Port inotify to vfs2, with support in tmpfs.

Support in other filesystem impls is still needed. Unlike in Linux and vfs1, we
need to plumb inotify down to each filesystem implementation in order to keep
track of links/inode structures properly.

IN_EXCL_UNLINK still needs to be implemented, as well as a few inotify hooks
that are not present in either vfs1 or vfs2. Those will be addressed in
subsequent changes.

Updates #1479.

PiperOrigin-RevId: 313781995

7 files changed, 131 insertions, 7 deletions

diff --git a/pkg/sentry/fsimpl/ext/dentry.go b/pkg/sentry/fsimpl/ext/dentry.go
index bfbd7c3d4..4d0deaf03 100644
--- a/pkg/sentry/fsimpl/ext/dentry.go
+++ b/pkg/sentry/fsimpl/ext/dentry.go
@@ -60,3 +60,15 @@ func (d *dentry) DecRef() {
 	// inode.decRef().
 	d.inode.decRef()
 }
+
+// InotifyWithParent implements vfs.DentryImpl.InotifyWithParent.
+//
+// TODO(gvisor.dev/issue/1479): Implement inotify.
+func (d *dentry) InotifyWithParent(events uint32, cookie uint32) {}
+
+// Watches implements vfs.DentryImpl.Watches.
+//
+// TODO(gvisor.dev/issue/1479): Implement inotify.
+func (d *dentry) Watches() *vfs.Watches {
+	return nil
+}
diff --git a/pkg/sentry/fsimpl/gofer/gofer.go b/pkg/sentry/fsimpl/gofer/gofer.go
index 131da332f..850482a19 100644
--- a/pkg/sentry/fsimpl/gofer/gofer.go
+++ b/pkg/sentry/fsimpl/gofer/gofer.go
@@ -1039,6 +1039,18 @@ func (d *dentry) decRefLocked() {
 	}
 }
 
+// InotifyWithParent implements vfs.DentryImpl.InotifyWithParent.
+//
+// TODO(gvisor.dev/issue/1479): Implement inotify.
+func (d *dentry) InotifyWithParent(events uint32, cookie uint32) {}
+
+// Watches implements vfs.DentryImpl.Watches.
+//
+// TODO(gvisor.dev/issue/1479): Implement inotify.
+func (d *dentry) Watches() *vfs.Watches {
+	return nil
+}
+
 // checkCachingLocked should be called after d's reference count becomes 0 or it
 // becomes disowned.
 //
diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go
index a83151ad3..682545994 100644
--- a/pkg/sentry/fsimpl/kernfs/kernfs.go
+++ b/pkg/sentry/fsimpl/kernfs/kernfs.go
@@ -225,6 +225,18 @@ func (d *Dentry) destroy() {
 	}
 }
 
+// InotifyWithParent implements vfs.DentryImpl.InotifyWithParent.
+//
+// TODO(gvisor.dev/issue/1479): Implement inotify.
+func (d *Dentry) InotifyWithParent(events uint32, cookie uint32) {}
+
+// Watches implements vfs.DentryImpl.Watches.
+//
+// TODO(gvisor.dev/issue/1479): Implement inotify.
+func (d *Dentry) Watches() *vfs.Watches {
+	return nil
+}
+
 // 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
 // it's own isn't sufficient to insert a child into a directory. InsertChild
diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD
index 007be1572..062321cbc 100644
--- a/pkg/sentry/fsimpl/tmpfs/BUILD
+++ b/pkg/sentry/fsimpl/tmpfs/BUILD
@@ -59,6 +59,7 @@ go_library(
         "//pkg/sentry/pgalloc",
         "//pkg/sentry/platform",
         "//pkg/sentry/socket/unix/transport",
+        "//pkg/sentry/uniqueid",
         "//pkg/sentry/usage",
         "//pkg/sentry/vfs",
         "//pkg/sentry/vfs/lock",
diff --git a/pkg/sentry/fsimpl/tmpfs/directory.go b/pkg/sentry/fsimpl/tmpfs/directory.go
index f2399981b..8bc475f88 100644
--- a/pkg/sentry/fsimpl/tmpfs/directory.go
+++ b/pkg/sentry/fsimpl/tmpfs/directory.go
@@ -112,6 +112,7 @@ func (fd *directoryFD) IterDirents(ctx context.Context, cb vfs.IterDirentsCallba
 	dir.iterMu.Lock()
 	defer dir.iterMu.Unlock()
 
