编辑：
此“错误”已报告here。在这个特定的用例中，它只是条件变量的未定义行为。没有稳健的条件变量，因此它们不能在具有共享内存的 IPC 中使用。线程取消会使条件变量处于不一致状态。
上一个答案如下：
我也有同样的问题。以下是导致 pthread_cond_broadcast 死锁的示例代码：
#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <pthread.h>
#include <sys/mman.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <errno.h>
#define TRUE 1
#define FALSE 0
typedef struct {
pthread_cond_t cond;
pthread_mutex_t mtx;
int predicate;
} channel_hdr_t;
typedef struct {
int fd;
channel_hdr_t *hdr;
} channel_t;
void printUsage() {
printf("usage: shm_comm_test2 channel_name1 channel_name2\n");
}
int robust_mutex_lock(pthread_mutex_t *mutex) {
// lock hdr mutex in the safe way
int lock_status = pthread_mutex_lock (mutex);
int acquired = FALSE;
int err = -18;
switch (lock_status)
{
case 0:
acquired = TRUE;
break;
case EINVAL:
printf("**** EINVAL ****\n");
err = -12;
break;
case EAGAIN:
printf("**** EAGAIN ****\n");
err = -13;
break;
case EDEADLK:
printf("**** EDEADLK ****\n");
err = -14;
break;
case EOWNERDEAD:
// the reader that acquired the mutex is dead
printf("**** EOWNERDEAD ****\n");
// recover the mutex
if (pthread_mutex_consistent(mutex) == EINVAL) {
printf("**** EOWNERDEAD, EINVAL ****\n");
err = -15;
break;
}
acquired = TRUE;
break;
default:
printf("**** OTHER ****\n");
// other error
err = -18;
break;
}
return acquired ? 0 : err;
}
int init_channel(char *shm_name, channel_t *out) {
int initialize = FALSE;
int shm_fd = shm_open (shm_name, O_RDWR | O_CREAT | O_EXCL, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
if (shm_fd < 0) {
if (errno == EEXIST) {
// open again, do not initialize
shm_fd = shm_open (shm_name, O_RDWR | O_CREAT, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
if (shm_fd < 0) {
printf( "ERROR: could not create %s, errno: %d\n", shm_name, errno );
return 1;
}
}
else {
printf( "ERROR: could not create %s, errno: %d\n", shm_name, errno );
return 2;
}
}
else {
// the shm object was created, so initialize it
initialize = TRUE;
printf("created shm object %s\n", shm_name);
if (ftruncate (shm_fd, sizeof(channel_hdr_t)) != 0)
{
printf( "ERROR: could not ftruncate %s, errno: %d\n", shm_name, errno );
close (shm_fd);
shm_unlink (shm_name);
return 3;
}
}
void *ptr_shm_hdr = mmap (NULL, sizeof(channel_hdr_t), PROT_READ | PROT_WRITE, MAP_SHARED, shm_fd, 0);
if (ptr_shm_hdr == MAP_FAILED)
{
printf( "ERROR: could not mmap %s, errno: %d\n", shm_name, errno );
close (shm_fd);
shm_unlink (shm_name);
return 4;
}
channel_hdr_t *shm_hdr = ptr_shm_hdr;
if (initialize) {
// set mutex shared between processes
pthread_mutexattr_t mutex_attr;
pthread_mutexattr_init(&mutex_attr);
pthread_mutexattr_setpshared (&mutex_attr, PTHREAD_PROCESS_SHARED);
pthread_mutexattr_setrobust (&mutex_attr, PTHREAD_MUTEX_ROBUST);
pthread_mutexattr_setprotocol(&mutex_attr, PTHREAD_PRIO_INHERIT);
