Thread (6 messages) flat view 6 messages, 3 authors, 2017-03-07

Re: [PATCH net 2/2] net: fix socket refcounting in skb_complete_tx_timestamp()

From: Soheil Hassas Yeganeh <hidden>
Date: 2017-03-04 05:51:54

On Fri, Mar 3, 2017 at 7:01 PM, Eric Dumazet [off-list ref] wrote:
TX skbs do not necessarily hold a reference on skb->sk->sk_refcnt
By the time TX completion happens, sk_refcnt might be already 0.

sock_hold()/sock_put() would then corrupt critical state, like
sk_wmem_alloc and lead to leaks or use after free.

Fixes: 62bccb8cdb69 ("net-timestamp: Make the clone operation stand-alone from phy timestamping")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Alexander Duyck <redacted>
Cc: Johannes Berg <johannes@sipsolutions.net>
Cc: Soheil Hassas Yeganeh <redacted>
Cc: Willem de Bruijn <willemb@google.com>
Acked-by: Soheil Hassas Yeganeh <redacted>
quoted hunk ↗ jump to hunk
---
 net/core/skbuff.c | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index e2f37a560ec43ccf60a71f190423bd265eccf594..cd4ba8c6b6091651403cf74de8c60ccf69aa3e7b 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -3828,13 +3828,14 @@ void skb_complete_tx_timestamp(struct sk_buff *skb,
        if (!skb_may_tx_timestamp(sk, false))
                return;

-       /* take a reference to prevent skb_orphan() from freeing the socket */
-       sock_hold(sk);
-
-       *skb_hwtstamps(skb) = *hwtstamps;
-       __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND);
-
-       sock_put(sk);
+       /* Take a reference to prevent skb_orphan() from freeing the socket,
+        * but only if the socket refcount is not zero.
+        */
+       if (likely(atomic_inc_not_zero(&sk->sk_refcnt))) {
+               *skb_hwtstamps(skb) = *hwtstamps;
+               __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND);
+               sock_put(sk);
+       }
 }
 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);

--
2.12.0.rc1.440.g5b76565f74-goog
Thanks for the fix!
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help