[pve-ha-manager.git] / src / watchdog-mux.c

#define _GNU_SOURCE
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <fcntl.h>
#include <string.h>
#include <errno.h>
#include <time.h>
#include <sys/ioctl.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/socket.h>
#include <sys/un.h>
#include <sys/epoll.h>
#include <signal.h>
#include <sys/signalfd.h>

#include <linux/types.h>
#include <linux/watchdog.h>

#define WD_SOCK_PATH "/run/watchdog-mux.sock"
#define WD_ACTIVE_MARKER "/run/watchdog-mux.active"

#define LISTEN_BACKLOG 32

#define MAX_EVENTS 10

#define WATCHDOG_DEV "/dev/watchdog"

#define JOURNALCTL_BIN "/bin/journalctl"

int watchdog_fd = -1;
int watchdog_timeout = 10;
int client_watchdog_timeout = 60;
int update_watchdog = 1; 

typedef struct {
    int fd;
    time_t time;
    int magic_close;
} wd_client_t;

#define MAX_CLIENTS 100

static wd_client_t client_list[MAX_CLIENTS];

static wd_client_t *
alloc_client(int fd, time_t time)
{
    int i;

    for (i = 0; i < MAX_CLIENTS; i++) {
        if (client_list[i].fd == 0) {
            client_list[i].fd = fd;
            client_list[i].time = time;
            client_list[i].magic_close = 0;
            return &client_list[i];
        }
    }

    return NULL;
}

static void
free_client(wd_client_t *wd_client)
{
    if (!wd_client)
        return;

    wd_client->time = 0;
    wd_client->fd = 0;
    wd_client->magic_close = 0;
}

static int
active_client_count(void)
{
    int i, count = 0;

    for (i = 0; i < MAX_CLIENTS; i++) {
        if (client_list[i].fd != 0 && client_list[i].time != 0) {
            count++;
        }
    }
    
    return count;
}

static void 
watchdog_close(void)
{
    if (watchdog_fd != -1) {
        if (write(watchdog_fd, "V", 1) == -1) {
            perror("write magic watchdog close");
        }
        if (close(watchdog_fd) == -1) {
            perror("write magic watchdog close");
        }
    }

    watchdog_fd = -1;
}

static void
sync_journal_unsafe(void)
{

    pid_t child = fork();

    // do not care about fork error or collecting the childs exit status,
    // we are resetting soon anyway and just want to sync out the journal
    if (child == 0) {
	execl(JOURNALCTL_BIN, JOURNALCTL_BIN, "--sync", NULL);
	exit(-1);
    }
}

int 
main(void)
{
    struct sockaddr_un my_addr, peer_addr;
    socklen_t peer_addr_size;
    struct epoll_event ev, events[MAX_EVENTS];
    int listen_sock, nfds, epollfd, sigfd;

    struct stat fs;

    if (stat(WD_ACTIVE_MARKER, &fs) == 0) {
        fprintf(stderr, "watchdog active - unable to restart watchdog-mux\n");
        exit(EXIT_FAILURE);       
    }

    /* if you want to debug, set options in /lib/modprobe.d/aliases.conf
     * options softdog soft_noboot=1
     */
    if (stat(WATCHDOG_DEV, &fs) == -1) {
        char *wd_module = getenv("WATCHDOG_MODULE");
        if (wd_module) {
            char *cmd = NULL;
            if ((asprintf(&cmd, "modprobe -q %s", wd_module) == -1)) {
                perror("assemble modprobe command failed");
                exit(EXIT_FAILURE);
            }
            fprintf(stderr, "Loading watchdog module '%s'\n", wd_module);
            system(cmd);
            free(cmd);
        } else {
            system("modprobe -q softdog"); // load softdog by default
        }
    }

    if ((watchdog_fd = open(WATCHDOG_DEV, O_WRONLY)) == -1) {
         perror("watchdog open");
         exit(EXIT_FAILURE);
    }
       
    if (ioctl(watchdog_fd, WDIOC_SETTIMEOUT, &watchdog_timeout) == -1) {
        perror("watchdog set timeout");
        watchdog_close();
        exit(EXIT_FAILURE);
    }

    /* read and log watchdog identity */
    struct watchdog_info wdinfo;
    if (ioctl(watchdog_fd, WDIOC_GETSUPPORT, &wdinfo) == -1) {
        perror("read watchdog info");
        watchdog_close();
        exit(EXIT_FAILURE);
    }

    wdinfo.identity[sizeof(wdinfo.identity) - 1] = 0; // just to be sure
    fprintf(stderr, "Watchdog driver '%s', version %x\n",
            wdinfo.identity, wdinfo.firmware_version);

