From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: [PATCH] tproxy: nf_tproxy_assign_sock() can handle tw sockets Date: Wed, 14 Jul 2010 05:21:18 +0200 Message-ID: <1279077678.2444.95.camel@edumazet-laptop> References: <1278626921.2435.73.camel@edumazet-laptop> <1278695580.2696.55.camel@edumazet-laptop> <1278742649.2538.17.camel@edumazet-laptop> <4C395459.6080407@redhat.com> <1278835332.2538.51.camel@edumazet-laptop> <1279032023.2634.384.camel@edumazet-laptop> <1279036193.2634.468.camel@edumazet-laptop> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Avi Kivity , David Miller , Patrick McHardy , linux-kernel@vger.kernel.org, netdev To: Felipe W Damasio Return-path: In-Reply-To: Sender: linux-kernel-owner@vger.kernel.org List-Id: netdev.vger.kernel.org Le mardi 13 juillet 2010 =C3=A0 17:55 -0300, Felipe W Damasio a =C3=A9c= rit : > Hi Mr. Dumazet, >=20 > I used the patched kernel on the production machine and squid frooze = again. >=20 > This is the dmesg message: >=20 >=20 > general protection fault: 0000 [#1] SMP > last sysfs file: /sys/devices/pci0000:00/0000:00:1f.3/i2c-0/name > CPU 1 > Modules linked in: >=20 > Pid: 5533, comm: squid Not tainted 2.6.34 #6 DX58SO/ > RIP: 0010:[] [] sock_rfree+0x26/= 0x37 > RSP: 0018:ffff88042287fc20 EFLAGS: 00010206 > RAX: 66c86f938964c696 RBX: ffff88034e8f9a00 RCX: 0000000000000720 > RDX: ffff8803f0ce05c0 RSI: ffff8803d441960c RDI: ffff88034e8f9a00 > RBP: ffff8803f0ee05c0 R08: ffffea000dcb9998 R09: 0000000000000000 > R10: 000000000003d830 R11: ffff8803f0ee05c0 R12: 00000000000005a8 > R13: 00000000000005a8 R14: 0000000000004378 R15: 0000000000000000 > FS: 00007f4cf33ee710(0000) GS:ffff880001840000(0000) knlGS:000000000= 0000000 > CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 > CR2: 00000000021d5fd0 CR3: 0000000422872000 CR4: 00000000000006e0 > DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 > DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400 > Process squid (pid: 5533, threadinfo ffff88042287e000, task ffff88042= eb61a40) > Stack: > ffffffff8136ecda ffff88034e8f9a00 ffffffff8136ea8c ffff88034e8f9a00 > <0> ffffffff813ab142 00000000000000d0 ffffffff8136f9f9 000000000eec60= e2 > <0> ffff88042eb61a40 ffff88042eb61a40 ffff88042eb61a40 00000000edca73= 00 > Call Trace: > [] ? skb_release_head_state+0x6d/0xb7 > [] ? __kfree_skb+0x9/0x7d > [] ? tcp_recvmsg+0x6a3/0x89a > [] ? __alloc_skb+0x5e/0x14e > [] ? sock_common_recvmsg+0x30/0x45 > [] ? sock_aio_read+0xdd/0xf1 > [] ? tcp_write_xmit+0x93e/0x96c > [] ? do_sync_read+0xb0/0xf2 > [] ? vfs_read+0xb9/0xff > [] ? sys_read+0x45/0x6e > [] ? system_call_fastpath+0x16/0x1b > Code: ff ff ff ff c3 48 8b 57 18 8b 87 d8 00 00 00 48 8d 8a ac 00 00 > 00 f0 29 82 ac 00 00 00 48 8b 57 18 8b 8f d8 00 00 00 48 8b 42 38 <48= > > 83 b8 b0 00 00 00 00 74 06 01 8a f4 00 00 00 c3 41 57 41 89 > RIP [] sock_rfree+0x26/0x37 > RSP > ---[ end trace 22e6ca9ef825c0e6 ]--- >=20 >=20 > Seems to be the same issue, right? >=20 Exactly the same. Only RAX value is different, its another chain. BTW, 0x720 is not skb->len like I said earlier, but skb->truesize, and 0x720 is OK on a 64 bit machine for a regular packet. 48 8b 57 18 mov 0x18(%rdi),%rdx skb->sk 8b 87 d8 00 00 00 mov 0xd8(%rdi),%eax skb->truesize 48 8d 8a ac 00 00 00 lea 0xac(%rdx),%rcx f0 29 82 ac 00 00 00 lock sub %eax,0xac(%rdx) 48 8b 57 18 mov 0x18(%rdi),%rdx skb->sk 8b 8f d8 00 00 00 mov 0xd8(%rdi),%ecx skb->truesize 48 8b 42 38 mov 0x38(%rdx),%rax sk->sk_prot <48> 83 b8 b0 00 00 00 00 cmpq $0x0,0xb0(%rax) 74 06 je .+6 01 8a fa 00 00 00 add %ecx,0xfa(%rdx) One thing to notice are the RDX and RBP values: RDX: ffff8803f0ce05c0=20 RBP: ffff8803f0ee05c0 RDX being the sk pointer (and sk+0x38 contains the corrupted "sk_prot" = value) , we notice RBP contains same "sk" value + 0x200000 (2 Mbytes). (same remark on your initial bug report) Could you enable CONFIG_FRAME_POINTER=3Dy in your config ?