Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1471145 > unrolled thread

Re: kernel BUG at net/unix/garbage.c:149!"

Started byMiklos Szeredi <mszeredi@redhat.com>
First post2016-08-27 12:00 +0200
Last post2016-09-01 11:20 +0200
Articles 6 — 3 participants

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: kernel BUG at net/unix/garbage.c:149!" Miklos Szeredi <mszeredi@redhat.com> - 2016-08-27 12:00 +0200
    Re: kernel BUG at net/unix/garbage.c:149!" Miklos Szeredi <mszeredi@redhat.com> - 2016-08-30 00:40 +0200
      Re: kernel BUG at net/unix/garbage.c:149!" Miklos Szeredi <mszeredi@redhat.com> - 2016-08-30 11:20 +0200
        Re: kernel BUG at net/unix/garbage.c:149!" Nikolay Borisov <kernel@kyup.com> - 2016-08-30 11:40 +0200
          Re: kernel BUG at net/unix/garbage.c:149!" Miklos Szeredi <mszeredi@redhat.com> - 2016-08-30 11:50 +0200
        Re: kernel BUG at net/unix/garbage.c:149!" Hannes Frederic Sowa <hannes@stressinduktion.org> - 2016-09-01 11:20 +0200

#1471145 — Re: kernel BUG at net/unix/garbage.c:149!"

FromMiklos Szeredi <mszeredi@redhat.com>
Date2016-08-27 12:00 +0200
SubjectRe: kernel BUG at net/unix/garbage.c:149!"
Message-ID<saIz8-2Vv-15@gated-at.bofh.it>
On Wed, Aug 24, 2016 at 11:40 PM, Hannes Frederic Sowa
<hannes@stressinduktion.org> wrote:
> On 24.08.2016 16:24, Nikolay Borisov wrote:
>> Hello,
>>
>> I hit the following BUG:
>>
>> [1851513.239831] ------------[ cut here ]------------
>> [1851513.240079] kernel BUG at net/unix/garbage.c:149!
>> [1851513.240313] invalid opcode: 0000 [#1] SMP
>> [1851513.248320] CPU: 37 PID: 11683 Comm: nginx Tainted: G           O    4.4.14-clouder3 #26
>> [1851513.248719] Hardware name: Supermicro X10DRi/X10DRi, BIOS 1.1 04/14/2015
>> [1851513.248966] task: ffff883b0f6f0000 ti: ffff880189cf0000 task.ti: ffff880189cf0000
>> [1851513.249361] RIP: 0010:[<ffffffff815f895d>]  [<ffffffff815f895d>] unix_notinflight+0x8d/0x90
>> [1851513.249846] RSP: 0018:ffff880189cf3cf8  EFLAGS: 00010246
>> [1851513.250082] RAX: ffff883b05491968 RBX: ffff883b05491680 RCX: ffff8807f9967330
>> [1851513.250476] RDX: 0000000000000001 RSI: ffff882e6d8bae00 RDI: ffffffff82073f10
>> [1851513.250886] RBP: ffff880189cf3d08 R08: ffff880cbc70e200 R09: 0000000180200001
>> [1851513.251280] R10: ffff883fff3b9dc0 R11: ffffea0032f1c380 R12: ffff883fbaf50000
>> [1851513.251674] R13: ffffffff815f6354 R14: ffff881a7c77b140 R15: ffff881a7c7792c0
>> [1851513.252083] FS:  00007f4f19573720(0000) GS:ffff883fff3a0000(0000) knlGS:0000000000000000
>> [1851513.252481] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>> [1851513.252717] CR2: 00000000013062d8 CR3: 0000001712f32000 CR4: 00000000001406e0
>> [1851513.253116] Stack:
>> [1851513.253345]  00000000ffffffff ffff880189cf3d40 ffff880189cf3d28 ffffffff815f4383
>> [1851513.254022]  ffff8839ee11a800 ffff8839ee11a800 ffff880189cf3d60 ffffffff815f53b8
>> [1851513.254685]  0000000000000000 ffff883406788de0 0000000000000000 0000000000000000
>> [1851513.255360] Call Trace:
>> [1851513.255594]  [<ffffffff815f4383>] unix_detach_fds.isra.19+0x43/0x50
>> [1851513.255851]  [<ffffffff815f53b8>] unix_destruct_scm+0x48/0x80
>> [1851513.256090]  [<ffffffff815384af>] skb_release_head_state+0x4f/0xb0
>> [1851513.256328]  [<ffffffff81538522>] skb_release_all+0x12/0x30
>> [1851513.256564]  [<ffffffff81538592>] kfree_skb+0x32/0xa0
>> [1851513.256810]  [<ffffffff815f6354>] unix_release_sock+0x1e4/0x2c0
>> [1851513.257046]  [<ffffffff815f6450>] unix_release+0x20/0x30
>> [1851513.257284]  [<ffffffff8152fbcf>] sock_release+0x1f/0x80
>> [1851513.257521]  [<ffffffff8152fc42>] sock_close+0x12/0x20
>> [1851513.257769]  [<ffffffff8119a8aa>] __fput+0xea/0x1f0
>> [1851513.258005]  [<ffffffff8119a9ee>] ____fput+0xe/0x10
>> [1851513.258244]  [<ffffffff8106fccf>] task_work_run+0x7f/0xb0
>> [1851513.258488]  [<ffffffff81002210>] exit_to_usermode_loop+0xc0/0xd0
>> [1851513.258728]  [<ffffffff81002a90>] syscall_return_slowpath+0x80/0xf0
>> [1851513.258983]  [<ffffffff816147b4>] int_ret_from_sys_call+0x25/0x9f
>> [1851513.259222] Code: 7e 5b 41 5c 5d c3 48 8b 8b e8 02 00 00 48 8b 93 f0 02 00 00 48 89 51 08 48 89 0a 48 89 83 e8 02 00 00 48 89 83 f0 02 00 00 eb b8 <0f> 0b 90 0f 1f 44 00 00 55 48 c7 c7 10 3f 07 82 48 89 e5 41 54
>> [1851513.268473] RIP  [<ffffffff815f895d>] unix_notinflight+0x8d/0x90
>> [1851513.268793]  RSP <ffff880189cf3cf8>
>>
>> That's essentially BUG_ON(list_empty(&u->link));
>>
>> I see that all the code involving the ->link member hasn't really been
>> touched since it was introduced in 2007. So this must be a latent bug.
>> This is the first time I've observed it. The state
>> of the struct unix_sock can be found here http://sprunge.us/WCMW . Evidently,
>> there are no inflight sockets.

