WARNING: at net/ipv4/tcp_input.c:3418

Resending this mail because of earlier failure...


On 25 February 2012 10:34, Markus Trippelsdorf [off-list ref] wrote:

I've hit the following warnings today while using bittorrent:

------------[ cut here ]------------
WARNING: at net/ipv4/tcp_input.c:3418 tcp_ack+0xf8f/0x11a0()
Hardware name: System Product Name
Pid: 14513, comm: cc1plus Not tainted 3.3.0-rc4-00091-gbb4c7e9-dirty #3
Call Trace:
 <IRQ>  [<ffffffff81055c40>] ? warn_slowpath_common+0x60/0xa0
 [<ffffffff8142430f>] ? tcp_ack+0xf8f/0x11a0
 [<ffffffff81424e4d>] ? tcp_rcv_established+0x28d/0x720
 [<ffffffff8142c4ed>] ? tcp_v4_do_rcv+0x8d/0x2a0
 [<ffffffff8142e0d9>] ? tcp_v4_rcv+0x5f9/0x720
 [<ffffffff8140dcbd>] ? ip_local_deliver_finish+0x5d/0x140
 [<ffffffff813d9af6>] ? __netif_receive_skb+0x2d6/0x3a0
 [<ffffffff813dcf9a>] ? netif_receive_skb+0x1a/0x80
 [<ffffffff813d0d79>] ? __netdev_alloc_skb+0x19/0x40
 [<ffffffff8131baec>] ? atl1e_clean+0x1ac/0x3c0
 [<ffffffff813dd9d9>] ? net_rx_action+0x79/0x120
 [<ffffffff8105b3ed>] ? __do_softirq+0x8d/0x120
 [<ffffffff810935f9>] ? handle_irq_event_percpu+0x79/0x140
 [<ffffffff8147928c>] ? call_softirq+0x1c/0x26
 [<ffffffff810341bd>] ? do_softirq+0x3d/0x80
 [<ffffffff8105b6e5>] ? irq_exit+0xa5/0xc0
 [<ffffffff8103408d>] ? do_IRQ+0x4d/0xc0
 [<ffffffff81477a2b>] ? common_interrupt+0x6b/0x6b
 <EOI>
---[ end trace 6bd1602ca76fa623 ]---
------------[ cut here ]------------
WARNING: at net/ipv4/tcp_input.c:3048 tcp_fastretrans_alert+0xe9a/0xee0()
Hardware name: System Product Name
Pid: 14513, comm: cc1plus Tainted: G        W    3.3.0-rc4-00091-gbb4c7e9-dirty #3
Call Trace:
 <IRQ>  [<ffffffff81055c40>] ? warn_slowpath_common+0x60/0xa0
 [<ffffffff8142315a>] ? tcp_fastretrans_alert+0xe9a/0xee0
 [<ffffffff814238ac>] ? tcp_ack+0x52c/0x11a0
 [<ffffffff81424e4d>] ? tcp_rcv_established+0x28d/0x720
 [<ffffffff8142c4ed>] ? tcp_v4_do_rcv+0x8d/0x2a0
 [<ffffffff8142e0d9>] ? tcp_v4_rcv+0x5f9/0x720
 [<ffffffff8140dcbd>] ? ip_local_deliver_finish+0x5d/0x140
 [<ffffffff813d9af6>] ? __netif_receive_skb+0x2d6/0x3a0
 [<ffffffff813dcf9a>] ? netif_receive_skb+0x1a/0x80
 [<ffffffff813d0d79>] ? __netdev_alloc_skb+0x19/0x40
 [<ffffffff8131baec>] ? atl1e_clean+0x1ac/0x3c0
 [<ffffffff813dd9d9>] ? net_rx_action+0x79/0x120
 [<ffffffff8105b3ed>] ? __do_softirq+0x8d/0x120
 [<ffffffff810935f9>] ? handle_irq_event_percpu+0x79/0x140
 [<ffffffff8147928c>] ? call_softirq+0x1c/0x26
 [<ffffffff810341bd>] ? do_softirq+0x3d/0x80
 [<ffffffff8105b6e5>] ? irq_exit+0xa5/0xc0
 [<ffffffff8103408d>] ? do_IRQ+0x4d/0xc0
 [<ffffffff81477a2b>] ? common_interrupt+0x6b/0x6b
 <EOI>
---[ end trace 6bd1602ca76fa624 ]---

 Hi Markus,
This looks similar to the errors that were reported in the following threads
http://lists.openwall.net/netdev/2012/02/06/13
http://lists.openwall.net/netdev/2012/02/06/23

Maybe you are hitting the same bug. It was hoped in the discussion
that the following commit would fix it but looks like it does not
(assuming it is the same bug). From your kernel version  (3.3.0-rc4),
it looks like you should have this commit. So, maybe it did not fix
this bug.


commit 5b35e1e6e9ca651e6b291c96d1106043c9af314a
Author: Neal Cardwell [off-list ref]
Date:   Sat Jan 28 17:29:46 2012 +0000

   tcp: fix tcp_trim_head() to adjust segment count with skb MSS


CCing Neal and Ilpo.

Regards,
Vijay