    /* always unlink socket path then create socket */
    unlink(WD_SOCK_PATH);

    listen_sock = socket(AF_UNIX, SOCK_STREAM, 0);
    if (listen_sock == -1) {
      perror("socket create");
      exit(EXIT_FAILURE);
    }
    memset(&my_addr, 0, sizeof(struct sockaddr_un));
    my_addr.sun_family = AF_UNIX;
    strncpy(my_addr.sun_path, WD_SOCK_PATH, sizeof(my_addr.sun_path) - 1);

    if (bind(listen_sock, (struct sockaddr *) &my_addr,
	     sizeof(struct sockaddr_un)) == -1) {
      perror("socket bind");
      exit(EXIT_FAILURE);
    }

    if (listen(listen_sock, LISTEN_BACKLOG) == -1) {
      perror("socket listen");
      goto err;
    }

    epollfd = epoll_create(10);
    if (epollfd == -1) {
        perror("epoll_create");
        goto err;
    }

    ev.events = EPOLLIN;
    ev.data.ptr = alloc_client(listen_sock, 0);
    if (epoll_ctl(epollfd, EPOLL_CTL_ADD, listen_sock, &ev) == -1) {
        perror("epoll_ctl add listen_sock");
        goto err;
    }

    sigset_t mask;
    sigemptyset(&mask);
    sigaddset(&mask, SIGINT);
    sigaddset(&mask, SIGTERM);
    sigaddset(&mask, SIGHUP);
    
    sigprocmask(SIG_BLOCK, &mask, NULL);
   
    if ((sigfd = signalfd(-1, &mask, SFD_NONBLOCK)) < 0) {
        perror("unable to open signalfd");
        goto err;
    }

    ev.events = EPOLLIN;
    ev.data.ptr = alloc_client(sigfd, 0);
    if (epoll_ctl(epollfd, EPOLL_CTL_ADD, sigfd, &ev) == -1) {
        perror("epoll_ctl add sigfd");
        goto err;
    }
  
    for (;;) {
        nfds = epoll_wait(epollfd, events, MAX_EVENTS, 1000);
        if (nfds == -1) {
            if (errno == EINTR)
                continue;
            
            perror("epoll_pwait");
            goto err;
        }

        if (nfds == 0) { // timeout

            // check for timeouts
            if (update_watchdog) {
                int i;
                time_t ctime = time(NULL);
                for (i = 0; i < MAX_CLIENTS; i++) {
                    if (client_list[i].fd != 0 && client_list[i].time != 0 &&
                        ((ctime - client_list[i].time) > client_watchdog_timeout)) {
                        update_watchdog = 0;
                        fprintf(stderr, "client watchdog expired - disable watchdog updates\n"); 
                    }
                }
            }

            if (update_watchdog) {
                if (ioctl(watchdog_fd, WDIOC_KEEPALIVE, 0) == -1) {
                    perror("watchdog update failed");
                }
            }
            
            continue;
        }

        if (!update_watchdog)
            break;
            
        int terminate = 0;
        
        int n;
        for (n = 0; n < nfds; ++n) {
            wd_client_t *wd_client = events[n].data.ptr;
            if (wd_client->fd == listen_sock) {
                int conn_sock = accept(listen_sock, (struct sockaddr *) &peer_addr, &peer_addr_size);
                if (conn_sock == -1) {
                    perror("accept");
                    goto err; // fixme
                }
                if (fcntl(conn_sock, F_SETFL, O_NONBLOCK) == -1) {
                    perror("setnonblocking");
                    goto err; // fixme
                }

                wd_client_t *new_client = alloc_client(conn_sock, time(NULL));
                if (new_client == NULL) {
                    fprintf(stderr, "unable to alloc wd_client structure\n");
                    goto err; // fixme;
                }
                
                mkdir(WD_ACTIVE_MARKER, 0600);
                
                ev.events = EPOLLIN;
                ev.data.ptr = new_client;
                if (epoll_ctl(epollfd, EPOLL_CTL_ADD, conn_sock, &ev) == -1) {
                    perror("epoll_ctl: add conn_sock");
                    goto err; // fixme                   
                }
            } else if (wd_client->fd == sigfd) {

                /* signal handling */

                int rv = 0;
                struct signalfd_siginfo si;

                if ((rv = read(sigfd, &si, sizeof(si))) && rv >= 0) {
                    if (si.ssi_signo == SIGHUP) {
                        perror("got SIGHUP - ignored");
                    } else {
                        terminate = 1;
                        fprintf(stderr, "got terminate request\n");
                    }
                }
                
