]>
git.proxmox.com Git - pve-ha-manager.git/blob - src/watchdog-mux.c
11 #include <sys/socket.h>
13 #include <sys/epoll.h>
15 #include <sys/signalfd.h>
17 #include <linux/types.h>
18 #include <linux/watchdog.h>
20 #include <systemd/sd-daemon.h>
22 #define MY_SOCK_PATH "/run/watchdog-mux.sock"
23 #define WD_ACTIVE_MARKER "/run/watchdog-mux.active"
25 #define LISTEN_BACKLOG 50
28 #define WATCHDOG_DEV "/dev/watchdog"
31 int watchdog_timeout
= 20;
39 #define MAX_CLIENTS 100
41 static wd_client_t client_list
[MAX_CLIENTS
];
44 alloc_client(int fd
, time_t time
)
48 for (i
= 0; i
< MAX_CLIENTS
; i
++) {
49 if (client_list
[i
].fd
== 0) {
50 client_list
[i
].fd
= fd
;
51 client_list
[i
].time
= time
;
52 return &client_list
[i
];
60 free_client(wd_client_t
*wd_client
)
70 active_client_count(void)
74 for (i
= 0; i
< MAX_CLIENTS
; i
++) {
75 if (client_list
[i
].fd
!= 0 && client_list
[i
].time
!= 0) {
86 if (watchdog_fd
!= -1) {
87 if (write(watchdog_fd
, "V", 1) == -1) {
88 perror("write magic watchdog close");
90 if (close(watchdog_fd
) == -1) {
91 perror("write magic watchdog close");
101 struct sockaddr_un my_addr
, peer_addr
;
102 socklen_t peer_addr_size
;
103 struct epoll_event ev
, events
[MAX_EVENTS
];
104 int socket_count
, listen_sock
, nfds
, epollfd
, sigfd
;
109 if (stat(WD_ACTIVE_MARKER
, &fs
) == 0) {
110 fprintf(stderr
, "watchdog active - unable to restart watchdog-mux\n");
114 if (stat(WATCHDOG_DEV
, &fs
) == -1) {
115 system("modprobe -q softdog soft_noboot=1"); // fixme
118 if ((watchdog_fd
= open(WATCHDOG_DEV
, O_WRONLY
)) == -1) {
119 perror("watchdog open");
123 if (ioctl(watchdog_fd
, WDIOC_SETTIMEOUT
, &watchdog_timeout
) == -1) {
124 perror("watchdog set timeout");
129 /* read and log watchdog identity */
130 struct watchdog_info wdinfo
;
131 if (ioctl(watchdog_fd
, WDIOC_GETSUPPORT
, &wdinfo
) == -1) {
132 perror("read watchdog info");
137 wdinfo
.identity
[sizeof(wdinfo
.identity
) - 1] = 0; // just to be sure
138 fprintf(stderr
, "Watchdog driver '%s', version %x\n",
139 wdinfo
.identity
, wdinfo
.firmware_version
);
141 socket_count
= sd_listen_fds(0);
143 if (socket_count
> 1) {
145 perror("too many file descriptors received.\n");
148 } else if (socket_count
== 1) {
150 listen_sock
= SD_LISTEN_FDS_START
+ 0;
154 unlink(MY_SOCK_PATH
);
156 listen_sock
= socket(AF_UNIX
, SOCK_STREAM
, 0);
157 if (listen_sock
== -1) {
158 perror("socket create");
162 memset(&my_addr
, 0, sizeof(struct sockaddr_un
));
163 my_addr
.sun_family
= AF_UNIX
;
164 strncpy(my_addr
.sun_path
, MY_SOCK_PATH
, sizeof(my_addr
.sun_path
) - 1);
166 if (bind(listen_sock
, (struct sockaddr
*) &my_addr
,
167 sizeof(struct sockaddr_un
)) == -1) {
168 perror("socket bind");
