Race condition in pthread_mutex_timedlock?
Luca Bacci
luca.bacci@outlook.com
Fri Oct 23 11:57:26 GMT 2026
Yeah, that's a good idea :) See the attached STC -- that sample triggers
the error quite frequently here (YMMV!)
Best Regards!
Luca
On 10/23/2026 1:22 AM, Mark Geisert wrote:
> Hello Luca,
>
> On 10/20/2026 7:36 AM, Luca Bacci wrote:
>> Hello,
>>
>> Current implementation of pthread_mutex_timelock() seems racy; if
>> timedwait() times
>> while unlock() is running on another thread, the mutex can be left
>> unlocked and
>> with the internal event set.
>>
>> Consider this:
>>
>> Thread A Thread B
>>
>> timedlock()
>> counter 0 -> 1
>>
>> timedlock()
>> counter 1 -> 2
>> (waits with timeout)
>>
>>
>>
>> unlock()
>> counter 2 -> 1
>> (resumes due to timeout)
>> counter 1 -> 0
>> set event
>>
>> Now the mutex is unlocked but the internal event is left in the
>> signaled state. Whenever contention occurs
>> again, two threads will acquire the lock.
>>
>> Am I missing something?
>
> Just speaking for myself, I don't know offhand if you're missing
> something or not and I don't have time to investigate.
>
> I invite you to write (i.e., code) what we call an STC (Simple Test
> Case) that tests the situation you raise. In other words, try to
> break the existing Cygwin code. Write it so you can vary the number
> of threads. Use assert() to test conditions along the way. If you do
> find an issue then we can investigate. I can also then offer to test
> your STC on a 16C/32T machine with dozens or hundreds of threads :-).
> Regards,
>
> ..mark
-------------- next part --------------
#include <stdlib.h>
#include <stdbool.h>
#include <time.h>
#include <errno.h>
#include <assert.h>
#include <pthread.h>
#define NUM_THREADS 5
#define ITERATIONS 100000
#define MILLISECONDS 1
static pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
static _Atomic bool owned;
static void
sleep_milliseconds (int milliseconds)
{
struct timespec ts = {
.tv_sec = (milliseconds / 1000),
.tv_nsec = (milliseconds % 1000) * 1000 * 1000
};
nanosleep (&ts, NULL);
}
static void *
worker_main (void *user_data)
{
for (int i = 0; i < ITERATIONS; i++)
{
struct timespec ts;
int ret;
ret = clock_gettime (CLOCK_REALTIME, &ts);
assert (ret == 0);
ts.tv_sec += MILLISECONDS / 1000;
ts.tv_nsec += (MILLISECONDS % 1000) * 1000 * 1000;
if (ts.tv_nsec >= 1000 * 1000 * 1000)
{
ts.tv_nsec -= 1000 * 1000 * 1000;
ts.tv_sec++;
}
ret = pthread_mutex_timedlock (&mutex, &ts);
if (ret == ETIMEDOUT)
continue;
assert (ret == 0);
assert (owned == false);
owned = true;
sleep_milliseconds (MILLISECONDS);
assert (owned == true);
owned = false;
ret = pthread_mutex_unlock (&mutex);
assert (ret == 0);
}
}
int
main (void)
{
pthread_t threads[NUM_THREADS];
for (int i = 0; i < NUM_THREADS; i++)
pthread_create (&threads[i], NULL, worker_main, NULL);
for (int i = 0; i < NUM_THREADS; i++)
pthread_join (threads[i], NULL);
return 0;
}
More information about the Cygwin-developers
mailing list