            } else {
                char buf[4096];
                int cfd = wd_client->fd;
                
                ssize_t bytes = read(cfd, buf, sizeof(buf));
                if (bytes == -1) {
                    perror("read");
                    goto err; // fixme                   
                } else if (bytes > 0) {
                    int i;
                    for (i = 0; i < bytes; i++) {
                        if (buf[i] == 'V') {
                            wd_client->magic_close = 1;
                        } else {
                            wd_client->magic_close = 0;
                        }
                    }
                    wd_client->time = time(NULL);
                } else {
                    if (events[n].events & EPOLLHUP || events[n].events & EPOLLERR) {
                        //printf("GOT %016x event\n", events[n].events);
                        if (epoll_ctl(epollfd, EPOLL_CTL_DEL, cfd, NULL) == -1) {
                            perror("epoll_ctl: del conn_sock");
                            goto err; // fixme                   
                        }
                        if (close(cfd) == -1) {
                            perror("close conn_sock");
                            goto err; // fixme                   
                        }

                        if (!wd_client->magic_close) {
                            fprintf(stderr, "client did not stop watchdog - disable watchdog updates\n");
                            sync_journal_unsafe();
                            update_watchdog = 0;
                        } else {
                            free_client(wd_client);
                        }
                        
                        if (!active_client_count()) {
                            rmdir(WD_ACTIVE_MARKER);
                        }
                    }
                }
            }
        }
        if (terminate)
            break;
    }

    int active_count = active_client_count();
    if (active_count > 0) {
        fprintf(stderr, "exit watchdog-mux with active connections\n");
        sync_journal_unsafe();
    } else {
        fprintf(stderr, "clean exit\n");
        watchdog_close();
    }

    unlink(WD_SOCK_PATH);

    exit(EXIT_SUCCESS);

err:
    unlink(WD_SOCK_PATH);

