Re: general protection fault in requeue_rx_msgs

[Kirill Tkhai]

On 31.05.2018 11:16, syzbot wrote:
> Hello,
> 
> syzbot found the following crash on:
> 
> HEAD commit:ÂÂÂ 0044cdeb7313 Merge branch 'for-linus' of git://git.kernel...
> git tree:ÂÂÂÂÂÂ upstream
> console output: https://syzkaller.appspot.com/x/log.txt?x=15aeff0f800000
> kernel config:Â https://syzkaller.appspot.com/x/.config?x=968b0b23c7854c0b
> dashboard link: https://syzkaller.appspot.com/bug?extid=554266c04a41d1f9754d
> compiler:ÂÂÂÂÂÂ gcc (GCC) 8.0.1 20180413 (experimental)
> syzkaller repro:https://syzkaller.appspot.com/x/repro.syz?x=131a208f800000
> 
> IMPORTANT: if you fix the bug, please add the following tag to the commit:
> Reported-by: syzbot+554266c04a41d1f9754d@xxxxxxxxxxxxxxxxxxxxxxxxx
> 
> kasan: CONFIG_KASAN_INLINE enabled
> kasan: GPF could be caused by NULL-ptr deref or user memory access
> general protection fault: 0000 [#1] SMP KASAN
> Dumping ftrace buffer:
> ÂÂ (ftrace buffer empty)
> Modules linked in:
> CPU: 0 PID: 4788 Comm: kworker/u4:3 Not tainted 4.17.0-rc7+ #74
> Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
> Workqueue: kstrp strp_work
> RIP: 0010:__skb_unlink include/linux/skbuff.h:1844 [inline]
> RIP: 0010:__skb_dequeue include/linux/skbuff.h:1861 [inline]
> RIP: 0010:requeue_rx_msgs+0x14d/0x620 net/kcm/kcmsock.c:226
> RSP: 0018:ffff8801aa97f0b8 EFLAGS: 00010202
> RAX: 0000000000000000 RBX: dffffc0000000000 RCX: ffffffff86d54ed3
> RDX: 0000000000000001 RSI: ffffffff86d531e2 RDI: 0000000000000008
> RBP: ffff8801aa97f1b8 R08: ffff8801aaa8e3c0 R09: ffffed0035f0a0e8
> R10: ffffed0035f0a0e8 R11: ffff8801af850743 R12: ffff8801d4407000
> R13: ffffed003552fe22 R14: 0000000000000000 R15: ffff8801a6bb06c0
> FS:Â 0000000000000000(0000) GS:ffff8801dae00000(0000) knlGS:0000000000000000
> CS:Â 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> CR2: 00007fba7c5099a0 CR3: 00000001af6f6000 CR4: 00000000001406f0
> DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
> Call Trace:
> Âunreserve_rx_kcm+0x471/0x520 net/kcm/kcmsock.c:334
> Âkcm_rcv_strparser+0x109/0x8d0 net/kcm/kcmsock.c:375
> Â__strp_recv+0x34b/0x2130 net/strparser/strparser.c:328
> Âstrp_recv+0xcf/0x110 net/strparser/strparser.c:362
> Âtcp_read_sock+0x2aa/0x810 net/ipv4/tcp.c:1652
> Âstrp_read_sock+0x1a1/0x2d0 net/strparser/strparser.c:385
> Âdo_strp_work net/strparser/strparser.c:440 [inline]
> Âstrp_work+0xcd/0x120 net/strparser/strparser.c:449
> Âprocess_one_work+0xc1e/0x1b50 kernel/workqueue.c:2145
> Âworker_thread+0x1cc/0x1440 kernel/workqueue.c:2279
> Âkthread+0x345/0x410 kernel/kthread.c:240
> Âret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:412
> Code: 80 3c 1a 00 0f 85 70 04 00 00 48 8d 78 08 4d 8b 74 24 08 49 c7 04 24 00 00 00 00 49 c7 44 24 08 00 00 00 00 48 89 fa 48 c1 ea 03 <80> 3c 1a 00 0f 85 a7 04 00 00 4c 89 f2 4c 89 70 08 48 c1 ea 03
> RIP: __skb_unlink include/linux/skbuff.h:1844 [inline] RSP: ffff8801aa97f0b8
> RIP: __skb_dequeue include/linux/skbuff.h:1861 [inline] RSP: ffff8801aa97f0b8
> RIP: requeue_rx_msgs+0x14d/0x620 net/kcm/kcmsock.c:226 RSP: ffff8801aa97f0b8
> ---[ end trace e4c0e45094907eaa ]---

This looks like the same as syzbot+5f1a04e374a635efc426@xxxxxxxxxxxxxxxxxxxxxxxxx:
"WARNING in kcm_exit_net (3)". This may confirm the theory. It looks like after async
pernet_operations the race window became bigger, and now the work has more chances
to have no a time to complete.

I'm not close to this code. Tom, could you please to say, whether kcm_done_work()
can be called for in-kernel kcm sockets (created via kcm_clone())?

Also, is there a possibility to create !kernel socket in kcm_clone()? I forgot
the reasons, why we can't do that in some places.

Thanks,
Kirill