractSockets().Bind(t, p[1:], bep, s); err != nil { // syserr.ErrPortInUse corresponds to EADDRINUSE. return syserr.ErrPortInUse } } else { path := fspath.Parse(p) root := t.FSContext().RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) start := root relPath := !path.Absolute if relPath { start = t.FSContext().WorkingDirectoryVFS2() - defer start.DecRef() + defer start.DecRef(t) } pop := vfs.PathOperation{ Root: root, @@ -333,7 +333,7 @@ func (*providerVFS2) Socket(t *kernel.Task, stype linux.SockType, protocol int) f, err := NewSockfsFile(t, ep, stype) if err != nil { - ep.Close() + ep.Close(t) return nil, err } return f, nil @@ -357,14 +357,14 @@ func (*providerVFS2) Pair(t *kernel.Task, stype linux.SockType, protocol int) (* ep1, ep2 := transport.NewPair(t, stype, t.Kernel()) s1, err := NewSockfsFile(t, ep1, stype) if err != nil { - ep1.Close() - ep2.Close() + ep1.Close(t) + ep2.Close(t) return nil, nil, err } s2, err := NewSockfsFile(t, ep2, stype) if err != nil { - s1.DecRef() - ep2.Close() + s1.DecRef(t) + ep2.Close(t) return nil, nil, err } diff --git a/pkg/sentry/state/state.go b/pkg/sentry/state/state.go index 9eb626b76..a06c9b8ab 100644 --- a/pkg/sentry/state/state.go +++ b/pkg/sentry/state/state.go @@ -60,6 +60,7 @@ type SaveOpts struct { func (opts SaveOpts) Save(k *kernel.Kernel, w *watchdog.Watchdog) error { log.Infof("Sandbox save started, pausing all tasks.") k.Pause() + k.ReceiveTaskStates() defer k.Unpause() defer log.Infof("Tasks resumed after save.") diff --git a/pkg/sentry/strace/socket.go b/pkg/sentry/strace/socket.go index c0512de89..b51c4c941 100644 --- a/pkg/sentry/strace/socket.go +++ b/pkg/sentry/strace/socket.go @@ -521,6 +521,7 @@ var sockOptNames = map[uint64]abi.ValueSet{ linux.IP_ROUTER_ALERT: "IP_ROUTER_ALERT", linux.IP_PKTOPTIONS: "IP_PKTOPTIONS", linux.IP_MTU: "IP_MTU", + linux.SO_ORIGINAL_DST: "SO_ORIGINAL_DST", }, linux.SOL_SOCKET: { linux.SO_ERROR: "SO_ERROR", diff --git a/pkg/sentry/strace/strace.go b/pkg/sentry/strace/strace.go index 68ca537c8..87b239730 100644 --- a/pkg/sentry/strace/strace.go +++ b/pkg/sentry/strace/strace.go @@ -147,14 +147,14 @@ func fd(t *kernel.Task, fd int32) string { root := t.FSContext().RootDirectory() if root != nil { - defer root.DecRef() + defer root.DecRef(t) } if fd == linux.AT_FDCWD { wd := t.FSContext().WorkingDirectory() var name string if wd != nil { - defer wd.DecRef() + defer wd.DecRef(t) name, _ = wd.FullName(root) } else { name = "(unknown cwd)" @@ -167,7 +167,7 @@ func fd(t *kernel.Task, fd int32) string { // Cast FD to uint64 to avoid printing negative hex. return fmt.Sprintf("%#x (bad FD)", uint64(fd)) } - defer file.DecRef() + defer file.DecRef(t) name, _ := file.Dirent.FullName(root) return fmt.Sprintf("%#x %s", fd, name) @@ -175,12 +175,12 @@ func fd(t *kernel.Task, fd int32) string { func fdVFS2(t *kernel.Task, fd int32) string { root := t.FSContext().RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) vfsObj := root.Mount().Filesystem().VirtualFilesystem() if fd == linux.AT_FDCWD { wd := t.FSContext().WorkingDirectoryVFS2() - defer wd.DecRef() + defer wd.DecRef(t) name, _ := vfsObj.PathnameWithDeleted(t, root, wd) return fmt.Sprintf("AT_FDCWD %s", name) @@ -191,7 +191,7 @@ func fdVFS2(t *kernel.Task, fd int32) string { // Cast FD to uint64 to avoid printing negative hex. return fmt.Sprintf("%#x (bad FD)", uint64(fd)) } - defer file.DecRef() + defer file.DecRef(t) name, _ := vfsObj.PathnameWithDeleted(t, root, file.VirtualDentry()) return fmt.Sprintf("%#x %s", fd, name) diff --git a/pkg/sentry/syscalls/epoll.go b/pkg/sentry/syscalls/epoll.go index d9fb808c0..d23a0068a 100644 --- a/pkg/sentry/syscalls/epoll.go +++ b/pkg/sentry/syscalls/epoll.go @@ -28,7 +28,7 @@ import ( // CreateEpoll implements the epoll_create(2) linux syscall. func CreateEpoll(t *kernel.Task, closeOnExec bool) (int32, error) { file := epoll.NewEventPoll(t) - defer file.DecRef() + defer file.DecRef(t) fd, err := t.NewFDFrom(0, file, kernel.FDFlags{ CloseOnExec: closeOnExec, @@ -47,14 +47,14 @@ func AddEpoll(t *kernel.Task, epfd int32, fd int32, flags epoll.EntryFlags, mask if epollfile == nil { return syserror.EBADF } - defer epollfile.DecRef() + defer epollfile.DecRef(t) // Get the target file id. file := t.GetFile(fd) if file == nil { return syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the epollPoll operations. e, ok := epollfile.FileOperations.(*epoll.EventPoll) @@ -73,14 +73,14 @@ func UpdateEpoll(t *kernel.Task, epfd int32, fd int32, flags epoll.EntryFlags, m if epollfile == nil { return syserror.EBADF } - defer epollfile.DecRef() + defer epollfile.DecRef(t) // Get the target file id. file := t.GetFile(fd) if file == nil { return syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the epollPoll operations. e, ok := epollfile.FileOperations.(*epoll.EventPoll) @@ -99,14 +99,14 @@ func RemoveEpoll(t *kernel.Task, epfd int32, fd int32) error { if epollfile == nil { return syserror.EBADF } - defer epollfile.DecRef() + defer epollfile.DecRef(t) // Get the target file id. file := t.GetFile(fd) if file == nil { return syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the epollPoll operations. e, ok := epollfile.FileOperations.(*epoll.EventPoll) @@ -115,7 +115,7 @@ func RemoveEpoll(t *kernel.Task, epfd int32, fd int32) error { } // Try to remove the entry. - return e.RemoveEntry(epoll.FileIdentifier{file, fd}) + return e.RemoveEntry(t, epoll.FileIdentifier{file, fd}) } // WaitEpoll implements the epoll_wait(2) linux syscall. @@ -125,7 +125,7 @@ func WaitEpoll(t *kernel.Task, fd int32, max int, timeout int) ([]linux.EpollEve if epollfile == nil { return nil, syserror.EBADF } - defer epollfile.DecRef() + defer epollfile.DecRef(t) // Extract the epollPoll operations. e, ok := epollfile.FileOperations.(*epoll.EventPoll) diff --git a/pkg/sentry/syscalls/linux/sys_aio.go b/pkg/sentry/syscalls/linux/sys_aio.go index ba2557c52..e9d64dec5 100644 --- a/pkg/sentry/syscalls/linux/sys_aio.go +++ b/pkg/sentry/syscalls/linux/sys_aio.go @@ -247,7 +247,7 @@ func getAIOCallback(t *kernel.Task, file *fs.File, cbAddr usermem.Addr, cb *linu ev.Result = -int64(kernel.ExtractErrno(err, 0)) } - file.DecRef() + file.DecRef(ctx) // Queue the result for delivery. actx.FinishRequest(ev) @@ -257,7 +257,7 @@ func getAIOCallback(t *kernel.Task, file *fs.File, cbAddr usermem.Addr, cb *linu // wake up. if eventFile != nil { eventFile.FileOperations.(*eventfd.EventOperations).Signal(1) - eventFile.DecRef() + eventFile.DecRef(ctx) } } } @@ -269,7 +269,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr user // File not found. return syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Was there an eventFD? Extract it. var eventFile *fs.File @@ -279,7 +279,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr user // Bad FD. return syserror.EBADF } - defer eventFile.DecRef() + defer eventFile.DecRef(t) // Check that it is an eventfd. if _, ok := eventFile.FileOperations.(*eventfd.EventOperations); !ok { diff --git a/pkg/sentry/syscalls/linux/sys_eventfd.go b/pkg/sentry/syscalls/linux/sys_eventfd.go index ed3413ca6..3b4f879e4 100644 --- a/pkg/sentry/syscalls/linux/sys_eventfd.go +++ b/pkg/sentry/syscalls/linux/sys_eventfd.go @@ -37,7 +37,7 @@ func Eventfd2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc event.SetFlags(fs.SettableFileFlags{ NonBlocking: flags&linux.EFD_NONBLOCK != 0, }) - defer event.DecRef() + defer event.DecRef(t) fd, err := t.NewFDFrom(0, event, kernel.FDFlags{ CloseOnExec: flags&linux.EFD_CLOEXEC != 0, diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 8cf6401e7..1bc9b184e 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -40,7 +40,7 @@ func fileOpAt(t *kernel.Task, dirFD int32, path string, fn func(root *fs.Dirent, // Common case: we are accessing a file in the root. root := t.FSContext().RootDirectory() err := fn(root, root, name, linux.MaxSymlinkTraversals) - root.DecRef() + root.DecRef(t) return err } else if dir == "." && dirFD == linux.AT_FDCWD { // Common case: we are accessing a file relative to the current @@ -48,8 +48,8 @@ func fileOpAt(t *kernel.Task, dirFD int32, path string, fn func(root *fs.Dirent, wd := t.FSContext().WorkingDirectory() root := t.FSContext().RootDirectory() err := fn(root, wd, name, linux.MaxSymlinkTraversals) - wd.DecRef() - root.DecRef() + wd.DecRef(t) + root.DecRef(t) return err } @@ -97,19 +97,19 @@ func fileOpOn(t *kernel.Task, dirFD int32, path string, resolve bool, fn func(ro } else { d, err = t.MountNamespace().FindLink(t, root, rel, path, &remainingTraversals) } - root.DecRef() + root.DecRef(t) if wd != nil { - wd.DecRef() + wd.DecRef(t) } if f != nil { - f.DecRef() + f.DecRef(t) } if err != nil { return err } err = fn(root, d, remainingTraversals) - d.DecRef() + d.DecRef(t) return err } @@ -186,7 +186,7 @@ func openAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint) (fd uint if err != nil { return syserror.ConvertIntr(err, kernel.ERESTARTSYS) } - defer file.DecRef() + defer file.DecRef(t) // Success. newFD, err := t.NewFDFrom(0, file, kernel.FDFlags{ @@ -242,7 +242,7 @@ func mknodAt(t *kernel.Task, dirFD int32, addr usermem.Addr, mode linux.FileMode if err != nil { return err } - file.DecRef() + file.DecRef(t) return nil case linux.ModeNamedPipe: @@ -332,7 +332,7 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l if err != nil { break } - defer found.DecRef() + defer found.DecRef(t) // We found something (possibly a symlink). If the // O_EXCL flag was passed, then we can immediately @@ -357,7 +357,7 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l resolved, err = found.Inode.Getlink(t) if err == nil { // No more resolution necessary. - defer resolved.DecRef() + defer resolved.DecRef(t) break } if err != fs.ErrResolveViaReadlink { @@ -384,7 +384,7 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l if err != nil { break } - defer newParent.DecRef() + defer newParent.DecRef(t) // Repeat the process with the parent and name of the // symlink target. @@ -416,7 +416,7 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l if err != nil { return syserror.ConvertIntr(err, kernel.ERESTARTSYS) } - defer newFile.DecRef() + defer newFile.DecRef(t) case syserror.ENOENT: // File does not exist. Proceed with creation. @@ -432,7 +432,7 @@ func createAt(t *kernel.Task, dirFD int32, addr usermem.Addr, flags uint, mode l // No luck, bail. return err } - defer newFile.DecRef() + defer newFile.DecRef(t) found = newFile.Dirent default: return err @@ -596,7 +596,7 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Shared flags between file and socket. switch request { @@ -671,9 +671,9 @@ func Getcwd(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal addr := args[0].Pointer() size := args[1].SizeT() cwd := t.FSContext().WorkingDirectory() - defer cwd.DecRef() + defer cwd.DecRef(t) root := t.FSContext().RootDirectory() - defer root.DecRef() + defer root.DecRef(t) // Get our fullname from the root and preprend unreachable if the root was // unreachable from our current dirent this is the same behavior as on linux. @@ -722,7 +722,7 @@ func Chroot(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return err } - t.FSContext().SetRootDirectory(d) + t.FSContext().SetRootDirectory(t, d) return nil }) } @@ -747,7 +747,7 @@ func Chdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return err } - t.FSContext().SetWorkingDirectory(d) + t.FSContext().SetWorkingDirectory(t, d) return nil }) } @@ -760,7 +760,7 @@ func Fchdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Is it a directory? if !fs.IsDir(file.Dirent.Inode.StableAttr) { @@ -772,7 +772,7 @@ func Fchdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return 0, nil, err } - t.FSContext().SetWorkingDirectory(file.Dirent) + t.FSContext().SetWorkingDirectory(t, file.Dirent) return 0, nil, nil } @@ -791,7 +791,7 @@ func Close(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) err := file.Flush(t) return 0, nil, handleIOError(t, false /* partial */, err, syserror.EINTR, "close", file) @@ -805,7 +805,7 @@ func Dup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) newFD, err := t.NewFDFrom(0, file, kernel.FDFlags{}) if err != nil { @@ -826,7 +826,7 @@ func Dup2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if oldFile == nil { return 0, nil, syserror.EBADF } - defer oldFile.DecRef() + defer oldFile.DecRef(t) return uintptr(newfd), nil, nil } @@ -850,7 +850,7 @@ func Dup3(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if oldFile == nil { return 0, nil, syserror.EBADF } - defer oldFile.DecRef() + defer oldFile.DecRef(t) err := t.NewFDAt(newfd, oldFile, kernel.FDFlags{CloseOnExec: flags&linux.O_CLOEXEC != 0}) if err != nil { @@ -925,7 +925,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) switch cmd { case linux.F_DUPFD, linux.F_DUPFD_CLOEXEC: @@ -1132,7 +1132,7 @@ func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // If the FD refers to a pipe or FIFO, return error. if fs.IsPipe(file.Dirent.Inode.StableAttr) { @@ -1171,7 +1171,7 @@ func mkdirAt(t *kernel.Task, dirFD int32, addr usermem.Addr, mode linux.FileMode switch err { case nil: // The directory existed. - defer f.DecRef() + defer f.DecRef(t) return syserror.EEXIST case syserror.EACCES: // Permission denied while walking to the directory. @@ -1349,7 +1349,7 @@ func linkAt(t *kernel.Task, oldDirFD int32, oldAddr usermem.Addr, newDirFD int32 if target == nil { return syserror.EBADF } - defer target.DecRef() + defer target.DecRef(t) if err := mayLinkAt(t, target.Dirent.Inode); err != nil { return err } @@ -1602,7 +1602,7 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Reject truncation if the file flags do not permit this operation. // This is different from truncate(2) above. @@ -1730,7 +1730,7 @@ func chownAt(t *kernel.Task, fd int32, addr usermem.Addr, resolve, allowEmpty bo if file == nil { return syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return chown(t, file.Dirent, uid, gid) } @@ -1768,7 +1768,7 @@ func Fchown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, chown(t, file.Dirent, uid, gid) } @@ -1833,7 +1833,7 @@ func Fchmod(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, chmod(t, file.Dirent, mode) } @@ -1893,10 +1893,10 @@ func utimes(t *kernel.Task, dirFD int32, addr usermem.Addr, ts fs.TimeSpec, reso if f == nil { return syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) root := t.FSContext().RootDirectory() - defer root.DecRef() + defer root.DecRef(t) return setTimestamp(root, f.Dirent, linux.MaxSymlinkTraversals) } @@ -2088,7 +2088,7 @@ func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) if offset < 0 || length <= 0 { return 0, nil, syserror.EINVAL @@ -2141,7 +2141,7 @@ func Flock(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // flock(2): EBADF fd is not an open file descriptor. return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) nonblocking := operation&linux.LOCK_NB != 0 operation &^= linux.LOCK_NB @@ -2224,8 +2224,8 @@ func MemfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S return 0, nil, err } - defer dirent.DecRef() - defer file.DecRef() + defer dirent.DecRef(t) + defer file.DecRef(t) newFD, err := t.NewFDFrom(0, file, kernel.FDFlags{ CloseOnExec: cloExec, diff --git a/pkg/sentry/syscalls/linux/sys_futex.go b/pkg/sentry/syscalls/linux/sys_futex.go index f04d78856..9d1b2edb1 100644 --- a/pkg/sentry/syscalls/linux/sys_futex.go +++ b/pkg/sentry/syscalls/linux/sys_futex.go @@ -73,7 +73,7 @@ func futexWaitAbsolute(t *kernel.Task, clockRealtime bool, ts linux.Timespec, fo err = t.BlockWithDeadline(w.C, true, ktime.FromTimespec(ts)) } - t.Futex().WaitComplete(w) + t.Futex().WaitComplete(w, t) return 0, syserror.ConvertIntr(err, kernel.ERESTARTSYS) } @@ -95,7 +95,7 @@ func futexWaitDuration(t *kernel.Task, duration time.Duration, forever bool, add } remaining, err := t.BlockWithTimeout(w.C, !forever, duration) - t.Futex().WaitComplete(w) + t.Futex().WaitComplete(w, t) if err == nil { return 0, nil } @@ -148,7 +148,7 @@ func futexLockPI(t *kernel.Task, ts linux.Timespec, forever bool, addr usermem.A timer.Destroy() } - t.Futex().WaitComplete(w) + t.Futex().WaitComplete(w, t) return syserror.ConvertIntr(err, kernel.ERESTARTSYS) } diff --git a/pkg/sentry/syscalls/linux/sys_getdents.go b/pkg/sentry/syscalls/linux/sys_getdents.go index b126fecc0..f5699e55d 100644 --- a/pkg/sentry/syscalls/linux/sys_getdents.go +++ b/pkg/sentry/syscalls/linux/sys_getdents.go @@ -68,7 +68,7 @@ func getdents(t *kernel.Task, fd int32, addr usermem.Addr, size int, f func(*dir if dir == nil { return 0, syserror.EBADF } - defer dir.DecRef() + defer dir.DecRef(t) w := &usermem.IOReadWriter{ Ctx: t, diff --git a/pkg/sentry/syscalls/linux/sys_inotify.go b/pkg/sentry/syscalls/linux/sys_inotify.go index b2c7b3444..cf47bb9dd 100644 --- a/pkg/sentry/syscalls/linux/sys_inotify.go +++ b/pkg/sentry/syscalls/linux/sys_inotify.go @@ -40,7 +40,7 @@ func InotifyInit1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. NonBlocking: flags&linux.IN_NONBLOCK != 0, } n := fs.NewFile(t, dirent, fileFlags, fs.NewInotify(t)) - defer n.DecRef() + defer n.DecRef(t) fd, err := t.NewFDFrom(0, n, kernel.FDFlags{ CloseOnExec: flags&linux.IN_CLOEXEC != 0, @@ -71,7 +71,7 @@ func fdToInotify(t *kernel.Task, fd int32) (*fs.Inotify, *fs.File, error) { ino, ok := file.FileOperations.(*fs.Inotify) if !ok { // Not an inotify fd. - file.DecRef() + file.DecRef(t) return nil, nil, syserror.EINVAL } @@ -98,7 +98,7 @@ func InotifyAddWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kern if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) path, _, err := copyInPath(t, addr, false /* allowEmpty */) if err != nil { @@ -128,6 +128,6 @@ func InotifyRmWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if err != nil { return 0, nil, err } - defer file.DecRef() - return 0, nil, ino.RmWatch(wd) + defer file.DecRef(t) + return 0, nil, ino.RmWatch(t, wd) } diff --git a/pkg/sentry/syscalls/linux/sys_lseek.go b/pkg/sentry/syscalls/linux/sys_lseek.go index 3f7691eae..1c38f8f4f 100644 --- a/pkg/sentry/syscalls/linux/sys_lseek.go +++ b/pkg/sentry/syscalls/linux/sys_lseek.go @@ -33,7 +33,7 @@ func Lseek(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) var sw fs.SeekWhence switch whence { diff --git a/pkg/sentry/syscalls/linux/sys_mmap.go b/pkg/sentry/syscalls/linux/sys_mmap.go index 91694d374..72786b032 100644 --- a/pkg/sentry/syscalls/linux/sys_mmap.go +++ b/pkg/sentry/syscalls/linux/sys_mmap.go @@ -75,7 +75,7 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC } defer func() { if opts.MappingIdentity != nil { - opts.MappingIdentity.DecRef() + opts.MappingIdentity.DecRef(t) } }() @@ -85,7 +85,7 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) flags := file.Flags() // mmap unconditionally requires that the FD is readable. diff --git a/pkg/sentry/syscalls/linux/sys_mount.go b/pkg/sentry/syscalls/linux/sys_mount.go index eb5ff48f5..bd0633564 100644 --- a/pkg/sentry/syscalls/linux/sys_mount.go +++ b/pkg/sentry/syscalls/linux/sys_mount.go @@ -115,7 +115,7 @@ func Mount(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall }); err != nil { // Something went wrong. Drop our ref on rootInode before // returning the error. - rootInode.DecRef() + rootInode.DecRef(t) return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_pipe.go b/pkg/sentry/syscalls/linux/sys_pipe.go index 43c510930..3149e4aad 100644 --- a/pkg/sentry/syscalls/linux/sys_pipe.go +++ b/pkg/sentry/syscalls/linux/sys_pipe.go @@ -34,10 +34,10 @@ func pipe2(t *kernel.Task, addr usermem.Addr, flags uint) (uintptr, error) { r, w := pipe.NewConnectedPipe(t, pipe.DefaultPipeSize, usermem.PageSize) r.SetFlags(linuxToFlags(flags).Settable()) - defer r.DecRef() + defer r.DecRef(t) w.SetFlags(linuxToFlags(flags).Settable()) - defer w.DecRef() + defer w.DecRef(t) fds, err := t.NewFDs(0, []*fs.File{r, w}, kernel.FDFlags{ CloseOnExec: flags&linux.O_CLOEXEC != 0, @@ -49,7 +49,7 @@ func pipe2(t *kernel.Task, addr usermem.Addr, flags uint) (uintptr, error) { if _, err := t.CopyOut(addr, fds); err != nil { for _, fd := range fds { if file, _ := t.FDTable().Remove(fd); file != nil { - file.DecRef() + file.DecRef(t) } } return 0, err diff --git a/pkg/sentry/syscalls/linux/sys_poll.go b/pkg/sentry/syscalls/linux/sys_poll.go index f0198141c..3435bdf77 100644 --- a/pkg/sentry/syscalls/linux/sys_poll.go +++ b/pkg/sentry/syscalls/linux/sys_poll.go @@ -70,7 +70,7 @@ func initReadiness(t *kernel.Task, pfd *linux.PollFD, state *pollState, ch chan } if ch == nil { - defer file.DecRef() + defer file.DecRef(t) } else { state.file = file state.waiter, _ = waiter.NewChannelEntry(ch) @@ -82,11 +82,11 @@ func initReadiness(t *kernel.Task, pfd *linux.PollFD, state *pollState, ch chan } // releaseState releases all the pollState in "state". -func releaseState(state []pollState) { +func releaseState(t *kernel.Task, state []pollState) { for i := range state { if state[i].file != nil { state[i].file.EventUnregister(&state[i].waiter) - state[i].file.DecRef() + state[i].file.DecRef(t) } } } @@ -107,7 +107,7 @@ func pollBlock(t *kernel.Task, pfd []linux.PollFD, timeout time.Duration) (time. // result, we stop registering for events but still go through all files // to get their ready masks. state := make([]pollState, len(pfd)) - defer releaseState(state) + defer releaseState(t, state) n := uintptr(0) for i := range pfd { initReadiness(t, &pfd[i], &state[i], ch) @@ -266,7 +266,7 @@ func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Add if file == nil { return 0, syserror.EBADF } - file.DecRef() + file.DecRef(t) var mask int16 if (rV & m) != 0 { diff --git a/pkg/sentry/syscalls/linux/sys_prctl.go b/pkg/sentry/syscalls/linux/sys_prctl.go index f92bf8096..64a725296 100644 --- a/pkg/sentry/syscalls/linux/sys_prctl.go +++ b/pkg/sentry/syscalls/linux/sys_prctl.go @@ -128,7 +128,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // They trying to set exe to a non-file? if !fs.IsFile(file.Dirent.Inode.StableAttr) { @@ -136,7 +136,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } // Set the underlying executable. - t.MemoryManager().SetExecutable(fsbridge.NewFSFile(file)) + t.MemoryManager().SetExecutable(t, fsbridge.NewFSFile(file)) case linux.PR_SET_MM_AUXV, linux.PR_SET_MM_START_CODE, diff --git a/pkg/sentry/syscalls/linux/sys_read.go b/pkg/sentry/syscalls/linux/sys_read.go index 071b4bacc..3bbc3fa4b 100644 --- a/pkg/sentry/syscalls/linux/sys_read.go +++ b/pkg/sentry/syscalls/linux/sys_read.go @@ -48,7 +48,7 @@ func Read(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is readable. if !file.Flags().Read { @@ -84,7 +84,7 @@ func Readahead(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is readable. if !file.Flags().Read { @@ -118,7 +118,7 @@ func Pread64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate and does not overflow. if offset < 0 || offset+int64(size) < 0 { @@ -164,7 +164,7 @@ func Readv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is readable. if !file.Flags().Read { @@ -195,7 +195,7 @@ func Preadv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < 0 { @@ -244,7 +244,7 @@ func Preadv2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < -1 { diff --git a/pkg/sentry/syscalls/linux/sys_shm.go b/pkg/sentry/syscalls/linux/sys_shm.go index 4a8bc24a2..f0ae8fa8e 100644 --- a/pkg/sentry/syscalls/linux/sys_shm.go +++ b/pkg/sentry/syscalls/linux/sys_shm.go @@ -39,7 +39,7 @@ func Shmget(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - defer segment.DecRef() + defer segment.DecRef(t) return uintptr(segment.ID), nil, nil } @@ -66,7 +66,7 @@ func Shmat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if err != nil { return 0, nil, syserror.EINVAL } - defer segment.DecRef() + defer segment.DecRef(t) opts, err := segment.ConfigureAttach(t, addr, shm.AttachOpts{ Execute: flag&linux.SHM_EXEC == linux.SHM_EXEC, @@ -108,7 +108,7 @@ func Shmctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, syserror.EINVAL } - defer segment.DecRef() + defer segment.DecRef(t) stat, err := segment.IPCStat(t) if err == nil { @@ -132,7 +132,7 @@ func Shmctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, syserror.EINVAL } - defer segment.DecRef() + defer segment.DecRef(t) switch cmd { case linux.IPC_SET: @@ -145,7 +145,7 @@ func Shmctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return 0, nil, err case linux.IPC_RMID: - segment.MarkDestroyed() + segment.MarkDestroyed(t) return 0, nil, nil case linux.SHM_LOCK, linux.SHM_UNLOCK: diff --git a/pkg/sentry/syscalls/linux/sys_signal.go b/pkg/sentry/syscalls/linux/sys_signal.go index d2b0012ae..20cb1a5cb 100644 --- a/pkg/sentry/syscalls/linux/sys_signal.go +++ b/pkg/sentry/syscalls/linux/sys_signal.go @@ -536,7 +536,7 @@ func sharedSignalfd(t *kernel.Task, fd int32, sigset usermem.Addr, sigsetsize ui if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Is this a signalfd? if s, ok := file.FileOperations.(*signalfd.SignalOperations); ok { @@ -553,7 +553,7 @@ func sharedSignalfd(t *kernel.Task, fd int32, sigset usermem.Addr, sigsetsize ui if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) // Set appropriate flags. file.SetFlags(fs.SettableFileFlags{ diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index 414fce8e3..fec1c1974 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -200,7 +200,7 @@ func Socket(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal s.SetFlags(fs.SettableFileFlags{ NonBlocking: stype&linux.SOCK_NONBLOCK != 0, }) - defer s.DecRef() + defer s.DecRef(t) fd, err := t.NewFDFrom(0, s, kernel.FDFlags{ CloseOnExec: stype&linux.SOCK_CLOEXEC != 0, @@ -235,8 +235,8 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } s1.SetFlags(fileFlags) s2.SetFlags(fileFlags) - defer s1.DecRef() - defer s2.DecRef() + defer s1.DecRef(t) + defer s2.DecRef(t) // Create the FDs for the sockets. fds, err := t.NewFDs(0, []*fs.File{s1, s2}, kernel.FDFlags{ @@ -250,7 +250,7 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if _, err := t.CopyOut(socks, fds); err != nil { for _, fd := range fds { if file, _ := t.FDTable().Remove(fd); file != nil { - file.DecRef() + file.DecRef(t) } } return 0, nil, err @@ -270,7 +270,7 @@ func Connect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -301,7 +301,7 @@ func accept(t *kernel.Task, fd int32, addr usermem.Addr, addrLen usermem.Addr, f if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -360,7 +360,7 @@ func Bind(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -387,7 +387,7 @@ func Listen(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -416,7 +416,7 @@ func Shutdown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -447,7 +447,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -529,7 +529,7 @@ func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -567,7 +567,7 @@ func GetSockName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -595,7 +595,7 @@ func GetPeerName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -628,7 +628,7 @@ func RecvMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -681,7 +681,7 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -775,7 +775,7 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i } if !cms.Unix.Empty() { mflags |= linux.MSG_CTRUNC - cms.Release() + cms.Release(t) } if int(msg.Flags) != mflags { @@ -795,7 +795,7 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i if e != nil { return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) } - defer cms.Release() + defer cms.Release(t) controlData := make([]byte, 0, msg.ControlLen) controlData = control.PackControlMessages(t, cms, controlData) @@ -851,7 +851,7 @@ func recvFrom(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flag if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -880,7 +880,7 @@ func recvFrom(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flag } n, _, sender, senderLen, cm, e := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, nameLenPtr != 0, 0) - cm.Release() + cm.Release(t) if e != nil { return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) } @@ -924,7 +924,7 @@ func SendMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -962,7 +962,7 @@ func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) @@ -1066,7 +1066,7 @@ func sendSingleMsg(t *kernel.Task, s socket.Socket, file *fs.File, msgPtr userme n, e := s.SendMsg(t, src, to, int(flags), haveDeadline, deadline, controlMessages) err = handleIOError(t, n != 0, e.ToError(), kernel.ERESTARTSYS, "sendmsg", file) if err != nil { - controlMessages.Release() + controlMessages.Release(t) } return uintptr(n), err } @@ -1084,7 +1084,7 @@ func sendTo(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flags if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.FileOperations.(socket.Socket) diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go index 77c78889d..b8846a10a 100644 --- a/pkg/sentry/syscalls/linux/sys_splice.go +++ b/pkg/sentry/syscalls/linux/sys_splice.go @@ -101,7 +101,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) if !inFile.Flags().Read { return 0, nil, syserror.EBADF @@ -111,7 +111,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) if !outFile.Flags().Write { return 0, nil, syserror.EBADF @@ -192,13 +192,13 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) inFile := t.GetFile(inFD) if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) // The operation is non-blocking if anything is non-blocking. // @@ -300,13 +300,13 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) inFile := t.GetFile(inFD) if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) // All files must be pipes. if !fs.IsPipe(inFile.Dirent.Inode.StableAttr) || !fs.IsPipe(outFile.Dirent.Inode.StableAttr) { diff --git a/pkg/sentry/syscalls/linux/sys_stat.go b/pkg/sentry/syscalls/linux/sys_stat.go index 46ebf27a2..a5826f2dd 100644 --- a/pkg/sentry/syscalls/linux/sys_stat.go +++ b/pkg/sentry/syscalls/linux/sys_stat.go @@ -58,7 +58,7 @@ func Fstatat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, fstat(t, file, statAddr) } @@ -100,7 +100,7 @@ func Fstat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, fstat(t, file, statAddr) } @@ -158,7 +158,7 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) uattr, err := file.UnstableAttr(t) if err != nil { return 0, nil, err @@ -249,7 +249,7 @@ func Fstatfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, statfsImpl(t, file.Dirent, statfsAddr) } diff --git a/pkg/sentry/syscalls/linux/sys_sync.go b/pkg/sentry/syscalls/linux/sys_sync.go index 5ad465ae3..f2c0e5069 100644 --- a/pkg/sentry/syscalls/linux/sys_sync.go +++ b/pkg/sentry/syscalls/linux/sys_sync.go @@ -39,7 +39,7 @@ func Syncfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Use "sync-the-world" for now, it's guaranteed that fd is at least // on the root filesystem. @@ -54,7 +54,7 @@ func Fsync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) err := file.Fsync(t, 0, fs.FileMaxOffset, fs.SyncAll) return 0, nil, syserror.ConvertIntr(err, kernel.ERESTARTSYS) @@ -70,7 +70,7 @@ func Fdatasync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) err := file.Fsync(t, 0, fs.FileMaxOffset, fs.SyncData) return 0, nil, syserror.ConvertIntr(err, kernel.ERESTARTSYS) @@ -103,7 +103,7 @@ func SyncFileRange(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // SYNC_FILE_RANGE_WAIT_BEFORE waits upon write-out of all pages in the // specified range that have already been submitted to the device diff --git a/pkg/sentry/syscalls/linux/sys_thread.go b/pkg/sentry/syscalls/linux/sys_thread.go index 00915fdde..2d16e4933 100644 --- a/pkg/sentry/syscalls/linux/sys_thread.go +++ b/pkg/sentry/syscalls/linux/sys_thread.go @@ -117,7 +117,7 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user resolveFinal := flags&linux.AT_SYMLINK_NOFOLLOW == 0 root := t.FSContext().RootDirectory() - defer root.DecRef() + defer root.DecRef(t) var wd *fs.Dirent var executable fsbridge.File @@ -133,7 +133,7 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) closeOnExec = fdFlags.CloseOnExec if atEmptyPath && len(pathname) == 0 { @@ -155,7 +155,7 @@ func execveat(t *kernel.Task, dirFD int32, pathnameAddr, argvAddr, envvAddr user } } if wd != nil { - defer wd.DecRef() + defer wd.DecRef(t) } // Load the new TaskContext. diff --git a/pkg/sentry/syscalls/linux/sys_timerfd.go b/pkg/sentry/syscalls/linux/sys_timerfd.go index cf49b43db..34b03e4ee 100644 --- a/pkg/sentry/syscalls/linux/sys_timerfd.go +++ b/pkg/sentry/syscalls/linux/sys_timerfd.go @@ -43,7 +43,7 @@ func TimerfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel return 0, nil, syserror.EINVAL } f := timerfd.NewFile(t, c) - defer f.DecRef() + defer f.DecRef(t) f.SetFlags(fs.SettableFileFlags{ NonBlocking: flags&linux.TFD_NONBLOCK != 0, }) @@ -73,7 +73,7 @@ func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) tf, ok := f.FileOperations.(*timerfd.TimerOperations) if !ok { @@ -107,7 +107,7 @@ func TimerfdGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) tf, ok := f.FileOperations.(*timerfd.TimerOperations) if !ok { diff --git a/pkg/sentry/syscalls/linux/sys_write.go b/pkg/sentry/syscalls/linux/sys_write.go index 6ec0de96e..485526e28 100644 --- a/pkg/sentry/syscalls/linux/sys_write.go +++ b/pkg/sentry/syscalls/linux/sys_write.go @@ -48,7 +48,7 @@ func Write(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is writable. if !file.Flags().Write { @@ -85,7 +85,7 @@ func Pwrite64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate and does not overflow. if offset < 0 || offset+int64(size) < 0 { @@ -131,7 +131,7 @@ func Writev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is writable. if !file.Flags().Write { @@ -162,7 +162,7 @@ func Pwritev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < 0 { @@ -215,7 +215,7 @@ func Pwritev2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < -1 { diff --git a/pkg/sentry/syscalls/linux/sys_xattr.go b/pkg/sentry/syscalls/linux/sys_xattr.go index c24946160..97474fd3c 100644 --- a/pkg/sentry/syscalls/linux/sys_xattr.go +++ b/pkg/sentry/syscalls/linux/sys_xattr.go @@ -49,7 +49,7 @@ func FGetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) n, err := getXattr(t, f.Dirent, nameAddr, valueAddr, size) if err != nil { @@ -153,7 +153,7 @@ func FSetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) return 0, nil, setXattr(t, f.Dirent, nameAddr, valueAddr, uint64(size), flags) } @@ -270,7 +270,7 @@ func FListXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) n, err := listXattr(t, f.Dirent, listAddr, size) if err != nil { @@ -384,7 +384,7 @@ func FRemoveXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. if f == nil { return 0, nil, syserror.EBADF } - defer f.DecRef() + defer f.DecRef(t) return 0, nil, removeXattr(t, f.Dirent, nameAddr) } diff --git a/pkg/sentry/syscalls/linux/vfs2/aio.go b/pkg/sentry/syscalls/linux/vfs2/aio.go index e5cdefc50..399b4f60c 100644 --- a/pkg/sentry/syscalls/linux/vfs2/aio.go +++ b/pkg/sentry/syscalls/linux/vfs2/aio.go @@ -88,7 +88,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr user if fd == nil { return syserror.EBADF } - defer fd.DecRef() + defer fd.DecRef(t) // Was there an eventFD? Extract it. var eventFD *vfs.FileDescription @@ -97,7 +97,7 @@ func submitCallback(t *kernel.Task, id uint64, cb *linux.IOCallback, cbAddr user if eventFD == nil { return syserror.EBADF } - defer eventFD.DecRef() + defer eventFD.DecRef(t) // Check that it is an eventfd. if _, ok := eventFD.Impl().(*eventfd.EventFileDescription); !ok { @@ -169,7 +169,7 @@ func getAIOCallback(t *kernel.Task, fd, eventFD *vfs.FileDescription, cbAddr use ev.Result = -int64(kernel.ExtractErrno(err, 0)) } - fd.DecRef() + fd.DecRef(ctx) // Queue the result for delivery. aioCtx.FinishRequest(ev) @@ -179,7 +179,7 @@ func getAIOCallback(t *kernel.Task, fd, eventFD *vfs.FileDescription, cbAddr use // wake up. if eventFD != nil { eventFD.Impl().(*eventfd.EventFileDescription).Signal(1) - eventFD.DecRef() + eventFD.DecRef(ctx) } } } diff --git a/pkg/sentry/syscalls/linux/vfs2/epoll.go b/pkg/sentry/syscalls/linux/vfs2/epoll.go index 34c90ae3e..c62f03509 100644 --- a/pkg/sentry/syscalls/linux/vfs2/epoll.go +++ b/pkg/sentry/syscalls/linux/vfs2/epoll.go @@ -37,11 +37,11 @@ func EpollCreate1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. return 0, nil, syserror.EINVAL } - file, err := t.Kernel().VFS().NewEpollInstanceFD() + file, err := t.Kernel().VFS().NewEpollInstanceFD(t) if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{ CloseOnExec: flags&linux.EPOLL_CLOEXEC != 0, @@ -62,11 +62,11 @@ func EpollCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S return 0, nil, syserror.EINVAL } - file, err := t.Kernel().VFS().NewEpollInstanceFD() + file, err := t.Kernel().VFS().NewEpollInstanceFD(t) if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{}) if err != nil { @@ -86,7 +86,7 @@ func EpollCtl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if epfile == nil { return 0, nil, syserror.EBADF } - defer epfile.DecRef() + defer epfile.DecRef(t) ep, ok := epfile.Impl().(*vfs.EpollInstance) if !ok { return 0, nil, syserror.EINVAL @@ -95,7 +95,7 @@ func EpollCtl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) if epfile == file { return 0, nil, syserror.EINVAL } @@ -135,7 +135,7 @@ func EpollWait(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if epfile == nil { return 0, nil, syserror.EBADF } - defer epfile.DecRef() + defer epfile.DecRef(t) ep, ok := epfile.Impl().(*vfs.EpollInstance) if !ok { return 0, nil, syserror.EINVAL diff --git a/pkg/sentry/syscalls/linux/vfs2/eventfd.go b/pkg/sentry/syscalls/linux/vfs2/eventfd.go index aff1a2070..807f909da 100644 --- a/pkg/sentry/syscalls/linux/vfs2/eventfd.go +++ b/pkg/sentry/syscalls/linux/vfs2/eventfd.go @@ -38,11 +38,11 @@ func Eventfd2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc fileFlags |= linux.O_NONBLOCK } semMode := flags&linux.EFD_SEMAPHORE != 0 - eventfd, err := eventfd.New(vfsObj, initVal, semMode, fileFlags) + eventfd, err := eventfd.New(t, vfsObj, initVal, semMode, fileFlags) if err != nil { return 0, nil, err } - defer eventfd.DecRef() + defer eventfd.DecRef(t) fd, err := t.NewFDFromVFS2(0, eventfd, kernel.FDFlags{ CloseOnExec: flags&linux.EFD_CLOEXEC != 0, diff --git a/pkg/sentry/syscalls/linux/vfs2/execve.go b/pkg/sentry/syscalls/linux/vfs2/execve.go index aef0078a8..066ee0863 100644 --- a/pkg/sentry/syscalls/linux/vfs2/execve.go +++ b/pkg/sentry/syscalls/linux/vfs2/execve.go @@ -71,7 +71,7 @@ func execveat(t *kernel.Task, dirfd int32, pathnameAddr, argvAddr, envvAddr user } root := t.FSContext().RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) var executable fsbridge.File closeOnExec := false if path := fspath.Parse(pathname); dirfd != linux.AT_FDCWD && !path.Absolute { @@ -90,7 +90,7 @@ func execveat(t *kernel.Task, dirfd int32, pathnameAddr, argvAddr, envvAddr user } start := dirfile.VirtualDentry() start.IncRef() - dirfile.DecRef() + dirfile.DecRef(t) closeOnExec = dirfileFlags.CloseOnExec file, err := t.Kernel().VFS().OpenAt(t, t.Credentials(), &vfs.PathOperation{ Root: root, @@ -101,19 +101,19 @@ func execveat(t *kernel.Task, dirfd int32, pathnameAddr, argvAddr, envvAddr user Flags: linux.O_RDONLY, FileExec: true, }) - start.DecRef() + start.DecRef(t) if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) executable = fsbridge.NewVFSFile(file) } // Load the new TaskContext. mntns := t.MountNamespaceVFS2() // FIXME(jamieliu): useless refcount change - defer mntns.DecRef() + defer mntns.DecRef(t) wd := t.FSContext().WorkingDirectoryVFS2() - defer wd.DecRef() + defer wd.DecRef(t) remainingTraversals := uint(linux.MaxSymlinkTraversals) loadArgs := loader.LoadArgs{ Opener: fsbridge.NewVFSLookup(mntns, root, wd), diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go index 67f191551..4856554fe 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fd.go +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go @@ -38,7 +38,7 @@ func Close(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) err := file.OnClose(t) return 0, nil, slinux.HandleIOErrorVFS2(t, false /* partial */, err, syserror.EINTR, "close", file) @@ -52,7 +52,7 @@ func Dup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) newFD, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{}) if err != nil { @@ -72,7 +72,7 @@ func Dup2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - file.DecRef() + file.DecRef(t) return uintptr(newfd), nil, nil } @@ -101,7 +101,7 @@ func dup3(t *kernel.Task, oldfd, newfd int32, flags uint32) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) err := t.NewFDAtVFS2(newfd, file, kernel.FDFlags{ CloseOnExec: flags&linux.O_CLOEXEC != 0, @@ -121,7 +121,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) switch cmd { case linux.F_DUPFD, linux.F_DUPFD_CLOEXEC: @@ -208,7 +208,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall 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. + // Everything else is not yet supported. return 0, nil, syserror.EINVAL } } @@ -332,7 +332,7 @@ func Fadvise64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // If the FD refers to a pipe or FIFO, return error. if _, isPipe := file.Impl().