Re: sound: deadlock between snd_rawmidi_kernel_open/snd_seq_port_connect

From: Dmitry Vyukov
Date: Tue Feb 02 2016 - 16:24:22 EST


On Mon, Jan 25, 2016 at 11:47 AM, Takashi Iwai <tiwai@xxxxxxx> wrote:
> On Sun, 24 Jan 2016 10:44:34 +0100,
> Dmitry Vyukov wrote:
>>
>> Hello,
>>
>> While running syzkaller fuzzer I've got the following lockdep report:
>>
>> ======================================================
>> [ INFO: possible circular locking dependency detected ]
>> 4.4.0+ #276 Not tainted
>> -------------------------------------------------------
>> syz-executor/21025 is trying to acquire lock:
>> (register_mutex#5){+.+.+.}, at: [<ffffffff84f889cb>]
>> snd_rawmidi_kernel_open+0x4b/0x260 sound/core/rawmidi.c:341
>>
>> but task is already holding lock:
>> (&grp->list_mutex/1){+.+...}, at: [<ffffffff84fc4afa>]
>> snd_seq_port_connect+0x1ba/0x840 sound/core/seq/seq_ports.c:506
>>
>> which lock already depends on the new lock.
>>
>>
>> the existing dependency chain (in reverse order) is:
>>
>> -> #2 (&grp->list_mutex/1){+.+...}:
>> [<ffffffff8145ad8c>] lock_acquire+0x1dc/0x430
>> kernel/locking/lockdep.c:3585
>> [<ffffffff8144851a>] down_write_nested+0x4a/0xa0
>> kernel/locking/rwsem.c:149
>> [<ffffffff84fc4afa>] snd_seq_port_connect+0x1ba/0x840
>> sound/core/seq/seq_ports.c:506
>> [<ffffffff84fb6f54>] snd_seq_ioctl_subscribe_port+0x1c4/0x290
>> sound/core/seq/seq_clientmgr.c:1464
>> [<ffffffff84fb126d>] snd_seq_do_ioctl+0x19d/0x1c0
>> sound/core/seq/seq_clientmgr.c:2209
>> [<ffffffff84fb2c2b>] snd_seq_kernel_client_ctl+0xdb/0x170
>> sound/core/seq/seq_clientmgr.c:2423
>> [<ffffffff88413c20>] snd_seq_oss_create_client+0x253/0x2d5
>> sound/core/seq/oss/seq_oss_init.c:119
>> [<ffffffff8841393e>] alsa_seq_oss_init+0x1af/0x23e
>> sound/core/seq/oss/seq_oss.c:89
>> [<ffffffff81002259>] do_one_initcall+0x159/0x380 init/main.c:794
>> [< inline >] do_initcall_level init/main.c:859
>> [< inline >] do_initcalls init/main.c:867
>> [< inline >] do_basic_setup init/main.c:885
>> [<ffffffff88315c1a>] kernel_init_freeable+0x474/0x52d init/main.c:1010
>> [<ffffffff86312683>] kernel_init+0x13/0x150 init/main.c:936
>> [<ffffffff86336fef>] ret_from_fork+0x3f/0x70
>> arch/x86/entry/entry_64.S:468
>>
>> -> #1 (&grp->list_mutex){++++.+}:
>> [<ffffffff8145ad8c>] lock_acquire+0x1dc/0x430
>> kernel/locking/lockdep.c:3585
>> [<ffffffff863323a7>] down_read+0x47/0x60 kernel/locking/rwsem.c:22
>> [< inline >] deliver_to_subscribers
>> sound/core/seq/seq_clientmgr.c:679
>> [<ffffffff84fb5509>] snd_seq_deliver_event+0x5a9/0x800
>> sound/core/seq/seq_clientmgr.c:817
>> [<ffffffff84fb6466>] snd_seq_kernel_client_dispatch+0x126/0x170
>> sound/core/seq/seq_clientmgr.c:2401
>> [<ffffffff84fc1e62>] snd_seq_system_broadcast+0xb2/0xf0
>> sound/core/seq/seq_system.c:101
