This is a note to let you know that I've just added the patch titled
staging: comedi: comedi_test: fix timer race conditions
to my staging git tree which can be found at
git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging.git
in the staging-linus branch.
The patch will show up in the next release of the linux-next tree
(usually sometime within the next 24 hours during the week.)
The patch will hopefully also be merged in Linus's tree for the
next -rc kernel release.
If you have any questions about this process, please let me know.
>From 403fe7f34e3327ddac2e06a15e76a293d613381e Mon Sep 17 00:00:00 2001
From: Ian Abbott <abbotti@xxxxxxxxx>
Date: Thu, 30 Jun 2016 19:58:32 +0100
Subject: staging: comedi: comedi_test: fix timer race conditions
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Commit 73e0e4dfed4c ("staging: comedi: comedi_test: fix timer lock-up")
fixed a lock-up in the timer routine `waveform_ai_timer()` (which was
called `waveform_ai_interrupt()` at the time) caused by
commit 240512474424 ("staging: comedi: comedi_test: use
comedi_handle_events()"). However, it introduced a race condition that
can result in the timer routine misbehaving, such as accessing freed
memory or dereferencing a NULL pointer.
73e0... changed the timer routine to do nothing unless a
`WAVEFORM_AI_RUNNING` flag was set, and changed `waveform_ai_cancel()`
to clear the flag and replace a call to `del_timer_sync()` with a call
to `del_timer()`. `waveform_ai_cancel()` may be called from the timer
routine itself (via `comedi_handle_events()`), or from `do_cancel()`.
(`do_cancel()` is called as a result of a file operation (usually a
`COMEDI_CANCEL` ioctl command, or a release), or during device removal.)
When called from `do_cancel()`, the call to `waveform_ai_cancel()` is
followed by a call to `do_become_nonbusy()`, which frees up stuff for
the current asynchronous command under the assumption that it is now
safe to do so. The race condition occurs when the timer routine
`waveform_ai_timer()` checks the `WAVEFORM_AI_RUNNING` flag just before
it is cleared by `waveform_ai_cancel()`, and is still running during the
call to `do_become_nonbusy()`. In particular, it can lead to a NULL
pointer dereference:
BUG: unable to handle kernel NULL pointer dereference at (null)
IP: [<ffffffffc0c63add>] waveform_ai_timer+0x17d/0x290 [comedi_test]
That corresponds to this line in `waveform_ai_timer()`:
unsigned int chanspec = cmd->chanlist[async->cur_chan];
but `do_become_nonbusy()` frees `cmd->chanlist` and sets it to `NULL`.
Fix the race by calling `del_timer_sync()` instead of `del_timer()` in
`waveform_ai_cancel()` when not in an interrupt context. The only time
`waveform_ai_cancel()` is called in an interrupt context is when it is
called from the timer routine itself, via `comedi_handle_events()`.
There is no longer any need for the `WAVEFORM_AI_RUNNING` flag, so get
rid of it.
The bug was copied from the AI subdevice to the AO when support for
commands on the AO subdevice was added by commit 0cf55bbef2f9 ("staging:
comedi: comedi_test: implement commands on AO subdevice"). That
involves the timer routine `waveform_ao_timer()`, the comedi "cancel"
routine `waveform_ao_cancel()`, and the flag `WAVEFORM_AO_RUNNING`. Fix
it in the same way as for the AI subdevice.