(*pipe.VFSPipeFD); isPipe { diff --git a/pkg/sentry/syscalls/linux/vfs2/filesystem.go b/pkg/sentry/syscalls/linux/vfs2/filesystem.go index 6b14c2bef..01e0f9010 100644 --- a/pkg/sentry/syscalls/linux/vfs2/filesystem.go +++ b/pkg/sentry/syscalls/linux/vfs2/filesystem.go @@ -18,7 +18,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" @@ -57,7 +56,7 @@ func linkat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd i if err != nil { return err } - defer oldtpop.Release() + defer oldtpop.Release(t) newpath, err := copyInPath(t, newpathAddr) if err != nil { @@ -67,7 +66,7 @@ func linkat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd i if err != nil { return err } - defer newtpop.Release() + defer newtpop.Release(t) return t.Kernel().VFS().LinkAt(t, t.Credentials(), &oldtpop.pop, &newtpop.pop) } @@ -96,7 +95,7 @@ func mkdirat(t *kernel.Task, dirfd int32, addr usermem.Addr, mode uint) error { if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) return t.Kernel().VFS().MkdirAt(t, t.Credentials(), &tpop.pop, &vfs.MkdirOptions{ Mode: linux.FileMode(mode & (0777 | linux.S_ISVTX) &^ t.FSContext().Umask()), }) @@ -128,7 +127,7 @@ func mknodat(t *kernel.Task, dirfd int32, addr usermem.Addr, mode linux.FileMode if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) // "Zero file type is equivalent to type S_IFREG." - mknod(2) if mode.FileType() == 0 { @@ -175,7 +174,7 @@ func openat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr, flags uint32, mo if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) file, err := t.Kernel().VFS().OpenAt(t, t.Credentials(), &tpop.pop, &vfs.OpenOptions{ Flags: flags | linux.O_LARGEFILE, @@ -184,7 +183,7 @@ func openat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr, flags uint32, mo if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{ CloseOnExec: flags&linux.O_CLOEXEC != 0, @@ -228,7 +227,7 @@ func renameat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd if err != nil { return err } - defer oldtpop.Release() + defer oldtpop.Release(t) newpath, err := copyInPath(t, newpathAddr) if err != nil { @@ -238,62 +237,13 @@ func renameat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd if err != nil { return err } - defer newtpop.Release() + defer newtpop.Release(t) return t.Kernel().VFS().RenameAt(t, t.Credentials(), &oldtpop.pop, &newtpop.pop, &vfs.RenameOptions{ Flags: flags, }) } -// 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 - } - - return 0, nil, file.Impl().Allocate(t, mode, uint64(offset), uint64(length)) - - // File length modified, generate notification. - // TODO(gvisor.dev/issue/1479): Reenable when Inotify is ported. - // file.Dirent.InotifyEvent(linux.IN_MODIFY, 0) -} - // Rmdir implements Linux syscall rmdir(2). func Rmdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { pathAddr := args[0].Pointer() @@ -309,7 +259,7 @@ func rmdirat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr) error { if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) return t.Kernel().VFS().RmdirAt(t, t.Credentials(), &tpop.pop) } @@ -328,7 +278,7 @@ func unlinkat(t *kernel.Task, dirfd int32, pathAddr usermem.Addr) error { if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) return t.Kernel().VFS().UnlinkAt(t, t.Credentials(), &tpop.pop) } @@ -379,6 +329,6 @@ func symlinkat(t *kernel.Task, targetAddr usermem.Addr, newdirfd int32, linkpath if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) return t.Kernel().VFS().SymlinkAt(t, t.Credentials(), &tpop.pop, target) } diff --git a/pkg/sentry/syscalls/linux/vfs2/fscontext.go b/pkg/sentry/syscalls/linux/vfs2/fscontext.go index 317409a18..a7d4d2a36 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fscontext.go +++ b/pkg/sentry/syscalls/linux/vfs2/fscontext.go @@ -31,8 +31,8 @@ func Getcwd(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal root := t.FSContext().RootDirectoryVFS2() wd := t.FSContext().WorkingDirectoryVFS2() s, err := t.Kernel().VFS().PathnameForGetcwd(t, root, wd) - root.DecRef() - wd.DecRef() + root.DecRef(t) + wd.DecRef(t) if err != nil { return 0, nil, err } @@ -67,7 +67,7 @@ func Chdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) vd, err := t.Kernel().VFS().GetDentryAt(t, t.Credentials(), &tpop.pop, &vfs.GetDentryOptions{ CheckSearchable: true, @@ -75,8 +75,8 @@ func Chdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if err != nil { return 0, nil, err } - t.FSContext().SetWorkingDirectoryVFS2(vd) - vd.DecRef() + t.FSContext().SetWorkingDirectoryVFS2(t, vd) + vd.DecRef(t) return 0, nil, nil } @@ -88,7 +88,7 @@ func Fchdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) vd, err := t.Kernel().VFS().GetDentryAt(t, t.Credentials(), &tpop.pop, &vfs.GetDentryOptions{ CheckSearchable: true, @@ -96,8 +96,8 @@ func Fchdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - t.FSContext().SetWorkingDirectoryVFS2(vd) - vd.DecRef() + t.FSContext().SetWorkingDirectoryVFS2(t, vd) + vd.DecRef(t) return 0, nil, nil } @@ -117,7 +117,7 @@ func Chroot(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) vd, err := t.Kernel().VFS().GetDentryAt(t, t.Credentials(), &tpop.pop, &vfs.GetDentryOptions{ CheckSearchable: true, @@ -125,7 +125,7 @@ func Chroot(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - t.FSContext().SetRootDirectoryVFS2(vd) - vd.DecRef() + t.FSContext().SetRootDirectoryVFS2(t, vd) + vd.DecRef(t) return 0, nil, nil } diff --git a/pkg/sentry/syscalls/linux/vfs2/getdents.go b/pkg/sentry/syscalls/linux/vfs2/getdents.go index c7c7bf7ce..5517595b5 100644 --- a/pkg/sentry/syscalls/linux/vfs2/getdents.go +++ b/pkg/sentry/syscalls/linux/vfs2/getdents.go @@ -44,7 +44,7 @@ func getdents(t *kernel.Task, args arch.SyscallArguments, isGetdents64 bool) (ui if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) cb := getGetdentsCallback(t, addr, size, isGetdents64) err := file.IterDirents(t, cb) diff --git a/pkg/sentry/syscalls/linux/vfs2/inotify.go b/pkg/sentry/syscalls/linux/vfs2/inotify.go index 5d98134a5..11753d8e5 100644 --- a/pkg/sentry/syscalls/linux/vfs2/inotify.go +++ b/pkg/sentry/syscalls/linux/vfs2/inotify.go @@ -35,7 +35,7 @@ func InotifyInit1(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. if err != nil { return 0, nil, err } - defer ino.DecRef() + defer ino.DecRef(t) fd, err := t.NewFDFromVFS2(0, ino, kernel.FDFlags{ CloseOnExec: flags&linux.IN_CLOEXEC != 0, @@ -66,7 +66,7 @@ func fdToInotify(t *kernel.Task, fd int32) (*vfs.Inotify, *vfs.FileDescription, ino, ok := f.Impl().(*vfs.Inotify) if !ok { // Not an inotify fd. - f.DecRef() + f.DecRef(t) return nil, nil, syserror.EINVAL } @@ -96,7 +96,7 @@ func InotifyAddWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kern if err != nil { return 0, nil, err } - defer f.DecRef() + defer f.DecRef(t) path, err := copyInPath(t, addr) if err != nil { @@ -109,12 +109,12 @@ func InotifyAddWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kern if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) d, err := t.Kernel().VFS().GetDentryAt(t, t.Credentials(), &tpop.pop, &vfs.GetDentryOptions{}) if err != nil { return 0, nil, err } - defer d.DecRef() + defer d.DecRef(t) fd, err = ino.AddWatch(d.Dentry(), mask) if err != nil { @@ -132,6 +132,6 @@ func InotifyRmWatch(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if err != nil { return 0, nil, err } - defer f.DecRef() - return 0, nil, ino.RmWatch(wd) + defer f.DecRef(t) + return 0, nil, ino.RmWatch(t, wd) } diff --git a/pkg/sentry/syscalls/linux/vfs2/ioctl.go b/pkg/sentry/syscalls/linux/vfs2/ioctl.go index fd6ab94b2..38778a388 100644 --- a/pkg/sentry/syscalls/linux/vfs2/ioctl.go +++ b/pkg/sentry/syscalls/linux/vfs2/ioctl.go @@ -29,7 +29,7 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Handle ioctls that apply to all FDs. switch args[1].Int() { diff --git a/pkg/sentry/syscalls/linux/vfs2/lock.go b/pkg/sentry/syscalls/linux/vfs2/lock.go index bf19028c4..b910b5a74 100644 --- a/pkg/sentry/syscalls/linux/vfs2/lock.go +++ b/pkg/sentry/syscalls/linux/vfs2/lock.go @@ -32,7 +32,7 @@ func Flock(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // flock(2): EBADF fd is not an open file descriptor. return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) nonblocking := operation&linux.LOCK_NB != 0 operation &^= linux.LOCK_NB diff --git a/pkg/sentry/syscalls/linux/vfs2/memfd.go b/pkg/sentry/syscalls/linux/vfs2/memfd.go index bbe248d17..519583e4e 100644 --- a/pkg/sentry/syscalls/linux/vfs2/memfd.go +++ b/pkg/sentry/syscalls/linux/vfs2/memfd.go @@ -47,7 +47,7 @@ func MemfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S } shmMount := t.Kernel().ShmMount() - file, err := tmpfs.NewMemfd(shmMount, t.Credentials(), allowSeals, memfdPrefix+name) + file, err := tmpfs.NewMemfd(t, t.Credentials(), shmMount, allowSeals, memfdPrefix+name) if err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/vfs2/mmap.go b/pkg/sentry/syscalls/linux/vfs2/mmap.go index 60a43f0a0..dc05c2994 100644 --- a/pkg/sentry/syscalls/linux/vfs2/mmap.go +++ b/pkg/sentry/syscalls/linux/vfs2/mmap.go @@ -61,7 +61,7 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC } defer func() { if opts.MappingIdentity != nil { - opts.MappingIdentity.DecRef() + opts.MappingIdentity.DecRef(t) } }() @@ -71,7 +71,7 @@ func Mmap(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // mmap unconditionally requires that the FD is readable. if !file.IsReadable() { diff --git a/pkg/sentry/syscalls/linux/vfs2/mount.go b/pkg/sentry/syscalls/linux/vfs2/mount.go index ea337de7c..4bd5c7ca2 100644 --- a/pkg/sentry/syscalls/linux/vfs2/mount.go +++ b/pkg/sentry/syscalls/linux/vfs2/mount.go @@ -108,7 +108,7 @@ func Mount(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if err != nil { return 0, nil, err } - defer target.Release() + defer target.Release(t) return 0, nil, t.Kernel().VFS().MountAt(t, creds, source, &target.pop, fsType, &opts) } @@ -140,7 +140,7 @@ func Umount2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) opts := vfs.UmountOptions{ Flags: uint32(flags), diff --git a/pkg/sentry/syscalls/linux/vfs2/path.go b/pkg/sentry/syscalls/linux/vfs2/path.go index 97da6c647..90a511d9a 100644 --- a/pkg/sentry/syscalls/linux/vfs2/path.go +++ b/pkg/sentry/syscalls/linux/vfs2/path.go @@ -42,7 +42,7 @@ func getTaskPathOperation(t *kernel.Task, dirfd int32, path fspath.Path, shouldA haveStartRef := false if !path.Absolute { if !path.HasComponents() && !bool(shouldAllowEmptyPath) { - root.DecRef() + root.DecRef(t) return taskPathOperation{}, syserror.ENOENT } if dirfd == linux.AT_FDCWD { @@ -51,13 +51,13 @@ func getTaskPathOperation(t *kernel.Task, dirfd int32, path fspath.Path, shouldA } else { dirfile := t.GetFileVFS2(dirfd) if dirfile == nil { - root.DecRef() + root.DecRef(t) return taskPathOperation{}, syserror.EBADF } start = dirfile.VirtualDentry() start.IncRef() haveStartRef = true - dirfile.DecRef() + dirfile.DecRef(t) } } return taskPathOperation{ @@ -71,10 +71,10 @@ func getTaskPathOperation(t *kernel.Task, dirfd int32, path fspath.Path, shouldA }, nil } -func (tpop *taskPathOperation) Release() { - tpop.pop.Root.DecRef() +func (tpop *taskPathOperation) Release(t *kernel.Task) { + tpop.pop.Root.DecRef(t) if tpop.haveStartRef { - tpop.pop.Start.DecRef() + tpop.pop.Start.DecRef(t) tpop.haveStartRef = false } } diff --git a/pkg/sentry/syscalls/linux/vfs2/pipe.go b/pkg/sentry/syscalls/linux/vfs2/pipe.go index 4a01e4209..9b4848d9e 100644 --- a/pkg/sentry/syscalls/linux/vfs2/pipe.go +++ b/pkg/sentry/syscalls/linux/vfs2/pipe.go @@ -42,8 +42,8 @@ func pipe2(t *kernel.Task, addr usermem.Addr, flags int32) error { return syserror.EINVAL } r, w := pipefs.NewConnectedPipeFDs(t, t.Kernel().PipeMount(), uint32(flags&linux.O_NONBLOCK)) - defer r.DecRef() - defer w.DecRef() + defer r.DecRef(t) + defer w.DecRef(t) fds, err := t.NewFDsVFS2(0, []*vfs.FileDescription{r, w}, kernel.FDFlags{ CloseOnExec: flags&linux.O_CLOEXEC != 0, @@ -54,7 +54,7 @@ func pipe2(t *kernel.Task, addr usermem.Addr, flags int32) error { if _, err := t.CopyOut(addr, fds); err != nil { for _, fd := range fds { if _, file := t.FDTable().Remove(fd); file != nil { - file.DecRef() + file.DecRef(t) } } return err diff --git a/pkg/sentry/syscalls/linux/vfs2/poll.go b/pkg/sentry/syscalls/linux/vfs2/poll.go index ff1b25d7b..7b9d5e18a 100644 --- a/pkg/sentry/syscalls/linux/vfs2/poll.go +++ b/pkg/sentry/syscalls/linux/vfs2/poll.go @@ -73,7 +73,7 @@ func initReadiness(t *kernel.Task, pfd *linux.PollFD, state *pollState, ch chan } if ch == nil { - defer file.DecRef() + defer file.DecRef(t) } else { state.file = file state.waiter, _ = waiter.NewChannelEntry(ch) @@ -85,11 +85,11 @@ func initReadiness(t *kernel.Task, pfd *linux.PollFD, state *pollState, ch chan } // releaseState releases all the pollState in "state". -func releaseState(state []pollState) { +func releaseState(t *kernel.Task, state []pollState) { for i := range state { if state[i].file != nil { state[i].file.EventUnregister(&state[i].waiter) - state[i].file.DecRef() + state[i].file.DecRef(t) } } } @@ -110,7 +110,7 @@ func pollBlock(t *kernel.Task, pfd []linux.PollFD, timeout time.Duration) (time. // result, we stop registering for events but still go through all files // to get their ready masks. state := make([]pollState, len(pfd)) - defer releaseState(state) + defer releaseState(t, state) n := uintptr(0) for i := range pfd { initReadiness(t, &pfd[i], &state[i], ch) @@ -269,7 +269,7 @@ func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Add if file == nil { return 0, syserror.EBADF } - file.DecRef() + file.DecRef(t) var mask int16 if (rV & m) != 0 { diff --git a/pkg/sentry/syscalls/linux/vfs2/read_write.go b/pkg/sentry/syscalls/linux/vfs2/read_write.go index cd25597a7..a905dae0a 100644 --- a/pkg/sentry/syscalls/linux/vfs2/read_write.go +++ b/pkg/sentry/syscalls/linux/vfs2/read_write.go @@ -44,7 +44,7 @@ func Read(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the size is legitimate. si := int(size) @@ -75,7 +75,7 @@ func Readv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Get the destination of the read. dst, err := t.IovecsIOSequence(addr, iovcnt, usermem.IOOpts{ @@ -94,7 +94,7 @@ func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opt n, err := file.Read(t, dst, opts) if err != syserror.ErrWouldBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -102,7 +102,7 @@ func read(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, opt allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -135,7 +135,7 @@ 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) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return total, err } @@ -151,7 +151,7 @@ func Pread64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate and does not overflow. if offset < 0 || offset+int64(size) < 0 { @@ -188,7 +188,7 @@ func Preadv(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < 0 { @@ -226,7 +226,7 @@ func Preadv2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < -1 { @@ -258,7 +258,7 @@ func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, of n, err := file.PRead(t, dst, offset, opts) if err != syserror.ErrWouldBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -266,7 +266,7 @@ func pread(t *kernel.Task, file *vfs.FileDescription, dst usermem.IOSequence, of allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -299,7 +299,7 @@ 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) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return total, err } @@ -314,7 +314,7 @@ func Write(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the size is legitimate. si := int(size) @@ -345,7 +345,7 @@ func Writev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Get the source of the write. src, err := t.IovecsIOSequence(addr, iovcnt, usermem.IOOpts{ @@ -364,7 +364,7 @@ func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, op n, err := file.Write(t, src, opts) if err != syserror.ErrWouldBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) } return n, err } @@ -372,7 +372,7 @@ func write(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, op allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) } return n, err } @@ -405,7 +405,7 @@ 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) + file.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) } return total, err } @@ -421,7 +421,7 @@ func Pwrite64(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate and does not overflow. if offset < 0 || offset+int64(size) < 0 { @@ -458,7 +458,7 @@ func Pwritev(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < 0 { @@ -496,7 +496,7 @@ func Pwritev2(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the offset is legitimate. if offset < -1 { @@ -528,7 +528,7 @@ func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, o n, err := file.PWrite(t, src, offset, opts) if err != syserror.ErrWouldBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) } return n, err } @@ -536,7 +536,7 @@ func pwrite(t *kernel.Task, file *vfs.FileDescription, src usermem.IOSequence, o allowBlock, deadline, hasDeadline := blockPolicy(t, file) if !allowBlock { if n > 0 { - file.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return n, err } @@ -569,7 +569,7 @@ 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) + file.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) } return total, err } @@ -601,7 +601,7 @@ func Lseek(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) newoff, err := file.Seek(t, offset, whence) return uintptr(newoff), nil, err @@ -617,7 +617,7 @@ func Readahead(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Check that the file is readable. if !file.IsReadable() { diff --git a/pkg/sentry/syscalls/linux/vfs2/setstat.go b/pkg/sentry/syscalls/linux/vfs2/setstat.go index 6daedd173..5e6eb13ba 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" @@ -65,7 +66,7 @@ func Fchmod(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, file.SetStat(t, vfs.SetStatOptions{ Stat: linux.Statx{ @@ -150,7 +151,7 @@ func Fchown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) var opts vfs.SetStatOptions if err := populateSetStatOptionsForChown(t, owner, group, &opts); err != nil { @@ -196,7 +197,7 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) if !file.IsWritable() { return 0, nil, syserror.EINVAL @@ -211,6 +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(t) + + 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(t, 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() @@ -387,7 +438,7 @@ func populateSetStatOptionsForUtimens(t *kernel.Task, timesAddr usermem.Addr, op func setstatat(t *kernel.Task, dirfd int32, path fspath.Path, shouldAllowEmptyPath shouldAllowEmptyPath, shouldFollowFinalSymlink shouldFollowFinalSymlink, opts *vfs.SetStatOptions) error { root := t.FSContext().RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) start := root if !path.Absolute { if !path.HasComponents() && !bool(shouldAllowEmptyPath) { @@ -395,7 +446,7 @@ func setstatat(t *kernel.Task, dirfd int32, path fspath.Path, shouldAllowEmptyPa } if dirfd == linux.AT_FDCWD { start = t.FSContext().WorkingDirectoryVFS2() - defer start.DecRef() + defer start.DecRef(t) } else { dirfile := t.GetFileVFS2(dirfd) if dirfile == nil { @@ -406,13 +457,13 @@ func setstatat(t *kernel.Task, dirfd int32, path fspath.Path, shouldAllowEmptyPa // VirtualFilesystem.SetStatAt(), since the former may be able // to use opened file state to expedite the SetStat. err := dirfile.SetStat(t, *opts) - dirfile.DecRef() + dirfile.DecRef(t) return err } start = dirfile.VirtualDentry() start.IncRef() - defer start.DecRef() - dirfile.DecRef() + defer start.DecRef(t) + dirfile.DecRef(t) } } return t.Kernel().VFS().SetStatAt(t, t.Credentials(), &vfs.PathOperation{ diff --git a/pkg/sentry/syscalls/linux/vfs2/signal.go b/pkg/sentry/syscalls/linux/vfs2/signal.go index 623992f6f..b89f34cdb 100644 --- a/pkg/sentry/syscalls/linux/vfs2/signal.go +++ b/pkg/sentry/syscalls/linux/vfs2/signal.go @@ -45,7 +45,7 @@ func sharedSignalfd(t *kernel.Task, fd int32, sigset usermem.Addr, sigsetsize ui if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Is this a signalfd? if sfd, ok := file.Impl().