Weird.  If it was the BUG_ON(!list_empty(&u->link)) I'd understand,
because the code in scan children looks fishy: what prevents "embryos"
from fledging to full socket status and going in-flight?

But this one, I cannot imagine any scenario.

Can we have access to the crashdump?

Thanks,
Miklos

[toc] | [next] | [standalone]

#1472147

FromMiklos Szeredi <mszeredi@redhat.com>
Date2016-08-30 00:40 +0200
Message-ID<sbDnJ-4Pc-79@gated-at.bofh.it>
In reply to#1471145
On Sat, Aug 27, 2016 at 11:55 AM, Miklos Szeredi <mszeredi@redhat.com> wrote:
>
> On Wed, Aug 24, 2016 at 11:40 PM, Hannes Frederic Sowa
> <hannes@stressinduktion.org> wrote:
> > On 24.08.2016 16:24, Nikolay Borisov wrote:
> >> Hello,
> >>
> >> I hit the following BUG:
> >>
> >> [1851513.239831] ------------[ cut here ]------------
> >> [1851513.240079] kernel BUG at net/unix/garbage.c:149!
> >> [1851513.240313] invalid opcode: 0000 [#1] SMP
> >> [1851513.248320] CPU: 37 PID: 11683 Comm: nginx Tainted: G           O    4.4.14-clouder3 #26
> >> [1851513.248719] Hardware name: Supermicro X10DRi/X10DRi, BIOS 1.1 04/14/2015
> >> [1851513.248966] task: ffff883b0f6f0000 ti: ffff880189cf0000 task.ti: ffff880189cf0000
> >> [1851513.249361] RIP: 0010:[<ffffffff815f895d>]  [<ffffffff815f895d>] unix_notinflight+0x8d/0x90
> >> [1851513.249846] RSP: 0018:ffff880189cf3cf8  EFLAGS: 00010246
> >> [1851513.250082] RAX: ffff883b05491968 RBX: ffff883b05491680 RCX: ffff8807f9967330
> >> [1851513.250476] RDX: 0000000000000001 RSI: ffff882e6d8bae00 RDI: ffffffff82073f10
> >> [1851513.250886] RBP: ffff880189cf3d08 R08: ffff880cbc70e200 R09: 0000000180200001
> >> [1851513.251280] R10: ffff883fff3b9dc0 R11: ffffea0032f1c380 R12: ffff883fbaf50000
> >> [1851513.251674] R13: ffffffff815f6354 R14: ffff881a7c77b140 R15: ffff881a7c7792c0
> >> [1851513.252083] FS:  00007f4f19573720(0000) GS:ffff883fff3a0000(0000) knlGS:0000000000000000
> >> [1851513.252481] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> >> [1851513.252717] CR2: 00000000013062d8 CR3: 0000001712f32000 CR4: 00000000001406e0
> >> [1851513.253116] Stack:
> >> [1851513.253345]  00000000ffffffff ffff880189cf3d40 ffff880189cf3d28 ffffffff815f4383
> >> [1851513.254022]  ffff8839ee11a800 ffff8839ee11a800 ffff880189cf3d60 ffffffff815f53b8
> >> [1851513.254685]  0000000000000000 ffff883406788de0 0000000000000000 0000000000000000
> >> [1851513.255360] Call Trace:
> >> [1851513.255594]  [<ffffffff815f4383>] unix_detach_fds.isra.19+0x43/0x50
