Support inotify in vfs2 gofer fs.

Because there is no inode structure stored in the sandbox, inotify watches
must be held on the dentry. This would be an issue in the presence of hard
links, where multiple dentries would need to share the same set of watches,
but in VFS2, we do not support the internal creation of hard links on gofer
fs. As a result, we make the assumption that every dentry corresponds to a
unique inode.

Furthermore, dentries can be cached and then evicted, even if the underlying
file has not be deleted. We must prevent this from occurring if there are any
watches that would be lost. Note that if the dentry was deleted or invalidated
(d.vfsd.IsDead()), we should still destroy it along with its watches.

Additionally, when a dentry’s last watch is removed, we cache it if it also
has zero references. This way, the dentry can eventually be evicted from
memory if it is no longer needed. This is accomplished with a new dentry
method, OnZeroWatches(), which is called by Inotify.RmWatch and
Inotify.Release. Note that it must be called after all inotify locks are
released to avoid violating lock order. Stress tests are added to make sure
that inotify operations don't deadlock with gofer.OnZeroWatches.

Updates #1479.

PiperOrigin-RevId: 317958034

1 files changed, 29 insertions, 3 deletions

diff --git a/pkg/sentry/vfs/dentry.go b/pkg/sentry/vfs/dentry.go
index 24af13eb1..cea3e6955 100644
--- a/pkg/sentry/vfs/dentry.go
+++ b/pkg/sentry/vfs/dentry.go
@@ -113,12 +113,29 @@ type DentryImpl interface {
 	//
 	// Note that the events may not actually propagate up to the user, depending
 	// on the event masks.
-	InotifyWithParent(events uint32, cookie uint32, et EventType)
+	InotifyWithParent(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
 	// share the same watches.
+	//
+	// Watches may return nil if the dentry belongs to a FilesystemImpl that
+	// does not support inotify. If an implementation returns a non-nil watch
+	// set, it must always return a non-nil watch set. Likewise, if an
+	// implementation returns a nil watch set, it must always return a nil watch
+	// set.
+	//
+	// The caller does not need to hold a reference on the dentry.
 	Watches() *Watches
+
+	// OnZeroWatches is called whenever the number of watches on a dentry drops
+	// to zero. This is needed by some FilesystemImpls (e.g. gofer) to manage
+	// dentry lifetime.
+	//
+	// 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()
 }
 
 // IncRef increments d's reference count.
@@ -149,17 +166,26 @@ func (d *Dentry) isMounted() bool {
 	return atomic.LoadUint32(&d.mounts) != 0
 }
 
-// InotifyWithParent notifies all watches on the inodes for this dentry and
+// InotifyWithParent notifies all watches on the targets represented by d and
 // its parent of events.
-func (d *Dentry) InotifyWithParent(events uint32, cookie uint32, et EventType) {
+func (d *Dentry) InotifyWithParent(events, cookie uint32, et EventType) {
 	d.impl.InotifyWithParent(events, cookie, et)
 }
 
 // Watches returns the set of inotify watches associated with d.
+//
+// Watches will return nil if d belongs to a FilesystemImpl that does not
+// support inotify.
 func (d *Dentry) Watches() *Watches {
 	return d.impl.Watches()
 }
 
+// OnZeroWatches performs cleanup tasks whenever the number of watches on a
+// dentry drops to zero.
+func (d *Dentry) OnZeroWatches() {
+	d.impl.OnZeroWatches()
+}
+
 // The following functions are exported so that filesystem implementations can
 // use them. The vfs package, and users of VFS, should not call these
 // functions.