pthread_mutex_timedlock()过早退出而不等待超时

Question

我想保护一个函数免受多线程访问。为此，我使用了一个pthread_mutex_t互斥锁。我尝试在函数开头锁定它，然后执行函数，然后再释放它。如果互斥锁正在使用中，它应该等待最多60秒钟，直到可用为止。如果在那之后仍然无法获得，函数就应该失败。

我遇到的问题是pthread_mutex_timedlock似乎完全忽略了我给它的超时值。虽然我指定了60秒的超时时间，但如果锁已被占用，则函数立即返回错误代码ETIMEDOUT - 而不是实际等待。

以下是一个最小化的示例，可以重现该问题。在这种情况下，无论我使用递归或非递归互斥锁都没有关系，因为我不会从同一线程多次尝试锁定它们。

#include <stdlib.h>
#include <stdio.h>
#include <stdint.h>
#include <stdbool.h>
#include <string.h>
#include <stddef.h> 
#include <unistd.h>
#include <time.h>
#include <errno.h>
#include <pthread.h>

pthread_mutex_t lock; /* exclusive lock */

//do some work to keep the processor busy..
int wut() {
    int x = 0;
    for(int i=0; i < 1024*1024*1024; i++)
        x += 1;
    return x;
}

void InitMutex(){
    /*pthread_mutexattr_t Attr;
    pthread_mutexattr_init(&Attr);
    pthread_mutexattr_settype(&Attr, PTHREAD_MUTEX_RECURSIVE);
    pthread_mutex_init(&lock, &Attr);*/
    pthread_mutex_init(&lock, NULL);
}

//lock mutex, wait at maximum 60 seconds, return sucesss
int LockMutex() {
    struct timespec timeoutTime;
    timeoutTime.tv_nsec = 0;
    timeoutTime.tv_sec = 60;
    printf("Nanoseconds: %lu, seconds %lu\n", timeoutTime.tv_nsec, timeoutTime.tv_sec);
    int retVal = pthread_mutex_timedlock(&lock, &timeoutTime);
    printf("pthread_mutex_timedlock(): %d\n", retVal);
    if(retVal != 0) {
        const char* errVal = NULL;
        switch(retVal) {
        case EINVAL: errVal = "EINVAL"; break;
        case EAGAIN: errVal = "EAGAIN"; break;
        case ETIMEDOUT: errVal = "ETIMEDOUT"; break;
        case EDEADLK: errVal = "EDEADLK"; break;
        default: errVal = "unknown.."; break;
        }
        printf("Error taking lock in thread %lu: %s (%s)\n", pthread_self(), errVal , strerror(retVal));
    }
    return retVal == 0; //indicate success/failure
}

void UnlockMutex() {
    pthread_mutex_unlock(&lock);
}

void TestLockNative() {
    uint64_t thread_id = pthread_self();
    printf("Trying to take lock in thread %lu.\n", thread_id);
    int ret = LockMutex();
    printf("Got lock in thread %lu. sucess=%d\n", thread_id, ret);
    wut();
    printf("Giving up lock now from thread %lu.\n", thread_id);
    UnlockMutex();

}

void* test_thread(void* arg) {
    //TestLock();
    TestLockNative();
    return NULL;
}

int main() {
    InitMutex();
    //create two threads which will try to access the protected function at once
    pthread_t t1, t2;
    pthread_create(&t1, NULL, &test_thread, NULL);
    pthread_create(&t2, NULL, &test_thread, NULL);

    //wait for threads to end
    pthread_join(t1, NULL);
    pthread_join(t2, NULL);

    return 0;
}

程序的输出示例如下：

Trying to take lock in thread 139845914396416.
Nanoseconds: 0, seconds 6000
pthread_mutex_timedlock(): 0
Got lock in thread 139845914396416. sucess=1
Trying to take lock in thread 139845906003712.
Nanoseconds: 0, seconds 6000
pthread_mutex_timedlock(): 110
Error taking lock in thread 139845906003712: ETIMEDOUT (Connection timed out) [<-- this occurs immediately, not after 60 seconds]
Got lock in thread 139845906003712. sucess=0
Giving up lock now from thread 139845906003712.

使用gcc -o test test.c -lpthread进行编译应该可以正常工作。

那么，有人知道这里发生了什么，为什么pthread_mutex_timedlock()忽略了我的超时值吗？ 它根本不按照文档中所述的方式行事。

我正在使用Ubuntu 16.04.2 LTS系统，并使用gcc编译。

- Maximilian Gerhardt

Answer 1

pthread_mutex_timedlock的手册页如下所示：

当由超时时间 abstime 指定的绝对时间到达时，基于超时的时钟度量的时间到期

因此，请使用实时时间来指定超时值：

int LockMutex() {
    struct timespec timeoutTime;
    clock_gettime(CLOCK_REALTIME, &timeoutTime);
    timeoutTime.tv_sec += 60;

    int retVal = pthread_mutex_timedlock(&lock, &timeoutTime);
    ....

- LWimsey

---

一个简单的单词，或者说，我无法阅读文档中参数值或此行导致了这个问题。非常感谢。 - Maximilian Gerhardt

很高兴能帮到你 :) - LWimsey