> >> [1851513.255851]  [<ffffffff815f53b8>] unix_destruct_scm+0x48/0x80
> >> [1851513.256090]  [<ffffffff815384af>] skb_release_head_state+0x4f/0xb0
> >> [1851513.256328]  [<ffffffff81538522>] skb_release_all+0x12/0x30
> >> [1851513.256564]  [<ffffffff81538592>] kfree_skb+0x32/0xa0
> >> [1851513.256810]  [<ffffffff815f6354>] unix_release_sock+0x1e4/0x2c0
> >> [1851513.257046]  [<ffffffff815f6450>] unix_release+0x20/0x30
> >> [1851513.257284]  [<ffffffff8152fbcf>] sock_release+0x1f/0x80
> >> [1851513.257521]  [<ffffffff8152fc42>] sock_close+0x12/0x20
> >> [1851513.257769]  [<ffffffff8119a8aa>] __fput+0xea/0x1f0
> >> [1851513.258005]  [<ffffffff8119a9ee>] ____fput+0xe/0x10
> >> [1851513.258244]  [<ffffffff8106fccf>] task_work_run+0x7f/0xb0
> >> [1851513.258488]  [<ffffffff81002210>] exit_to_usermode_loop+0xc0/0xd0
> >> [1851513.258728]  [<ffffffff81002a90>] syscall_return_slowpath+0x80/0xf0
> >> [1851513.258983]  [<ffffffff816147b4>] int_ret_from_sys_call+0x25/0x9f
> >> [1851513.259222] Code: 7e 5b 41 5c 5d c3 48 8b 8b e8 02 00 00 48 8b 93 f0 02 00 00 48 89 51 08 48 89 0a 48 89 83 e8 02 00 00 48 89 83 f0 02 00 00 eb b8 <0f> 0b 90 0f 1f 44 00 00 55 48 c7 c7 10 3f 07 82 48 89 e5 41 54
> >> [1851513.268473] RIP  [<ffffffff815f895d>] unix_notinflight+0x8d/0x90
> >> [1851513.268793]  RSP <ffff880189cf3cf8>
> >>
> >> That's essentially BUG_ON(list_empty(&u->link));
> >>
> >> I see that all the code involving the ->link member hasn't really been
> >> touched since it was introduced in 2007. So this must be a latent bug.
> >> This is the first time I've observed it. The state
> >> of the struct unix_sock can be found here http://sprunge.us/WCMW . Evidently,
> >> there are no inflight sockets.
>
> Weird.  If it was the BUG_ON(!list_empty(&u->link)) I'd understand,
> because the code in scan children looks fishy: what prevents "embryos"
> from fledging to full socket status and going in-flight?
>
> But this one, I cannot imagine any scenario.
>
> Can we have access to the crashdump?

crash> list -H gc_inflight_list unix_sock.link -s unix_sock.inflight |
grep counter | cut -d= -f2 | awk '{s+=$1} END {print s}'
130
crash> p unix_tot_inflight
unix_tot_inflight = $2 = 135

We've lost track of a total of five inflight sockets, so it's not a
one-off thing.  Really weird...  Now off to sleep, maybe I'll dream of
the solution.

Btw. I notice that this is a patched kernel.  Nothing in there that
could be relevant to this bug?

