lkml.org 
[lkml]   [2017]   [Oct]   [29]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: possible deadlock in snd_seq_deliver_event
On Fri, 27 Oct 2017 10:11:18 +0200,
Dmitry Vyukov wrote:
>
> On Fri, Oct 27, 2017 at 10:09 AM, syzbot
> <bot+7feb8de6b4d6bf810cf098bef942cc387e79d0ad@syzkaller.appspotmail.com>
> wrote:
> > Hello,
> >
> > syzkaller hit the following crash on
> > 2bd6bf03f4c1c59381d62c61d03f6cc3fe71f66e
> > git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next.git/master
> > compiler: gcc (GCC) 7.1.1 20170620
> > .config is attached
> > Raw console output is attached.
> > C reproducer is attached
> > syzkaller reproducer is attached. See https://goo.gl/kgGztJ
> > for information about syzkaller reproducers
> >
> >
> > ============================================
> > WARNING: possible recursive locking detected
> > 4.14.0-rc1+ #88 Not tainted
> > --------------------------------------------
> > syzkaller883997/2981 is trying to acquire lock:
> > (&grp->list_mutex){++++}, at: [<ffffffff83d4dd49>] deliver_to_subscribers
> > sound/core/seq/seq_clientmgr.c:666 [inline]
> > (&grp->list_mutex){++++}, at: [<ffffffff83d4dd49>]
> > snd_seq_deliver_event+0x279/0x790 sound/core/seq/seq_clientmgr.c:807
> >
> > but task is already holding lock:
> > (&grp->list_mutex){++++}, at: [<ffffffff83d4dd49>] deliver_to_subscribers
> > sound/core/seq/seq_clientmgr.c:666 [inline]
> > (&grp->list_mutex){++++}, at: [<ffffffff83d4dd49>]
> > snd_seq_deliver_event+0x279/0x790 sound/core/seq/seq_clientmgr.c:807
> >
> > other info that might help us debug this:
> > Possible unsafe locking scenario:
> >
> > CPU0
> > ----
> > lock(&grp->list_mutex);
> > lock(&grp->list_mutex);
> >
> > *** DEADLOCK ***
> >
> > May be due to missing lock nesting notation
> >
> > 2 locks held by syzkaller883997/2981:
> > #0: (register_mutex#4){+.+.}, at: [<ffffffff83d60ada>]
> > odev_release+0x4a/0x70 sound/core/seq/oss/seq_oss.c:152
> > #1: (&grp->list_mutex){++++}, at: [<ffffffff83d4dd49>]
> > deliver_to_subscribers sound/core/seq/seq_clientmgr.c:666 [inline]
> > #1: (&grp->list_mutex){++++}, at: [<ffffffff83d4dd49>]
> > snd_seq_deliver_event+0x279/0x790 sound/core/seq/seq_clientmgr.c:807
> >
> > stack backtrace:
> > CPU: 1 PID: 2981 Comm: syzkaller883997 Not tainted 4.14.0-rc1+ #88
> > Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS
> > Google 01/01/2011
> > Call Trace:
> > __dump_stack lib/dump_stack.c:16 [inline]
> > dump_stack+0x194/0x257 lib/dump_stack.c:52
> > print_deadlock_bug kernel/locking/lockdep.c:1797 [inline]
> > check_deadlock kernel/locking/lockdep.c:1844 [inline]
> > validate_chain kernel/locking/lockdep.c:2453 [inline]
> > __lock_acquire+0x1232/0x4620 kernel/locking/lockdep.c:3498
> > lock_acquire+0x1d5/0x580 kernel/locking/lockdep.c:4002
> > down_read+0x96/0x150 kernel/locking/rwsem.c:23
> > deliver_to_subscribers sound/core/seq/seq_clientmgr.c:666 [inline]
> > snd_seq_deliver_event+0x279/0x790 sound/core/seq/seq_clientmgr.c:807
> > snd_seq_kernel_client_dispatch+0x11e/0x150
> > sound/core/seq/seq_clientmgr.c:2309
> > dummy_input+0x2c4/0x400 sound/core/seq/seq_dummy.c:104
> > snd_seq_deliver_single_event.constprop.11+0x2fb/0x940
> > sound/core/seq/seq_clientmgr.c:621
> > deliver_to_subscribers sound/core/seq/seq_clientmgr.c:676 [inline]
> > snd_seq_deliver_event+0x318/0x790 sound/core/seq/seq_clientmgr.c:807
> > snd_seq_kernel_client_dispatch+0x11e/0x150