>> [<ffffffff84fb248e>] snd_seq_create_kernel_client+0x21e/0x300
>> sound/core/seq/seq_clientmgr.c:2280
>> [< inline >] snd_virmidi_dev_attach_seq
>> sound/core/seq/seq_virmidi.c:372
>> [<ffffffff84fdc55f>] snd_virmidi_dev_register+0x29f/0x750
>> sound/core/seq/seq_virmidi.c:439
>> [<ffffffff84f824ec>] snd_rawmidi_dev_register+0x30c/0xd40
>> sound/core/rawmidi.c:1589
>> [<ffffffff84f33583>] __snd_device_register.part.0+0x63/0xc0
>> sound/core/device.c:164
>> [< inline >] __snd_device_register sound/core/device.c:162
>> [<ffffffff84f341ed>] snd_device_register_all+0xad/0x110
>> sound/core/device.c:212
>> [<ffffffff84f277cf>] snd_card_register+0xef/0x6a0 sound/core/init.c:749
>> [<ffffffff84fefc0f>] snd_virmidi_probe+0x3ef/0x590
>> sound/drivers/virmidi.c:123
>> [<ffffffff832feb6c>] platform_drv_probe+0x8c/0x160
>> drivers/base/platform.c:562
>> [< inline >] really_probe drivers/base/dd.c:377
>> [<ffffffff832f862e>] driver_probe_device+0x37e/0xc90
>> drivers/base/dd.c:499
>> [<ffffffff832f926e>] __device_attach_driver+0x19e/0x250
>> drivers/base/dd.c:584
>> [<ffffffff832f23df>] bus_for_each_drv+0x13f/0x1d0 drivers/base/bus.c:464
>> [<ffffffff832f809f>] __device_attach+0x1ef/0x2e0 drivers/base/dd.c:641
>> [<ffffffff832f93ba>] device_initial_probe+0x1a/0x20 drivers/base/dd.c:688
>> [<ffffffff832f5939>] bus_probe_device+0x1e9/0x290 drivers/base/bus.c:558
>> [<ffffffff832ef11b>] device_add+0x84b/0x1490 drivers/base/core.c:1120
>> [<ffffffff832fe2c9>] platform_device_add+0x389/0x790
>> drivers/base/platform.c:403
>> [<ffffffff832fff26>] platform_device_register_full+0x396/0x4c0
>> drivers/base/platform.c:535
>> [< inline >] platform_device_register_resndata
>> include/linux/platform_device.h:111
>> [< inline >] platform_device_register_simple
>> include/linux/platform_device.h:140
>> [<ffffffff884146d1>] alsa_card_virmidi_init+0x104/0x1da
>> sound/drivers/virmidi.c:172
>> [<ffffffff81002259>] do_one_initcall+0x159/0x380 init/main.c:794
>> [< inline >] do_initcall_level init/main.c:859
>> [< inline >] do_initcalls init/main.c:867
>> [< inline >] do_basic_setup init/main.c:885
>> [<ffffffff88315c1a>] kernel_init_freeable+0x474/0x52d init/main.c:1010
>> [<ffffffff86312683>] kernel_init+0x13/0x150 init/main.c:936
>> [<ffffffff86336fef>] ret_from_fork+0x3f/0x70
>> arch/x86/entry/entry_64.S:468
>>
>> -> #0 (register_mutex#5){+.+.+.}:
>> [< inline >] check_prev_add kernel/locking/lockdep.c:1853
>> [< inline >] check_prevs_add kernel/locking/lockdep.c:1958
>> [< inline >] validate_chain kernel/locking/lockdep.c:2144
>> [<ffffffff8145742b>] __lock_acquire+0x31eb/0x4700
>> kernel/locking/lockdep.c:3206
>> [<ffffffff8145ad8c>] lock_acquire+0x1dc/0x430
>> kernel/locking/lockdep.c:3585
>> [< inline >] __mutex_lock_common kernel/locking/mutex.c:518
