一朵浊浪映天地
0%

Linux下的wait()

Posted on Edited on In

Background

写了点代码，用于测试开启seccomp机制下的系统调用是否正常工作，思路很简单——先创建子进程，再执行系统调用测试用例，通过使用测试用例返回的errno打印测试用例错误信息，考虑到非测试系统调用的报错，决定返回-errno。核心代码如下：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
// main function
#define REPORT(func, status)    \
    do {    \
        if ((status) < 0) { \
            fprintf(stdout, "run '%s' failed during other syscall: %s\n",   \
                    (func), strerror(-status)); \
        } else if ((status) != EXIT_SUCCESS) {  \
            fprintf(stdout, "'%s' test failed: %s\n",   \
                            (func), strerror(status));  \
        } else {    \
            fprintf(stdout, "'%s' test pass!\n", (func));   \
        }   \
    } while(0)

int main(int argc, char *argv[]) {
    char *syscalls[] = {};
    ...
    for (int i = 0; i < sizeof(syscalls)/sizeof(char *); i++) {
        /* 构建执行路径 */
        retval = snprintf(path, sizeof(path), "%s/%s", prefix, syscalls[i]);
        path[retval] = '\0';

        /* fork子进程并execv */
        retval = posix_spawnp(&prog, path, NULL, NULL, argv, environ);
        if (retval != 0) {
            fprintf(stderr, "posix_spawnp('%s') failed: %s\n",
                            path, strerror(errno));
        }

        waitpid(prog, &status, 0);

        if (retval == 0)
            REPORT(syscalls[i], status);
    }
    ...

// write测试用例
int main(int argc, char *argv[]) {
    ...
    fd = syscall(__NR_open, "/dev/null", O_WRONLY);
    if (fd < 0)
        return -errno;

    syscall(__NR_write, fd, buffer, strlen(buffer));

    return errno;
}

上面的代码有两个大问题：

  1. waitpid()的返回
  2. waitpid(prog, &status, 0)执行后对status的处理

Question Answer

waitpid()的返回

waitpid()返回的是进程pid，印象中存在不成功等待子进程退出便返回的情况(被信号中断)，便查阅了《Linux环境编程》与man page，修改成了这样：

1
2
while (waitpid(prog, &status, 0) == -1 && errno == EINTR)
    continue;

子进程的返回值(exit code)

waitpid()的参数status记录了子进程的返回值(即main func最后一行的return xxx，或exit(xxx)中的xxx)。根据man page与bits/waitstatus.h头文件，可以看出status的有效位只有16 bits，高8 bit (status & 0xff00)代表子进程的返回值，低8 bit (status & 0x00ff)指示进程因哪个信号而终止(terminate)，其中0代表正常结束。

查阅到errno的值当前最大是124，因此修改成了这样：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
#define ERRNO_OFFSET    0x80

// write测试用例
int main(int argc, char *argv[]) {
    ...
    fd = syscall(__NR_open, "/dev/null", O_WRONLY);
    if (fd < 0)
        return errno+ERRNO_OFFSET;
    ...

// main function
#define REPORT(func, status)    \
    do {    \
        if (WIFEXITED((status)) &&  \
                WEXITSTATUS((status)) > ERRNO_OFFSET) { \
            fprintf(stdout, "run '%s' failed during other syscall: %s\n",   \
                    (func), strerror(WEXITSTATUS((status))-ERRNO_OFFSET));  \
        } else if (WIFEXITED((status)) &&   \
                    WEXITSTATUS((status)) != EXIT_SUCCESS) {    \
            fprintf(stdout, "'%s' test failed: %s\n",   \
                    (func), strerror(WEXITSTATUS((status))));   \
        } else if (WIFEXITED((status)) &&   \
                    WEXITSTATUS((status)) == EXIT_SUCCESS) {    \
            fprintf(stdout, "'%s' test pass!\n", (func));   \
        } else {    \
            fprintf(stdout, "'%s' test terminated abnormally\n", func); \
        }   \
    } while(0)

details

有个疑问：wait()参数wstatus为int类型，为什么只用到了16 bits？在Stack Overflow看到了下面的回答：

POSIX requires that the full exit value be passed in the si_status member of the siginfo_t structure passed to the SIGCHLD handler, if it is appropriately established via a call to sigaction with SA_SIGINFO specified in the flags:

If si_code is equal to CLD_EXITED, then si_status holds the exit value of the process; otherwise, it is equal to the signal that caused the process to change state. The exit value in si_status shall be equal to the full exit value (that is, the value passed to _exit(), _Exit(), or exit(), or returned from main()); it shall not be limited to the least significant eight bits of the value.
(Emphasis mine).

Note that upon testing, it appears that Linux does not honour this requirement and returns only the lower 8 bits of the exit code in the si_status member. Other operating systems may correctly return the full status; FreeBSD does. See test program here.

Be wary, though, that is not completely clear that you will receive an individual SIGCHLD signal for every child process termination (multiple pending instances of a signal can be merged), so this technique is not completely infallible. It is probably better to find another way to communicate a value between processes if you need more than 8 bits.

大意是：POSIX标准并没有限制进程退出时候的返回值bit数，Linux自己做了8 bits的限制，FreeBSD没有这种限制。每个子进程结束后父进程都会收到SIGCHLD信号，也可以通过处理SIGCHLD信号获取子程序的返回值。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
sig_atomic_t exit_status;

void sigchld_handler(int n, siginfo_t *si, void *v) {
    exit_status = si->si_status;
}

int main(int argc, char **argv) {
    ...
    struct sigaction act;

    act.sa_sigaction = sigchld_handler;
    act.sa_flags = SA_SIGINFO;
    sigemptyset(&act.sa_mask);

    sigaction(SIGCHLD, &act, NULL);
    ...
}

更进一步地，Linux是怎么限制进程退出时候的返回值为8 bits的？

系统调用exit()与exit_group()中有代码，还有wait4()获取exit code：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
/* in kernel/exit.c */
SYSCALL_DEFINE1(exit, int, error_code)
{
    do_exit((error_code&0xff)<<8);
}

SYSCALL_DEFINE1(exit_group, int, error_code)
{
    do_group_exit((error_code & 0xff) << 8);
    /* NOTREACHED */
    return 0;
}

// wait4
static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
{
    ...
    status = (p->signal->flags & SIGNAL_GROUP_EXIT)
        ? p->signal->group_exit_code : p->exit_code;
    wo->wo_stat = status;
    ...
out_info:
    infop = wo->wo_info;
    if (infop) {
        if ((status & 0x7f) == 0) {
            infop->cause = CLD_EXITED;
            infop->status = status >> 8;
        } else {
            infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
            infop->status = status & 0x7f;
        }
    ...

至于wait()获取的进程的低8 bit，或许与这一行代码有关：

1
2
3
4
5
6
7
8
/* in kernel/signal.c */
static void complete_signal(int sig, struct task_struct *p, enum pid_type type)
{
    ...
    signal->flags = SIGNAL_GROUP_EXIT;
    signal->group_exit_code = sig;
    signal->group_stop_count = 0;
    ...

完整代码

完整代码在这里。

Reference