Fixes: 73e0e4dfed4c ("staging: comedi: comedi_test: fix timer lock-up")
Fixes: 0cf55bbef2f9 ("staging: comedi: comedi_test: implement commands
on AO subdevice")
Reported-by: Éric Piel <piel@xxxxxxxxxx>
Signed-off-by: Ian Abbott <abbotti@xxxxxxxxx>
Cc: <stable@xxxxxxxxxxxxxxx> # 4.4+
Cc: Éric Piel <piel@xxxxxxxxxx>
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
---
drivers/staging/comedi/drivers/comedi_test.c | 46 ++++++++--------------------
1 file changed, 12 insertions(+), 34 deletions(-)
diff --git a/drivers/staging/comedi/drivers/comedi_test.c b/drivers/staging/comedi/drivers/comedi_test.c
index 4ab186669f0c..ec5b9a23494d 100644
--- a/drivers/staging/comedi/drivers/comedi_test.c
+++ b/drivers/staging/comedi/drivers/comedi_test.c
@@ -56,11 +56,6 @@
#define N_CHANS 8
-enum waveform_state_bits {
- WAVEFORM_AI_RUNNING,
- WAVEFORM_AO_RUNNING
-};
-
/* Data unique to this driver */
struct waveform_private {
struct timer_list ai_timer; /* timer for AI commands */
@@ -68,7 +63,6 @@ struct waveform_private {
unsigned int wf_amplitude; /* waveform amplitude in microvolts */
unsigned int wf_period; /* waveform period in microseconds */
unsigned int wf_current; /* current time in waveform period */
- unsigned long state_bits;
unsigned int ai_scan_period; /* AI scan period in usec */
unsigned int ai_convert_period; /* AI conversion period in usec */
struct timer_list ao_timer; /* timer for AO commands */
@@ -191,10 +185,6 @@ static void waveform_ai_timer(unsigned long arg)
unsigned int nsamples;
unsigned int time_increment;
- /* check command is still active */
- if (!test_bit(WAVEFORM_AI_RUNNING, &devpriv->state_bits))
- return;
-
now = ktime_to_us(ktime_get());
nsamples = comedi_nsamples_left(s, UINT_MAX);
@@ -386,11 +376,6 @@ static int waveform_ai_cmd(struct comedi_device *dev,
*/
devpriv->ai_timer.expires =
jiffies + usecs_to_jiffies(devpriv->ai_convert_period) + 1;
-
- /* mark command as active */
- smp_mb__before_atomic();
- set_bit(WAVEFORM_AI_RUNNING, &devpriv->state_bits);
- smp_mb__after_atomic();
add_timer(&devpriv->ai_timer);
return 0;
}
@@ -400,11 +385,12 @@ static int waveform_ai_cancel(struct comedi_device *dev,
{
struct waveform_private *devpriv = dev->private;
- /* mark command as no longer active */
- clear_bit(WAVEFORM_AI_RUNNING, &devpriv->state_bits);
- smp_mb__after_atomic();
- /* cannot call del_timer_sync() as may be called from timer routine */
- del_timer(&devpriv->ai_timer);
+ if (in_softirq()) {
+ /* Assume we were called from the timer routine itself. */
+ del_timer(&devpriv->ai_timer);
+ } else {
+ del_timer_sync(&devpriv->ai_timer);
+ }
return 0;
}
@@ -436,10 +422,6 @@ static void waveform_ao_timer(unsigned long arg)
u64 scans_since;
unsigned int scans_avail = 0;
- /* check command is still active */
- if (!test_bit(WAVEFORM_AO_RUNNING, &devpriv->state_bits))
- return;
-
/* determine number of scan periods since last time */
now = ktime_to_us(ktime_get());
scans_since = now - devpriv->ao_last_scan_time;
@@ -518,11 +500,6 @@ static int waveform_ao_inttrig_start(struct comedi_device *dev,
devpriv->ao_last_scan_time = ktime_to_us(ktime_get());
devpriv->ao_timer.expires =
jiffies + usecs_to_jiffies(devpriv->ao_scan_period);
-
- /* mark command as active */
- smp_mb__before_atomic();
- set_bit(WAVEFORM_AO_RUNNING, &devpriv->state_bits);
- smp_mb__after_atomic();
add_timer(&devpriv->ao_timer);
return 1;
@@ -608,11 +585,12 @@ static int waveform_ao_cancel(struct comedi_device *dev,
struct waveform_private *devpriv = dev->private;
s->async->inttrig = NULL;
- /* mark command as no longer active */
- clear_bit(WAVEFORM_AO_RUNNING, &devpriv->state_bits);
- smp_mb__after_atomic();
- /* cannot call del_timer_sync() as may be called from timer routine */
- del_timer(&devpriv->ao_timer);
+ if (in_softirq()) {
+ /* Assume we were called from the timer routine itself. */
+ del_timer(&devpriv->ao_timer);
+ } else {
+ del_timer_sync(&devpriv->ao_timer);
+ }
return 0;
}
--
2.9.3
--
To unsubscribe from this list: send the line "unsubscribe stable" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html