(*signalfd.SignalFileDescription); ok { @@ -68,7 +68,7 @@ func sharedSignalfd(t *kernel.Task, fd int32, sigset usermem.Addr, sigsetsize ui if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) // Create a new descriptor. fd, err = t.NewFDFromVFS2(0, file, kernel.FDFlags{ diff --git a/pkg/sentry/syscalls/linux/vfs2/socket.go b/pkg/sentry/syscalls/linux/vfs2/socket.go index 8096a8f9c..4a68c64f3 100644 --- a/pkg/sentry/syscalls/linux/vfs2/socket.go +++ b/pkg/sentry/syscalls/linux/vfs2/socket.go @@ -196,7 +196,7 @@ func Socket(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if e != nil { return 0, nil, e.ToError() } - defer s.DecRef() + defer s.DecRef(t) if err := s.SetStatusFlags(t, t.Credentials(), uint32(stype&linux.SOCK_NONBLOCK)); err != nil { return 0, nil, err @@ -230,8 +230,8 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy return 0, nil, e.ToError() } // Adding to the FD table will cause an extra reference to be acquired. - defer s1.DecRef() - defer s2.DecRef() + defer s1.DecRef(t) + defer s2.DecRef(t) nonblocking := uint32(stype & linux.SOCK_NONBLOCK) if err := s1.SetStatusFlags(t, t.Credentials(), nonblocking); err != nil { @@ -253,7 +253,7 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if _, err := t.CopyOut(addr, fds); err != nil { for _, fd := range fds { if _, file := t.FDTable().Remove(fd); file != nil { - file.DecRef() + file.DecRef(t) } } return 0, nil, err @@ -273,7 +273,7 @@ func Connect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -304,7 +304,7 @@ func accept(t *kernel.Task, fd int32, addr usermem.Addr, addrLen usermem.Addr, f if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -363,7 +363,7 @@ func Bind(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -390,7 +390,7 @@ func Listen(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -419,7 +419,7 @@ func Shutdown(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -450,7 +450,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -532,7 +532,7 @@ func SetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -570,7 +570,7 @@ func GetSockName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -598,7 +598,7 @@ func GetPeerName(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -631,7 +631,7 @@ func RecvMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -684,7 +684,7 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -778,7 +778,7 @@ func recvSingleMsg(t *kernel.Task, s socket.SocketVFS2, msgPtr usermem.Addr, fla } if !cms.Unix.Empty() { mflags |= linux.MSG_CTRUNC - cms.Release() + cms.Release(t) } if int(msg.Flags) != mflags { @@ -798,7 +798,7 @@ func recvSingleMsg(t *kernel.Task, s socket.SocketVFS2, msgPtr usermem.Addr, fla if e != nil { return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) } - defer cms.Release() + defer cms.Release(t) controlData := make([]byte, 0, msg.ControlLen) controlData = control.PackControlMessages(t, cms, controlData) @@ -854,7 +854,7 @@ func recvFrom(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flag if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -883,7 +883,7 @@ func recvFrom(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flag } n, _, sender, senderLen, cm, e := s.RecvMsg(t, dst, int(flags), haveDeadline, deadline, nameLenPtr != 0, 0) - cm.Release() + cm.Release(t) if e != nil { return 0, syserror.ConvertIntr(e.ToError(), kernel.ERESTARTSYS) } @@ -927,7 +927,7 @@ func SendMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -965,7 +965,7 @@ func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) @@ -1069,7 +1069,7 @@ func sendSingleMsg(t *kernel.Task, s socket.SocketVFS2, file *vfs.FileDescriptio n, e := s.SendMsg(t, src, to, int(flags), haveDeadline, deadline, controlMessages) err = slinux.HandleIOErrorVFS2(t, n != 0, e.ToError(), kernel.ERESTARTSYS, "sendmsg", file) if err != nil { - controlMessages.Release() + controlMessages.Release(t) } return uintptr(n), err } @@ -1087,7 +1087,7 @@ func sendTo(t *kernel.Task, fd int32, bufPtr usermem.Addr, bufLen uint64, flags if file == nil { return 0, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) // Extract the socket. s, ok := file.Impl().(socket.SocketVFS2) diff --git a/pkg/sentry/syscalls/linux/vfs2/splice.go b/pkg/sentry/syscalls/linux/vfs2/splice.go index 63ab11f8c..16f59fce9 100644 --- a/pkg/sentry/syscalls/linux/vfs2/splice.go +++ b/pkg/sentry/syscalls/linux/vfs2/splice.go @@ -53,12 +53,12 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) outFile := t.GetFileVFS2(outFD) if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) // Check that both files support the required directionality. if !inFile.IsReadable() || !outFile.IsWritable() { @@ -175,7 +175,7 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // 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) + outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) return uintptr(n), nil, nil } @@ -203,12 +203,12 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) outFile := t.GetFileVFS2(outFD) if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) // Check that both files support the required directionality. if !inFile.IsReadable() || !outFile.IsWritable() { @@ -251,7 +251,7 @@ func Tee(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallCo if n == 0 { return 0, nil, err } - outFile.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) return uintptr(n), nil, nil } @@ -266,7 +266,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if inFile == nil { return 0, nil, syserror.EBADF } - defer inFile.DecRef() + defer inFile.DecRef(t) if !inFile.IsReadable() { return 0, nil, syserror.EBADF } @@ -275,7 +275,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if outFile == nil { return 0, nil, syserror.EBADF } - defer outFile.DecRef() + defer outFile.DecRef(t) if !outFile.IsWritable() { return 0, nil, syserror.EBADF } @@ -419,8 +419,8 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc return 0, nil, err } - inFile.Dentry().InotifyWithParent(linux.IN_ACCESS, 0, vfs.PathEvent) - outFile.Dentry().InotifyWithParent(linux.IN_MODIFY, 0, vfs.PathEvent) + inFile.Dentry().InotifyWithParent(t, linux.IN_ACCESS, 0, vfs.PathEvent) + outFile.Dentry().InotifyWithParent(t, linux.IN_MODIFY, 0, vfs.PathEvent) return uintptr(n), nil, nil } diff --git a/pkg/sentry/syscalls/linux/vfs2/stat.go b/pkg/sentry/syscalls/linux/vfs2/stat.go index bb1d5cac4..0f5d5189c 100644 --- a/pkg/sentry/syscalls/linux/vfs2/stat.go +++ b/pkg/sentry/syscalls/linux/vfs2/stat.go @@ -65,7 +65,7 @@ func fstatat(t *kernel.Task, dirfd int32, pathAddr, statAddr usermem.Addr, flags } root := t.FSContext().RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) start := root if !path.Absolute { if !path.HasComponents() && flags&linux.AT_EMPTY_PATH == 0 { @@ -73,7 +73,7 @@ func fstatat(t *kernel.Task, dirfd int32, pathAddr, statAddr usermem.Addr, flags } if dirfd == linux.AT_FDCWD { start = t.FSContext().WorkingDirectoryVFS2() - defer start.DecRef() + defer start.DecRef(t) } else { dirfile := t.GetFileVFS2(dirfd) if dirfile == nil { @@ -85,7 +85,7 @@ func fstatat(t *kernel.Task, dirfd int32, pathAddr, statAddr usermem.Addr, flags // former may be able to use opened file state to expedite the // Stat. statx, err := dirfile.Stat(t, opts) - dirfile.DecRef() + dirfile.DecRef(t) if err != nil { return err } @@ -96,8 +96,8 @@ func fstatat(t *kernel.Task, dirfd int32, pathAddr, statAddr usermem.Addr, flags } start = dirfile.VirtualDentry() start.IncRef() - defer start.DecRef() - dirfile.DecRef() + defer start.DecRef(t) + dirfile.DecRef(t) } } @@ -132,7 +132,7 @@ func Fstat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) statx, err := file.Stat(t, vfs.StatOptions{ Mask: linux.STATX_BASIC_STATS, @@ -177,7 +177,7 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } root := t.FSContext().RootDirectoryVFS2() - defer root.DecRef() + defer root.DecRef(t) start := root if !path.Absolute { if !path.HasComponents() && flags&linux.AT_EMPTY_PATH == 0 { @@ -185,7 +185,7 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } if dirfd == linux.AT_FDCWD { start = t.FSContext().WorkingDirectoryVFS2() - defer start.DecRef() + defer start.DecRef(t) } else { dirfile := t.GetFileVFS2(dirfd) if dirfile == nil { @@ -197,7 +197,7 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // former may be able to use opened file state to expedite the // Stat. statx, err := dirfile.Stat(t, opts) - dirfile.DecRef() + dirfile.DecRef(t) if err != nil { return 0, nil, err } @@ -207,8 +207,8 @@ func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } start = dirfile.VirtualDentry() start.IncRef() - defer start.DecRef() - dirfile.DecRef() + defer start.DecRef(t) + dirfile.DecRef(t) } } @@ -282,7 +282,7 @@ func accessAt(t *kernel.Task, dirfd int32, pathAddr usermem.Addr, mode uint) err if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) // access(2) and faccessat(2) check permissions using real // UID/GID, not effective UID/GID. @@ -328,7 +328,7 @@ func readlinkat(t *kernel.Task, dirfd int32, pathAddr, bufAddr usermem.Addr, siz if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) target, err := t.Kernel().VFS().ReadlinkAt(t, t.Credentials(), &tpop.pop) if err != nil { @@ -358,7 +358,7 @@ func Statfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) statfs, err := t.Kernel().VFS().StatFSAt(t, t.Credentials(), &tpop.pop) if err != nil { @@ -377,7 +377,7 @@ func Fstatfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) statfs, err := t.Kernel().VFS().StatFSAt(t, t.Credentials(), &tpop.pop) if err != nil { diff --git a/pkg/sentry/syscalls/linux/vfs2/sync.go b/pkg/sentry/syscalls/linux/vfs2/sync.go index 0d0ebf46a..a6491ac37 100644 --- a/pkg/sentry/syscalls/linux/vfs2/sync.go +++ b/pkg/sentry/syscalls/linux/vfs2/sync.go @@ -34,7 +34,7 @@ func Syncfs(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, file.SyncFS(t) } @@ -47,7 +47,7 @@ func Fsync(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) return 0, nil, file.Sync(t) } @@ -77,7 +77,7 @@ func SyncFileRange(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(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 diff --git a/pkg/sentry/syscalls/linux/vfs2/timerfd.go b/pkg/sentry/syscalls/linux/vfs2/timerfd.go index 5ac79bc09..7a26890ef 100644 --- a/pkg/sentry/syscalls/linux/vfs2/timerfd.go +++ b/pkg/sentry/syscalls/linux/vfs2/timerfd.go @@ -50,11 +50,11 @@ func TimerfdCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel return 0, nil, syserror.EINVAL } vfsObj := t.Kernel().VFS() - file, err := timerfd.New(vfsObj, clock, fileFlags) + file, err := timerfd.New(t, vfsObj, clock, fileFlags) if err != nil { return 0, nil, err } - defer file.DecRef() + defer file.DecRef(t) fd, err := t.NewFDFromVFS2(0, file, kernel.FDFlags{ CloseOnExec: flags&linux.TFD_CLOEXEC != 0, }) @@ -79,7 +79,7 @@ func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) tfd, ok := file.Impl().(*timerfd.TimerFileDescription) if !ok { @@ -113,7 +113,7 @@ func TimerfdGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) tfd, ok := file.Impl().(*timerfd.TimerFileDescription) if !ok { diff --git a/pkg/sentry/syscalls/linux/vfs2/xattr.go b/pkg/sentry/syscalls/linux/vfs2/xattr.go index af455d5c1..ef99246ed 100644 --- a/pkg/sentry/syscalls/linux/vfs2/xattr.go +++ b/pkg/sentry/syscalls/linux/vfs2/xattr.go @@ -49,7 +49,7 @@ func listxattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSyml if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) names, err := t.Kernel().VFS().ListxattrAt(t, t.Credentials(), &tpop.pop, uint64(size)) if err != nil { @@ -72,7 +72,7 @@ func Flistxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) names, err := file.Listxattr(t, uint64(size)) if err != nil { @@ -109,7 +109,7 @@ func getxattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSymli if err != nil { return 0, nil, err } - defer tpop.Release() + defer tpop.Release(t) name, err := copyInXattrName(t, nameAddr) if err != nil { @@ -141,7 +141,7 @@ func Fgetxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) name, err := copyInXattrName(t, nameAddr) if err != nil { @@ -188,7 +188,7 @@ func setxattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSymli if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) name, err := copyInXattrName(t, nameAddr) if err != nil { @@ -222,7 +222,7 @@ func Fsetxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) name, err := copyInXattrName(t, nameAddr) if err != nil { @@ -262,7 +262,7 @@ func removexattr(t *kernel.Task, args arch.SyscallArguments, shouldFollowFinalSy if err != nil { return err } - defer tpop.Release() + defer tpop.Release(t) name, err := copyInXattrName(t, nameAddr) if err != nil { @@ -281,7 +281,7 @@ func Fremovexattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. if file == nil { return 0, nil, syserror.EBADF } - defer file.DecRef() + defer file.DecRef(t) name, err := copyInXattrName(t, nameAddr) if err != nil { diff --git a/pkg/sentry/vfs/anonfs.go b/pkg/sentry/vfs/anonfs.go index 641e3e502..5a0e3e6b5 100644 --- a/pkg/sentry/vfs/anonfs.go +++ b/pkg/sentry/vfs/anonfs.go @@ -82,7 +82,7 @@ type anonDentry struct { } // Release implements FilesystemImpl.Release. -func (fs *anonFilesystem) Release() { +func (fs *anonFilesystem) Release(ctx context.Context) { } // Sync implements FilesystemImpl.Sync. @@ -294,7 +294,7 @@ func (d *anonDentry) TryIncRef() bool { } // DecRef implements DentryImpl.DecRef. -func (d *anonDentry) DecRef() { +func (d *anonDentry) DecRef(ctx context.Context) { // no-op } @@ -303,7 +303,7 @@ func (d *anonDentry) DecRef() { // Although Linux technically supports inotify on pseudo filesystems (inotify // is implemented at the vfs layer), it is not particularly useful. It is left // unimplemented until someone actually needs it. -func (d *anonDentry) InotifyWithParent(events, cookie uint32, et EventType) {} +func (d *anonDentry) InotifyWithParent(ctx context.Context, events, cookie uint32, et EventType) {} // Watches implements DentryImpl.Watches. func (d *anonDentry) Watches() *Watches { @@ -311,4 +311,4 @@ func (d *anonDentry) Watches() *Watches { } // OnZeroWatches implements Dentry.OnZeroWatches. -func (d *anonDentry) OnZeroWatches() {} +func (d *anonDentry) OnZeroWatches(context.Context) {} diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go index cea3e6955..bc7ea93ea 100644 --- a/pkg/sentry/vfs/dentry.go +++ b/pkg/sentry/vfs/dentry.go @@ -17,6 +17,7 @@ package vfs import ( "sync/atomic" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" ) @@ -102,7 +103,7 @@ type DentryImpl interface { TryIncRef() bool // DecRef decrements the Dentry's reference count. - DecRef() + DecRef(ctx context.Context) // InotifyWithParent notifies all watches on the targets represented by this // dentry and its parent. The parent's watches are notified first, followed @@ -113,7 +114,7 @@ type DentryImpl interface { // // Note that the events may not actually propagate up to the user, depending // on the event masks. - InotifyWithParent(events, cookie uint32, et EventType) + InotifyWithParent(ctx context.Context, events, cookie uint32, et EventType) // Watches returns the set of inotify watches for the file corresponding to // the Dentry. Dentries that are hard links to the same underlying file @@ -135,7 +136,7 @@ type DentryImpl interface { // The caller does not need to hold a reference on the dentry. OnZeroWatches // may acquire inotify locks, so to prevent deadlock, no inotify locks should // be held by the caller. - OnZeroWatches() + OnZeroWatches(ctx context.Context) } // IncRef increments d's reference count. @@ -150,8 +151,8 @@ func (d *Dentry) TryIncRef() bool { } // DecRef decrements d's reference count. -func (d *Dentry) DecRef() { - d.impl.DecRef() +func (d *Dentry) DecRef(ctx context.Context) { + d.impl.DecRef(ctx) } // IsDead returns true if d has been deleted or invalidated by its owning @@ -168,8 +169,8 @@ func (d *Dentry) isMounted() bool { // InotifyWithParent notifies all watches on the targets represented by d and // its parent of events. -func (d *Dentry) InotifyWithParent(events, cookie uint32, et EventType) { - d.impl.InotifyWithParent(events, cookie, et) +func (d *Dentry) InotifyWithParent(ctx context.Context, events, cookie uint32, et EventType) { + d.impl.InotifyWithParent(ctx, events, cookie, et) } // Watches returns the set of inotify watches associated with d. @@ -182,8 +183,8 @@ func (d *Dentry) Watches() *Watches { // OnZeroWatches performs cleanup tasks whenever the number of watches on a // dentry drops to zero. -func (d *Dentry) OnZeroWatches() { - d.impl.OnZeroWatches() +func (d *Dentry) OnZeroWatches(ctx context.Context) { + d.impl.OnZeroWatches(ctx) } // The following functions are exported so that filesystem implementations can @@ -214,11 +215,11 @@ func (vfs *VirtualFilesystem) AbortDeleteDentry(d *Dentry) { // CommitDeleteDentry must be called after PrepareDeleteDentry if the deletion // succeeds. -func (vfs *VirtualFilesystem) CommitDeleteDentry(d *Dentry) { +func (vfs *VirtualFilesystem) CommitDeleteDentry(ctx context.Context, d *Dentry) { d.dead = true d.mu.Unlock() if d.isMounted() { - vfs.forgetDeadMountpoint(d) + vfs.forgetDeadMountpoint(ctx, d) } } @@ -226,12 +227,12 @@ func (vfs *VirtualFilesystem) CommitDeleteDentry(d *Dentry) { // did for reasons outside of VFS' control (e.g. d represents the local state // of a file on a remote filesystem on which the file has already been // deleted). -func (vfs *VirtualFilesystem) InvalidateDentry(d *Dentry) { +func (vfs *VirtualFilesystem) InvalidateDentry(ctx context.Context, d *Dentry) { d.mu.Lock() d.dead = true d.mu.Unlock() if d.isMounted() { - vfs.forgetDeadMountpoint(d) + vfs.forgetDeadMountpoint(ctx, d) } } @@ -278,13 +279,13 @@ func (vfs *VirtualFilesystem) AbortRenameDentry(from, to *Dentry) { // that was replaced by from. // // Preconditions: PrepareRenameDentry was previously called on from and to. -func (vfs *VirtualFilesystem) CommitRenameReplaceDentry(from, to *Dentry) { +func (vfs *VirtualFilesystem) CommitRenameReplaceDentry(ctx context.Context, from, to *Dentry) { from.mu.Unlock() if to != nil { to.dead = true to.mu.Unlock() if to.isMounted() { - vfs.forgetDeadMountpoint(to) + vfs.forgetDeadMountpoint(ctx, to) } } } @@ -303,7 +304,7 @@ func (vfs *VirtualFilesystem) CommitRenameExchangeDentry(from, to *Dentry) { // // forgetDeadMountpoint is analogous to Linux's // fs/namespace.c:__detach_mounts(). -func (vfs *VirtualFilesystem) forgetDeadMountpoint(d *Dentry) { +func (vfs *VirtualFilesystem) forgetDeadMountpoint(ctx context.Context, d *Dentry) { var ( vdsToDecRef []VirtualDentry mountsToDecRef []*Mount @@ -316,9 +317,9 @@ func (vfs *VirtualFilesystem) forgetDeadMountpoint(d *Dentry) { vfs.mounts.seq.EndWrite() vfs.mountMu.Unlock() for _, vd := range vdsToDecRef { - vd.DecRef() + vd.DecRef(ctx) } for _, mnt := range mountsToDecRef { - mnt.DecRef() + mnt.DecRef(ctx) } } diff --git a/pkg/sentry/vfs/epoll.go b/pkg/sentry/vfs/epoll.go index 5b009b928..1b5af9f73 100644 --- a/pkg/sentry/vfs/epoll.go +++ b/pkg/sentry/vfs/epoll.go @@ -93,9 +93,9 @@ type epollInterest struct { // NewEpollInstanceFD returns a FileDescription representing a new epoll // instance. A reference is taken on the returned FileDescription. -func (vfs *VirtualFilesystem) NewEpollInstanceFD() (*FileDescription, error) { +func (vfs *VirtualFilesystem) NewEpollInstanceFD(ctx context.Context) (*FileDescription, error) { vd := vfs.NewAnonVirtualDentry("[eventpoll]") - defer vd.DecRef() + defer vd.DecRef(ctx) ep := &EpollInstance{ interest: make(map[epollInterestKey]*epollInterest), } @@ -110,7 +110,7 @@ func (vfs *VirtualFilesystem) NewEpollInstanceFD() (*FileDescription, error) { } // Release implements FileDescriptionImpl.Release. -func (ep *EpollInstance) Release() { +func (ep *EpollInstance) Release(ctx context.Context) { // Unregister all polled fds. ep.interestMu.Lock() defer ep.interestMu.Unlock() diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go index 0c42574db..576ab3920 100644 --- a/pkg/sentry/vfs/file_description.go +++ b/pkg/sentry/vfs/file_description.go @@ -171,7 +171,7 @@ func (fd *FileDescription) TryIncRef() bool { } // DecRef decrements fd's reference count. -func (fd *FileDescription) DecRef() { +func (fd *FileDescription) DecRef(ctx context.Context) { if refs := atomic.AddInt64(&fd.refs, -1); refs == 0 { // Unregister fd from all epoll instances. fd.epollMu.Lock() @@ -196,11 +196,11 @@ func (fd *FileDescription) DecRef() { } // Release implementation resources. - fd.impl.Release() + fd.impl.Release(ctx) if fd.writable { fd.vd.mount.EndWrite() } - fd.vd.DecRef() + fd.vd.DecRef(ctx) fd.flagsMu.Lock() // TODO(gvisor.dev/issue/1663): We may need to unregister during save, as we do in VFS1. if fd.statusFlags&linux.O_ASYNC != 0 && fd.asyncHandler != nil { @@ -335,7 +335,7 @@ func (fd *FileDescription) Impl() FileDescriptionImpl { type FileDescriptionImpl interface { // Release is called when the associated FileDescription reaches zero // references. - Release() + Release(ctx context.Context) // OnClose is called when a file descriptor representing the // FileDescription is closed. Note that returning a non-nil error does not @@ -354,7 +354,7 @@ type FileDescriptionImpl interface { // represented by the FileDescription. StatFS(ctx context.Context) (linux.Statfs, error) - // Allocate grows file represented by FileDescription to offset + length bytes. + // Allocate grows the file to offset + length bytes. // Only mode == 0 is supported currently. Allocate(ctx context.Context, mode, offset, length uint64) error @@ -526,7 +526,7 @@ func (fd *FileDescription) Stat(ctx context.Context, opts StatOptions) (linux.St Start: fd.vd, }) stat, err := fd.vd.mount.fs.impl.StatAt(ctx, rp, opts) - vfsObj.putResolvingPath(rp) + vfsObj.putResolvingPath(ctx, rp) return stat, err } return fd.impl.Stat(ctx, opts) @@ -541,7 +541,7 @@ func (fd *FileDescription) SetStat(ctx context.Context, opts SetStatOptions) err Start: fd.vd, }) err := fd.vd.mount.fs.impl.SetStatAt(ctx, rp, opts) - vfsObj.putResolvingPath(rp) + vfsObj.putResolvingPath(ctx, rp) return err } return fd.impl.SetStat(ctx, opts) @@ -557,12 +557,17 @@ func (fd *FileDescription) StatFS(ctx context.Context) (linux.Statfs, error) { Start: fd.vd, }) statfs, err := fd.vd.mount.fs.impl.StatFSAt(ctx, rp) - vfsObj.putResolvingPath(rp) + vfsObj.putResolvingPath(ctx, rp) return statfs, err } return fd.impl.StatFS(ctx) } +// Allocate grows file represented by FileDescription to offset + length bytes. +func (fd *FileDescription) Allocate(ctx context.Context, mode, offset, length uint64) error { + return fd.impl.Allocate(ctx, mode, offset, length) +} + // Readiness implements waiter.Waitable.Readiness. // // It returns fd's I/O readiness. @@ -669,7 +674,7 @@ func (fd *FileDescription) Listxattr(ctx context.Context, size uint64) ([]string Start: fd.vd, }) names, err := fd.vd.mount.fs.impl.ListxattrAt(ctx, rp, size) - vfsObj.putResolvingPath(rp) + vfsObj.putResolvingPath(ctx, rp) return names, err } names, err := fd.impl.Listxattr(ctx, size) @@ -698,7 +703,7 @@ func (fd *FileDescription) Getxattr(ctx context.Context, opts *GetxattrOptions) Start: fd.vd, }) val, err := fd.vd.mount.fs.impl.GetxattrAt(ctx, rp, *opts) - vfsObj.putResolvingPath(rp) + vfsObj.putResolvingPath(ctx, rp) return val, err } return fd.impl.Getxattr(ctx, *opts) @@ -714,7 +719,7 @@ func (fd *FileDescription) Setxattr(ctx context.Context, opts *SetxattrOptions) Start: fd.vd, }) err := fd.vd.mount.fs.impl.SetxattrAt(ctx, rp, *opts) - vfsObj.putResolvingPath(rp) + vfsObj.putResolvingPath(ctx, rp) return err } return fd.impl.Setxattr(ctx, *opts) @@ -730,7 +735,7 @@ func (fd *FileDescription) Removexattr(ctx context.Context, name string) error { Start: fd.vd, }) err := fd.vd.mount.fs.impl.RemovexattrAt(ctx, rp, name) - vfsObj.putResolvingPath(rp) + vfsObj.putResolvingPath(ctx, rp) return err } return fd.impl.Removexattr(ctx, name) @@ -747,7 +752,7 @@ func (fd *FileDescription) MappedName(ctx context.Context) string { vfsroot := RootFromContext(ctx) s, _ := fd.vd.mount.vfs.PathnameWithDeleted(ctx, vfsroot, fd.vd) if vfsroot.Ok() { - vfsroot.DecRef() + vfsroot.DecRef(ctx) } return s } diff --git a/pkg/sentry/vfs/file_description_impl_util_test.go b/pkg/sentry/vfs/file_description_impl_util_test.go index 3b7e1c273..1cd607c0a 100644 --- a/pkg/sentry/vfs/file_description_impl_util_test.go +++ b/pkg/sentry/vfs/file_description_impl_util_test.go @@ -80,9 +80,9 @@ type testFD struct { data DynamicBytesSource } -func newTestFD(vfsObj *VirtualFilesystem, statusFlags uint32, data DynamicBytesSource) *FileDescription { +func newTestFD(ctx context.Context, vfsObj *VirtualFilesystem, statusFlags uint32, data DynamicBytesSource) *FileDescription { vd := vfsObj.NewAnonVirtualDentry("genCountFD") - defer vd.DecRef() + defer vd.DecRef(ctx) var fd testFD fd.vfsfd.Init(&fd, statusFlags, vd.Mount(), vd.Dentry(), &FileDescriptionOptions{}) fd.DynamicBytesFileDescriptionImpl.SetDataSource(data) @@ -90,7 +90,7 @@ func newTestFD(vfsObj *VirtualFilesystem, statusFlags uint32, data DynamicBytesS } // Release implements FileDescriptionImpl.Release. -func (fd *testFD) Release() { +func (fd *testFD) Release(context.Context) { } // SetStatusFlags implements FileDescriptionImpl.SetStatusFlags. @@ -109,11 +109,11 @@ func TestGenCountFD(t *testing.T) { ctx := contexttest.Context(t) vfsObj := &VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { t.Fatalf("VFS init: %v", err) } - fd := newTestFD(vfsObj, linux.O_RDWR, &genCount{}) - defer fd.DecRef() + fd := newTestFD(ctx, vfsObj, linux.O_RDWR, &genCount{}) + defer fd.DecRef(ctx) // The first read causes Generate to be called to fill the FD's buffer. buf := make([]byte, 2) @@ -167,11 +167,11 @@ func TestWritable(t *testing.T) { ctx := contexttest.Context(t) vfsObj := &VirtualFilesystem{} - if err := vfsObj.Init(); err != nil { + if err := vfsObj.Init(ctx); err != nil { t.Fatalf("VFS init: %v", err) } - fd := newTestFD(vfsObj, linux.O_RDWR, &storeData{data: "init"}) - defer fd.DecRef() + fd := newTestFD(ctx, vfsObj, linux.O_RDWR, &storeData{data: "init"}) + defer fd.DecRef(ctx) buf := make([]byte, 10) ioseq := usermem.BytesIOSequence(buf) diff --git a/pkg/sentry/vfs/filesystem.go b/pkg/sentry/vfs/filesystem.go index 6bb9ca180..df3758fd1 100644 --- a/pkg/sentry/vfs/filesystem.go +++ b/pkg/sentry/vfs/filesystem.go @@ -100,12 +100,12 @@ func (fs *Filesystem) TryIncRef() bool { } // DecRef decrements fs' reference count. -func (fs *Filesystem) DecRef() { +func (fs *Filesystem) DecRef(ctx context.Context) { if refs := atomic.AddInt64(&fs.refs, -1); refs == 0 { fs.vfs.filesystemsMu.Lock() delete(fs.vfs.filesystems, fs) fs.vfs.filesystemsMu.Unlock() - fs.impl.Release() + fs.impl.Release(ctx) } else if refs < 0 { panic("Filesystem.decRef() called without holding a reference") } @@ -149,7 +149,7 @@ func (fs *Filesystem) DecRef() { type FilesystemImpl interface { // Release is called when the associated Filesystem reaches zero // references. - Release() + Release(ctx context.Context) // Sync "causes all pending modifications to filesystem metadata and cached // file data to be written to the underlying [filesystem]", as by syncfs(2). diff --git a/pkg/sentry/vfs/inotify.go b/pkg/sentry/vfs/inotify.go index 167b731ac..aff220a61 100644 --- a/pkg/sentry/vfs/inotify.go +++ b/pkg/sentry/vfs/inotify.go @@ -100,7 +100,7 @@ func NewInotifyFD(ctx context.Context, vfsObj *VirtualFilesystem, flags uint32) id := uniqueid.GlobalFromContext(ctx) vd := vfsObj.NewAnonVirtualDentry(fmt.Sprintf("[inotifyfd:%d]", id)) - defer vd.DecRef() + defer vd.DecRef(ctx) fd := &Inotify{ id: id, scratch: make([]byte, inotifyEventBaseSize), @@ -118,7 +118,7 @@ func NewInotifyFD(ctx context.Context, vfsObj *VirtualFilesystem, flags uint32) // Release implements FileDescriptionImpl.Release. Release removes all // watches and frees all resources for an inotify instance. -func (i *Inotify) Release() { +func (i *Inotify) Release(ctx context.Context) { var ds []*Dentry // We need to hold i.mu to avoid a race with concurrent calls to @@ -144,7 +144,7 @@ func (i *Inotify) Release() { i.mu.Unlock() for _, d := range ds { - d.OnZeroWatches() + d.OnZeroWatches(ctx) } } @@ -350,7 +350,7 @@ func (i *Inotify) AddWatch(target *Dentry, mask uint32) (int32, error) { // RmWatch looks up an inotify watch for the given 'wd' and configures the // target to stop sending events to this inotify instance. -func (i *Inotify) RmWatch(wd int32) error { +func (i *Inotify) RmWatch(ctx context.Context, wd int32) error { i.mu.Lock() // Find the watch we were asked to removed. @@ -374,7 +374,7 @@ func (i *Inotify) RmWatch(wd int32) error { i.mu.Unlock() if remaining == 0 { - w.target.OnZeroWatches() + w.target.OnZeroWatches(ctx) } // Generate the event for the removal. @@ -462,7 +462,7 @@ func (w *Watches) Remove(id uint64) { // Notify queues a new event with watches in this set. Watches with // IN_EXCL_UNLINK are skipped if the event is coming from a child that has been // unlinked. -func (w *Watches) Notify(name string, events, cookie uint32, et EventType, unlinked bool) { +func (w *Watches) Notify(ctx context.Context, name string, events, cookie uint32, et EventType, unlinked bool) { var hasExpired bool w.mu.RLock() for _, watch := range w.ws { @@ -476,13 +476,13 @@ func (w *Watches) Notify(name string, events, cookie uint32, et EventType, unlin w.mu.RUnlock() if hasExpired { - w.cleanupExpiredWatches() + w.cleanupExpiredWatches(ctx) } } // This function is relatively expensive and should only be called where there // are expired watches. -func (w *Watches) cleanupExpiredWatches() { +func (w *Watches) cleanupExpiredWatches(ctx context.Context) { // Because of lock ordering, we cannot acquire Inotify.mu for each watch // owner while holding w.mu. As a result, store expired watches locally // before removing. @@ -495,15 +495,15 @@ func (w *Watches) cleanupExpiredWatches() { } w.mu.RUnlock() for _, watch := range toRemove { - watch.owner.RmWatch(watch.wd) + watch.owner.RmWatch(ctx, watch.wd) } } // HandleDeletion is called when the watch target is destroyed. Clear the // watch set, detach watches from the inotify instances they belong to, and // generate the appropriate events. -func (w *Watches) HandleDeletion() { - w.Notify("", linux.IN_DELETE_SELF, 0, InodeEvent, true /* unlinked */) +func (w *Watches) HandleDeletion(ctx context.Context) { + w.Notify(ctx, "", linux.IN_DELETE_SELF, 0, InodeEvent, true /* unlinked */) // As in Watches.Notify, we can't hold w.mu while acquiring Inotify.mu for // the owner of each watch being deleted. Instead, atomically store the @@ -744,12 +744,12 @@ func InotifyEventFromStatMask(mask uint32) uint32 { // InotifyRemoveChild sends the appriopriate notifications to the watch sets of // the child being removed and its parent. Note that unlike most pairs of // parent/child notifications, the child is notified first in this case. -func InotifyRemoveChild(self, parent *Watches, name string) { +func InotifyRemoveChild(ctx context.Context, self, parent *Watches, name string) { if self != nil { - self.Notify("", linux.IN_ATTRIB, 0, InodeEvent, true /* unlinked */) + self.Notify(ctx, "", linux.IN_ATTRIB, 0, InodeEvent, true /* unlinked */) } if parent != nil { - parent.Notify(name, linux.IN_DELETE, 0, InodeEvent, true /* unlinked */) + parent.Notify(ctx, name, linux.IN_DELETE, 0, InodeEvent, true /* unlinked */) } } @@ -762,13 +762,13 @@ func InotifyRename(ctx context.Context, renamed, oldParent, newParent *Watches, } cookie := uniqueid.InotifyCookie(ctx) if oldParent != nil { - oldParent.Notify(oldName, dirEv|linux.IN_MOVED_FROM, cookie, InodeEvent, false /* unlinked */) + oldParent.Notify(ctx, oldName, dirEv|linux.IN_MOVED_FROM, cookie, InodeEvent, false /* unlinked */) } if newParent != nil { - newParent.Notify(newName, dirEv|linux.IN_MOVED_TO, cookie, InodeEvent, false /* unlinked */) + newParent.Notify(ctx, newName, dirEv|linux.IN_MOVED_TO, cookie, InodeEvent, false /* unlinked */) } // Somewhat surprisingly, self move events do not have a cookie. if renamed != nil { - renamed.Notify("", linux.IN_MOVE_SELF, 0, InodeEvent, false /* unlinked */) + renamed.Notify(ctx, "", linux.IN_MOVE_SELF, 0, InodeEvent, false /* unlinked */) } } diff --git a/pkg/sentry/vfs/mount.go b/pkg/sentry/vfs/mount.go index 32f901bd8..d1d29d0cd 100644 --- a/pkg/sentry/vfs/mount.go +++ b/pkg/sentry/vfs/mount.go @@ -200,8 +200,8 @@ func (vfs *VirtualFilesystem) MountDisconnected(ctx context.Context, creds *auth if err != nil { return nil, err } - defer root.DecRef() - defer fs.DecRef() + defer root.DecRef(ctx) + defer fs.DecRef(ctx) return vfs.NewDisconnectedMount(fs, root, opts) } @@ -221,7 +221,7 @@ func (vfs *VirtualFilesystem) ConnectMountAt(ctx context.Context, creds *auth.Cr if vd.dentry.dead { vd.dentry.mu.Unlock() vfs.mountMu.Unlock() - vd.DecRef() + vd.DecRef(ctx) return syserror.ENOENT } // vd might have been mounted over between vfs.GetDentryAt() and @@ -243,7 +243,7 @@ func (vfs *VirtualFilesystem) ConnectMountAt(ctx context.Context, creds *auth.Cr // This can't fail since we're holding vfs.mountMu. nextmnt.root.IncRef() vd.dentry.mu.Unlock() - vd.DecRef() + vd.DecRef(ctx) vd = VirtualDentry{ mount: nextmnt, dentry: nextmnt.root, @@ -268,7 +268,7 @@ func (vfs *VirtualFilesystem) MountAt(ctx context.Context, creds *auth.Credentia if err != nil { return err } - defer mnt.DecRef() + defer mnt.DecRef(ctx) if err := vfs.ConnectMountAt(ctx, creds, mnt, target); err != nil { return err } @@ -293,13 +293,13 @@ func (vfs *VirtualFilesystem) UmountAt(ctx context.Context, creds *auth.Credenti if err != nil { return err } - defer vd.DecRef() + defer vd.DecRef(ctx) if vd.dentry != vd.mount.root { return syserror.EINVAL } vfs.mountMu.Lock() if mntns := MountNamespaceFromContext(ctx); mntns != nil { - defer mntns.DecRef() + defer mntns.DecRef(ctx) if mntns != vd.mount.ns { vfs.mountMu.Unlock() return syserror.EINVAL @@ -335,10 +335,10 @@ func (vfs *VirtualFilesystem) UmountAt(ctx context.Context, creds *auth.Credenti vfs.mounts.seq.EndWrite() vfs.mountMu.Unlock() for _, vd := range vdsToDecRef { - vd.DecRef() + vd.DecRef(ctx) } for _, mnt := range mountsToDecRef { - mnt.DecRef() + mnt.DecRef(ctx) } return nil } @@ -479,7 +479,7 @@ func (mnt *Mount) IncRef() { } // DecRef decrements mnt's reference count. -func (mnt *Mount) DecRef() { +func (mnt *Mount) DecRef(ctx context.Context) { refs := atomic.AddInt64(&mnt.refs, -1) if refs&^math.MinInt64 == 0 { // mask out MSB var vd VirtualDentry @@ -490,10 +490,10 @@ func (mnt *Mount) DecRef() { mnt.vfs.mounts.seq.EndWrite() mnt.vfs.mountMu.Unlock() } - mnt.root.DecRef() - mnt.fs.DecRef() + mnt.root.DecRef(ctx) + mnt.fs.DecRef(ctx) if vd.Ok() { - vd.DecRef() + vd.DecRef(ctx) } } } @@ -506,7 +506,7 @@ func (mntns *MountNamespace) IncRef() { } // DecRef decrements mntns' reference count. -func (mntns *MountNamespace) DecRef() { +func (mntns *MountNamespace) DecRef(ctx context.Context) { vfs := mntns.root.fs.VirtualFilesystem() if refs := atomic.AddInt64(&mntns.refs, -1); refs == 0 { vfs.mountMu.Lock() @@ -517,10 +517,10 @@ func (mntns *MountNamespace) DecRef() { vfs.mounts.seq.EndWrite() vfs.mountMu.Unlock() for _, vd := range vdsToDecRef { - vd.DecRef() + vd.DecRef(ctx) } for _, mnt := range mountsToDecRef { - mnt.DecRef() + mnt.DecRef(ctx) } } else if refs < 0 { panic("MountNamespace.DecRef() called without holding a reference") @@ -534,7 +534,7 @@ func (mntns *MountNamespace) DecRef() { // getMountAt is analogous to Linux's fs/namei.c:follow_mount(). // // Preconditions: References are held on mnt and d. -func (vfs *VirtualFilesystem) getMountAt(mnt *Mount, d *Dentry) *Mount { +func (vfs *VirtualFilesystem) getMountAt(ctx context.Context, mnt *Mount, d *Dentry) *Mount { // The first mount is special-cased: // // - The caller is assumed to have checked d.isMounted() already. (This @@ -565,7 +565,7 @@ retryFirst: // Raced with umount. continue } - mnt.DecRef() + mnt.DecRef(ctx) mnt = next d = next.root } @@ -578,7 +578,7 @@ retryFirst: // // Preconditions: References are held on mnt and root. vfsroot is not (mnt, // mnt.root). -func (vfs *VirtualFilesystem) getMountpointAt(mnt *Mount, vfsroot VirtualDentry) VirtualDentry { +func (vfs *VirtualFilesystem) getMountpointAt(ctx context.Context, mnt *Mount, vfsroot VirtualDentry) VirtualDentry { // The first mount is special-cased: // // - The caller must have already checked mnt against vfsroot. @@ -602,12 +602,12 @@ retryFirst: if !point.TryIncRef() { // Since Mount holds a reference on Mount.key.point, this can only // happen due to a racing change to Mount.key. - parent.DecRef() + parent.DecRef(ctx) goto retryFirst } if !vfs.mounts.seq.ReadOk(epoch) { - point.DecRef() - parent.DecRef() + point.DecRef(ctx) + parent.DecRef(ctx) goto retryFirst } mnt = parent @@ -635,16 +635,16 @@ retryFirst: if !point.TryIncRef() { // Since Mount holds a reference on Mount.key.point, this can // only happen due to a racing change to Mount.key. - parent.DecRef() + parent.DecRef(ctx) goto retryNotFirst } if !vfs.mounts.seq.ReadOk(epoch) { - point.DecRef() - parent.DecRef() + point.DecRef(ctx) + parent.DecRef(ctx) goto retryNotFirst } - d.DecRef() - mnt.DecRef() + d.DecRef(ctx) + mnt.DecRef(ctx) mnt = parent d = point } diff --git a/pkg/sentry/vfs/pathname.go b/pkg/sentry/vfs/pathname.go index cd78d66bc..e4da15009 100644 --- a/pkg/sentry/vfs/pathname.go +++ b/pkg/sentry/vfs/pathname.go @@ -47,7 +47,7 @@ func (vfs *VirtualFilesystem) PathnameWithDeleted(ctx context.Context, vfsroot, haveRef := false defer func() { if haveRef { - vd.DecRef() + vd.DecRef(ctx) } }() @@ -64,12 +64,12 @@ loop: // of FilesystemImpl.PrependPath() may return nil instead. break loop } - nextVD := vfs.getMountpointAt(vd.mount, vfsroot) + nextVD := vfs.getMountpointAt(ctx, vd.mount, vfsroot) if !nextVD.Ok() { break loop } if haveRef { - vd.DecRef() + vd.DecRef(ctx) } vd = nextVD haveRef = true @@ -101,7 +101,7 @@ func (vfs *VirtualFilesystem) PathnameReachable(ctx context.Context, vfsroot, vd haveRef := false defer func() { if haveRef { - vd.DecRef() + vd.DecRef(ctx) } }() loop: @@ -112,12 +112,12 @@ loop: if vd.mount == vfsroot.mount && vd.mount.root == vfsroot.dentry { break loop } - nextVD := vfs.getMountpointAt(vd.mount, vfsroot) + nextVD := vfs.getMountpointAt(ctx, vd.mount, vfsroot) if !nextVD.Ok() { return "", nil } if haveRef { - vd.DecRef() + vd.DecRef(ctx) } vd = nextVD haveRef = true @@ -145,7 +145,7 @@ func (vfs *VirtualFilesystem) PathnameForGetcwd(ctx context.Context, vfsroot, vd haveRef := false defer func() { if haveRef { - vd.DecRef() + vd.DecRef(ctx) } }() unreachable := false @@ -157,13 +157,13 @@ loop: if vd.mount == vfsroot.mount && vd.mount.root == vfsroot.dentry { break loop } - nextVD := vfs.getMountpointAt(vd.mount, vfsroot) + nextVD := vfs.getMountpointAt(ctx, vd.mount, vfsroot) if !nextVD.Ok() { unreachable = true break loop } if haveRef { - vd.DecRef() + vd.DecRef(ctx) } vd = nextVD haveRef = true diff --git a/pkg/sentry/vfs/resolving_path.go b/pkg/sentry/vfs/resolving_path.go index 9d047ff88..3304372d9 100644 --- a/pkg/sentry/vfs/resolving_path.go +++ b/pkg/sentry/vfs/resolving_path.go @@ -18,6 +18,7 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sync" @@ -136,31 +137,31 @@ func (vfs *VirtualFilesystem) getResolvingPath(creds *auth.Credentials, pop *Pat return rp } -func (vfs *VirtualFilesystem) putResolvingPath(rp *ResolvingPath) { +func (vfs *VirtualFilesystem) putResolvingPath(ctx context.Context, rp *ResolvingPath) { rp.root = VirtualDentry{} - rp.decRefStartAndMount() + rp.decRefStartAndMount(ctx) rp.mount = nil rp.start = nil - rp.releaseErrorState() + rp.releaseErrorState(ctx) resolvingPathPool.Put(rp) } -func (rp *ResolvingPath) decRefStartAndMount() { +func (rp *ResolvingPath) decRefStartAndMount(ctx context.Context) { if rp.flags&rpflagsHaveStartRef != 0 { - rp.start.DecRef() + rp.start.DecRef(ctx) } if rp.flags&rpflagsHaveMountRef != 0 { - rp.mount.DecRef() + rp.mount.DecRef(ctx) } } -func (rp *ResolvingPath) releaseErrorState() { +func (rp *ResolvingPath) releaseErrorState(ctx context.Context) { if rp.nextStart != nil { - rp.nextStart.DecRef() + rp.nextStart.DecRef(ctx) rp.nextStart = nil } if rp.nextMount != nil { - rp.nextMount.DecRef() + rp.nextMount.DecRef(ctx) rp.nextMount = nil } } @@ -236,13 +237,13 @@ func (rp *ResolvingPath) Advance() { // Restart resets the stream of path components represented by rp to its state // on entry to the current FilesystemImpl method. -func (rp *ResolvingPath) Restart() { +func (rp *ResolvingPath) Restart(ctx context.Context) { rp.pit = rp.origParts[rp.numOrigParts-1] rp.mustBeDir = rp.mustBeDirOrig rp.symlinks = rp.symlinksOrig rp.curPart = rp.numOrigParts - 1 copy(rp.parts[:], rp.origParts[:rp.numOrigParts]) - rp.releaseErrorState() + rp.releaseErrorState(ctx) } func (rp *ResolvingPath) relpathCommit() { @@ -260,13 +261,13 @@ func (rp *ResolvingPath) relpathCommit() { // Mount, CheckRoot returns (unspecified, non-nil error). Otherwise, path // resolution should resolve d's parent normally, and CheckRoot returns (false, // nil). -func (rp *ResolvingPath) CheckRoot(d *Dentry) (bool, error) { +func (rp *ResolvingPath) CheckRoot(ctx context.Context, d *Dentry) (bool, error) { if d == rp.root.dentry && rp.mount == rp.root.mount { // At contextual VFS root (due to e.g. chroot(2)). return true, nil } else if d == rp.mount.root { // At mount root ... - vd := rp.vfs.getMountpointAt(rp.mount, rp.root) + vd := rp.vfs.getMountpointAt(ctx, rp.mount, rp.root) if vd.Ok() { // ... of non-root mount. rp.nextMount = vd.mount @@ -283,11 +284,11 @@ func (rp *ResolvingPath) CheckRoot(d *Dentry) (bool, error) { // to d. If d is a mount point, such that path resolution should switch to // another Mount, CheckMount returns a non-nil error. Otherwise, CheckMount // returns nil. -func (rp *ResolvingPath) CheckMount(d *Dentry) error { +func (rp *ResolvingPath) CheckMount(ctx context.Context, d *Dentry) error { if !d.isMounted() { return nil } - if mnt := rp.vfs.getMountAt(rp.mount, d); mnt != nil { + if mnt := rp.vfs.getMountAt(ctx, rp.mount, d); mnt != nil { rp.nextMount = mnt return resolveMountPointError{} } @@ -389,11 +390,11 @@ func (rp *ResolvingPath) HandleJump(target VirtualDentry) error { return resolveMountRootOrJumpError{} } -func (rp *ResolvingPath) handleError(err error) bool { +func (rp *ResolvingPath) handleError(ctx context.Context, err error) bool { switch err.