pthread_mutex_init (&shm_hdr->mtx, &mutex_attr);
// set condition shared between processes
pthread_condattr_t cond_attr;
pthread_condattr_init(&cond_attr);
pthread_condattr_setpshared (&cond_attr, PTHREAD_PROCESS_SHARED);
pthread_cond_init (&shm_hdr->cond, &cond_attr);
}
shm_hdr->predicate = 0;
out->fd = shm_fd;
out->hdr = shm_hdr;
return 0;
}
int main(int argc, char **argv) {
if (argc != 3) {
printUsage();
return 0;
}
char *shm_1_name = argv[1];
char *shm_2_name = argv[2];
channel_t ch_1;
if (init_channel(shm_1_name, &ch_1) != 0) {
return 1;
}
channel_t ch_2;
if (init_channel(shm_2_name, &ch_2) != 0) {
munmap( ch_1.hdr, sizeof(channel_hdr_t) );
close( ch_1.fd );
return 2;
}
int counter = 0;
int counter2 = 0;
while (TRUE) {
++counter;
if (counter == 100000) {
printf("alive %d\n", counter2);
++counter2;
counter = 0;
}
int ret = robust_mutex_lock(&ch_1.hdr->mtx);
if (ret != 0) {
return ret;
}
ch_1.hdr->predicate = 1;
pthread_cond_broadcast (&ch_1.hdr->cond); // deadlock here
pthread_mutex_unlock (&ch_1.hdr->mtx);
ret = robust_mutex_lock(&ch_2.hdr->mtx);
if (ret != 0) {
return ret;
}
while (ch_2.hdr->predicate == 0 && ret == 0)
{
ret = pthread_cond_wait (&ch_2.hdr->cond, &ch_2.hdr->mtx); // deadlock here
}
ch_2.hdr->predicate = 0;
pthread_mutex_unlock (&ch_2.hdr->mtx);
}
munmap( ch_1.hdr, sizeof(channel_hdr_t) );
close( ch_1.fd );
munmap( ch_2.hdr, sizeof(channel_hdr_t) );
close( ch_2.fd );
return 0;
}
重现死锁：
- 使用 args 运行程序的第一个实例：channel1 channel2
- 使用 args 运行程序的第二个实例：channel2 channel1
- 使用 Ctrl+C 中断两个程序
- 再次运行这两个程序
Ubuntu 16.04 中不存在该问题。
但是，它发生在 18.04。
两个死锁程序的回溯：
第一：
#0 0x00007f9802d989f3 in futex_wait_cancelable (private=<optimized out>, expected=0, futex_word=0x7f98031cd02c)
at ../sysdeps/unix/sysv/linux/futex-internal.h:88
#1 __pthread_cond_wait_common (abstime=0x0, mutex=0x7f98031cd030, cond=0x7f98031cd000) at pthread_cond_wait.c:502
#2 __pthread_cond_wait (cond=0x7f98031cd000, mutex=0x7f98031cd030) at pthread_cond_wait.c:655
#3 0x00005648bc2af081 in main (argc=<optimized out>, argv=<optimized out>)
at /home/dseredyn/ws_velma/ws_fabric/src/shm_comm/src/test2.c:198
第二：
#0 0x00007f1a3434b724 in futex_wait (private=<optimized out>, expected=3, futex_word=0x7f1a34780010)
at ../sysdeps/unix/sysv/linux/futex-internal.h:61
#1 futex_wait_simple (private=<optimized out>, expected=3, futex_word=0x7f1a34780010)
at ../sysdeps/nptl/futex-internal.h:135
#2 __condvar_quiesce_and_switch_g1 (private=<optimized out>, g1index=<synthetic pointer>, wseq=<optimized out>,
cond=0x7f1a34780000) at pthread_cond_common.c:412
#3 __pthread_cond_broadcast (cond=0x7f1a34780000) at pthread_cond_broadcast.c:73
#4 0x0000557a978b2043 in main (argc=<optimized out>, argv=<optimized out>)
at /home/dseredyn/ws_velma/ws_fabric/src/shm_comm/src/test2.c:185