sem_timedwait Fails at Random

Question

I've written a small class that encapsulates shared locks and shared semaphores. These classes pass all of my unit tests, except for one: about 60% of the time locking a shared memory semaphore using sem_timedwait fails. Here's my code:

SharedLock::SharedLock(const SharedSemaphore& semaphore, unsigned int wait_ns) :
    Succeeded(false),
    Semaphore(semaphore)
{
    timespec wait;
    if(clock_gettime(CLOCK_REALTIME, &wait) < 0)
    {
        throw exception_ty(SHARED_LOCK_RUNTIME_ERROR,
            "SharedLock::SharedLock(...) failed: clock_gettime(...) encountered an error.");
    }

    unsigned long ns = wait_ns % 1000000000;
    wait.tv_sec += (wait_ns - ns) / 1000000000;
    wait.tv_nsec += ns;
    int result = sem_timedwait(const_cast<sem_t*>(Semaphore.Semaphore.get_data()), &wait);

    if(result < 0 && errno != ETIMEDOUT)
    {
        throw_on_sem_wait();
    }

    else if(result == 0)
    {
        Succeeded = true;
    }
}

I am able to lock and unlock this semaphore without any errors using sem_wait and sem_trywait; only sem_timedwait fails randomly. When it fails it always sets errno to EINVAL. I am running these tests on Ubuntu Linux 12.10 (64 bit).

Could anyone suggest why I am having these problems, and how I may solve them?

Answer 1

EINVAL is returned either if sem is not a livad semaphore, or the value of abs_timeout.tv_nsecs is less than 0, or greater than or equal to 1000 million.

In your case sem should be a valid one, so the possible reason is that the wait.tv_nsec exceeds 1000 million, because it looks possible for the code wait.tv_nsec += ns; does not check the value.

You could print its value for debugging purpose, to see if it's the root cause.