(type) { case resolveMountRootOrJumpError: // Switch to the new Mount. We hold references on the Mount and Dentry. - rp.decRefStartAndMount() + rp.decRefStartAndMount(ctx) rp.mount = rp.nextMount rp.start = rp.nextStart rp.flags |= rpflagsHaveMountRef | rpflagsHaveStartRef @@ -412,7 +413,7 @@ func (rp *ResolvingPath) handleError(err error) bool { case resolveMountPointError: // Switch to the new Mount. We hold a reference on the Mount, but // borrow the reference on the mount root from the Mount. - rp.decRefStartAndMount() + rp.decRefStartAndMount(ctx) rp.mount = rp.nextMount rp.start = rp.nextMount.root rp.flags = rp.flags&^rpflagsHaveStartRef | rpflagsHaveMountRef @@ -423,12 +424,12 @@ func (rp *ResolvingPath) handleError(err error) bool { // path. rp.relpathCommit() // Restart path resolution on the new Mount. - rp.releaseErrorState() + rp.releaseErrorState(ctx) return true case resolveAbsSymlinkError: // Switch to the new Mount. References are borrowed from rp.root. - rp.decRefStartAndMount() + rp.decRefStartAndMount(ctx) rp.mount = rp.root.mount rp.start = rp.root.dentry rp.flags &^= rpflagsHaveMountRef | rpflagsHaveStartRef @@ -440,7 +441,7 @@ func (rp *ResolvingPath) handleError(err error) bool { // path, including the symlink target we just prepended. rp.relpathCommit() // Restart path resolution on the new Mount. - rp.releaseErrorState() + rp.releaseErrorState(ctx) return true default: diff --git a/pkg/sentry/vfs/vfs.go b/pkg/sentry/vfs/vfs.go index 522e27475..9c2420683 100644 --- a/pkg/sentry/vfs/vfs.go +++ b/pkg/sentry/vfs/vfs.go @@ -122,7 +122,7 @@ type VirtualFilesystem struct { } // Init initializes a new VirtualFilesystem with no mounts or FilesystemTypes. -func (vfs *VirtualFilesystem) Init() error { +func (vfs *VirtualFilesystem) Init(ctx context.Context) error { if vfs.mountpoints != nil { panic("VFS already initialized") } @@ -145,7 +145,7 @@ func (vfs *VirtualFilesystem) Init() error { devMinor: anonfsDevMinor, } anonfs.vfsfs.Init(vfs, &anonFilesystemType{}, &anonfs) - defer anonfs.vfsfs.DecRef() + defer anonfs.vfsfs.DecRef(ctx) anonMount, err := vfs.NewDisconnectedMount(&anonfs.vfsfs, nil, &MountOptions{}) if err != nil { // We should not be passing any MountOptions that would cause @@ -192,11 +192,11 @@ func (vfs *VirtualFilesystem) AccessAt(ctx context.Context, creds *auth.Credenti for { err := rp.mount.fs.impl.AccessAt(ctx, rp, creds, ats) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -214,11 +214,11 @@ func (vfs *VirtualFilesystem) GetDentryAt(ctx context.Context, creds *auth.Crede dentry: d, } rp.mount.IncRef() - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return vd, nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return VirtualDentry{}, err } } @@ -236,7 +236,7 @@ func (vfs *VirtualFilesystem) getParentDirAndName(ctx context.Context, creds *au } rp.mount.IncRef() name := rp.Component() - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return parentVD, name, nil } if checkInvariants { @@ -244,8 +244,8 @@ func (vfs *VirtualFilesystem) getParentDirAndName(ctx context.Context, creds *au panic(fmt.Sprintf("%T.GetParentDentryAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return VirtualDentry{}, "", err } } @@ -260,14 +260,14 @@ func (vfs *VirtualFilesystem) LinkAt(ctx context.Context, creds *auth.Credential } if !newpop.Path.Begin.Ok() { - oldVD.DecRef() + oldVD.DecRef(ctx) if newpop.Path.Absolute { return syserror.EEXIST } return syserror.ENOENT } if newpop.FollowFinalSymlink { - oldVD.DecRef() + oldVD.DecRef(ctx) ctx.Warningf("VirtualFilesystem.LinkAt: file creation paths can't follow final symlink") return syserror.EINVAL } @@ -276,8 +276,8 @@ func (vfs *VirtualFilesystem) LinkAt(ctx context.Context, creds *auth.Credential for { err := rp.mount.fs.impl.LinkAt(ctx, rp, oldVD) if err == nil { - vfs.putResolvingPath(rp) - oldVD.DecRef() + vfs.putResolvingPath(ctx, rp) + oldVD.DecRef(ctx) return nil } if checkInvariants { @@ -285,9 +285,9 @@ func (vfs *VirtualFilesystem) LinkAt(ctx context.Context, creds *auth.Credential panic(fmt.Sprintf("%T.LinkAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) - oldVD.DecRef() + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) + oldVD.DecRef(ctx) return err } } @@ -313,7 +313,7 @@ func (vfs *VirtualFilesystem) MkdirAt(ctx context.Context, creds *auth.Credentia for { err := rp.mount.fs.impl.MkdirAt(ctx, rp, *opts) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } if checkInvariants { @@ -321,8 +321,8 @@ func (vfs *VirtualFilesystem) MkdirAt(ctx context.Context, creds *auth.Credentia panic(fmt.Sprintf("%T.MkdirAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -346,7 +346,7 @@ func (vfs *VirtualFilesystem) MknodAt(ctx context.Context, creds *auth.Credentia for { err := rp.mount.fs.impl.MknodAt(ctx, rp, *opts) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } if checkInvariants { @@ -354,8 +354,8 @@ func (vfs *VirtualFilesystem) MknodAt(ctx context.Context, creds *auth.Credentia panic(fmt.Sprintf("%T.MknodAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -408,31 +408,31 @@ func (vfs *VirtualFilesystem) OpenAt(ctx context.Context, creds *auth.Credential for { fd, err := rp.mount.fs.impl.OpenAt(ctx, rp, *opts) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) if opts.FileExec { if fd.Mount().Flags.NoExec { - fd.DecRef() + fd.DecRef(ctx) return nil, syserror.EACCES } // Only a regular file can be executed. stat, err := fd.Stat(ctx, StatOptions{Mask: linux.STATX_TYPE}) if err != nil { - fd.DecRef() + fd.DecRef(ctx) return nil, err } if stat.Mask&linux.STATX_TYPE == 0 || stat.Mode&linux.S_IFMT != linux.S_IFREG { - fd.DecRef() + fd.DecRef(ctx) return nil, syserror.EACCES } } - fd.Dentry().InotifyWithParent(linux.IN_OPEN, 0, PathEvent) + fd.Dentry().InotifyWithParent(ctx, linux.IN_OPEN, 0, PathEvent) return fd, nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return nil, err } } @@ -444,11 +444,11 @@ func (vfs *VirtualFilesystem) ReadlinkAt(ctx context.Context, creds *auth.Creden for { target, err := rp.mount.fs.impl.ReadlinkAt(ctx, rp) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return target, nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return "", err } } @@ -472,19 +472,19 @@ func (vfs *VirtualFilesystem) RenameAt(ctx context.Context, creds *auth.Credenti return err } if oldName == "." || oldName == ".." { - oldParentVD.DecRef() + oldParentVD.DecRef(ctx) return syserror.EBUSY } if !newpop.Path.Begin.Ok() { - oldParentVD.DecRef() + oldParentVD.DecRef(ctx) if newpop.Path.Absolute { return syserror.EBUSY } return syserror.ENOENT } if newpop.FollowFinalSymlink { - oldParentVD.DecRef() + oldParentVD.DecRef(ctx) ctx.Warningf("VirtualFilesystem.RenameAt: destination path can't follow final symlink") return syserror.EINVAL } @@ -497,8 +497,8 @@ func (vfs *VirtualFilesystem) RenameAt(ctx context.Context, creds *auth.Credenti for { err := rp.mount.fs.impl.RenameAt(ctx, rp, oldParentVD, oldName, renameOpts) if err == nil { - vfs.putResolvingPath(rp) - oldParentVD.DecRef() + vfs.putResolvingPath(ctx, rp) + oldParentVD.DecRef(ctx) return nil } if checkInvariants { @@ -506,9 +506,9 @@ func (vfs *VirtualFilesystem) RenameAt(ctx context.Context, creds *auth.Credenti panic(fmt.Sprintf("%T.RenameAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) - oldParentVD.DecRef() + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) + oldParentVD.DecRef(ctx) return err } } @@ -531,7 +531,7 @@ func (vfs *VirtualFilesystem) RmdirAt(ctx context.Context, creds *auth.Credentia for { err := rp.mount.fs.impl.RmdirAt(ctx, rp) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } if checkInvariants { @@ -539,8 +539,8 @@ func (vfs *VirtualFilesystem) RmdirAt(ctx context.Context, creds *auth.Credentia panic(fmt.Sprintf("%T.RmdirAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -552,11 +552,11 @@ func (vfs *VirtualFilesystem) SetStatAt(ctx context.Context, creds *auth.Credent for { err := rp.mount.fs.impl.SetStatAt(ctx, rp, *opts) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -568,11 +568,11 @@ func (vfs *VirtualFilesystem) StatAt(ctx context.Context, creds *auth.Credential for { stat, err := rp.mount.fs.impl.StatAt(ctx, rp, *opts) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return stat, nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return linux.Statx{}, err } } @@ -585,11 +585,11 @@ func (vfs *VirtualFilesystem) StatFSAt(ctx context.Context, creds *auth.Credenti for { statfs, err := rp.mount.fs.impl.StatFSAt(ctx, rp) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return statfs, nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return linux.Statfs{}, err } } @@ -612,7 +612,7 @@ func (vfs *VirtualFilesystem) SymlinkAt(ctx context.Context, creds *auth.Credent for { err := rp.mount.fs.impl.SymlinkAt(ctx, rp, target) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } if checkInvariants { @@ -620,8 +620,8 @@ func (vfs *VirtualFilesystem) SymlinkAt(ctx context.Context, creds *auth.Credent panic(fmt.Sprintf("%T.SymlinkAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -644,7 +644,7 @@ func (vfs *VirtualFilesystem) UnlinkAt(ctx context.Context, creds *auth.Credenti for { err := rp.mount.fs.impl.UnlinkAt(ctx, rp) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } if checkInvariants { @@ -652,8 +652,8 @@ func (vfs *VirtualFilesystem) UnlinkAt(ctx context.Context, creds *auth.Credenti panic(fmt.Sprintf("%T.UnlinkAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -671,7 +671,7 @@ func (vfs *VirtualFilesystem) BoundEndpointAt(ctx context.Context, creds *auth.C for { bep, err := rp.mount.fs.impl.BoundEndpointAt(ctx, rp, *opts) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return bep, nil } if checkInvariants { @@ -679,8 +679,8 @@ func (vfs *VirtualFilesystem) BoundEndpointAt(ctx context.Context, creds *auth.C panic(fmt.Sprintf("%T.BoundEndpointAt() consumed all path components and returned %v", rp.mount.fs.impl, err)) } } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return nil, err } } @@ -693,7 +693,7 @@ func (vfs *VirtualFilesystem) ListxattrAt(ctx context.Context, creds *auth.Crede for { names, err := rp.mount.fs.impl.ListxattrAt(ctx, rp, size) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return names, nil } if err == syserror.ENOTSUP { @@ -701,11 +701,11 @@ func (vfs *VirtualFilesystem) ListxattrAt(ctx context.Context, creds *auth.Crede // fs/xattr.c:vfs_listxattr() falls back to allowing the security // subsystem to return security extended attributes, which by // default don't exist. - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil, nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return nil, err } } @@ -718,11 +718,11 @@ func (vfs *VirtualFilesystem) GetxattrAt(ctx context.Context, creds *auth.Creden for { val, err := rp.mount.fs.impl.GetxattrAt(ctx, rp, *opts) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return val, nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return "", err } } @@ -735,11 +735,11 @@ func (vfs *VirtualFilesystem) SetxattrAt(ctx context.Context, creds *auth.Creden for { err := rp.mount.fs.impl.SetxattrAt(ctx, rp, *opts) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -751,11 +751,11 @@ func (vfs *VirtualFilesystem) RemovexattrAt(ctx context.Context, creds *auth.Cre for { err := rp.mount.fs.impl.RemovexattrAt(ctx, rp, name) if err == nil { - vfs.putResolvingPath(rp) + vfs.putResolvingPath(ctx, rp) return nil } - if !rp.handleError(err) { - vfs.putResolvingPath(rp) + if !rp.handleError(ctx, err) { + vfs.putResolvingPath(ctx, rp) return err } } @@ -777,7 +777,7 @@ func (vfs *VirtualFilesystem) SyncAllFilesystems(ctx context.Context) error { if err := fs.impl.Sync(ctx); err != nil && retErr == nil { retErr = err } - fs.DecRef() + fs.DecRef(ctx) } return retErr } @@ -831,9 +831,9 @@ func (vd VirtualDentry) IncRef() { // DecRef decrements the reference counts on the Mount and Dentry represented // by vd. -func (vd VirtualDentry) DecRef() { - vd.dentry.DecRef() - vd.mount.DecRef() +func (vd VirtualDentry) DecRef(ctx context.Context) { + vd.dentry.DecRef(ctx) + vd.mount.DecRef(ctx) } // Mount returns the Mount associated with vd. It does not take a reference on diff --git a/pkg/syncevent/waiter_amd64.s b/pkg/syncevent/waiter_amd64.s index 985b56ae5..5e216b045 100644 --- a/pkg/syncevent/waiter_amd64.s +++ b/pkg/syncevent/waiter_amd64.s @@ -16,9 +16,9 @@ // See waiter_noasm_unsafe.go for a description of waiterUnlock. // -// func waiterUnlock(g unsafe.Pointer, wg *unsafe.Pointer) bool +// func waiterUnlock(ptr unsafe.Pointer, wg *unsafe.Pointer) bool TEXT ·waiterUnlock(SB),NOSPLIT,$0-24 - MOVQ g+0(FP), DI + MOVQ ptr+0(FP), DI MOVQ wg+8(FP), SI MOVQ $·preparingG(SB), AX diff --git a/pkg/syncevent/waiter_arm64.s b/pkg/syncevent/waiter_arm64.s index 20d7ac23b..f4c06f194 100644 --- a/pkg/syncevent/waiter_arm64.s +++ b/pkg/syncevent/waiter_arm64.s @@ -16,11 +16,11 @@ // See waiter_noasm_unsafe.go for a description of waiterUnlock. // -// func waiterUnlock(g unsafe.Pointer, wg *unsafe.Pointer) bool +// func waiterUnlock(ptr unsafe.Pointer, wg *unsafe.Pointer) bool TEXT ·waiterUnlock(SB),NOSPLIT,$0-24 MOVD wg+8(FP), R0 MOVD $·preparingG(SB), R1 - MOVD g+0(FP), R2 + MOVD ptr+0(FP), R2 again: LDAXR (R0), R3 CMP R1, R3 diff --git a/pkg/syncevent/waiter_asm_unsafe.go b/pkg/syncevent/waiter_asm_unsafe.go index 0995e9053..19d6b0b15 100644 --- a/pkg/syncevent/waiter_asm_unsafe.go +++ b/pkg/syncevent/waiter_asm_unsafe.go @@ -21,4 +21,4 @@ import ( ) // See waiter_noasm_unsafe.go for a description of waiterUnlock. -func waiterUnlock(g unsafe.Pointer, wg *unsafe.Pointer) bool +func waiterUnlock(ptr unsafe.Pointer, wg *unsafe.Pointer) bool diff --git a/pkg/syncevent/waiter_noasm_unsafe.go b/pkg/syncevent/waiter_noasm_unsafe.go index 1c4b0e39a..0f74a689c 100644 --- a/pkg/syncevent/waiter_noasm_unsafe.go +++ b/pkg/syncevent/waiter_noasm_unsafe.go @@ -32,8 +32,8 @@ import ( // should be aborted. // //go:nosplit -func waiterUnlock(g unsafe.Pointer, wg *unsafe.Pointer) bool { +func waiterUnlock(ptr unsafe.Pointer, wg *unsafe.Pointer) bool { // The only way this CAS can fail is if a call to Waiter.NotifyPending() // has replaced *wg with nil, in which case we should not sleep. - return atomic.CompareAndSwapPointer(wg, (unsafe.Pointer)(&preparingG), g) + return atomic.CompareAndSwapPointer(wg, (unsafe.Pointer)(&preparingG), ptr) } diff --git a/pkg/tcpip/link/tun/BUILD b/pkg/tcpip/link/tun/BUILD index e0db6cf54..6c137f693 100644 --- a/pkg/tcpip/link/tun/BUILD +++ b/pkg/tcpip/link/tun/BUILD @@ -12,6 +12,7 @@ go_library( visibility = ["//visibility:public"], deps = [ "//pkg/abi/linux", + "//pkg/context", "//pkg/refs", "//pkg/sync", "//pkg/syserror", diff --git a/pkg/tcpip/link/tun/device.go b/pkg/tcpip/link/tun/device.go index 04ae58e59..22b0a12bd 100644 --- a/pkg/tcpip/link/tun/device.go +++ b/pkg/tcpip/link/tun/device.go @@ -18,6 +18,7 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" @@ -64,14 +65,14 @@ func (d *Device) beforeSave() { } // Release implements fs.FileOperations.Release. -func (d *Device) Release() { +func (d *Device) Release(ctx context.Context) { d.mu.Lock() defer d.mu.Unlock() // Decrease refcount if there is an endpoint associated with this file. if d.endpoint != nil { d.endpoint.RemoveNotify(d.notifyHandle) - d.endpoint.DecRef() + d.endpoint.DecRef(ctx) d.endpoint = nil } } @@ -341,8 +342,8 @@ type tunEndpoint struct { } // DecRef decrements refcount of e, removes NIC if refcount goes to 0. -func (e *tunEndpoint) DecRef() { - e.DecRefWithDestructor(func() { +func (e *tunEndpoint) DecRef(ctx context.Context) { + e.DecRefWithDestructor(ctx, func(context.Context) { e.stack.RemoveNIC(e.nicID) }) } diff --git a/pkg/tcpip/network/fragmentation/fragmentation.go b/pkg/tcpip/network/fragmentation/fragmentation.go index ffbadb6e2..1827666c5 100644 --- a/pkg/tcpip/network/fragmentation/fragmentation.go +++ b/pkg/tcpip/network/fragmentation/fragmentation.go @@ -23,6 +23,7 @@ import ( "time" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" ) @@ -52,13 +53,30 @@ var ( ErrInvalidArgs = errors.New("invalid args") ) +// FragmentID is the identifier for a fragment. +type FragmentID struct { + // Source is the source address of the fragment. + Source tcpip.Address + + // Destination is the destination address of the fragment. + Destination tcpip.Address + + // ID is the identification value of the fragment. + // + // This is a uint32 because IPv6 uses a 32-bit identification value. + ID uint32 + + // The protocol for the packet. + Protocol uint8 +} + // Fragmentation is the main structure that other modules // of the stack should use to implement IP Fragmentation. type Fragmentation struct { mu sync.Mutex highLimit int lowLimit int - reassemblers map[uint32]*reassembler + reassemblers map[FragmentID]*reassembler rList reassemblerList size int timeout time.Duration @@ -93,7 +111,7 @@ func NewFragmentation(blockSize uint16, highMemoryLimit, lowMemoryLimit int, rea } return &Fragmentation{ - reassemblers: make(map[uint32]*reassembler), + reassemblers: make(map[FragmentID]*reassembler), highLimit: highMemoryLimit, lowLimit: lowMemoryLimit, timeout: reassemblingTimeout, @@ -109,7 +127,7 @@ func NewFragmentation(blockSize uint16, highMemoryLimit, lowMemoryLimit int, rea // first must be a multiple of the block size f is configured with. The size // of the fragment data must be a multiple of the block size, unless there are // no fragments following this fragment (more set to false). -func (f *Fragmentation) Process(id uint32, first, last uint16, more bool, vv buffer.VectorisedView) (buffer.VectorisedView, bool, error) { +func (f *Fragmentation) Process(id FragmentID, first, last uint16, more bool, vv buffer.VectorisedView) (buffer.VectorisedView, bool, error) { if first > last { return buffer.VectorisedView{}, false, fmt.Errorf("first=%d is greater than last=%d: %w", first, last, ErrInvalidArgs) } diff --git a/pkg/tcpip/network/fragmentation/fragmentation_test.go b/pkg/tcpip/network/fragmentation/fragmentation_test.go index ebc3232e5..9eedd33c4 100644 --- a/pkg/tcpip/network/fragmentation/fragmentation_test.go +++ b/pkg/tcpip/network/fragmentation/fragmentation_test.go @@ -34,7 +34,7 @@ func vv(size int, pieces ...string) buffer.VectorisedView { } type processInput struct { - id uint32 + id FragmentID first uint16 last uint16 more bool @@ -54,8 +54,8 @@ var processTestCases = []struct { { comment: "One ID", in: []processInput{ - {id: 0, first: 0, last: 1, more: true, vv: vv(2, "01")}, - {id: 0, first: 2, last: 3, more: false, vv: vv(2, "23")}, + {id: FragmentID{ID: 0}, first: 0, last: 1, more: true, vv: vv(2, "01")}, + {id: FragmentID{ID: 0}, first: 2, last: 3, more: false, vv: vv(2, "23")}, }, out: []processOutput{ {vv: buffer.VectorisedView{}, done: false}, @@ -65,10 +65,10 @@ var processTestCases = []struct { { comment: "Two IDs", in: []processInput{ - {id: 0, first: 0, last: 1, more: true, vv: vv(2, "01")}, - {id: 1, first: 0, last: 1, more: true, vv: vv(2, "ab")}, - {id: 1, first: 2, last: 3, more: false, vv: vv(2, "cd")}, - {id: 0, first: 2, last: 3, more: false, vv: vv(2, "23")}, + {id: FragmentID{ID: 0}, first: 0, last: 1, more: true, vv: vv(2, "01")}, + {id: FragmentID{ID: 1}, first: 0, last: 1, more: true, vv: vv(2, "ab")}, + {id: FragmentID{ID: 1}, first: 2, last: 3, more: false, vv: vv(2, "cd")}, + {id: FragmentID{ID: 0}, first: 2, last: 3, more: false, vv: vv(2, "23")}, }, out: []processOutput{ {vv: buffer.VectorisedView{}, done: false}, @@ -113,12 +113,12 @@ func TestReassemblingTimeout(t *testing.T) { timeout := time.Millisecond f := NewFragmentation(minBlockSize, 1024, 512, timeout) // Send first fragment with id = 0, first = 0, last = 0, and more = true. - f.Process(0, 0, 0, true, vv(1, "0")) + f.Process(FragmentID{}, 0, 0, true, vv(1, "0")) // Sleep more than the timeout. time.Sleep(2 * timeout) // Send another fragment that completes a packet. // However, no packet should be reassembled because the fragment arrived after the timeout. - _, done, err := f.Process(0, 1, 1, false, vv(1, "1")) + _, done, err := f.Process(FragmentID{}, 1, 1, false, vv(1, "1")) if err != nil { t.Fatalf("f.Process(0, 1, 1, false, vv(1, \"1\")) failed: %v", err) } @@ -130,23 +130,23 @@ func TestReassemblingTimeout(t *testing.T) { func TestMemoryLimits(t *testing.T) { f := NewFragmentation(minBlockSize, 3, 1, DefaultReassembleTimeout) // Send first fragment with id = 0. - f.Process(0, 0, 0, true, vv(1, "0")) + f.Process(FragmentID{ID: 0}, 0, 0, true, vv(1, "0")) // Send first fragment with id = 1. - f.Process(1, 0, 0, true, vv(1, "1")) + f.Process(FragmentID{ID: 1}, 0, 0, true, vv(1, "1")) // Send first fragment with id = 2. - f.Process(2, 0, 0, true, vv(1, "2")) + f.Process(FragmentID{ID: 2}, 0, 0, true, vv(1, "2")) // Send first fragment with id = 3. This should caused id = 0 and id = 1 to be // evicted. - f.Process(3, 0, 0, true, vv(1, "3")) + f.Process(FragmentID{ID: 3}, 0, 0, true, vv(1, "3")) - if _, ok := f.reassemblers[0]; ok { + if _, ok := f.reassemblers[FragmentID{ID: 0}]; ok { t.Errorf("Memory limits are not respected: id=0 has not been evicted.") } - if _, ok := f.reassemblers[1]; ok { + if _, ok := f.reassemblers[FragmentID{ID: 1}]; ok { t.Errorf("Memory limits are not respected: id=1 has not been evicted.") } - if _, ok := f.reassemblers[3]; !ok { + if _, ok := f.reassemblers[FragmentID{ID: 3}]; !ok { t.Errorf("Implementation of memory limits is wrong: id=3 is not present.") } } @@ -154,9 +154,9 @@ func TestMemoryLimits(t *testing.T) { func TestMemoryLimitsIgnoresDuplicates(t *testing.T) { f := NewFragmentation(minBlockSize, 1, 0, DefaultReassembleTimeout) // Send first fragment with id = 0. - f.Process(0, 0, 0, true, vv(1, "0")) + f.Process(FragmentID{}, 0, 0, true, vv(1, "0")) // Send the same packet again. - f.Process(0, 0, 0, true, vv(1, "0")) + f.Process(FragmentID{}, 0, 0, true, vv(1, "0")) got := f.size want := 1 @@ -166,8 +166,6 @@ func TestMemoryLimitsIgnoresDuplicates(t *testing.T) { } func TestErrors(t *testing.T) { - const fragID = 5 - tests := []struct { name string blockSize uint16 @@ -250,12 +248,12 @@ func TestErrors(t *testing.T) { for _, test := range tests { t.Run(test.name, func(t *testing.T) { f := NewFragmentation(test.blockSize, HighFragThreshold, LowFragThreshold, DefaultReassembleTimeout) - _, done, err := f.Process(fragID, test.first, test.last, test.more, vv(len(test.data), test.data)) + _, done, err := f.Process(FragmentID{}, test.first, test.last, test.more, vv(len(test.data), test.data)) if !errors.Is(err, test.err) { - t.Errorf("got Proceess(%d, %d, %d, %t, %q) = (_, _, %v), want = (_, _, %v)", fragID, test.first, test.last, test.more, test.data, err, test.err) + t.Errorf("got Proceess(_, %d, %d, %t, %q) = (_, _, %v), want = (_, _, %v)", test.first, test.last, test.more, test.data, err, test.err) } if done { - t.Errorf("got Proceess(%d, %d, %d, %t, %q) = (_, true, _), want = (_, false, _)", fragID, test.first, test.last, test.more, test.data) + t.Errorf("got Proceess(_, %d, %d, %t, %q) = (_, true, _), want = (_, false, _)", test.first, test.last, test.more, test.data) } }) } diff --git a/pkg/tcpip/network/fragmentation/reassembler.go b/pkg/tcpip/network/fragmentation/reassembler.go index 0a83d81f2..50d30bbf0 100644 --- a/pkg/tcpip/network/fragmentation/reassembler.go +++ b/pkg/tcpip/network/fragmentation/reassembler.go @@ -32,7 +32,7 @@ type hole struct { type reassembler struct { reassemblerEntry - id uint32 + id FragmentID size int mu sync.Mutex holes []hole @@ -42,7 +42,7 @@ type reassembler struct { creationTime time.Time } -func newReassembler(id uint32) *reassembler { +func newReassembler(id FragmentID) *reassembler { r := &reassembler{ id: id, holes: make([]hole, 0, 16), diff --git a/pkg/tcpip/network/fragmentation/reassembler_test.go b/pkg/tcpip/network/fragmentation/reassembler_test.go index 7eee0710d..dff7c9dcb 100644 --- a/pkg/tcpip/network/fragmentation/reassembler_test.go +++ b/pkg/tcpip/network/fragmentation/reassembler_test.go @@ -94,7 +94,7 @@ var holesTestCases = []struct { func TestUpdateHoles(t *testing.T) { for _, c := range holesTestCases { - r := newReassembler(0) + r := newReassembler(FragmentID{}) for _, i := range c.in { r.updateHoles(i.first, i.last, i.more) } diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index 0b5a35cce..d5f5d38f7 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -442,7 +442,18 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { } var ready bool var err error - pkt.Data, ready, err = e.fragmentation.Process(hash.IPv4FragmentHash(h), h.FragmentOffset(), last, h.More(), pkt.Data) + pkt.Data, ready, err = e.fragmentation.Process( + fragmentation.FragmentID{ + Source: h.SourceAddress(), + Destination: h.DestinationAddress(), + ID: uint32(h.ID()), + Protocol: h.Protocol(), + }, + h.FragmentOffset(), + last, + h.More(), + pkt.Data, + ) if err != nil { r.Stats().IP.MalformedPacketsReceived.Increment() r.Stats().IP.MalformedFragmentsReceived.Increment() diff --git a/pkg/tcpip/network/ipv4/ipv4_test.go b/pkg/tcpip/network/ipv4/ipv4_test.go index 4f82c45e2..ded97ac64 100644 --- a/pkg/tcpip/network/ipv4/ipv4_test.go +++ b/pkg/tcpip/network/ipv4/ipv4_test.go @@ -486,12 +486,16 @@ func TestInvalidFragments(t *testing.T) { // TestReceiveFragments feeds fragments in through the incoming packet path to // test reassembly func TestReceiveFragments(t *testing.T) { - const addr1 = "\x0c\xa8\x00\x01" // 192.168.0.1 - const addr2 = "\x0c\xa8\x00\x02" // 192.168.0.2 - const nicID = 1 + const ( + nicID = 1 + + addr1 = "\x0c\xa8\x00\x01" // 192.168.0.1 + addr2 = "\x0c\xa8\x00\x02" // 192.168.0.2 + addr3 = "\x0c\xa8\x00\x03" // 192.168.0.3 + ) // Build and return a UDP header containing payload. - udpGen := func(payloadLen int, multiplier uint8) buffer.View { + udpGen := func(payloadLen int, multiplier uint8, src, dst tcpip.Address) buffer.View { payload := buffer.NewView(payloadLen) for i := 0; i < len(payload); i++ { payload[i] = uint8(i) * multiplier @@ -507,25 +511,29 @@ func TestReceiveFragments(t *testing.T) { Length: uint16(udpLength), }) copy(u.Payload(), payload) - sum := header.PseudoHeaderChecksum(udp.ProtocolNumber, addr1, addr2, uint16(udpLength)) + sum := header.PseudoHeaderChecksum(udp.ProtocolNumber, src, dst, uint16(udpLength)) sum = header.Checksum(payload, sum) u.SetChecksum(^u.CalculateChecksum(sum)) return hdr.View() } // UDP header plus a payload of 0..256 - ipv4Payload1 := udpGen(256, 1) - udpPayload1 := ipv4Payload1[header.UDPMinimumSize:] + ipv4Payload1Addr1ToAddr2 := udpGen(256, 1, addr1, addr2) + udpPayload1Addr1ToAddr2 := ipv4Payload1Addr1ToAddr2[header.UDPMinimumSize:] + ipv4Payload1Addr3ToAddr2 := udpGen(256, 1, addr3, addr2) + udpPayload1Addr3ToAddr2 := ipv4Payload1Addr3ToAddr2[header.UDPMinimumSize:] // UDP header plus a payload of 0..256 in increments of 2. - ipv4Payload2 := udpGen(128, 2) - udpPayload2 := ipv4Payload2[header.UDPMinimumSize:] + ipv4Payload2Addr1ToAddr2 := udpGen(128, 2, addr1, addr2) + udpPayload2Addr1ToAddr2 := ipv4Payload2Addr1ToAddr2[header.UDPMinimumSize:] // UDP header plus a payload of 0..256 in increments of 3. // Used to test cases where the fragment blocks are not a multiple of // the fragment block size of 8 (RFC 791 section 3.1 page 14). - ipv4Payload3 := udpGen(127, 3) - udpPayload3 := ipv4Payload3[header.UDPMinimumSize:] + ipv4Payload3Addr1ToAddr2 := udpGen(127, 3, addr1, addr2) + udpPayload3Addr1ToAddr2 := ipv4Payload3Addr1ToAddr2[header.UDPMinimumSize:] type fragmentData struct { + srcAddr tcpip.Address + dstAddr tcpip.Address id uint16 flags uint8 fragmentOffset uint16 @@ -541,34 +549,40 @@ func TestReceiveFragments(t *testing.T) { name: "No fragmentation", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, id: 1, flags: 0, fragmentOffset: 0, - payload: ipv4Payload1, + payload: ipv4Payload1Addr1ToAddr2, }, }, - expectedPayloads: [][]byte{udpPayload1}, + expectedPayloads: [][]byte{udpPayload1Addr1ToAddr2}, }, { name: "No fragmentation with size not a multiple of fragment block size", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, id: 1, flags: 0, fragmentOffset: 0, - payload: ipv4Payload3, + payload: ipv4Payload3Addr1ToAddr2, }, }, - expectedPayloads: [][]byte{udpPayload3}, + expectedPayloads: [][]byte{udpPayload3Addr1ToAddr2}, }, { name: "More fragments without payload", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, id: 1, flags: header.IPv4FlagMoreFragments, fragmentOffset: 0, - payload: ipv4Payload1, + payload: ipv4Payload1Addr1ToAddr2, }, }, expectedPayloads: nil, @@ -577,10 +591,12 @@ func TestReceiveFragments(t *testing.T) { name: "Non-zero fragment offset without payload", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, id: 1, flags: 0, fragmentOffset: 8, - payload: ipv4Payload1, + payload: ipv4Payload1Addr1ToAddr2, }, }, expectedPayloads: nil, @@ -589,52 +605,64 @@ func TestReceiveFragments(t *testing.T) { name: "Two fragments", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, id: 1, flags: header.IPv4FlagMoreFragments, fragmentOffset: 0, - payload: ipv4Payload1[:64], + payload: ipv4Payload1Addr1ToAddr2[:64], }, { + srcAddr: addr1, + dstAddr: addr2, id: 1, flags: 0, fragmentOffset: 64, - payload: ipv4Payload1[64:], + payload: ipv4Payload1Addr1ToAddr2[64:], }, }, - expectedPayloads: [][]byte{udpPayload1}, + expectedPayloads: [][]byte{udpPayload1Addr1ToAddr2}, }, { name: "Two fragments with last fragment size not a multiple of fragment block size", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, id: 1, flags: header.IPv4FlagMoreFragments, fragmentOffset: 0, - payload: ipv4Payload3[:64], + payload: ipv4Payload3Addr1ToAddr2[:64], }, { + srcAddr: addr1, + dstAddr: addr2, id: 1, flags: 0, fragmentOffset: 64, - payload: ipv4Payload3[64:], + payload: ipv4Payload3Addr1ToAddr2[64:], }, }, - expectedPayloads: [][]byte{udpPayload3}, + expectedPayloads: [][]byte{udpPayload3Addr1ToAddr2}, }, { name: "Two fragments with first fragment size not a multiple of fragment block size", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, id: 1, flags: header.IPv4FlagMoreFragments, fragmentOffset: 0, - payload: ipv4Payload3[:63], + payload: ipv4Payload3Addr1ToAddr2[:63], }, { + srcAddr: addr1, + dstAddr: addr2, id: 1, flags: 0, fragmentOffset: 63, - payload: ipv4Payload3[63:], + payload: ipv4Payload3Addr1ToAddr2[63:], }, }, expectedPayloads: nil, @@ -643,16 +671,20 @@ func TestReceiveFragments(t *testing.T) { name: "Second fragment has MoreFlags set", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, id: 1, flags: header.IPv4FlagMoreFragments, fragmentOffset: 0, - payload: ipv4Payload1[:64], + payload: ipv4Payload1Addr1ToAddr2[:64], }, { + srcAddr: addr1, + dstAddr: addr2, id: 1, flags: header.IPv4FlagMoreFragments, fragmentOffset: 64, - payload: ipv4Payload1[64:], + payload: ipv4Payload1Addr1ToAddr2[64:], }, }, expectedPayloads: nil, @@ -661,16 +693,20 @@ func TestReceiveFragments(t *testing.T) { name: "Two fragments with different IDs", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, id: 1, flags: header.IPv4FlagMoreFragments, fragmentOffset: 0, - payload: ipv4Payload1[:64], + payload: ipv4Payload1Addr1ToAddr2[:64], }, { + srcAddr: addr1, + dstAddr: addr2, id: 2, flags: 0, fragmentOffset: 64, - payload: ipv4Payload1[64:], + payload: ipv4Payload1Addr1ToAddr2[64:], }, }, expectedPayloads: nil, @@ -679,40 +715,88 @@ func TestReceiveFragments(t *testing.T) { name: "Two interleaved fragmented packets", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, id: 1, flags: header.IPv4FlagMoreFragments, fragmentOffset: 0, - payload: ipv4Payload1[:64], + payload: ipv4Payload1Addr1ToAddr2[:64], }, { + srcAddr: addr1, + dstAddr: addr2, id: 2, flags: header.IPv4FlagMoreFragments, fragmentOffset: 0, - payload: ipv4Payload2[:64], + payload: ipv4Payload2Addr1ToAddr2[:64], }, { + srcAddr: addr1, + dstAddr: addr2, id: 1, flags: 0, fragmentOffset: 64, - payload: ipv4Payload1[64:], + payload: ipv4Payload1Addr1ToAddr2[64:], }, { + srcAddr: addr1, + dstAddr: addr2, id: 2, flags: 0, fragmentOffset: 64, - payload: ipv4Payload2[64:], + payload: ipv4Payload2Addr1ToAddr2[64:], + }, + }, + expectedPayloads: [][]byte{udpPayload1Addr1ToAddr2, udpPayload2Addr1ToAddr2}, + }, + { + name: "Two interleaved fragmented packets from different sources but with same ID", + fragments: []fragmentData{ + { + srcAddr: addr1, + dstAddr: addr2, + id: 1, + flags: header.IPv4FlagMoreFragments, + fragmentOffset: 0, + payload: ipv4Payload1Addr1ToAddr2[:64], + }, + { + srcAddr: addr3, + dstAddr: addr2, + id: 1, + flags: header.IPv4FlagMoreFragments, + fragmentOffset: 0, + payload: ipv4Payload1Addr3ToAddr2[:32], + }, + { + srcAddr: addr1, + dstAddr: addr2, + id: 1, + flags: 0, + fragmentOffset: 64, + payload: ipv4Payload1Addr1ToAddr2[64:], + }, + { + srcAddr: addr3, + dstAddr: addr2, + id: 1, + flags: 0, + fragmentOffset: 32, + payload: ipv4Payload1Addr3ToAddr2[32:], }, }, - expectedPayloads: [][]byte{udpPayload1, udpPayload2}, + expectedPayloads: [][]byte{udpPayload1Addr1ToAddr2, udpPayload1Addr3ToAddr2}, }, { name: "Fragment without followup", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, id: 1, flags: header.IPv4FlagMoreFragments, fragmentOffset: 0, - payload: ipv4Payload1[:64], + payload: ipv4Payload1Addr1ToAddr2[:64], }, }, expectedPayloads: nil, @@ -764,8 +848,8 @@ func TestReceiveFragments(t *testing.T) { FragmentOffset: frag.fragmentOffset, TTL: 64, Protocol: uint8(header.UDPProtocolNumber), - SrcAddr: addr1, - DstAddr: addr2, + SrcAddr: frag.srcAddr, + DstAddr: frag.dstAddr, }) vv := hdr.View().ToVectorisedView() diff --git a/pkg/tcpip/network/ipv6/BUILD b/pkg/tcpip/network/ipv6/BUILD index feada63dc..bcc64994e 100644 --- a/pkg/tcpip/network/ipv6/BUILD +++ b/pkg/tcpip/network/ipv6/BUILD @@ -14,7 +14,6 @@ go_library( "//pkg/tcpip/buffer", "//pkg/tcpip/header", "//pkg/tcpip/network/fragmentation", - "//pkg/tcpip/network/hash", "//pkg/tcpip/stack", ], ) diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 5483ae4ee..a0a5c9c01 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -28,7 +28,6 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/buffer" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/network/fragmentation" - "gvisor.dev/gvisor/pkg/tcpip/network/hash" "gvisor.dev/gvisor/pkg/tcpip/stack" ) @@ -343,7 +342,19 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt *stack.PacketBuffer) { var ready bool // Note that pkt doesn't have its transport header set after reassembly, // and won't until DeliverNetworkPacket sets it. - pkt.Data, ready, err = e.fragmentation.Process(hash.IPv6FragmentHash(h, extHdr.ID()), start, last, extHdr.More(), rawPayload.Buf) + pkt.Data, ready, err = e.fragmentation.Process( + // IPv6 ignores the Protocol field since the ID only needs to be unique + // across source-destination pairs, as per RFC 8200 section 4.5. + fragmentation.FragmentID{ + Source: h.SourceAddress(), + Destination: h.DestinationAddress(), + ID: extHdr.ID(), + }, + start, + last, + extHdr.More(), + rawPayload.Buf, + ) if err != nil { r.Stats().IP.MalformedPacketsReceived.Increment() r.Stats().IP.MalformedFragmentsReceived.Increment() diff --git a/pkg/tcpip/network/ipv6/ipv6_test.go b/pkg/tcpip/network/ipv6/ipv6_test.go index 84bac14ff..3d65814de 100644 --- a/pkg/tcpip/network/ipv6/ipv6_test.go +++ b/pkg/tcpip/network/ipv6/ipv6_test.go @@ -673,6 +673,8 @@ func TestReceiveIPv6ExtHdrs(t *testing.T) { // fragmentData holds the IPv6 payload for a fragmented IPv6 packet. type fragmentData struct { + srcAddr tcpip.Address + dstAddr tcpip.Address nextHdr uint8 data buffer.VectorisedView } @@ -691,7 +693,7 @@ func TestReceiveIPv6Fragments(t *testing.T) { routingExtHdrLen = 8 ) - udpGen := func(payload []byte, multiplier uint8) buffer.View { + udpGen := func(payload []byte, multiplier uint8, src, dst tcpip.Address) buffer.View { payloadLen := len(payload) for i := 0; i < payloadLen; i++ { payload[i] = uint8(i) * multiplier @@ -707,23 +709,27 @@ func TestReceiveIPv6Fragments(t *testing.T) { Length: uint16(udpLength), }) copy(u.Payload(), payload) - sum := header.PseudoHeaderChecksum(udp.ProtocolNumber, addr1, addr2, uint16(udpLength)) + sum := header.PseudoHeaderChecksum(udp.ProtocolNumber, src, dst, uint16(udpLength)) sum = header.Checksum(payload, sum) u.SetChecksum(^u.CalculateChecksum(sum)) return hdr.View() } - var udpPayload1Buf [udpPayload1Length]byte - udpPayload1 := udpPayload1Buf[:] - ipv6Payload1 := udpGen(udpPayload1, 1) + var udpPayload1Addr1ToAddr2Buf [udpPayload1Length]byte + udpPayload1Addr1ToAddr2 := udpPayload1Addr1ToAddr2Buf[:] + ipv6Payload1Addr1ToAddr2 := udpGen(udpPayload1Addr1ToAddr2, 1, addr1, addr2) - var udpPayload2Buf [udpPayload2Length]byte - udpPayload2 := udpPayload2Buf[:] - ipv6Payload2 := udpGen(udpPayload2, 2) + var udpPayload1Addr3ToAddr2Buf [udpPayload1Length]byte + udpPayload1Addr3ToAddr2 := udpPayload1Addr3ToAddr2Buf[:] + ipv6Payload1Addr3ToAddr2 := udpGen(udpPayload1Addr3ToAddr2, 4, addr3, addr2) - var udpPayload3Buf [udpPayload3Length]byte - udpPayload3 := udpPayload3Buf[:] - ipv6Payload3 := udpGen(udpPayload3, 3) + var udpPayload2Addr1ToAddr2Buf [udpPayload2Length]byte + udpPayload2Addr1ToAddr2 := udpPayload2Addr1ToAddr2Buf[:] + ipv6Payload2Addr1ToAddr2 := udpGen(udpPayload2Addr1ToAddr2, 2, addr1, addr2) + + var udpPayload3Addr1ToAddr2Buf [udpPayload3Length]byte + udpPayload3Addr1ToAddr2 := udpPayload3Addr1ToAddr2Buf[:] + ipv6Payload3Addr1ToAddr2 := udpGen(udpPayload3Addr1ToAddr2, 3, addr1, addr2) tests := []struct { name string @@ -735,52 +741,60 @@ func TestReceiveIPv6Fragments(t *testing.T) { name: "No fragmentation", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, nextHdr: uint8(header.UDPProtocolNumber), - data: ipv6Payload1.ToVectorisedView(), + data: ipv6Payload1Addr1ToAddr2.ToVectorisedView(), }, }, - expectedPayloads: [][]byte{udpPayload1}, + expectedPayloads: [][]byte{udpPayload1Addr1ToAddr2}, }, { name: "Atomic fragment", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( - fragmentExtHdrLen+len(ipv6Payload1), + fragmentExtHdrLen+len(ipv6Payload1Addr1ToAddr2), []buffer.View{ // Fragment extension header. buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 0, 0, 0, 0, 0}), - ipv6Payload1, + ipv6Payload1Addr1ToAddr2, }, ), }, }, - expectedPayloads: [][]byte{udpPayload1}, + expectedPayloads: [][]byte{udpPayload1Addr1ToAddr2}, }, { name: "Atomic fragment with size not a multiple of fragment block size", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( - fragmentExtHdrLen+len(ipv6Payload3), + fragmentExtHdrLen+len(ipv6Payload3Addr1ToAddr2), []buffer.View{ // Fragment extension header. buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 0, 0, 0, 0, 0}), - ipv6Payload3, + ipv6Payload3Addr1ToAddr2, }, ), }, }, - expectedPayloads: [][]byte{udpPayload3}, + expectedPayloads: [][]byte{udpPayload3Addr1ToAddr2}, }, { name: "Two fragments", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( fragmentExtHdrLen+64, @@ -790,31 +804,35 @@ func TestReceiveIPv6Fragments(t *testing.T) { // Fragment offset = 0, More = true, ID = 1 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 1, 0, 0, 0, 1}), - ipv6Payload1[:64], + ipv6Payload1Addr1ToAddr2[:64], }, ), }, { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( - fragmentExtHdrLen+len(ipv6Payload1)-64, + fragmentExtHdrLen+len(ipv6Payload1Addr1ToAddr2)-64, []buffer.View{ // Fragment extension header. // // Fragment offset = 8, More = false, ID = 1 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 64, 0, 0, 0, 1}), - ipv6Payload1[64:], + ipv6Payload1Addr1ToAddr2[64:], }, ), }, }, - expectedPayloads: [][]byte{udpPayload1}, + expectedPayloads: [][]byte{udpPayload1Addr1ToAddr2}, }, { name: "Two fragments with last fragment size not a multiple of fragment block size", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( fragmentExtHdrLen+64, @@ -824,31 +842,35 @@ func TestReceiveIPv6Fragments(t *testing.T) { // Fragment offset = 0, More = true, ID = 1 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 1, 0, 0, 0, 1}), - ipv6Payload3[:64], + ipv6Payload3Addr1ToAddr2[:64], }, ), }, { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( - fragmentExtHdrLen+len(ipv6Payload3)-64, + fragmentExtHdrLen+len(ipv6Payload3Addr1ToAddr2)-64, []buffer.View{ // Fragment extension header. // // Fragment offset = 8, More = false, ID = 1 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 64, 0, 0, 0, 1}), - ipv6Payload3[64:], + ipv6Payload3Addr1ToAddr2[64:], }, ), }, }, - expectedPayloads: [][]byte{udpPayload3}, + expectedPayloads: [][]byte{udpPayload3Addr1ToAddr2}, }, { name: "Two fragments with first fragment size not a multiple of fragment block size", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( fragmentExtHdrLen+63, @@ -858,21 +880,23 @@ func TestReceiveIPv6Fragments(t *testing.T) { // Fragment offset = 0, More = true, ID = 1 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 1, 0, 0, 0, 1}), - ipv6Payload3[:63], + ipv6Payload3Addr1ToAddr2[:63], }, ), }, { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( - fragmentExtHdrLen+len(ipv6Payload3)-63, + fragmentExtHdrLen+len(ipv6Payload3Addr1ToAddr2)-63, []buffer.View{ // Fragment extension header. // // Fragment offset = 8, More = false, ID = 1 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 64, 0, 0, 0, 1}), - ipv6Payload3[63:], + ipv6Payload3Addr1ToAddr2[63:], }, ), }, @@ -883,6 +907,8 @@ func TestReceiveIPv6Fragments(t *testing.T) { name: "Two fragments with different IDs", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( fragmentExtHdrLen+64, @@ -892,21 +918,23 @@ func TestReceiveIPv6Fragments(t *testing.T) { // Fragment offset = 0, More = true, ID = 1 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 1, 0, 0, 0, 1}), - ipv6Payload1[:64], + ipv6Payload1Addr1ToAddr2[:64], }, ), }, { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( - fragmentExtHdrLen+len(ipv6Payload1)-64, + fragmentExtHdrLen+len(ipv6Payload1Addr1ToAddr2)-64, []buffer.View{ // Fragment extension header. // // Fragment offset = 8, More = false, ID = 2 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 