[ Upstream commit 05530ef7cf ]
With clang's kernel control flow integrity (kCFI, CONFIG_CFI_CLANG),
indirect call targets are validated against the expected function
pointer prototype to make sure the call target is valid to help mitigate
ROP attacks. If they are not identical, there is a failure at run time,
which manifests as either a kernel panic or thread getting killed.
seq_copy_in_user() and seq_copy_in_kernel() did not have prototypes
matching snd_seq_dump_func_t. Adjust this and remove the casts. There
are not resulting binary output differences.
This was found as a result of Clang's new -Wcast-function-type-strict
flag, which is more sensitive than the simpler -Wcast-function-type,
which only checks for type width mismatches.
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/lkml/202211041527.HD8TLSE1-lkp@intel.com
Cc: Jaroslav Kysela <perex@perex.cz>
Cc: Takashi Iwai <tiwai@suse.com>
Cc: "Gustavo A. R. Silva" <gustavoars@kernel.org>
Cc: alsa-devel@alsa-project.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Link: https://lore.kernel.org/r/20221118232346.never.380-kees@kernel.org
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 3e7e04b747 upstream.
It's been reported that there is a possible data-race accessing to the
global card_requested[] array at ALSA sequencer core, which is used
for determining whether to call request_module() for the card or not.
This data race itself is almost harmless, as it might end up with one
extra request_module() call for the already loaded module at most.
But it's still better to fix.
This patch addresses the possible data race of card_requested[] and
client_requested[] arrays by replacing them with bitmask.
It's an atomic operation and can work without locks.
Reported-by: Abhishek Shah <abhishek.shah@columbia.edu>
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/CAEHB24_ay6YzARpA1zgCsE7=H9CSJJzux618E=Ka4h0YdKn=qA@mail.gmail.com
Link: https://lore.kernel.org/r/20220823072717.1706-2-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 22dec134db upstream.
ALSA OSS sequencer refers to a global variable max_midi_devs at
creating a new port, storing it to its own field. Meanwhile this
variable may be changed by other sequencer events at
snd_seq_oss_midi_check_exit_port() in parallel, which may cause a data
race.
OTOH, this data race itself is almost harmless, as the access to the
MIDI device is done via get_mdev() and it's protected with a refcount,
hence its presence is guaranteed.
Though, it's sill better to address the data-race from the code sanity
POV, and this patch adds the proper spinlock for the protection.
Reported-by: Abhishek Shah <abhishek.shah@columbia.edu>
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/CAEHB2493pZRXs863w58QWnUTtv3HHfg85aYhLn5HJHCwxqtHQg@mail.gmail.com
Link: https://lore.kernel.org/r/20220823072717.1706-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 6fadb494a6 ]
Currently ALSA sequencer core tries to process the queued events as
much as possible when they become dispatchable. If applications try
to queue too massive events to be processed at the very same timing,
the sequencer core would still try to process such all events, either
in the interrupt context or via some notifier; in either away, it
might be a cause of RCU stall or such problems.
As a potential workaround for those problems, this patch adds the
upper limit of the amount of events to be processed. The remaining
events are processed in the next batch, so they won't be lost.
For the time being, it's limited up to 1000 events per queue, which
should be high enough for any normal usages.
Reported-by: Zqiang <qiang.zhang1211@gmail.com>
Reported-by: syzbot+bb950e68b400ab4f65f8@syzkaller.appspotmail.com
Link: https://lore.kernel.org/r/20211102033222.3849-1-qiang.zhang1211@gmail.com
Link: https://lore.kernel.org/r/20211207165146.2888-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 97367c9722 upstream.
It turned out that the current implementation of the port subscription
is racy. The subscription contains two linked lists, and we have to
add to or delete from both lists. Since both connection and
disconnection procedures perform the same order for those two lists
(i.e. src list, then dest list), when a deletion happens during a
connection procedure, the src list may be deleted before the dest list
addition completes, and this may lead to a use-after-free or an Oops,
even though the access to both lists are protected via mutex.
The simple workaround for this race is to change the access order for
the disconnection, namely, dest list, then src list. This assures
that the connection has been established when disconnecting, and also
the concurrent deletion can be avoided.