> > sound/core/seq/seq_clientmgr.c:2309
> > dummy_input+0x2c4/0x400 sound/core/seq/seq_dummy.c:104
> > snd_seq_deliver_single_event.constprop.11+0x2fb/0x940
> > sound/core/seq/seq_clientmgr.c:621
> > snd_seq_deliver_event+0x12c/0x790 sound/core/seq/seq_clientmgr.c:818
> > snd_seq_kernel_client_dispatch+0x11e/0x150
> > sound/core/seq/seq_clientmgr.c:2309
> > snd_seq_oss_dispatch sound/core/seq/oss/seq_oss_device.h:150 [inline]
> > snd_seq_oss_midi_reset+0x44b/0x700 sound/core/seq/oss/seq_oss_midi.c:481
> > snd_seq_oss_synth_reset+0x398/0x980 sound/core/seq/oss/seq_oss_synth.c:416
> > snd_seq_oss_reset+0x6c/0x260 sound/core/seq/oss/seq_oss_init.c:448
> > snd_seq_oss_release+0x71/0x120 sound/core/seq/oss/seq_oss_init.c:425
> > odev_release+0x52/0x70 sound/core/seq/oss/seq_oss.c:153
> > __fput+0x333/0x7f0 fs/file_table.c:210
> > ____fput+0x15/0x20 fs/file_table.c:244
> > task_work_run+0x199/0x270 kernel/task_work.c:112
> > exit_task_work include/linux/task_work.h:21 [inline]
> > do_exit+0xa52/0x1b40 kernel/exit.c:865
> > do_group_exit+0x149/0x400 kernel/exit.c:968
> > SYSC_exit_group kernel/exit.c:979 [inline]
> > SyS_exit_group+0x1d/0x20 kernel/exit.c:977
> > entry_SYSCALL_64_fastpath+0x1f/0xbe
> > RIP: 0033:0x442c58
> > RSP: 002b:00007ffd15d4f8d8 EFLAGS: 00000246 ORIG_RAX: 00000000000000e7
> > RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 0000000000442c58
> > RDX: 0000000000000000 RSI: 000000000000003c RDI: 0000000000000000
> > RBP: 0000000000000082 R08: 00000000000000e7 R09: ffffffffffffffd0
> > R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000401ca0
> > R13: 0000000000401d30 R14
>
> I've just re-reproduced this on upstream
> 15f859ae5c43c7f0a064ed92d33f7a5bc5de6de0 (Oct 26):
>
> ============================================
> WARNING: possible recursive locking detected
> 4.14.0-rc6+ #10 Not tainted
> --------------------------------------------
> a.out/3062 is trying to acquire lock:
> (&grp->list_mutex){++++}, at: [<ffffffff83d28879>]
> deliver_to_subscribers sound/core/seq/seq_clientmgr.c:666 [inline]
> (&grp->list_mutex){++++}, at: [<ffffffff83d28879>]
> snd_seq_deliver_event+0x279/0x790 sound/core/seq/seq_clientmgr.c:807
>
> but task is already holding lock:
> (&grp->list_mutex){++++}, at: [<ffffffff83d28879>]
> deliver_to_subscribers sound/core/seq/seq_clientmgr.c:666 [inline]
> (&grp->list_mutex){++++}, at: [<ffffffff83d28879>]
> snd_seq_deliver_event+0x279/0x790 sound/core/seq/seq_clientmgr.c:807
>
> other info that might help us debug this:
> Possible unsafe locking scenario:
>
> CPU0
> ----
> lock(&grp->list_mutex);
> lock(&grp->list_mutex);
>
> *** DEADLOCK ***
>
> May be due to missing lock nesting notation

Indeed, this looks more like a simply missing nesting annotation.
A totally untested patch is below.


thanks,

Takashi

---
diff --git a/sound/core/seq/seq_clientmgr.c b/sound/core/seq/seq_clientmgr.c
index 6c9cba2166d9..d10c780dfd54 100644
--- a/sound/core/seq/seq_clientmgr.c
+++ b/sound/core/seq/seq_clientmgr.c
@@ -663,7 +663,7 @@ static int deliver_to_subscribers(struct snd_seq_client *client,
if (atomic)
read_lock(&grp->list_lock);
else
- down_read(&grp->list_mutex);
+ down_read_nested(&grp->list_mutex, hop);
list_for_each_entry(subs, &grp->list_head, src_list) {
/* both ports ready? */
if (atomic_read(&subs->ref_count) != 2)
\
 
 \ /
  Last update: 2017-10-29 10:59    [W:0.042 / U:0.988 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site