Skip to contentSkip to conversation

ALSA: seq: Fix use-after-free at creating a port

PATCH In review Takashi Iwai 2 participants 3 messages Oct 11, 2017 Sound

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ALSA: seq: Fix use-after-free at creating a port

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.

---
 sound/core/seq/seq_clientmgr.c | 6 +++++-
 sound/core/seq/seq_ports.c     | 7 +++++--
 2 files changed, 10 insertions(+), 3 deletions(-)
sound/core/seq/seq_clientmgr.c +5 -1
index ea2d0ae85bd3..6c9cba2166d9 100644
@@ -1259,6 +1259,7 @@ static int snd_seq_ioctl_create_port(struct snd_seq_client *client, void *arg)
 	struct snd_seq_port_info *info = arg;
 	struct snd_seq_client_port *port;
 	struct snd_seq_port_callback *callback;
+	int port_idx;
 
 	/* it is not allowed to create the port for an another client */
 	if (info->addr.client != client->number)
@@ -1269,7 +1270,9 @@ static int snd_seq_ioctl_create_port(struct snd_seq_client *client, void *arg)
 		return -ENOMEM;
 
 	if (client->type == USER_CLIENT && info->kernel) {
-		snd_seq_delete_port(client, port->addr.port);
+		port_idx = port->addr.port;
+		snd_seq_port_unlock(port);
+		snd_seq_delete_port(client, port_idx);
 		return -EINVAL;
 	}
 	if (client->type == KERNEL_CLIENT) {
@@ -1290,6 +1293,7 @@ static int snd_seq_ioctl_create_port(struct snd_seq_client *client, void *arg)
 
 	snd_seq_set_port_info(port, info);
 	snd_seq_system_client_ev_port_start(port->addr.client, port->addr.port);
+	snd_seq_port_unlock(port);
 
 	return 0;
 }
sound/core/seq/seq_ports.c +5 -2
index 0a7020c82bfc..d21ece9f8d73 100644
@@ -122,7 +122,9 @@ static void port_subs_info_init(struct snd_seq_port_subs_info *grp)
 }
 
 
-/* create a port, port number is returned (-1 on failure) */
+/* create a port, port number is returned (-1 on failure);
+ * the caller needs to unref the port via snd_seq_port_unlock() appropriately
+ */
 struct snd_seq_client_port *snd_seq_create_port(struct snd_seq_client *client,
 						int port)
 {
@@ -151,6 +153,7 @@ struct snd_seq_client_port *snd_seq_create_port(struct snd_seq_client *client,
 	snd_use_lock_init(&new_port->use_lock);
 	port_subs_info_init(&new_port->c_src);
 	port_subs_info_init(&new_port->c_dest);
+	snd_use_lock_use(&new_port->use_lock);
 
 	num = port >= 0 ? port : 0;
 	mutex_lock(&client->ports_mutex);
@@ -165,9 +168,9 @@ struct snd_seq_client_port *snd_seq_create_port(struct snd_seq_client *client,
 	list_add_tail(&new_port->list, &p->list);
 	client->num_ports++;
 	new_port->addr.port = num;	/* store the port number in the port */
+	sprintf(new_port->name, "port-%d", num);
 	write_unlock_irqrestore(&client->ports_lock, flags);
 	mutex_unlock(&client->ports_mutex);
-	sprintf(new_port->name, "port-%d", num);
 
 	return new_port;
 }
Signature · 1 line
2.14.2
#1 lore raw
Quoted 2 lines · Takashi Iwai

There is a potential race window opened at creating and deleting a
port via ioctl, as spotted by fuzzing. [..]

I'm assuming I'll just get this through the normal sound fixes pull request?

Also, just for future issues, I really wish people would edit the call traces to be a bit more legible. You did remove some of the excessive KASAN noise, but the stack trace hex numbers are a bit distracting too.

Of course, the only reason those hex numbers are there in the first place was that the oops was generated with an older kernel - we don't even print them out any more.

Not a big deal, but I thought I'd mention it just because I generally encourage people to try to edit down oopses etc to the really relevant parts (the kernel often spits out a lot of stuff that isn't really relevant for the particular bug, because it *might* be)

Thanks,
Linus

#2 lore raw
Quoted 6 lines · Linus Torvalds
Quoted 2 lines · Takashi Iwai

There is a potential race window opened at creating and deleting a
port via ioctl, as spotted by fuzzing. [..]

I'm assuming I'll just get this through the normal sound fixes pull request?

Yes, I'll send you a pull request tomorrow together with other fixes.

Quoted 5 lines

Also, just for future issues, I really wish people would edit the call traces to be a bit more legible. You did remove some of the excessive KASAN noise, but the stack trace hex numbers are a bit distracting too.

Quoted 8 lines

Of course, the only reason those hex numbers are there in the first place was that the oops was generated with an older kernel - we don't even print them out any more.

Not a big deal, but I thought I'd mention it just because I generally encourage people to try to edit down oopses etc to the really relevant parts (the kernel often spits out a lot of stuff that isn't really relevant for the particular bug, because it *might* be)

OK, noted.

Now after seeing lots of commits with stack trace messages in this week (I warn you beforehand!), I share the same feeling. Will tidy up harder at the next time.

Thanks!

Takashi