Reported-and-tested-by: folkert <folkert@vanheusden.com>
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20210801182754.GP890690@belle.intranet.vanheusden.com
Link: https://lore.kernel.org/r/20210803114312.2536-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 217bfbb8b0 upstream.
snd_seq_oss_synth_make_info() didn't check the error code from
snd_seq_oss_midi_make_info(), and this leads to the call of strlcpy()
with the uninitialized string as the source, which may lead to the
access over the limit.
Add the proper error check for avoiding the failure.
Reported-by: syzbot+e42504ff21cff05a595f@syzkaller.appspotmail.com
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20210115093428.15882-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 4ebd470370 upstream.
The snd_seq_queue struct contains various flags in the bit fields.
Those are categorized to two different use cases, both of which are
protected by different spinlocks. That implies that there are still
potential risks of the bad operations for bit fields by concurrent
accesses.
For addressing the problem, this patch rearranges those flags to be
a standard bool instead of a bit field.
Reported-by: syzbot+63cbe31877bb80ef58f5@syzkaller.appspotmail.com
Link: https://lore.kernel.org/r/20201206083456.21110-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 2759caad26 ]
Recently we applied a fix to cover the whole OSS sequencer ioctls with
the mutex for dealing with the possible races. This works fine in
general, but in theory, this may lead to unexpectedly long stall if an
ioctl like SNDCTL_SEQ_SYNC is issued and an event with the far future
timestamp was queued.
For fixing such a potential stall, this patch changes the mutex lock
applied conditionally excluding such an ioctl command. Also, change
the mutex_lock() with the interruptible version for user to allow
escaping from the big-hammer mutex.
Fixes: 80982c7e83 ("ALSA: seq: oss: Serialize ioctls")
Suggested-by: Pavel Machek <pavel@ucw.cz>
Link: https://lore.kernel.org/r/20200922083856.28572-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 80982c7e83 upstream.
Some ioctls via OSS sequencer API may race and lead to UAF when the
port create and delete are performed concurrently, as spotted by a
couple of syzkaller cases. This patch is an attempt to address it by
serializing the ioctls with the existing register_mutex.
Basically OSS sequencer API is an obsoleted interface and was designed
without much consideration of the concurrency. There are very few
applications with it, and the concurrent performance isn't asked,
hence this "big hammer" approach should be good enough.
Reported-by: syzbot+1a54a94bd32716796edd@syzkaller.appspotmail.com
Reported-by: syzbot+9d2abfef257f3e2d4713@syzkaller.appspotmail.com
Suggested-by: Hillf Danton <hdanton@sina.com>
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20200804185815.2453-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 4384f167ce upstream.
The virmidi driver handles sysex event exceptionally in a short-cut
snd_seq_dump_var_event() call, but this missed the reset of the
running status. As a result, it may lead to an incomplete command
right after the sysex when an event with the same running status was
queued.
Fix it by clearing the running status properly via alling
snd_midi_event_reset_decode() for that code path.
Reported-by: Andreas Steinmetz <ast@domdv.de>
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/3b4a4e0f232b7afbaf0a843f63d0e538e3029bfd.camel@domdv.de
Link: https://lore.kernel.org/r/20200316090506.23966-2-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit dc7497795e upstream.
snd_seq_check_queue() passes the current tick and time of the given
queue as a pointer to snd_seq_prioq_cell_out(), but those might be
updated concurrently by the seq timer update.
Fix it by retrieving the current tick and time via the proper helper
functions at first, and pass those values to snd_seq_prioq_cell_out()
later in the loops.
snd_seq_timer_get_cur_time() takes a new argument and adjusts with the
current system time only when it's requested so; this update isn't
needed for snd_seq_check_queue(), as it's called either from the
interrupt handler or right after queuing.
Also, snd_seq_timer_get_cur_tick() is changed to read the value in the
spinlock for the concurrency, too.
Reported-by: syzbot+fd5e0eaa1a32999173b2@syzkaller.appspotmail.com
Link: https://lore.kernel.org/r/20200214111316.26939-3-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit bb51e669fa upstream.
The queue flags are represented in bit fields and the concurrent
access may result in unexpected results. Although the current code
should be mostly OK as it's only reading a field while writing other
fields as KCSAN reported, it's safer to cover both with a proper
spinlock protection.
This patch fixes the possible concurrent read by protecting with
q->owner_lock. Also the queue owner field is protected as well since
it's the field to be protected by the lock itself.