>> [<ffffffff8632cf61>] mutex_lock_nested+0xb1/0xa50
>> kernel/locking/mutex.c:618
>> [<ffffffff84f889cb>] snd_rawmidi_kernel_open+0x4b/0x260
>> sound/core/rawmidi.c:341
>> [<ffffffff84fdda77>] midisynth_subscribe+0xf7/0x340
>> sound/core/seq/seq_midi.c:188
>> [<ffffffff84fc2a4e>] subscribe_port.isra.2+0x14e/0x2b0
>> sound/core/seq/seq_ports.c:426
>> [<ffffffff84fc4dd0>] snd_seq_port_connect+0x490/0x840
>> sound/core/seq/seq_ports.c:527
>> [<ffffffff84fb6f54>] snd_seq_ioctl_subscribe_port+0x1c4/0x290
>> sound/core/seq/seq_clientmgr.c:1464
>> [<ffffffff84fb126d>] snd_seq_do_ioctl+0x19d/0x1c0
>> sound/core/seq/seq_clientmgr.c:2209
>> [<ffffffff84fb2c2b>] snd_seq_kernel_client_ctl+0xdb/0x170
>> sound/core/seq/seq_clientmgr.c:2423
>> [<ffffffff84fd69e4>] snd_seq_oss_midi_open+0x3b4/0x610
>> sound/core/seq/oss/seq_oss_midi.c:375
>> [<ffffffff84fd6ccb>] snd_seq_oss_midi_open_all+0x8b/0xd0
>> sound/core/seq/oss/seq_oss_midi.c:306
>> [<ffffffff84fca1d5>] snd_seq_oss_open+0x5c5/0x8d0
>> sound/core/seq/oss/seq_oss_init.c:276
>> [<ffffffff84fc903a>] odev_open+0x6a/0x90 sound/core/seq/oss/seq_oss.c:138
>> [<ffffffff84f2279f>] soundcore_open+0x30f/0x640 sound/sound_core.c:639
>> [<ffffffff817be1fa>] chrdev_open+0x22a/0x4c0 fs/char_dev.c:388
>> [<ffffffff817a9c02>] do_dentry_open+0x6a2/0xcb0 fs/open.c:736
>> [<ffffffff817ad2db>] vfs_open+0x17b/0x1f0 fs/open.c:853
>> [< inline >] do_last fs/namei.c:3254
>> [<ffffffff817e00d9>] path_openat+0xde9/0x5e30 fs/namei.c:3386
>> [<ffffffff817e895e>] do_filp_open+0x18e/0x250 fs/namei.c:3421
>> [<ffffffff817ada5c>] do_sys_open+0x1fc/0x420 fs/open.c:1022
>> [< inline >] SYSC_open fs/open.c:1040
>> [<ffffffff817adcad>] SyS_open+0x2d/0x40 fs/open.c:1035
>> [<ffffffff86336c36>] entry_SYSCALL_64_fastpath+0x16/0x7a
>> arch/x86/entry/entry_64.S:185
>>
>> other info that might help us debug this:
>>
>> Chain exists of:
>> register_mutex#5 --> &grp->list_mutex --> &grp->list_mutex/1
>>
>> Possible unsafe locking scenario:
>>
>> CPU0 CPU1
>> ---- ----
>> lock(&grp->list_mutex/1);
>> lock(&grp->list_mutex);
>> lock(&grp->list_mutex/1);
>> lock(register_mutex#5);
>>
>> *** DEADLOCK ***
>>
>> 3 locks held by syz-executor/21025:
>> #0: (register_mutex#4){+.+.+.}, at: [<ffffffff84fc902f>]
>> odev_open+0x5f/0x90 sound/core/seq/oss/seq_oss.c:137
>> #1: (&grp->list_mutex){++++.+}, at: [<ffffffff84fc4ae2>]
>> snd_seq_port_connect+0x1a2/0x840 sound/core/seq/seq_ports.c:505
>> #2: (&grp->list_mutex/1){+.+...}, at: [<ffffffff84fc4afa>]
>> snd_seq_port_connect+0x1ba/0x840 sound/core/seq/seq_ports.c:506
>
> This looks like a false-positive report to me. Of course, we should
> annotate the mutex there for nested locks, though.


Takashi, can you please annotate it for lockdep? I hit it on every run.
Thanks in advance