4 files changed, 143 insertions, 23 deletions
diff --git a/pkg/sentry/fsimpl/tmpfs/inode_refs.go b/pkg/sentry/fsimpl/tmpfs/inode_refs.go
new file mode 100644
index 000000000..38eddde7e
--- /dev/null
+++ b/pkg/sentry/fsimpl/tmpfs/inode_refs.go
@@ -0,0 +1,111 @@
+package tmpfs
+
+import (
+	"gvisor.dev/gvisor/pkg/log"
+	refs_vfs1 "gvisor.dev/gvisor/pkg/refs"
+	"runtime"
+	"sync/atomic"
+)
+
+// ownerType is used to customize logging. Note that we use a pointer to T so
+// that we do not copy the entire object when passed as a format parameter.
+var inodeownerType *inode
+
+// Refs implements refs.RefCounter. It keeps a reference count using atomic
+// operations and calls the destructor when the count reaches zero.
+//
+// Note that the number of references is actually refCount + 1 so that a default
+// zero-value Refs object contains one reference.
+//
+// +stateify savable
+type inodeRefs struct {
+	// refCount is composed of two fields:
+	//
+	//	[32-bit speculative references]:[32-bit real references]
+	//
+	// Speculative references are used for TryIncRef, to avoid a CompareAndSwap
+	// loop. See IncRef, DecRef and TryIncRef for details of how these fields are
+	// used.
+	refCount int64
+}
+
+func (r *inodeRefs) finalize() {
+	var note string
+	switch refs_vfs1.GetLeakMode() {
+	case refs_vfs1.NoLeakChecking:
+		return
+	case refs_vfs1.UninitializedLeakChecking:
+		note = "(Leak checker uninitialized): "
+	}
+	if n := r.ReadRefs(); n != 0 {
+		log.Warningf("%sRefs %p owned by %T garbage collected with ref count of %d (want 0)", note, r, inodeownerType, n)
+	}
+}
+
+// EnableLeakCheck checks for reference leaks when Refs gets garbage collected.
+func (r *inodeRefs) EnableLeakCheck() {
+	if refs_vfs1.GetLeakMode() != refs_vfs1.NoLeakChecking {
+		runtime.SetFinalizer(r, (*inodeRefs).finalize)
+	}
+}
+
+// ReadRefs returns the current number of references. The returned count is
+// inherently racy and is unsafe to use without external synchronization.
+func (r *inodeRefs) ReadRefs() int64 {
+
+	return atomic.LoadInt64(&r.refCount) + 1
+}
+
+// IncRef implements refs.RefCounter.IncRef.
+//
+//go:nosplit
+func (r *inodeRefs) IncRef() {
+	if v := atomic.AddInt64(&r.refCount, 1); v <= 0 {
+		panic("Incrementing non-positive ref count")
+	}
+}
+
+// TryIncRef implements refs.RefCounter.TryIncRef.
+//
+// To do this safely without a loop, a speculative reference is first acquired
+// on the object. This allows multiple concurrent TryIncRef calls to distinguish
+// other TryIncRef calls from genuine references held.
+//
+//go:nosplit
+func (r *inodeRefs) TryIncRef() bool {
+	const speculativeRef = 1 << 32
+	v := atomic.AddInt64(&r.refCount, speculativeRef)
+	if int32(v) < 0 {
+
+		atomic.AddInt64(&r.refCount, -speculativeRef)
+		return false
+	}
+
+	atomic.AddInt64(&r.refCount, -speculativeRef+1)
+	return true
+}
+
+// DecRef implements refs.RefCounter.DecRef.
+//
+// Note that speculative references are counted here. Since they were added
+// prior to real references reaching zero, they will successfully convert to
+// real references. In other words, we see speculative references only in the
+// following case:
+//
+//	A: TryIncRef [speculative increase => sees non-negative references]
+//	B: DecRef [real decrease]
+//	A: TryIncRef [transform speculative to real]
+//
+//go:nosplit
+func (r *inodeRefs) DecRef(destroy func()) {
+	switch v := atomic.AddInt64(&r.refCount, -1); {
+	case v < -1:
+		panic("Decrementing non-positive ref count")
+
+	case v == -1:
+
+		if destroy != nil {
+			destroy()
+		}
+	}
+}
diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go
index 4681a2f52..de2af6d01 100644
--- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go
+++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go
@@ -285,13 +285,10 @@ type inode struct {
 	// fs is the owning filesystem. fs is immutable.
 	fs *filesystem
 