Thanks,
Miklos

[toc] | [prev] | [next] | [standalone]

#1472344

FromMiklos Szeredi <mszeredi@redhat.com>
Date2016-08-30 11:20 +0200
Message-ID<sbNn3-2ST-7@gated-at.bofh.it>
In reply to#1472147

[Multipart message — attachments visible in raw view] — view raw

On Tue, Aug 30, 2016 at 12:37 AM, Miklos Szeredi <mszeredi@redhat.com> wrote:
> On Sat, Aug 27, 2016 at 11:55 AM, Miklos Szeredi <mszeredi@redhat.com> wrote:

> crash> list -H gc_inflight_list unix_sock.link -s unix_sock.inflight |
> grep counter | cut -d= -f2 | awk '{s+=$1} END {print s}'
> 130
> crash> p unix_tot_inflight
> unix_tot_inflight = $2 = 135
>
> We've lost track of a total of five inflight sockets, so it's not a
> one-off thing.  Really weird...  Now off to sleep, maybe I'll dream of
> the solution.

Okay, found one bug: gc assumes that in-flight sockets that don't have
an external ref can't gain one while unix_gc_lock is held.  That is
true because unix_notinflight() will be called before detaching fds,
which takes unix_gc_lock.  Only MSG_PEEK was somehow overlooked.  That
one also clones the fds, also keeping them in the skb.  But through
MSG_PEEK an external reference can definitely be gained without ever
touching unix_gc_lock.

Not sure whether the reported bug can be explained by this.  Can you
confirm the MSG_PEEK was used in the setup?

Does someone want to write a stress test for SCM_RIGHTS + MSG_PEEK?

Anyway, attaching a fix that works by acquiring unix_gc_lock in case
of MSG_PEEK also.  It is trivially correct, but I haven't tested it.

Thanks,
Miklos

[toc] | [prev] | [next] | [standalone]

#1472365

FromNikolay Borisov <kernel@kyup.com>
Date2016-08-30 11:40 +0200
Message-ID<sbNGp-2Zk-15@gated-at.bofh.it>
In reply to#1472344

On 08/30/2016 12:18 PM, Miklos Szeredi wrote:
> On Tue, Aug 30, 2016 at 12:37 AM, Miklos Szeredi <mszeredi@redhat.com> wrote:
>> On Sat, Aug 27, 2016 at 11:55 AM, Miklos Szeredi <mszeredi@redhat.com> wrote:
> 
>> crash> list -H gc_inflight_list unix_sock.link -s unix_sock.inflight |
>> grep counter | cut -d= -f2 | awk '{s+=$1} END {print s}'
>> 130
>> crash> p unix_tot_inflight
>> unix_tot_inflight = $2 = 135
>>
>> We've lost track of a total of five inflight sockets, so it's not a
>> one-off thing.  Really weird...  Now off to sleep, maybe I'll dream of
>> the solution.
> 
> Okay, found one bug: gc assumes that in-flight sockets that don't have
> an external ref can't gain one while unix_gc_lock is held.  That is
> true because unix_notinflight() will be called before detaching fds,
> which takes unix_gc_lock.  Only MSG_PEEK was somehow overlooked.  That
> one also clones the fds, also keeping them in the skb.  But through
> MSG_PEEK an external reference can definitely be gained without ever
> touching unix_gc_lock.
> 
> Not sure whether the reported bug can be explained by this.  Can you
> confirm the MSG_PEEK was used in the setup?
> 
> Does someone want to write a stress test for SCM_RIGHTS + MSG_PEEK?
> 
> Anyway, attaching a fix that works by acquiring unix_gc_lock in case
> of MSG_PEEK also.  It is trivially correct, but I haven't tested it.

I have no way of being 100% sure but looking through nginx's source code
it seems they do utilize MSG_PEEK on several occasions. This issue has
been apparently very hard to reproduce since I have 100s of servers
running a lot of  NGINX processes and this has been triggered only once.

On a different note - if I inspect a live node without this patch should
the discrepancy between the gc_inflight_list and the unix_tot_inflight
be present VS with this patch applied?

