net: add a READ_ONCE() in skb_peek_tail()
authorEric Dumazet <edumazet@google.com>
Fri, 8 Nov 2019 02:49:43 +0000 (18:49 -0800)
committerDavid S. Miller <davem@davemloft.net>
Fri, 8 Nov 2019 04:08:14 +0000 (20:08 -0800)
skb_peek_tail() can be used without protection of a lock,
as spotted by KCSAN [1]

In order to avoid load-stearing, add a READ_ONCE()

Note that the corresponding WRITE_ONCE() are already there.

[1]
BUG: KCSAN: data-race in sk_wait_data / skb_queue_tail

read to 0xffff8880b36a4118 of 8 bytes by task 20426 on cpu 1:
 skb_peek_tail include/linux/skbuff.h:1784 [inline]
 sk_wait_data+0x15b/0x250 net/core/sock.c:2477
 kcm_wait_data+0x112/0x1f0 net/kcm/kcmsock.c:1103
 kcm_recvmsg+0xac/0x320 net/kcm/kcmsock.c:1130
 sock_recvmsg_nosec net/socket.c:871 [inline]
 sock_recvmsg net/socket.c:889 [inline]
 sock_recvmsg+0x92/0xb0 net/socket.c:885
 ___sys_recvmsg+0x1a0/0x3e0 net/socket.c:2480
 do_recvmmsg+0x19a/0x5c0 net/socket.c:2601
 __sys_recvmmsg+0x1ef/0x200 net/socket.c:2680
 __do_sys_recvmmsg net/socket.c:2703 [inline]
 __se_sys_recvmmsg net/socket.c:2696 [inline]
 __x64_sys_recvmmsg+0x89/0xb0 net/socket.c:2696
 do_syscall_64+0xcc/0x370 arch/x86/entry/common.c:290
 entry_SYSCALL_64_after_hwframe+0x44/0xa9

write to 0xffff8880b36a4118 of 8 bytes by task 451 on cpu 0:
 __skb_insert include/linux/skbuff.h:1852 [inline]
 __skb_queue_before include/linux/skbuff.h:1958 [inline]
 __skb_queue_tail include/linux/skbuff.h:1991 [inline]
 skb_queue_tail+0x7e/0xc0 net/core/skbuff.c:3145
 kcm_queue_rcv_skb+0x202/0x310 net/kcm/kcmsock.c:206
 kcm_rcv_strparser+0x74/0x4b0 net/kcm/kcmsock.c:370
 __strp_recv+0x348/0xf50 net/strparser/strparser.c:309
 strp_recv+0x84/0xa0 net/strparser/strparser.c:343
 tcp_read_sock+0x174/0x5c0 net/ipv4/tcp.c:1639
 strp_read_sock+0xd4/0x140 net/strparser/strparser.c:366
 do_strp_work net/strparser/strparser.c:414 [inline]
 strp_work+0x9a/0xe0 net/strparser/strparser.c:423
 process_one_work+0x3d4/0x890 kernel/workqueue.c:2269
 worker_thread+0xa0/0x800 kernel/workqueue.c:2415
 kthread+0x1d4/0x200 drivers/block/aoe/aoecmd.c:1253
 ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:352

Reported by Kernel Concurrency Sanitizer on:
CPU: 0 PID: 451 Comm: kworker/u4:3 Not tainted 5.4.0-rc3+ #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Workqueue: kstrp strp_work

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/linux/skbuff.h

index 53238ac725a326a1f9c41f915c51b1e9cfdb0217..dfe02b658829d93b297fee0249c5997b3c10465b 100644 (file)
@@ -1795,7 +1795,7 @@ static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
  */
 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
 {
-       struct sk_buff *skb = list_->prev;
+       struct sk_buff *skb = READ_ONCE(list_->prev);
 
        if (skb == (struct sk_buff *)list_)
                skb = NULL;
@@ -1861,7 +1861,9 @@ static inline void __skb_insert(struct sk_buff *newsk,
                                struct sk_buff *prev, struct sk_buff *next,
                                struct sk_buff_head *list)
 {
-       /* see skb_queue_empty_lockless() for the opposite READ_ONCE() */
+       /* See skb_queue_empty_lockless() and skb_peek_tail()
+        * for the opposite READ_ONCE()
+        */
        WRITE_ONCE(newsk->next, next);
        WRITE_ONCE(newsk->prev, prev);
        WRITE_ONCE(next->prev, newsk);