64, 0, 0, 0, 2}), - ipv6Payload1[64:], + ipv6Payload1Addr1ToAddr2[64:], }, ), }, @@ -917,6 +945,8 @@ func TestReceiveIPv6Fragments(t *testing.T) { name: "Two fragments with per-fragment routing header with zero segments left", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, nextHdr: routingExtHdrID, data: buffer.NewVectorisedView( routingExtHdrLen+fragmentExtHdrLen+64, @@ -931,14 +961,16 @@ func TestReceiveIPv6Fragments(t *testing.T) { // Fragment offset = 0, More = true, ID = 1 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 1, 0, 0, 0, 1}), - ipv6Payload1[:64], + ipv6Payload1Addr1ToAddr2[:64], }, ), }, { + srcAddr: addr1, + dstAddr: addr2, nextHdr: routingExtHdrID, data: buffer.NewVectorisedView( - routingExtHdrLen+fragmentExtHdrLen+len(ipv6Payload1)-64, + routingExtHdrLen+fragmentExtHdrLen+len(ipv6Payload1Addr1ToAddr2)-64, []buffer.View{ // Routing extension header. // @@ -950,17 +982,19 @@ func TestReceiveIPv6Fragments(t *testing.T) { // Fragment offset = 8, More = false, ID = 1 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 64, 0, 0, 0, 1}), - ipv6Payload1[64:], + ipv6Payload1Addr1ToAddr2[64:], }, ), }, }, - expectedPayloads: [][]byte{udpPayload1}, + expectedPayloads: [][]byte{udpPayload1Addr1ToAddr2}, }, { name: "Two fragments with per-fragment routing header with non-zero segments left", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, nextHdr: routingExtHdrID, data: buffer.NewVectorisedView( routingExtHdrLen+fragmentExtHdrLen+64, @@ -975,14 +1009,16 @@ func TestReceiveIPv6Fragments(t *testing.T) { // Fragment offset = 0, More = true, ID = 1 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 1, 0, 0, 0, 1}), - ipv6Payload1[:64], + ipv6Payload1Addr1ToAddr2[:64], }, ), }, { + srcAddr: addr1, + dstAddr: addr2, nextHdr: routingExtHdrID, data: buffer.NewVectorisedView( - routingExtHdrLen+fragmentExtHdrLen+len(ipv6Payload1)-64, + routingExtHdrLen+fragmentExtHdrLen+len(ipv6Payload1Addr1ToAddr2)-64, []buffer.View{ // Routing extension header. // @@ -994,7 +1030,7 @@ func TestReceiveIPv6Fragments(t *testing.T) { // Fragment offset = 9, More = false, ID = 1 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 72, 0, 0, 0, 1}), - ipv6Payload1[64:], + ipv6Payload1Addr1ToAddr2[64:], }, ), }, @@ -1005,6 +1041,8 @@ func TestReceiveIPv6Fragments(t *testing.T) { name: "Two fragments with routing header with zero segments left", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( routingExtHdrLen+fragmentExtHdrLen+64, @@ -1019,31 +1057,35 @@ func TestReceiveIPv6Fragments(t *testing.T) { // Segments left = 0. buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 1, 0, 2, 3, 4, 5}), - ipv6Payload1[:64], + ipv6Payload1Addr1ToAddr2[:64], }, ), }, { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( - fragmentExtHdrLen+len(ipv6Payload1)-64, + fragmentExtHdrLen+len(ipv6Payload1Addr1ToAddr2)-64, []buffer.View{ // Fragment extension header. // // Fragment offset = 9, More = false, ID = 1 buffer.View([]byte{routingExtHdrID, 0, 0, 72, 0, 0, 0, 1}), - ipv6Payload1[64:], + ipv6Payload1Addr1ToAddr2[64:], }, ), }, }, - expectedPayloads: [][]byte{udpPayload1}, + expectedPayloads: [][]byte{udpPayload1Addr1ToAddr2}, }, { name: "Two fragments with routing header with non-zero segments left", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( routingExtHdrLen+fragmentExtHdrLen+64, @@ -1058,21 +1100,23 @@ func TestReceiveIPv6Fragments(t *testing.T) { // Segments left = 1. buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 1, 1, 2, 3, 4, 5}), - ipv6Payload1[:64], + ipv6Payload1Addr1ToAddr2[:64], }, ), }, { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( - fragmentExtHdrLen+len(ipv6Payload1)-64, + fragmentExtHdrLen+len(ipv6Payload1Addr1ToAddr2)-64, []buffer.View{ // Fragment extension header. // // Fragment offset = 9, More = false, ID = 1 buffer.View([]byte{routingExtHdrID, 0, 0, 72, 0, 0, 0, 1}), - ipv6Payload1[64:], + ipv6Payload1Addr1ToAddr2[64:], }, ), }, @@ -1083,6 +1127,8 @@ func TestReceiveIPv6Fragments(t *testing.T) { name: "Two fragments with routing header with zero segments left across fragments", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( // The length of this payload is fragmentExtHdrLen+8 because the @@ -1103,12 +1149,14 @@ func TestReceiveIPv6Fragments(t *testing.T) { ), }, { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( // The length of this payload is - // fragmentExtHdrLen+8+len(ipv6Payload1) because the last 8 bytes of + // fragmentExtHdrLen+8+len(ipv6Payload1Addr1ToAddr2) because the last 8 bytes of // the 16 byte routing extension header is in this fagment. - fragmentExtHdrLen+8+len(ipv6Payload1), + fragmentExtHdrLen+8+len(ipv6Payload1Addr1ToAddr2), []buffer.View{ // Fragment extension header. // @@ -1118,7 +1166,7 @@ func TestReceiveIPv6Fragments(t *testing.T) { // Routing extension header (part 2) buffer.View([]byte{6, 7, 8, 9, 10, 11, 12, 13}), - ipv6Payload1, + ipv6Payload1Addr1ToAddr2, }, ), }, @@ -1129,6 +1177,8 @@ func TestReceiveIPv6Fragments(t *testing.T) { name: "Two fragments with routing header with non-zero segments left across fragments", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( // The length of this payload is fragmentExtHdrLen+8 because the @@ -1149,12 +1199,14 @@ func TestReceiveIPv6Fragments(t *testing.T) { ), }, { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( // The length of this payload is - // fragmentExtHdrLen+8+len(ipv6Payload1) because the last 8 bytes of + // fragmentExtHdrLen+8+len(ipv6Payload1Addr1ToAddr2) because the last 8 bytes of // the 16 byte routing extension header is in this fagment. - fragmentExtHdrLen+8+len(ipv6Payload1), + fragmentExtHdrLen+8+len(ipv6Payload1Addr1ToAddr2), []buffer.View{ // Fragment extension header. // @@ -1164,7 +1216,7 @@ func TestReceiveIPv6Fragments(t *testing.T) { // Routing extension header (part 2) buffer.View([]byte{6, 7, 8, 9, 10, 11, 12, 13}), - ipv6Payload1, + ipv6Payload1Addr1ToAddr2, }, ), }, @@ -1177,6 +1229,8 @@ func TestReceiveIPv6Fragments(t *testing.T) { name: "Two fragments with atomic", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( fragmentExtHdrLen+64, @@ -1186,47 +1240,53 @@ func TestReceiveIPv6Fragments(t *testing.T) { // Fragment offset = 0, More = true, ID = 1 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 1, 0, 0, 0, 1}), - ipv6Payload1[:64], + ipv6Payload1Addr1ToAddr2[:64], }, ), }, // This fragment has the same ID as the other fragments but is an atomic // fragment. It should not interfere with the other fragments. { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( - fragmentExtHdrLen+len(ipv6Payload2), + fragmentExtHdrLen+len(ipv6Payload2Addr1ToAddr2), []buffer.View{ // Fragment extension header. // // Fragment offset = 0, More = false, ID = 1 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 0, 0, 0, 0, 1}), - ipv6Payload2, + ipv6Payload2Addr1ToAddr2, }, ), }, { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( - fragmentExtHdrLen+len(ipv6Payload1)-64, + fragmentExtHdrLen+len(ipv6Payload1Addr1ToAddr2)-64, []buffer.View{ // Fragment extension header. // // Fragment offset = 8, More = false, ID = 1 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 64, 0, 0, 0, 1}), - ipv6Payload1[64:], + ipv6Payload1Addr1ToAddr2[64:], }, ), }, }, - expectedPayloads: [][]byte{udpPayload2, udpPayload1}, + expectedPayloads: [][]byte{udpPayload2Addr1ToAddr2, udpPayload1Addr1ToAddr2}, }, { name: "Two interleaved fragmented packets", fragments: []fragmentData{ { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( fragmentExtHdrLen+64, @@ -1236,11 +1296,13 @@ func TestReceiveIPv6Fragments(t *testing.T) { // Fragment offset = 0, More = true, ID = 1 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 1, 0, 0, 0, 1}), - ipv6Payload1[:64], + ipv6Payload1Addr1ToAddr2[:64], }, ), }, { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( fragmentExtHdrLen+32, @@ -1250,40 +1312,114 @@ func TestReceiveIPv6Fragments(t *testing.T) { // Fragment offset = 0, More = true, ID = 2 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 1, 0, 0, 0, 2}), - ipv6Payload2[:32], + ipv6Payload2Addr1ToAddr2[:32], }, ), }, { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( - fragmentExtHdrLen+len(ipv6Payload1)-64, + fragmentExtHdrLen+len(ipv6Payload1Addr1ToAddr2)-64, []buffer.View{ // Fragment extension header. // // Fragment offset = 8, More = false, ID = 1 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 64, 0, 0, 0, 1}), - ipv6Payload1[64:], + ipv6Payload1Addr1ToAddr2[64:], }, ), }, { + srcAddr: addr1, + dstAddr: addr2, nextHdr: fragmentExtHdrID, data: buffer.NewVectorisedView( - fragmentExtHdrLen+len(ipv6Payload2)-32, + fragmentExtHdrLen+len(ipv6Payload2Addr1ToAddr2)-32, []buffer.View{ // Fragment extension header. // // Fragment offset = 4, More = false, ID = 2 buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 32, 0, 0, 0, 2}), - ipv6Payload2[32:], + ipv6Payload2Addr1ToAddr2[32:], + }, + ), + }, + }, + expectedPayloads: [][]byte{udpPayload1Addr1ToAddr2, udpPayload2Addr1ToAddr2}, + }, + { + name: "Two interleaved fragmented packets from different sources but with same ID", + fragments: []fragmentData{ + { + srcAddr: addr1, + dstAddr: addr2, + nextHdr: fragmentExtHdrID, + data: buffer.NewVectorisedView( + fragmentExtHdrLen+64, + []buffer.View{ + // Fragment extension header. + // + // Fragment offset = 0, More = true, ID = 1 + buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 1, 0, 0, 0, 1}), + + ipv6Payload1Addr1ToAddr2[:64], + }, + ), + }, + { + srcAddr: addr3, + dstAddr: addr2, + nextHdr: fragmentExtHdrID, + data: buffer.NewVectorisedView( + fragmentExtHdrLen+32, + []buffer.View{ + // Fragment extension header. + // + // Fragment offset = 0, More = true, ID = 1 + buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 1, 0, 0, 0, 1}), + + ipv6Payload1Addr3ToAddr2[:32], + }, + ), + }, + { + srcAddr: addr1, + dstAddr: addr2, + nextHdr: fragmentExtHdrID, + data: buffer.NewVectorisedView( + fragmentExtHdrLen+len(ipv6Payload1Addr1ToAddr2)-64, + []buffer.View{ + // Fragment extension header. + // + // Fragment offset = 8, More = false, ID = 1 + buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 64, 0, 0, 0, 1}), + + ipv6Payload1Addr1ToAddr2[64:], + }, + ), + }, + { + srcAddr: addr3, + dstAddr: addr2, + nextHdr: fragmentExtHdrID, + data: buffer.NewVectorisedView( + fragmentExtHdrLen+len(ipv6Payload1Addr1ToAddr2)-32, + []buffer.View{ + // Fragment extension header. + // + // Fragment offset = 4, More = false, ID = 1 + buffer.View([]byte{uint8(header.UDPProtocolNumber), 0, 0, 32, 0, 0, 0, 1}), + + ipv6Payload1Addr3ToAddr2[32:], }, ), }, }, - expectedPayloads: [][]byte{udpPayload1, udpPayload2}, + expectedPayloads: [][]byte{udpPayload1Addr1ToAddr2, udpPayload1Addr3ToAddr2}, }, } @@ -1326,8 +1462,8 @@ func TestReceiveIPv6Fragments(t *testing.T) { PayloadLength: uint16(f.data.Size()), NextHeader: f.nextHdr, HopLimit: 255, - SrcAddr: addr1, - DstAddr: addr2, + SrcAddr: f.srcAddr, + DstAddr: f.dstAddr, }) vv := hdr.View().ToVectorisedView() diff --git a/pkg/tcpip/stack/conntrack.go b/pkg/tcpip/stack/conntrack.go index 559a1c4dd..470c265aa 100644 --- a/pkg/tcpip/stack/conntrack.go +++ b/pkg/tcpip/stack/conntrack.go @@ -240,7 +240,10 @@ func (ct *ConnTrack) connFor(pkt *PacketBuffer) (*conn, direction) { if err != nil { return nil, dirOriginal } + return ct.connForTID(tid) +} +func (ct *ConnTrack) connForTID(tid tupleID) (*conn, direction) { bucket := ct.bucket(tid) now := time.Now() @@ -604,3 +607,26 @@ func (ct *ConnTrack) reapTupleLocked(tuple *tuple, bucket int, now time.Time) bo return true } + +func (ct *ConnTrack) originalDst(epID TransportEndpointID) (tcpip.Address, uint16, *tcpip.Error) { + // Lookup the connection. The reply's original destination + // describes the original address. + tid := tupleID{ + srcAddr: epID.LocalAddress, + srcPort: epID.LocalPort, + dstAddr: epID.RemoteAddress, + dstPort: epID.RemotePort, + transProto: header.TCPProtocolNumber, + netProto: header.IPv4ProtocolNumber, + } + conn, _ := ct.connForTID(tid) + if conn == nil { + // Not a tracked connection. + return "", 0, tcpip.ErrNotConnected + } else if conn.manip == manipNone { + // Unmanipulated connection. + return "", 0, tcpip.ErrInvalidOptionValue + } + + return conn.original.dstAddr, conn.original.dstPort, nil +} diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go index cbbae4224..110ba073d 100644 --- a/pkg/tcpip/stack/iptables.go +++ b/pkg/tcpip/stack/iptables.go @@ -218,19 +218,16 @@ func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, addr // Many users never configure iptables. Spare them the cost of rule // traversal if rules have never been set. it.mu.RLock() + defer it.mu.RUnlock() if !it.modified { - it.mu.RUnlock() return true } - it.mu.RUnlock() // Packets are manipulated only if connection and matching // NAT rule exists. shouldTrack := it.connections.handlePacket(pkt, hook, gso, r) // Go through each table containing the hook. - it.mu.RLock() - defer it.mu.RUnlock() priorities := it.priorities[hook] for _, tableID := range priorities { // If handlePacket already NATed the packet, we don't need to @@ -418,3 +415,9 @@ func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx // All the matchers matched, so run the target. return rule.Target.Action(pkt, &it.connections, hook, gso, r, address) } + +// OriginalDst returns the original destination of redirected connections. It +// returns an error if the connection doesn't exist or isn't redirected. +func (it *IPTables) OriginalDst(epID TransportEndpointID) (tcpip.Address, uint16, *tcpip.Error) { + return it.connections.originalDst(epID) +} diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index a634b9b60..45f59b60f 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -954,6 +954,10 @@ type DefaultTTLOption uint8 // classic BPF filter on a given endpoint. type SocketDetachFilterOption int +// OriginalDestinationOption is used to get the original destination address +// and port of a redirected packet. +type OriginalDestinationOption FullAddress + // IPPacketInfo is the message structure for IP_PKTINFO. // // +stateify savable diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 1798510bc..6e5e55b6f 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -1024,14 +1024,19 @@ func (e *endpoint) transitionToStateEstablishedLocked(h *handshake) { // delivered to this endpoint from the demuxer when the endpoint // is transitioned to StateClose. func (e *endpoint) transitionToStateCloseLocked() { - if e.EndpointState() == StateClose { + s := e.EndpointState() + if s == StateClose { return } + + if s.connected() { + e.stack.Stats().TCP.CurrentConnected.Decrement() + e.stack.Stats().TCP.EstablishedClosed.Increment() + } + // Mark the endpoint as fully closed for reads/writes. e.cleanupLocked() e.setEndpointState(StateClose) - e.stack.Stats().TCP.CurrentConnected.Decrement() - e.stack.Stats().TCP.EstablishedClosed.Increment() } // tryDeliverSegmentFromClosedEndpoint attempts to deliver the parsed diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index 0f7487963..682687ebe 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -2017,6 +2017,17 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { *o = tcpip.TCPDeferAcceptOption(e.deferAccept) e.UnlockUser() + case *tcpip.OriginalDestinationOption: + ipt := e.stack.IPTables() + addr, port, err := ipt.OriginalDst(e.ID) + if err != nil { + return err + } + *o = tcpip.OriginalDestinationOption{ + Addr: addr, + Port: port, + } + default: return tcpip.ErrUnknownProtocolOption } diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index e67ec42b1..fb25b86b9 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -146,6 +146,24 @@ func TestActiveFailedConnectionAttemptIncrement(t *testing.T) { } } +func TestCloseWithoutConnect(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() + + // Create TCP endpoint. + var err *tcpip.Error + c.EP, err = c.Stack().NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &c.WQ) + if err != nil { + t.Fatalf("NewEndpoint failed: %s", err) + } + + c.EP.Close() + + if got := c.Stack().Stats().TCP.CurrentConnected.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentConnected.Value() = %d, want = 0", got) + } +} + func TestTCPSegmentsSentIncrement(t *testing.T) { c := context.New(t, defaultMTU) defer c.Cleanup() @@ -1276,68 +1294,91 @@ func TestConnectBindToDevice(t *testing.T) { } } -func TestRstOnSynSent(t *testing.T) { - c := context.New(t, defaultMTU) - defer c.Cleanup() +func TestSynSent(t *testing.T) { + for _, test := range []struct { + name string + reset bool + }{ + {"RstOnSynSent", true}, + {"CloseOnSynSent", false}, + } { + t.Run(test.name, func(t *testing.T) { + c := context.New(t, defaultMTU) + defer c.Cleanup() - // Create an endpoint, don't handshake because we want to interfere with the - // handshake process. - c.Create(-1) + // Create an endpoint, don't handshake because we want to interfere with the + // handshake process. + c.Create(-1) - // Start connection attempt. - waitEntry, ch := waiter.NewChannelEntry(nil) - c.WQ.EventRegister(&waitEntry, waiter.EventOut) - defer c.WQ.EventUnregister(&waitEntry) + // Start connection attempt. + waitEntry, ch := waiter.NewChannelEntry(nil) + c.WQ.EventRegister(&waitEntry, waiter.EventOut) + defer c.WQ.EventUnregister(&waitEntry) - addr := tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort} - if err := c.EP.Connect(addr); err != tcpip.ErrConnectStarted { - t.Fatalf("got Connect(%+v) = %s, want %s", addr, err, tcpip.ErrConnectStarted) - } + addr := tcpip.FullAddress{Addr: context.TestAddr, Port: context.TestPort} + if err := c.EP.Connect(addr); err != tcpip.ErrConnectStarted { + t.Fatalf("got Connect(%+v) = %s, want %s", addr, err, tcpip.ErrConnectStarted) + } - // Receive SYN packet. - b := c.GetPacket() - checker.IPv4(t, b, - checker.TCP( - checker.DstPort(context.TestPort), - checker.TCPFlags(header.TCPFlagSyn), - ), - ) + // Receive SYN packet. + b := c.GetPacket() + checker.IPv4(t, b, + checker.TCP( + checker.DstPort(context.TestPort), + checker.TCPFlags(header.TCPFlagSyn), + ), + ) - // Ensure that we've reached SynSent state - if got, want := tcp.EndpointState(c.EP.State()), tcp.StateSynSent; got != want { - t.Fatalf("got State() = %s, want %s", got, want) - } - tcpHdr := header.TCP(header.IPv4(b).Payload()) - c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) + if got, want := tcp.EndpointState(c.EP.State()), tcp.StateSynSent; got != want { + t.Fatalf("got State() = %s, want %s", got, want) + } + tcpHdr := header.TCP(header.IPv4(b).Payload()) + c.IRS = seqnum.Value(tcpHdr.SequenceNumber()) - // Send a packet with a proper ACK and a RST flag to cause the socket - // to Error and close out - iss := seqnum.Value(789) - rcvWnd := seqnum.Size(30000) - c.SendPacket(nil, &context.Headers{ - SrcPort: tcpHdr.DestinationPort(), - DstPort: tcpHdr.SourcePort(), - Flags: header.TCPFlagRst | header.TCPFlagAck, - SeqNum: iss, - AckNum: c.IRS.Add(1), - RcvWnd: rcvWnd, - TCPOpts: nil, - }) + if test.reset { + // Send a packet with a proper ACK and a RST flag to cause the socket + // to error and close out. + iss := seqnum.Value(789) + rcvWnd := seqnum.Size(30000) + c.SendPacket(nil, &context.Headers{ + SrcPort: tcpHdr.DestinationPort(), + DstPort: tcpHdr.SourcePort(), + Flags: header.TCPFlagRst | header.TCPFlagAck, + SeqNum: iss, + AckNum: c.IRS.Add(1), + RcvWnd: rcvWnd, + TCPOpts: nil, + }) + } else { + c.EP.Close() + } - // Wait for receive to be notified. - select { - case <-ch: - case <-time.After(3 * time.Second): - t.Fatal("timed out waiting for packet to arrive") - } + // Wait for receive to be notified. + select { + case <-ch: + case <-time.After(3 * time.Second): + t.Fatal("timed out waiting for packet to arrive") + } - if _, _, err := c.EP.Read(nil); err != tcpip.ErrConnectionRefused { - t.Fatalf("got c.EP.Read(nil) = %s, want = %s", err, tcpip.ErrConnectionRefused) - } + if test.reset { + if _, _, err := c.EP.Read(nil); err != tcpip.ErrConnectionRefused { + t.Fatalf("got c.EP.Read(nil) = %s, want = %s", err, tcpip.ErrConnectionRefused) + } + } else { + if _, _, err := c.EP.Read(nil); err != tcpip.ErrAborted { + t.Fatalf("got c.EP.Read(nil) = %s, want = %s", err, tcpip.ErrAborted) + } + } - // Due to the RST the endpoint should be in an error state. - if got, want := tcp.EndpointState(c.EP.State()), tcp.StateError; got != want { - t.Fatalf("got State() = %s, want %s", got, want) + if got := c.Stack().Stats().TCP.CurrentConnected.Value(); got != 0 { + t.Errorf("got stats.TCP.CurrentConnected.Value() = %d, want = 0", got) + } + + // Due to the RST the endpoint should be in an error state. + if got, want := tcp.EndpointState(c.EP.State()), tcp.StateError; got != want { + t.Fatalf("got State() = %s, want %s", got, want) + } + }) } } |