ALSA: seq: Fix use-after-free at creating a port
authorTakashi Iwai <tiwai@suse.de>
Mon, 9 Oct 2017 09:09:20 +0000 (11:09 +0200)
committerBen Hutchings <ben@decadent.org.uk>
Sat, 11 Nov 2017 13:34:38 +0000 (13:34 +0000)
commitc3895a053b2505f9e409e6d6c57dcece714ab486
tree4e61d80b41a76fd00032cb801ab6a33cedd0d7f1
parentef810e7c3d2a8fb3bbd23726599c487c30ea747e
ALSA: seq: Fix use-after-free at creating a port

commit 71105998845fb012937332fe2e806d443c09e026 upstream.

There is a potential race window opened at creating and deleting a
port via ioctl, as spotted by fuzzing.  snd_seq_create_port() creates
a port object and returns its pointer, but it doesn't take the
refcount, thus it can be deleted immediately by another thread.
Meanwhile, snd_seq_ioctl_create_port() still calls the function
snd_seq_system_client_ev_port_start() with the created port object
that is being deleted, and this triggers use-after-free like:

 BUG: KASAN: use-after-free in snd_seq_ioctl_create_port+0x504/0x630 [snd_seq] at addr ffff8801f2241cb1
 =============================================================================
 BUG kmalloc-512 (Tainted: G    B          ): kasan: bad access detected
 -----------------------------------------------------------------------------
 INFO: Allocated in snd_seq_create_port+0x94/0x9b0 [snd_seq] age=1 cpu=3 pid=4511
  ___slab_alloc+0x425/0x460
  __slab_alloc+0x20/0x40
   kmem_cache_alloc_trace+0x150/0x190
snd_seq_create_port+0x94/0x9b0 [snd_seq]
snd_seq_ioctl_create_port+0xd1/0x630 [snd_seq]
  snd_seq_do_ioctl+0x11c/0x190 [snd_seq]
  snd_seq_ioctl+0x40/0x80 [snd_seq]
  do_vfs_ioctl+0x54b/0xda0
  SyS_ioctl+0x79/0x90
  entry_SYSCALL_64_fastpath+0x16/0x75
 INFO: Freed in port_delete+0x136/0x1a0 [snd_seq] age=1 cpu=2 pid=4717
  __slab_free+0x204/0x310
  kfree+0x15f/0x180
  port_delete+0x136/0x1a0 [snd_seq]
  snd_seq_delete_port+0x235/0x350 [snd_seq]
  snd_seq_ioctl_delete_port+0xc8/0x180 [snd_seq]
  snd_seq_do_ioctl+0x11c/0x190 [snd_seq]
  snd_seq_ioctl+0x40/0x80 [snd_seq]
  do_vfs_ioctl+0x54b/0xda0
  SyS_ioctl+0x79/0x90
  entry_SYSCALL_64_fastpath+0x16/0x75
 Call Trace:
  [<ffffffff81b03781>] dump_stack+0x63/0x82
  [<ffffffff81531b3b>] print_trailer+0xfb/0x160
  [<ffffffff81536db4>] object_err+0x34/0x40
  [<ffffffff815392d3>] kasan_report.part.2+0x223/0x520
  [<ffffffffa07aadf4>] ? snd_seq_ioctl_create_port+0x504/0x630 [snd_seq]
  [<ffffffff815395fe>] __asan_report_load1_noabort+0x2e/0x30
  [<ffffffffa07aadf4>] snd_seq_ioctl_create_port+0x504/0x630 [snd_seq]
  [<ffffffffa07aa8f0>] ? snd_seq_ioctl_delete_port+0x180/0x180 [snd_seq]
  [<ffffffff8136be50>] ? taskstats_exit+0xbc0/0xbc0
  [<ffffffffa07abc5c>] snd_seq_do_ioctl+0x11c/0x190 [snd_seq]
  [<ffffffffa07abd10>] snd_seq_ioctl+0x40/0x80 [snd_seq]
  [<ffffffff8136d433>] ? acct_account_cputime+0x63/0x80
  [<ffffffff815b515b>] do_vfs_ioctl+0x54b/0xda0
  .....

We may fix this in a few different ways, and in this patch, it's fixed
simply by taking the refcount properly at snd_seq_create_port() and
letting the caller unref the object after use.  Also, there is another
potential use-after-free by sprintf() call in snd_seq_create_port(),
and this is moved inside the lock.

This fix covers CVE-2017-15265.

Reported-and-tested-by: Michael23 Yu <ycqzsy@gmail.com>
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
sound/core/seq/seq_clientmgr.c
sound/core/seq/seq_ports.c