-	// refs is a reference count. refs is accessed using atomic memory
-	// operations.
-	//
 	// A reference is held on all inodes as long as they are reachable in the
 	// filesystem tree, i.e. nlink is nonzero. This reference is dropped when
 	// nlink reaches 0.
-	refs int64
+	refs inodeRefs
 
 	// xattrs implements extended attributes.
 	//
@@ -327,7 +324,6 @@ func (i *inode) init(impl interface{}, fs *filesystem, kuid auth.KUID, kgid auth
 		panic("file type is required in FileMode")
 	}
 	i.fs = fs
-	i.refs = 1
 	i.mode = uint32(mode)
 	i.uid = uint32(kuid)
 	i.gid = uint32(kgid)
@@ -339,6 +335,7 @@ func (i *inode) init(impl interface{}, fs *filesystem, kuid auth.KUID, kgid auth
 	i.mtime = now
 	// i.nlink initialized by caller
 	i.impl = impl
+	i.refs.EnableLeakCheck()
 }
 
 // incLinksLocked increments i's link count.
@@ -369,25 +366,15 @@ func (i *inode) decLinksLocked(ctx context.Context) {
 }
 
 func (i *inode) incRef() {
-	if atomic.AddInt64(&i.refs, 1) <= 1 {
-		panic("tmpfs.inode.incRef() called without holding a reference")
-	}
+	i.refs.IncRef()
 }
 
 func (i *inode) tryIncRef() bool {
-	for {
-		refs := atomic.LoadInt64(&i.refs)
-		if refs == 0 {
-			return false
-		}
-		if atomic.CompareAndSwapInt64(&i.refs, refs, refs+1) {
-			return true
-		}
-	}
+	return i.refs.TryIncRef()
 }
 
 func (i *inode) decRef(ctx context.Context) {
-	if refs := atomic.AddInt64(&i.refs, -1); refs == 0 {
+	i.refs.DecRef(func() {
 		i.watches.HandleDeletion(ctx)
 		if regFile, ok := i.impl.(*regularFile); ok {
 			// Release memory used by regFile to store data. Since regFile is
@@ -395,9 +382,7 @@ func (i *inode) decRef(ctx context.Context) {
 			// metadata.
 			regFile.data.DropAll(regFile.memFile)
 		}
-	} else if refs < 0 {
-		panic("tmpfs.inode.decRef() called without holding a reference")
-	}
+	})
 }
 
 func (i *inode) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error {
diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs_state_autogen.go b/pkg/sentry/fsimpl/tmpfs/tmpfs_state_autogen.go
index 1b617bd35..d88136656 100644
--- a/pkg/sentry/fsimpl/tmpfs/tmpfs_state_autogen.go
+++ b/pkg/sentry/fsimpl/tmpfs/tmpfs_state_autogen.go
@@ -58,7 +58,31 @@ func (x *dentryEntry) StateLoad(m state.Source) {
 	m.Load(1, &x.prev)
 }
 
+func (x *inodeRefs) StateTypeName() string {
+	return "pkg/sentry/fsimpl/tmpfs.inodeRefs"
+}
+
+func (x *inodeRefs) StateFields() []string {
+	return []string{
+		"refCount",
+	}
+}
+
+func (x *inodeRefs) beforeSave() {}
+
+func (x *inodeRefs) StateSave(m state.Sink) {
+	x.beforeSave()
+	m.Save(0, &x.refCount)
+}
+
+func (x *inodeRefs) afterLoad() {}
+
+func (x *inodeRefs) StateLoad(m state.Source) {
+	m.Load(0, &x.refCount)
+}
+
 func init() {
 	state.Register((*dentryList)(nil))
 	state.Register((*dentryEntry)(nil))
+	state.Register((*inodeRefs)(nil))
 }
diff --git a/pkg/sentry/platform/ring0/defs_impl_arm64.go b/pkg/sentry/platform/ring0/defs_impl_arm64.go
index 424b66f76..eba2eac30 100644
--- a/pkg/sentry/platform/ring0/defs_impl_arm64.go
+++ b/pkg/sentry/platform/ring0/defs_impl_arm64.go
@@ -3,11 +3,11 @@ package ring0
 import (
 	"gvisor.dev/gvisor/pkg/sentry/arch"
 	"gvisor.dev/gvisor/pkg/sentry/platform/ring0/pagetables"
+	"io"
+	"reflect"
 
 	"fmt"
 	"gvisor.dev/gvisor/pkg/usermem"
-	"io"
-	"reflect"
 )
 
 // Useful bits.