Reported-by: syzbot+65c6c92d04304d0a8efc@syzkaller.appspotmail.com
Reported-by: syzbot+e60ddfa48717579799dd@syzkaller.appspotmail.com
Link: https://lore.kernel.org/r/20200214111316.26939-2-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 60adcfde92 upstream.
snd_seq_info_timer_read() reads the information of the timer assigned
for each queue, but it's done in a racy way which may lead to UAF as
spotted by syzkaller.
This patch applies the missing q->timer_mutex lock while accessing the
timer object as well as a slight code change to adapt the standard
coding style.
Reported-by: syzbot+2b2ef983f973e5c40943@syzkaller.appspotmail.com
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20200115203733.26530-1-tiwai@suse.de
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit b8e131542b ]
snd_seq_system_client_init() doesn't check the errors returned from
its port creations. Let's do it properly and handle the error paths.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 75545304eb upstream.
The input pool of a client might be deleted via the resize ioctl, the
the access to it should be covered by the proper locks. Currently the
only missing place is the call in snd_seq_ioctl_get_client_pool(), and
this patch papers over it.
Reported-by: syzbot+4a75454b9ca2777f35c7@syzkaller.appspotmail.com
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit ede34f397d upstream.
The fix for the racy writes and ioctls to sequencer widened the
application of client->ioctl_mutex to the whole write loop. Although
it does unlock/relock for the lengthy operation like the event dup,
the loop keeps the ioctl_mutex for the whole time in other
situations. This may take quite long time if the user-space would
give a huge buffer, and this is a likely cause of some weird behavior
spotted by syzcaller fuzzer.
This patch puts a simple workaround, just adding a mutex break in the
loop when a large number of events have been processed. This
shouldn't hit any performance drop because the threshold is set high
enough for usual operations.
Fixes: 7bd8009156 ("ALSA: seq: More protection for concurrent write and ioctl races")
Reported-by: syzbot+97aae04ce27e39cbfca9@syzkaller.appspotmail.com
Reported-by: syzbot+4c595632b98bb8ffcc66@syzkaller.appspotmail.com
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit c3ea60c231 upstream.
There are two occurrances of a call to snd_seq_oss_fill_addr where
the dest_client and dest_port arguments are in the wrong order. Fix
this by swapping them around.
Addresses-Coverity: ("Arguments in wrong order")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit f0654ba94e ]
This reverts commit feb689025f.
The fix attempt was incorrect, leading to the mutex deadlock through
the close of OSS sequencer client. The proper fix needs more
consideration, so let's revert it now.
Fixes: feb689025f ("ALSA: seq: Protect in-kernel ioctl calls with mutex")
Reported-by: syzbot+47ded6c0f23016cde310@syzkaller.appspotmail.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 2eabc5ec8a ]
The snd_seq_ioctl_get_subscription() retrieves the port subscriber
information as a pointer, while the object isn't protected, hence it
may be deleted before the actual reference. This race was spotted by
syzkaller and may lead to a UAF.
The fix is simply copying the data in the lookup function that
performs in the rwsem to protect against the deletion.
Reported-by: syzbot+9437020c82413d00222d@syzkaller.appspotmail.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit feb689025f ]
ALSA OSS sequencer calls the ioctl function indirectly via
snd_seq_kernel_client_ctl(). While we already applied the protection
against races between the normal ioctls and writes via the client's
ioctl_mutex, this code path was left untouched. And this seems to be
the cause of still remaining some rare UAF as spontaneously triggered
by syzkaller.
For the sake of robustness, wrap the ioctl_mutex also for the call via
snd_seq_kernel_client_ctl(), too.
Reported-by: syzbot+e4c8abb920efa77bace9@syzkaller.appspotmail.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 7c32ae35fb upstream.
The call of unsubscribe_port() which manages the group count and
module refcount from delete_and_unsubscribe_port() looks racy; it's
not covered by the group list lock, and it's likely a cause of the
reported unbalance at port deletion. Let's move the call inside the
group list_mutex to plug the hole.
Reported-by: syzbot+e4c8abb920efa77bace9@syzkaller.appspotmail.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 212ac181c1 upstream.
When ioctl calls are made with non-null-terminated userspace strings,
strlcpy causes an OOB-read from within strlen. Fix by changing to use
strscpy instead.