+	fd.dentry().InotifyWithParent(linux.IN_ACCESS, 0)
 	fd.inode().touchAtime(fd.vfsfd.Mount())
 
 	if fd.off == 0 {
diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go
index 7c04570f1..b4159f904 100644
--- a/pkg/sentry/fsimpl/tmpfs/filesystem.go
+++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go
@@ -177,6 +177,12 @@ func (fs *filesystem) doCreateAt(rp *vfs.ResolvingPath, dir bool, create func(pa
 	if err := create(parentDir, name); err != nil {
 		return err
 	}
+
+	ev := linux.IN_CREATE
+	if dir {
+		ev |= linux.IN_ISDIR
+	}
+	parentDir.inode.watches.Notify(name, uint32(ev), 0)
 	parentDir.inode.touchCMtime()
 	return nil
 }
@@ -241,6 +247,7 @@ func (fs *filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs.
 			return syserror.EMLINK
 		}
 		d.inode.incLinksLocked()
+		d.inode.watches.Notify("", linux.IN_ATTRIB, 0)
 		parentDir.insertChildLocked(fs.newDentry(d.inode), name)
 		return nil
 	})
@@ -354,6 +361,7 @@ afterTrailingSymlink:
 		if err != nil {
 			return nil, err
 		}
+		parentDir.inode.watches.Notify(name, linux.IN_CREATE, 0)
 		parentDir.inode.touchCMtime()
 		return fd, nil
 	}
@@ -559,6 +567,8 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa
 		newParentDir.inode.touchCMtime()
 	}
 	renamed.inode.touchCtime()
+
+	vfs.InotifyRename(ctx, &renamed.inode.watches, &oldParentDir.inode.watches, &newParentDir.inode.watches, oldName, newName, renamed.inode.isDir())
 	return nil
 }
 
@@ -603,6 +613,7 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error
 		return err
 	}
 	parentDir.removeChildLocked(child)
+	parentDir.inode.watches.Notify(name, linux.IN_DELETE|linux.IN_ISDIR, 0)
 	// Remove links for child, child/., and child/..
 	child.inode.decLinksLocked()
 	child.inode.decLinksLocked()
@@ -620,7 +631,14 @@ func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts
 	if err != nil {
 		return err
 	}
-	return d.inode.setStat(ctx, rp.Credentials(), &opts.Stat)
+	if err := d.inode.setStat(ctx, rp.Credentials(), &opts.Stat); err != nil {
+		return err
+	}
+
+	if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 {
+		d.InotifyWithParent(ev, 0)
+	}
+	return nil
 }
 
 // StatAt implements vfs.FilesystemImpl.StatAt.
@@ -700,6 +718,12 @@ func (fs *filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error
 	if err := vfsObj.PrepareDeleteDentry(mntns, &child.vfsd); err != nil {
 		return err
 	}
+
+	// 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)
+
 	parentDir.removeChildLocked(child)
 	child.inode.decLinksLocked()
 	vfsObj.CommitDeleteDentry(&child.vfsd)
@@ -756,7 +780,12 @@ func (fs *filesystem) SetxattrAt(ctx context.Context, rp *vfs.ResolvingPath, opt
 	if err != nil {
 		return err
 	}
-	return d.inode.setxattr(rp.Credentials(), &opts)
+	if err := d.inode.setxattr(rp.Credentials(), &opts); err != nil {
+		return err
+	}
+
+	d.InotifyWithParent(linux.IN_ATTRIB, 0)
+	return nil
 }
 
 // RemovexattrAt implements vfs.FilesystemImpl.RemovexattrAt.
@@ -767,7 +796,12 @@ func (fs *filesystem) RemovexattrAt(ctx context.Context, rp *vfs.ResolvingPath,
 	if err != nil {
 		return err
 	}
-	return d.inode.removexattr(rp.Credentials(), name)
+	if err := d.inode.removexattr(rp.Credentials(), name); err != nil {
+		return err
+	}
+
+	d.InotifyWithParent(linux.IN_ATTRIB, 0)
+	return nil
 }
 
 // PrependPath implements vfs.FilesystemImpl.PrependPath.
diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go
index b739095b7..1d83b6840 100644
--- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go
+++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go
@@ -201,6 +201,26 @@ func (d *dentry) DecRef() {
 	d.inode.decRef()
 }
 
