[apue] 作为 daemon 启动, Unix Domain Socket 侦听失败?

前段时间写一个传递文件句柄的小 demo，有 server 端、有 client 端，之间通过 Unix Domain Socket 通讯。

在普通模式下，双方可以正常建立连接，当server端作为daemon启动时，则第一次启动成功，之后再启动， listen 会连接报 ENOTSUPP 错误，导致启动失败。

spipe.c

 1 int cli_conn(const char *name)
 2 {
 3     int fd, len, err, rval;
 4     struct sockaddr_un un;
 5
 6     if ((fd = socket (AF_UNIX, SOCK_STREAM, 0)) < 0) {
 7         printf ("create socket failed\n");
 8         return -1;
 9     }
10
11     printf ("create socket ok\n");
12     memset (&un, 0, sizeof (un));
13     un.sun_family = AF_UNIX;
14     strcpy (un.sun_path, name);
15     len = offsetof (struct sockaddr_un, sun_path) + strlen (name);
16     if (connect (fd, (struct sockaddr *)&un, len) < 0) {
17         err = errno;
18         printf ("connect failed\n");
19         rval = -4;
20         goto errout;
21     }
22
23     printf ("connect to server ok\n");
24     return fd;
25 errout:
26     close (fd);
27     errno = err;
28     return rval;
29 }
30
31
32 int serv_listen (const char *name)
33 {
34     int fd, len, err, rval;
35     struct sockaddr_un un;
36
37     if ((fd = socket (AF_UNIX, SOCK_STREAM, 0)) < 0) {
38         printf ("socket failed\n");
39         return -1;
40     }
41
42     printf ("create socket ok\n");
43     unlink (name);
44     memset (&un, 0, sizeof(un));
45     un.sun_family = AF_UNIX;
46     strcpy (un.sun_path, name);
47     len = offsetof (struct sockaddr_un, sun_path) + strlen (name);
48
49     if (bind (fd, (struct sockaddr *)&un, len) < 0) {
50         err = errno;
51         printf ("bind failed\n");
52         rval = -2;
53         goto errout;
54     }
55
56     printf ("bind socket to path ok\n");
57     if (listen (fd, QLEN) < 0) {
58         err = errno;
59         printf ("listen failed, errno %d\n", errno);
60         rval = -3;
61         goto errout;
62     }
63
64     printf ("start listen on socket ok\n");
65     return fd;
66 errout:
67     close (fd);
68     errno = err;
69     return rval;
70 }
71
72 int serv_accept (int listenfd, uid_t *uidptr)
73 {
74     int clifd, err, rval;
75     time_t staletime;
76     struct sockaddr_un un;
77     struct stat statbuf;
78
79     size_t len = sizeof (un);
80     if ((clifd = accept (listenfd, (struct sockaddr *)&un, &len)) < 0) {
81         printf ("accept failed\n");
82         return -1;
83     }
84
85     len -= offsetof (struct sockaddr_un, sun_path);
86     un.sun_path[len] = 0;
87     printf ("accept %s ok\n", un.sun_path);
88
89     unlink (un.sun_path);
90     return clifd;
91
92 errout:
93     close (clifd);
94     errno = err;
95     return rval;
96 }

出错的位置在 serv_listen (line 57) 处，出错时的 server 端输出为：

Jan 17 00:24:44 localhost opend: create socket ok
Jan 17 00:24:44 localhost opend: bind socket to path ok
Jan 17 00:24:44 localhost opend: listen failed, errno 95
Jan 17 00:24:44 localhost opend: serv_listen error: Operation not supported

errno 95 为 ENOTSUPP。不以 daemon 运行时正常的输出如下：

create socket ok
bind socket to path ok
start listen on socket ok
accept  ok
new connection: uid 0, fd 4

可能细心的读者会觉得，以 daemon 方式运行 printf 怎么还可以输出呢，是有以下宏定义做了处理：

1 #ifdef USE_APUE
2 #include "../apue.h"
3 #define printf log_msg
4 #endif

以 daemon 运行时会定义 USE_APUE 宏，从而将 printf 重定义为 log_msg 输出到 syslog。

下面是 server 端的代码：

csopend2.c

 1 int main (int argc, char *argv[])
 2 {
 3     int c = 0;
 4     log_open ("open.serv", LOG_PID, LOG_USER);
 5
 6     opterr = 0; // don't want getopt() writting to stderr !
 7     while ((c = getopt (argc, argv, "d")) != EOF) {
 8         switch (c) {
 9             case 'd':
10                 debug = log_to_stderr = 1;
11                 break;
12             case '?':
13                 err_quit ("unrecongnized option: -%c", optopt);
14         }
15     }
16
17     if (debug == 0)
18     {
19         log_to_stderr = 0;
20         daemonize ("opend");
21     }
22
23     loop ();
24     return 0;
25 }

不使用 -d 时表示 daemon 运行(与常识相反?)，上面标黄的代码就是。

对应的 client 端代码：

csopenc.c

一开始怀疑是用于 listen 的本地 socket 文件已经存在，于是去 /tmp 目录看了下，果然有 opend 这个文件，删除之，再运行，不行；

然后怀疑是没有复用端口(?)所致，于是在 listen 之前添加了以下代码段：

1     int opt = 1;
2     if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (const void *)&opt, sizeof(opt)) < 0) {
3         err = errno;
4         printf ("setsockopt failed\n");
5         rval = -3;
6         goto errout;
7     }

设置端口复用。编译、运行，输出如下：

Jan 17 00:43:11 localhost opend: create socket ok
Jan 17 00:43:11 localhost opend: bind socket to path ok
Jan 17 00:43:11 localhost opend: set socket option ok
Jan 17 00:43:11 localhost opend: listen failed, errno 95
Jan 17 00:43:11 localhost opend: serv_listen error: Operation not supported

设置成功了，但还是不行

难道 daemon 与普通进程使用 Unix 域套接字还有什么区别么?

暂时存疑……