Message ID | 20171011080019.16212-1-tiwai@suse.de (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On Wed, Oct 11, 2017 at 1:00 AM, Takashi Iwai <tiwai@suse.de> wrote: > 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
On Wed, 11 Oct 2017 19:35:46 +0200, Linus Torvalds wrote: > > On Wed, Oct 11, 2017 at 1:00 AM, Takashi Iwai <tiwai@suse.de> wrote: > > 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. > 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) 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
diff --git a/sound/core/seq/seq_clientmgr.c b/sound/core/seq/seq_clientmgr.c index ea2d0ae85bd3..6c9cba2166d9 100644 --- a/sound/core/seq/seq_clientmgr.c +++ b/sound/core/seq/seq_clientmgr.c @@ -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; } diff --git a/sound/core/seq/seq_ports.c b/sound/core/seq/seq_ports.c index 0a7020c82bfc..d21ece9f8d73 100644 --- a/sound/core/seq/seq_ports.c +++ b/sound/core/seq/seq_ports.c @@ -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; }
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> Cc: <stable@vger.kernel.org> Signed-off-by: Takashi Iwai <tiwai@suse.de> --- sound/core/seq/seq_clientmgr.c | 6 +++++- sound/core/seq/seq_ports.c | 7 +++++-- 2 files changed, 10 insertions(+), 3 deletions(-)