+// InotifyWithParent implements vfs.DentryImpl.InotifyWithParent.
+func (d *dentry) InotifyWithParent(events uint32, cookie uint32) {
+	if d.inode.isDir() {
+		events |= linux.IN_ISDIR
+	}
+
+	// The ordering below is important, Linux always notifies the parent first.
+	if d.parent != nil {
+		// Note that d.parent or d.name may be stale if there is a concurrent
+		// rename operation. Inotify does not provide consistency guarantees.
+		d.parent.inode.watches.Notify(d.name, events, cookie)
+	}
+	d.inode.watches.Notify("", events, cookie)
+}
+
+// Watches implements vfs.DentryImpl.Watches.
+func (d *dentry) Watches() *vfs.Watches {
+	return &d.inode.watches
+}
+
 // inode represents a filesystem object.
 type inode struct {
 	// fs is the owning filesystem. fs is immutable.
@@ -236,6 +256,9 @@ type inode struct {
 	// Advisory file locks, which lock at the inode level.
 	locks lock.FileLocks
 
+	// Inotify watches for this inode.
+	watches vfs.Watches
+
 	impl interface{} // immutable
 }
 
@@ -257,6 +280,7 @@ func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials,
 	i.ctime = now
 	i.mtime = now
 	// i.nlink initialized by caller
+	i.watches = vfs.Watches{}
 	i.impl = impl
 }
 
@@ -307,6 +331,7 @@ func (i *inode) tryIncRef() bool {
 
 func (i *inode) decRef() {
 	if refs := atomic.AddInt64(&i.refs, -1); refs == 0 {
+		i.watches.HandleDeletion()
 		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
@@ -628,8 +653,12 @@ func (fd *fileDescription) filesystem() *filesystem {
 	return fd.vfsfd.Mount().Filesystem().Impl().(*filesystem)
 }
 
+func (fd *fileDescription) dentry() *dentry {
+	return fd.vfsfd.Dentry().Impl().(*dentry)
+}
+
 func (fd *fileDescription) inode() *inode {
-	return fd.vfsfd.Dentry().Impl().(*dentry).inode
+	return fd.dentry().inode
 }
 
 // Stat implements vfs.FileDescriptionImpl.Stat.
@@ -642,7 +671,16 @@ func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linu
 // SetStat implements vfs.FileDescriptionImpl.SetStat.
 func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error {
 	creds := auth.CredentialsFromContext(ctx)
-	return fd.inode().setStat(ctx, creds, &opts.Stat)
+	d := fd.dentry()
+	if err := d.inode.setStat(ctx, creds, &opts.Stat); err != nil {
+		return err
+	}
+
+	// Generate inotify events.
+	if ev := vfs.InotifyEventFromStatMask(opts.Stat.Mask); ev != 0 {
+		d.InotifyWithParent(ev, 0)
+	}
+	return nil
 }
 
 // Listxattr implements vfs.FileDescriptionImpl.Listxattr.
@@ -657,12 +695,26 @@ func (fd *fileDescription) Getxattr(ctx context.Context, opts vfs.GetxattrOption
 
 // Setxattr implements vfs.FileDescriptionImpl.Setxattr.
 func (fd *fileDescription) Setxattr(ctx context.Context, opts vfs.SetxattrOptions) error {
-	return fd.inode().setxattr(auth.CredentialsFromContext(ctx), &opts)
+	d := fd.dentry()
+	if err := d.inode.setxattr(auth.CredentialsFromContext(ctx), &opts); err != nil {
+		return err
+	}
+
+	// Generate inotify events.
+	d.InotifyWithParent(linux.IN_ATTRIB, 0)
+	return nil
 }
 
 // Removexattr implements vfs.FileDescriptionImpl.Removexattr.
 func (fd *fileDescription) Removexattr(ctx context.Context, name string) error {
-	return fd.inode().removexattr(auth.CredentialsFromContext(ctx), name)
+	d := fd.dentry()
+	if err := d.inode.removexattr(auth.CredentialsFromContext(ctx), name); err != nil {
+		return err
+	}
+
+	// Generate inotify events.
+	d.InotifyWithParent(linux.IN_ATTRIB, 0)
+	return nil
 }
 
 // NewMemfd creates a new tmpfs regular file and file description that can back