Re: data-race in snd_seq_oss_midi_check_exit_port / snd_seq_oss_midi_setup

[Takashi Iwai <tiwai@suse.de>]

On Fri, 19 Aug 2022 03:00:00 +0200,
Abhishek Shah wrote:
> 
> 
> Hi all, 
> 
> We found a race involving the max_midi_devs variable. We see an interleaving
> where the following check here passes before the 
> snd_seq_oss_midi_check_exit_port() finishes, but this check should not pass
> if max_midi_devs will become zero, but we are not sure of its implications in
> terms of security impact. Please let us know what you think.

Through a quick glance, I guess it's rather harmless (although a bit
fragile from the code sanity POV).

A MIDI port could be closed at any time, and the dp->max_mididevs
holds locally the upper bound of currently possibly accessible ports.
The actual access to each port is done via get_mdev() in
seq_oss_midi.c, which is a sort of refcount managed, and it should be
fine that a port disappears meanwhile.

That said, it'd be even feasible just dropping dp->max_mididevs field
and scan all MIDI ports at each time, but it won't bring much benefit,
either.


thanks,

Takashi

> 
> Thanks!
> 
> -------------------Report---------------------
> 
> write to 0xffffffff88382f80 of 4 bytes by task 6541 on cpu 0:
>  snd_seq_oss_midi_check_exit_port+0x1a6/0x270 sound/core/seq/oss/
> seq_oss_midi.c:237
>  receive_announce+0x193/0x1b0 sound/core/seq/oss/seq_oss_init.c:143
>  snd_seq_deliver_single_event+0x30d/0x4e0 sound/core/seq/seq_clientmgr.c:640
>  deliver_to_subscribers sound/core/seq/seq_clientmgr.c:695 [inline]
>  snd_seq_deliver_event+0x38c/0x490 sound/core/seq/seq_clientmgr.c:830
>  snd_seq_kernel_client_dispatch+0x189/0x1a0 sound/core/seq/
> seq_clientmgr.c:2339
>  snd_seq_system_broadcast+0x98/0xd0 sound/core/seq/seq_system.c:86
>  snd_seq_ioctl_delete_port+0x9a/0xc0 sound/core/seq/seq_clientmgr.c:1356
>  snd_seq_ioctl+0x198/0x2d0 sound/core/seq/seq_clientmgr.c:2173
>  vfs_ioctl fs/ioctl.c:51 [inline]
>  __do_sys_ioctl fs/ioctl.c:870 [inline]
>  __se_sys_ioctl+0xe1/0x150 fs/ioctl.c:856
>  __x64_sys_ioctl+0x43/0x50 fs/ioctl.c:856
>  do_syscall_x64 arch/x86/entry/common.c:50 [inline]
>  do_syscall_64+0x3d/0x90 arch/x86/entry/common.c:80
>  entry_SYSCALL_64_after_hwframe+0x44/0xae
> 
> read to 0xffffffff88382f80 of 4 bytes by task 6542 on cpu 1:
>  snd_seq_oss_midi_setup+0x1b/0x40 sound/core/seq/oss/seq_oss_midi.c:273
>  snd_seq_oss_open+0x364/0x900 sound/core/seq/oss/seq_oss_init.c:198
>  odev_open+0x55/0x70 sound/core/seq/oss/seq_oss.c:128
>  soundcore_open+0x315/0x3a0 sound/sound_core.c:593
>  chrdev_open+0x373/0x3f0 fs/char_dev.c:414
>  do_dentry_open+0x543/0x8f0 fs/open.c:824
>  vfs_open+0x47/0x50 fs/open.c:958
>  do_open fs/namei.c:3476 [inline]
>  path_openat+0x1906/0x1dc0 fs/namei.c:3609
>  do_filp_open+0xef/0x200 fs/namei.c:3636
>  do_sys_openat2+0xa5/0x2a0 fs/open.c:1213
>  do_sys_open fs/open.c:1229 [inline]
>  __do_sys_openat fs/open.c:1245 [inline]
>  __se_sys_openat fs/open.c:1240 [inline]
>  __x64_sys_openat+0xf0/0x120 fs/open.c:1240
>  do_syscall_x64 arch/x86/entry/common.c:50 [inline]
>  do_syscall_64+0x3d/0x90 arch/x86/entry/common.c:80
>  entry_SYSCALL_64_after_hwframe+0x44/0xae
> 
> Reported by Kernel Concurrency Sanitizer on:
> CPU: 1 PID: 6542 Comm: syz-executor2-n Not tainted 5.18.0-rc5+ #107
> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.15.0-1 04/01/
> 2014
> 
> Reproducing Inputs
> 
> Input CPU 0:
> r0 = openat$sndseq(0xffffffffffffff9c, &(0x7f0000000040)='/dev/snd/seq\x00',
> 0x0)
> ioctl$SNDRV_SEQ_IOCTL_CREATE_PORT(r0, 0xc0a85320, &(0x7f0000000240)={{0x80},
> 'port1\x00', 0x10})
> ioctl$SNDRV_SEQ_IOCTL_SET_CLIENT_POOL(r0, 0x40a85321, &(0x7f0000000100)=
> {0x80})
> 
> Input CPU 1:
> r0 = openat$sequencer2(0xffffff9c, &(0x7f0000000000)='/dev/sequencer2\x00',
> 0x0, 0x0)
> ioctl$SNDCTL_SYNTH_INFO(r0, 0xc08c5102, &(0x7f0000000200)=
> {"02961a3ce6d4828f8b5559726313251b55fa11d8d65406f1f33c9af8e3f8", 0xffffffff})
> 
>