Signed-off-by: Zubin Mithra <zsm@chromium.org>
Reviewed-by: Guenter Roeck <groeck@chromium.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit c709f14f06 upstream.
dev is indirectly controlled by user-space, hence leading to
a potential exploitation of the Spectre variant 1 vulnerability.
This issue was detected with the help of Smatch:
sound/core/seq/oss/seq_oss_synth.c:626 snd_seq_oss_synth_make_info() warn: potential spectre issue 'dp->synths' [w] (local cap)
Fix this by sanitizing dev before using it to index dp->synths.
Notice that given that speculation windows are large, the policy is
to kill the speculation on the first load and not worry if it can be
completed with a dependent load/store [1].
[1] https://lore.kernel.org/lkml/20180423164740.GY17484@dhcp22.suse.cz/
Cc: stable@vger.kernel.org
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The recent change to move the virmidi output processing to a work
slightly modified the code to discard the unsubscribed outputs so that
it works without a temporary buffer. However, this is actually buggy,
and may spew a kernel warning due to the unexpected call of
snd_rawmidi_transmit_ack(), as triggered by syzbot.
This patch takes back to the original code in that part, use a
temporary buffer and simply repeat snd_rawmidi_transmit(), in order to
address the regression.
Fixes: f7debfe540 ("ALSA: seq: virmidi: Offload the output event processing")
Reported-by: syzbot+ec5f605c91812d200367@syzkaller.appspotmail.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>
In preparation to enabling -Wimplicit-fallthrough, mark switch cases
where we are expecting to fall through.
Warning level 2 was used: -Wimplicit-fallthrough=2
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
In preparation to enabling -Wimplicit-fallthrough, mark switch cases
where we are expecting to fall through.
Notice that in this particular case, I replaced the code comment with
a proper "fall through" annotation, which is what GCC is expecting
to find.
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
All usages of mutex in ALSA sequencer core would take too long, hence
we don't have to care about the user interruption that makes things
complicated. Let's replace them with simpler mutex_lock().
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The sequencer core module doesn't call some destructors in the error
path of the init code, which may leave some resources.
This patch mainly fix these leaks by calling the destructors
appropriately at alsa_seq_init(). Also the patch brings a few
cleanups along with it, namely:
- Expand the old "if ((err = xxx) < 0)" coding style
- Get rid of empty seq_queue_init() and its caller
- Change snd_seq_info_done() to void
Last but not least, a couple of functions lose __exit annotation since
they are called also in alsa_seq_init().
No functional changes but minor code cleanups.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
There are a few functions that have been commented out for ages.
And also there are functions that do nothing but placeholders.
Let's kill them.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
snd_midi_event_encode_byte() can never fail, and it can return rather
true/false. Change the return type to bool, adjust the argument to
receive a MIDI byte as unsigned char, and adjust the comment
accordingly. This allows callers to drop error checks, which
simplifies the code.
Meanwhile, snd_midi_event_encode() helper is used only in seq_midi.c,
and it can be better folded into it. This will reduce the total
amount of lines in the end.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The trigger flag in vmidi object can be referred in different contexts
concurrently, hence it's better to be put with READ_ONCE() and
WRITE_ONCE() macros to assure the accesses.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The virmidi sequencer stuff tries to translate the rawmidi bytes to
sequencer events and deliver the packets at trigger callback. The
amount of the whole process of these translations and deliveries
depends on the incoming rawmidi bytes, and we have no limit for that;
this was the cause of a CPU soft lockup that had been reported and
fixed recently.
Although we've fixed the soft lockup by putting the temporary unlock
and cond_resched(), it's rather a quick band aid. In this patch,
meanwhile, the event parsing and delivery process is offloaded to a
dedicated work, and the trigger callback just kicks it off. It has
three merits, at least:
- The processing is always done in a sleepable context, which can
assure the event delivery with non-atomic flag without hackish
is_atomic() usage.
- Other relevant codes can be simplified, reducing the lines
- It makes me happier
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The virmidi output trigger tries to parse the all available bytes and
process sequencer events as much as possible. In a normal situation,
this is supposed to be relatively short, but a program may give a huge
buffer and it'll take a long time in a single spin lock, which may
eventually lead to a soft lockup.
This patch simply adds a workaround, a cond_resched() call in the loop
if applicable. A better solution would be to move the event processor
into a work, but let's put a duct-tape quickly at first.