172 if (listen(listen_sock
, LISTEN_BACKLOG
) == -1) {
173 perror("socket listen");
178 epollfd
= epoll_create(10);
180 perror("epoll_create");
185 ev
.data
.ptr
= alloc_client(listen_sock
, 0);
186 if (epoll_ctl(epollfd
, EPOLL_CTL_ADD
, listen_sock
, &ev
) == -1) {
187 perror("epoll_ctl add listen_sock");
193 sigaddset(&mask
, SIGINT
);
194 sigaddset(&mask
, SIGTERM
);
195 sigaddset(&mask
, SIGHUP
);
197 sigprocmask(SIG_BLOCK
, &mask
, NULL
);
199 if ((sigfd
= signalfd(-1, &mask
, SFD_NONBLOCK
)) < 0) {
200 perror("unable to open signalfd");
205 ev
.data
.ptr
= alloc_client(sigfd
, 0);
206 if (epoll_ctl(epollfd
, EPOLL_CTL_ADD
, sigfd
, &ev
) == -1) {
207 perror("epoll_ctl add sigfd");
212 nfds
= epoll_wait(epollfd
, events
, MAX_EVENTS
, 1000);
217 perror("epoll_pwait");
221 if (nfds
== 0) { // timeout
223 if (ioctl(watchdog_fd
, WDIOC_KEEPALIVE
, 0) == -1) {
224 perror("watchdog update failed");
233 for (n
= 0; n
< nfds
; ++n
) {
234 wd_client_t
*wd_client
= events
[n
].data
.ptr
;
235 if (wd_client
->fd
== listen_sock
) {
236 int conn_sock
= accept(listen_sock
, (struct sockaddr
*) &peer_addr
, &peer_addr_size
);
237 if (conn_sock
== -1) {
241 if (fcntl(conn_sock
, F_SETFL
, O_NONBLOCK
) == -1) {
242 perror("setnonblocking");
246 wd_client_t
*new_client
= alloc_client(conn_sock
, time(NULL
));
247 if (new_client
== NULL
) {
248 fprintf(stderr
, "unable to alloc wd_client structure\n");
252 mkdir(WD_ACTIVE_MARKER
, 0600);
255 ev
.data
.ptr
= new_client
;
256 if (epoll_ctl(epollfd
, EPOLL_CTL_ADD
, conn_sock
, &ev
) == -1) {
257 perror("epoll_ctl: add conn_sock");
260 } else if (wd_client
->fd
== sigfd
) {
262 /* signal handling */
265 struct signalfd_siginfo si
;
267 if ((rv
= read(sigfd
, &si
, sizeof(si
))) && rv
>= 0) {
268 if (si
.ssi_signo
== SIGHUP
) {
269 perror("got SIGHUP - ignored");
272 fprintf(stderr
, "got terminate request\n");
278 int cfd
= wd_client
->fd
;
280 ssize_t bytes
= read(cfd
, buf
, sizeof(buf
));
284 } else if (bytes
> 0) {
285 fprintf(stderr
, "GOT %zd bytes\n", bytes
);
287 if (events
[n
].events
& EPOLLHUP
|| events
[n
].events
& EPOLLERR
) {
288 printf("GOT %016x event\n", events
[n
].events
);
289 if (epoll_ctl(epollfd
, EPOLL_CTL_DEL
, cfd
, NULL
) == -1) {
290 perror("epoll_ctl: del conn_sock");
293 if (close(cfd
) == -1) {
294 perror("close conn_sock");
297 fprintf(stderr
, "close client connection\n");
298 free_client(wd_client
);
300 if (!active_client_count()) {
301 rmdir(WD_ACTIVE_MARKER
);
311 int active_count
= active_client_count();
312 if (active_count
> 0) {
313 fprintf(stderr
, "exit watchdog-mux with active connections\n");
315 fprintf(stderr
, "clean exit\n");
319 unlink(MY_SOCK_PATH
);
323 unlink(MY_SOCK_PATH
);