Fix socket record leak in VFS2

VFS2 socket record is not removed from the system-wide
socket table when the socket is released, which will lead
to a memory leak. This patch fixes this issue.

Fixes: #3874

Signed-off-by: Tiwei Bie <tiwei.btw@antgroup.com>

3 files changed, 52 insertions, 20 deletions

diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD
index a89583dad..cc7408698 100644
--- a/pkg/sentry/socket/unix/BUILD
+++ b/pkg/sentry/socket/unix/BUILD
@@ -7,10 +7,21 @@ go_template_instance(
     name = "socket_refs",
     out = "socket_refs.go",
     package = "unix",
-    prefix = "socketOpsCommon",
+    prefix = "socketOperations",
     template = "//pkg/refs_vfs2:refs_template",
     types = {
-        "T": "socketOpsCommon",
+        "T": "SocketOperations",
+    },
+)
+
+go_template_instance(
+    name = "socket_vfs2_refs",
+    out = "socket_vfs2_refs.go",
+    package = "unix",
+    prefix = "socketVFS2",
+    template = "//pkg/refs_vfs2:refs_template",
+    types = {
+        "T": "SocketVFS2",
     },
 )
 
@@ -20,6 +31,7 @@ go_library(
         "device.go",
         "io.go",
         "socket_refs.go",
+        "socket_vfs2_refs.go",
         "unix.go",
         "unix_vfs2.go",
     ],
diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go
index 917055cea..f80011ce4 100644
--- a/pkg/sentry/socket/unix/unix.go
+++ b/pkg/sentry/socket/unix/unix.go
@@ -55,6 +55,7 @@ type SocketOperations struct {
 	fsutil.FileNoopFlush            `state:"nosave"`
 	fsutil.FileUseInodeUnstableAttr `state:"nosave"`
 
+	socketOperationsRefs
 	socketOpsCommon
 }
 
@@ -84,11 +85,27 @@ func NewWithDirent(ctx context.Context, d *fs.Dirent, ep transport.Endpoint, sty
 	return fs.NewFile(ctx, d, flags, &s)
 }
 
+// DecRef implements RefCounter.DecRef.
+func (s *SocketOperations) DecRef(ctx context.Context) {
+	s.socketOperationsRefs.DecRef(func() {
+		s.ep.Close(ctx)
+		if s.abstractNamespace != nil {
+			s.abstractNamespace.Remove(s.abstractName, s)
+		}
+	})
+}
+
+// Release implemements fs.FileOperations.Release.
+func (s *SocketOperations) Release(ctx context.Context) {
+	// Release only decrements a reference on s because s may be referenced in
+	// the abstract socket namespace.
+	s.DecRef(ctx)
+}
+
 // socketOpsCommon contains the socket operations common to VFS1 and VFS2.
 //
 // +stateify savable
 type socketOpsCommon struct {
-	socketOpsCommonRefs
 	socket.SendReceiveTimeout
 
 	ep    transport.Endpoint
@@ -101,23 +118,6 @@ type socketOpsCommon struct {
 	abstractNamespace *kernel.AbstractSocketNamespace
 }
 
-// DecRef implements RefCounter.DecRef.
-func (s *socketOpsCommon) DecRef(ctx context.Context) {
-	s.socketOpsCommonRefs.DecRef(func() {
-		s.ep.Close(ctx)
-		if s.abstractNamespace != nil {
-			s.abstractNamespace.Remove(s.abstractName, s)
-		}
-	})
-}
-
-// Release implemements fs.FileOperations.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(ctx)
-}
-
 func (s *socketOpsCommon) isPacket() bool {
 	switch s.stype {
 	case linux.SOCK_DGRAM, linux.SOCK_SEQPACKET:
diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go
index 3688f22d2..59f6fe44d 100644
--- a/pkg/sentry/socket/unix/unix_vfs2.go
+++ b/pkg/sentry/socket/unix/unix_vfs2.go
@@ -43,6 +43,7 @@ type SocketVFS2 struct {
 	vfs.DentryMetadataFileDescriptionImpl
 	vfs.LockFD
 
+	socketVFS2Refs
 	socketOpsCommon
 }
 
@@ -88,6 +89,25 @@ func NewFileDescription(ep transport.Endpoint, stype linux.SockType, flags uint3
 	return vfsfd, nil
 }
 
+// DecRef implements RefCounter.DecRef.
+func (s *SocketVFS2) DecRef(ctx context.Context) {
+	s.socketVFS2Refs.DecRef(func() {
+		t := kernel.TaskFromContext(ctx)
+		t.Kernel().DeleteSocketVFS2(&s.vfsfd)
+		s.ep.Close(ctx)
+		if s.abstractNamespace != nil {
+			s.abstractNamespace.Remove(s.abstractName, s)
+		}
+	})
+}
+
+// Release implements vfs.FileDescriptionImpl.Release.
+func (s *SocketVFS2) Release(ctx context.Context) {
+	// Release only decrements a reference on s because s may be referenced in
+	// the abstract socket namespace.
+	s.DecRef(ctx)
+}
+
 // GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by
 // a transport.Endpoint.
 func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) {