Reported-and-tested-by: Dae R. Jeong <threeearcat@gmail.com>
Reported-by: syzbot+619d9f40141d826b097e@syzkaller.appspotmail.com
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The sanity checks in ALSA sequencer and OSS sequencer emulation codes
return falsely -ENXIO from poll callback. They should be EPOLLERR
instead.
This was caught thanks to the recent change to the return value.
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The kernel may spew a WARNING with UBSAN undefined behavior at
handling ALSA sequencer ioctl SNDRV_SEQ_IOCTL_QUERY_NEXT_CLIENT:
UBSAN: Undefined behaviour in sound/core/seq/seq_clientmgr.c:2007:14
signed integer overflow:
2147483647 + 1 cannot be represented in type 'int'
Call Trace:
__dump_stack lib/dump_stack.c:77 [inline]
dump_stack+0x122/0x1c8 lib/dump_stack.c:113
ubsan_epilogue+0x12/0x86 lib/ubsan.c:159
handle_overflow+0x1c2/0x21f lib/ubsan.c:190
__ubsan_handle_add_overflow+0x2a/0x31 lib/ubsan.c:198
snd_seq_ioctl_query_next_client+0x1ac/0x1d0 sound/core/seq/seq_clientmgr.c:2007
snd_seq_ioctl+0x264/0x3d0 sound/core/seq/seq_clientmgr.c:2144
....
It happens only when INT_MAX is passed there, as we're incrementing it
unconditionally. So the fix is trivial, check the value with
INT_MAX. Although the bug itself is fairly harmless, it's better to
fix it so that fuzzers won't hit this again later.
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=200211
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Instead of open-coding for getting the timer resolution, use the
standard snd_timer_resolution() helper.
The original code falls back to the callback function when the
resolution is zero, but it must be always so when the callback
function is defined. So this should be no functional change.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The sequencer virmidi code has an open race at its output trigger
callback: namely, virmidi keeps only one event packet for processing
while it doesn't protect for concurrent output trigger calls.
snd_virmidi_output_trigger() tries to process the previously
unfinished event before starting encoding the given MIDI stream, but
this is done without any lock. Meanwhile, if another rawmidi stream
starts the output trigger, this proceeds further, and overwrites the
event package that is being processed in another thread. This
eventually corrupts and may lead to the invalid memory access if the
event type is like SYSEX.
The fix is just to move the spinlock to cover both the pending event
and the new stream.
The bug was spotted by a new fuzzer, RaceFuzzer.
BugLink: http://lkml.kernel.org/r/20180426045223.GA15307@dragonet.kaist.ac.kr
Reported-by: DaeRyong Jeong <threeearcat@gmail.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
As Smatch recently suggested, a few places in OSS sequencer codes may
expand the array directly from the user-space value with speculation,
namely there are a significant amount of references to either
info->ch[] or dp->synths[] array:
sound/core/seq/oss/seq_oss_event.c:315 note_on_event() warn: potential spectre issue 'info->ch' (local cap)
sound/core/seq/oss/seq_oss_event.c:362 note_off_event() warn: potential spectre issue 'info->ch' (local cap)
sound/core/seq/oss/seq_oss_synth.c:470 snd_seq_oss_synth_load_patch() warn: potential spectre issue 'dp->synths' (local cap)
sound/core/seq/oss/seq_oss_event.c:293 note_on_event() warn: potential spectre issue 'dp->synths'
sound/core/seq/oss/seq_oss_event.c:353 note_off_event() warn: potential spectre issue 'dp->synths'
sound/core/seq/oss/seq_oss_synth.c:506 snd_seq_oss_synth_sysex() warn: potential spectre issue 'dp->synths'
sound/core/seq/oss/seq_oss_synth.c:580 snd_seq_oss_synth_ioctl() warn: potential spectre issue 'dp->synths'
Although all these seem doing only the first load without further
reference, we may want to stay in a safer side, so hardening with
array_index_nospec() would still make sense.
We may put array_index_nospec() at each place, but here we take a
different approach:
- For dp->synths[], change the helpers to retrieve seq_oss_synthinfo
pointer directly instead of the array expansion at each place
- For info->ch[], harden in a normal way, as there are only a couple
of places
As a result, the existing helper, snd_seq_oss_synth_is_valid() is
replaced with snd_seq_oss_synth_info(). Also, we cover MIDI device
where a similar array expansion is done, too, although it wasn't
reported by Smatch.