> 
> Thanks,
> Miklos
>

[toc] | [prev] | [next] | [standalone]

#1472368

FromMiklos Szeredi <mszeredi@redhat.com>
Date2016-08-30 11:50 +0200
Message-ID<sbNQ5-32D-11@gated-at.bofh.it>
In reply to#1472365
On Tue, Aug 30, 2016 at 11:31 AM, Nikolay Borisov <kernel@kyup.com> wrote:
>
>
> On 08/30/2016 12:18 PM, Miklos Szeredi wrote:
>> On Tue, Aug 30, 2016 at 12:37 AM, Miklos Szeredi <mszeredi@redhat.com> wrote:
>>> On Sat, Aug 27, 2016 at 11:55 AM, Miklos Szeredi <mszeredi@redhat.com> wrote:
>>
>>> crash> list -H gc_inflight_list unix_sock.link -s unix_sock.inflight |
>>> grep counter | cut -d= -f2 | awk '{s+=$1} END {print s}'
>>> 130
>>> crash> p unix_tot_inflight
>>> unix_tot_inflight = $2 = 135
>>>
>>> We've lost track of a total of five inflight sockets, so it's not a
>>> one-off thing.  Really weird...  Now off to sleep, maybe I'll dream of
>>> the solution.
>>
>> Okay, found one bug: gc assumes that in-flight sockets that don't have
>> an external ref can't gain one while unix_gc_lock is held.  That is
>> true because unix_notinflight() will be called before detaching fds,
>> which takes unix_gc_lock.  Only MSG_PEEK was somehow overlooked.  That
>> one also clones the fds, also keeping them in the skb.  But through
>> MSG_PEEK an external reference can definitely be gained without ever
>> touching unix_gc_lock.
>>
>> Not sure whether the reported bug can be explained by this.  Can you
>> confirm the MSG_PEEK was used in the setup?
>>
>> Does someone want to write a stress test for SCM_RIGHTS + MSG_PEEK?
>>
>> Anyway, attaching a fix that works by acquiring unix_gc_lock in case
>> of MSG_PEEK also.  It is trivially correct, but I haven't tested it.
>
> I have no way of being 100% sure but looking through nginx's source code
> it seems they do utilize MSG_PEEK on several occasions. This issue has
> been apparently very hard to reproduce since I have 100s of servers
> running a lot of  NGINX processes and this has been triggered only once.
>
> On a different note - if I inspect a live node without this patch should
> the discrepancy between the gc_inflight_list and the unix_tot_inflight
> be present VS with this patch applied?

May well be, since in the vmcore 4 in-flight sockets were "lost"
before triggering the bug.  I guess the best way to check is with a
systemtap script that walks the list with the gc lock.

Thanks,
Miklos

[toc] | [prev] | [next] | [standalone]

#1474201

FromHannes Frederic Sowa <hannes@stressinduktion.org>
Date2016-09-01 11:20 +0200
Message-ID<scwk9-7Sa-3@gated-at.bofh.it>
In reply to#1472344
On 30.08.2016 11:18, Miklos Szeredi wrote:
> On Tue, Aug 30, 2016 at 12:37 AM, Miklos Szeredi <mszeredi@redhat.com> wrote:
>> On Sat, Aug 27, 2016 at 11:55 AM, Miklos Szeredi <mszeredi@redhat.com> wrote:
> 
>> crash> list -H gc_inflight_list unix_sock.link -s unix_sock.inflight |
>> grep counter | cut -d= -f2 | awk '{s+=$1} END {print s}'
>> 130
>> crash> p unix_tot_inflight
>> unix_tot_inflight = $2 = 135
>>
>> We've lost track of a total of five inflight sockets, so it's not a
>> one-off thing.  Really weird...  Now off to sleep, maybe I'll dream of
>> the solution.
> 
> Okay, found one bug: gc assumes that in-flight sockets that don't have
> an external ref can't gain one while unix_gc_lock is held.  That is
> true because unix_notinflight() will be called before detaching fds,
> which takes unix_gc_lock.  Only MSG_PEEK was somehow overlooked.  That
> one also clones the fds, also keeping them in the skb.  But through
> MSG_PEEK an external reference can definitely be gained without ever
> touching unix_gc_lock.
> 
> Not sure whether the reported bug can be explained by this.  Can you
> confirm the MSG_PEEK was used in the setup?
> 
> Does someone want to write a stress test for SCM_RIGHTS + MSG_PEEK?
> 
> Anyway, attaching a fix that works by acquiring unix_gc_lock in case
> of MSG_PEEK also.  It is trivially correct, but I haven't tested it.

You can use spin_unlock_wait in unix_gc_barrier to make it a bit more
lightweight.

Anyway, all of the scans on the socket receive queues are actually
protected by the appropriate locks again, I didn't see a way were we
could result in such a crash because of concurrent modification of the
receive queue. Do you have any hints or looked into this more closely?

Thanks,
Hannes

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web