BugLink: https://marc.info/?l=linux-kernel&m=152411496503418&w=2
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
When get_synthdev() is called for a MIDI device, it returns the fixed
midi_synth_dev without the use refcounting. OTOH, the caller is
supposed to unreference unconditionally after the usage, so this would
lead to unbalanced refcount.
This patch corrects the behavior and keep up the refcount balance also
for the MIDI synth device.
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
When releasing a client, we need to clear the clienttab[] entry at
first, then call snd_seq_queue_client_leave(). Otherwise, the
in-flight cell in the queue might be picked up by the timer interrupt
via snd_seq_check_queue() before calling snd_seq_queue_client_leave(),
and it's delivered to another queue while the client is clearing
queues. This may eventually result in an uncleared cell remaining in
a queue, and the later snd_seq_pool_delete() may need to wait for a
long time until the event gets really processed.
By moving the clienttab[] clearance at the beginning of release, any
event delivery of a cell belonging to this client will fail at a later
point, since snd_seq_client_ptr() returns NULL. Thus the cell that
was picked up by the timer interrupt will be returned immediately
without further delivery, and the long stall of snd_seq_delete_pool()
can be avoided, too.
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Although we've covered the races between concurrent write() and
ioctl() in the previous patch series, there is still a possible UAF in
the following scenario:
A: user client closed B: timer irq
-> snd_seq_release() -> snd_seq_timer_interrupt()
-> snd_seq_free_client() -> snd_seq_check_queue()
-> cell = snd_seq_prioq_cell_peek()
-> snd_seq_prioq_leave()
.... removing all cells
-> snd_seq_pool_done()
.... vfree()
-> snd_seq_compare_tick_time(cell)
... Oops
So the problem is that a cell is peeked and accessed without any
protection until it's retrieved from the queue again via
snd_seq_prioq_cell_out().
This patch tries to address it, also cleans up the code by a slight
refactoring. snd_seq_prioq_cell_out() now receives an extra pointer
argument. When it's non-NULL, the function checks the event timestamp
with the given pointer. The caller needs to pass the right reference
either to snd_seq_tick or snd_seq_realtime depending on the event
timestamp type.
A good news is that the above change allows us to remove the
snd_seq_prioq_cell_peek(), too, thus the patch actually reduces the
code size.
Reviewed-by: Nicolai Stange <nstange@suse.de>
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
With the previous two fixes for the write / ioctl races:
ALSA: seq: Don't allow resizing pool in use
ALSA: seq: More protection for concurrent write and ioctl races
the cells aren't any longer in queues at the point calling
snd_seq_pool_done() in snd_seq_ioctl_set_client_pool(). Hence the
function call snd_seq_queue_client_leave_cells() can be dropped safely
from there.
Suggested-by: Nicolai Stange <nstange@suse.de>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This patch is an attempt for further hardening against races between
the concurrent write and ioctls. The previous fix d15d662e89
("ALSA: seq: Fix racy pool initializations") covered the race of the
pool initialization at writer and the pool resize ioctl by the
client->ioctl_mutex (CVE-2018-1000004). However, basically this mutex
should be applied more widely to the whole write operation for
avoiding the unexpected pool operations by another thread.
The only change outside snd_seq_write() is the additional mutex
argument to helper functions, so that we can unlock / relock the given
mutex temporarily during schedule() call for blocking write.
Fixes: d15d662e89 ("ALSA: seq: Fix racy pool initializations")
Reported-by: 范龙飞 <long7573@126.com>
Reported-by: Nicolai Stange <nstange@suse.de>
Reviewed-and-tested-by: Nicolai Stange <nstange@suse.de>
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This is a fix for a (sort of) fallout in the recent commit
d15d662e89 ("ALSA: seq: Fix racy pool initializations") for
CVE-2018-1000004.
As the pool resize deletes the existing cells, it may lead to a race
when another thread is writing concurrently, eventually resulting a
UAF.
A simple workaround is not to allow the pool resizing when the pool is
in use. It's an invalid behavior in anyway.
Fixes: d15d662e89 ("ALSA: seq: Fix racy pool initializations")
Reported-by: 范龙飞 <long7573@126.com>
Reported-by: Nicolai Stange <nstange@suse.de>
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>