    exit(EXIT_FAILURE);
}
Commit	Line	Data
	1	#define _GNU_SOURCE
	2	#include <stdio.h>
	3	#include <stdlib.h>
	4	#include <unistd.h>
	5	#include <fcntl.h>
	6	#include <string.h>
	7	#include <errno.h>
	8	#include <time.h>
	9	#include <sys/ioctl.h>
	10	#include <sys/types.h>
	11	#include <sys/stat.h>
	12	#include <sys/socket.h>
	13	#include <sys/un.h>
	14	#include <sys/epoll.h>
	15	#include <signal.h>
	16	#include <sys/signalfd.h>
	17
	18	#include <linux/types.h>
	19	#include <linux/watchdog.h>
	20
	21	#define WD_SOCK_PATH "/run/watchdog-mux.sock"
	22	#define WD_ACTIVE_MARKER "/run/watchdog-mux.active"
	23
	24	#define LISTEN_BACKLOG 32
	25
	26	#define MAX_EVENTS 10
	27
	28	#define WATCHDOG_DEV "/dev/watchdog"
	29
	30	#define JOURNALCTL_BIN "/bin/journalctl"
	31
	32	int watchdog_fd = -1;
	33	int watchdog_timeout = 10;
	34	int client_watchdog_timeout = 60;
	35	int update_watchdog = 1;
	36
	37	typedef struct {
	38	int fd;
	39	time_t time;
	40	int magic_close;
	41	} wd_client_t;
	42
	43	#define MAX_CLIENTS 100
	44
	45	static wd_client_t client_list[MAX_CLIENTS];
	46
	47	static wd_client_t *
	48	alloc_client(int fd, time_t time)
	49	{
	50	int i;
	51
	52	for (i = 0; i < MAX_CLIENTS; i++) {
	53	if (client_list[i].fd == 0) {
	54	client_list[i].fd = fd;
	55	client_list[i].time = time;
	56	client_list[i].magic_close = 0;
	57	return &client_list[i];
	58	}
	59	}
	60
	61	return NULL;
	62	}
	63
	64	static void
	65	free_client(wd_client_t *wd_client)
	66	{
	67	if (!wd_client)
	68	return;
	69
	70	wd_client->time = 0;
	71	wd_client->fd = 0;
	72	wd_client->magic_close = 0;
	73	}
	74
	75	static int
	76	active_client_count(void)
	77	{
	78	int i, count = 0;
	79
	80	for (i = 0; i < MAX_CLIENTS; i++) {
	81	if (client_list[i].fd != 0 && client_list[i].time != 0) {
	82	count++;
	83	}
	84	}
	85
	86	return count;
	87	}
	88
	89	static void
	90	watchdog_close(void)
	91	{
	92	if (watchdog_fd != -1) {
	93	if (write(watchdog_fd, "V", 1) == -1) {
	94	perror("write magic watchdog close");
	95	}
	96	if (close(watchdog_fd) == -1) {
	97	perror("write magic watchdog close");
	98	}
	99	}
	100
	101	watchdog_fd = -1;
	102	}
	103
	104	static void
	105	sync_journal_unsafe(void)
	106	{
	107
	108	pid_t child = fork();
	109
	110	// do not care about fork error or collecting the childs exit status,
	111	// we are resetting soon anyway and just want to sync out the journal
	112	if (child == 0) {
	113	execl(JOURNALCTL_BIN, JOURNALCTL_BIN, "--sync", NULL);
	114	exit(-1);
	115	}
	116	}
	117
	118	int
	119	main(void)
	120	{
	121	struct sockaddr_un my_addr, peer_addr;
	122	socklen_t peer_addr_size;
	123	struct epoll_event ev, events[MAX_EVENTS];
	124	int listen_sock, nfds, epollfd, sigfd;
	125
	126	struct stat fs;
	127
	128	if (stat(WD_ACTIVE_MARKER, &fs) == 0) {
	129	fprintf(stderr, "watchdog active - unable to restart watchdog-mux\n");
	130	exit(EXIT_FAILURE);
	131	}
	132
	133	/* if you want to debug, set options in /lib/modprobe.d/aliases.conf
	134	* options softdog soft_noboot=1
	135	*/
	136	if (stat(WATCHDOG_DEV, &fs) == -1) {
	137	char *wd_module = getenv("WATCHDOG_MODULE");
	138	if (wd_module) {
	139	char *cmd = NULL;
	140	if ((asprintf(&cmd, "modprobe -q %s", wd_module) == -1)) {
	141	perror("assemble modprobe command failed");
	142	exit(EXIT_FAILURE);
	143	}
	144	fprintf(stderr, "Loading watchdog module '%s'\n", wd_module);
	145	system(cmd);
	146	free(cmd);
	147	} else {
	148	system("modprobe -q softdog"); // load softdog by default
	149	}
	150	}
	151
	152	if ((watchdog_fd = open(WATCHDOG_DEV, O_WRONLY)) == -1) {
	153	perror("watchdog open");
	154	exit(EXIT_FAILURE);
	155	}
	156
	157	if (ioctl(watchdog_fd, WDIOC_SETTIMEOUT, &watchdog_timeout) == -1) {
	158	perror("watchdog set timeout");
	159	watchdog_close();
	160	exit(EXIT_FAILURE);
	161	}
	162
	163	/* read and log watchdog identity */
	164	struct watchdog_info wdinfo;
	165	if (ioctl(watchdog_fd, WDIOC_GETSUPPORT, &wdinfo) == -1) {
	166	perror("read watchdog info");
	167	watchdog_close();
	168	exit(EXIT_FAILURE);
	169	}
	170
	171	wdinfo.identity[sizeof(wdinfo.identity) - 1] = 0; // just to be sure
	172	fprintf(stderr, "Watchdog driver '%s', version %x\n",
	173	wdinfo.identity, wdinfo.firmware_version);
	174
	175	/* always unlink socket path then create socket */
	176	unlink(WD_SOCK_PATH);
	177
	178	listen_sock = socket(AF_UNIX, SOCK_STREAM, 0);
	179	if (listen_sock == -1) {
	180	perror("socket create");
	181	exit(EXIT_FAILURE);
	182	}
	183	memset(&my_addr, 0, sizeof(struct sockaddr_un));
	184	my_addr.sun_family = AF_UNIX;
	185	strncpy(my_addr.sun_path, WD_SOCK_PATH, sizeof(my_addr.sun_path) - 1);
	186
	187	if (bind(listen_sock, (struct sockaddr *) &my_addr,
	188	sizeof(struct sockaddr_un)) == -1) {
	189	perror("socket bind");
	190	exit(EXIT_FAILURE);
	191	}
	192
	193	if (listen(listen_sock, LISTEN_BACKLOG) == -1) {
	194	perror("socket listen");
	195	goto err;
	196	}
	197
	198	epollfd = epoll_create(10);
	199	if (epollfd == -1) {
	200	perror("epoll_create");
	201	goto err;
	202	}
	203
	204	ev.events = EPOLLIN;
	205	ev.data.ptr = alloc_client(listen_sock, 0);
	206	if (epoll_ctl(epollfd, EPOLL_CTL_ADD, listen_sock, &ev) == -1) {
	207	perror("epoll_ctl add listen_sock");
	208	goto err;
	209	}
	210
	211	sigset_t mask;
	212	sigemptyset(&mask);
	213	sigaddset(&mask, SIGINT);
	214	sigaddset(&mask, SIGTERM);
	215	sigaddset(&mask, SIGHUP);
	216
	217	sigprocmask(SIG_BLOCK, &mask, NULL);
	218
	219	if ((sigfd = signalfd(-1, &mask, SFD_NONBLOCK)) < 0) {
	220	perror("unable to open signalfd");
	221	goto err;
	222	}
	223
	224	ev.events = EPOLLIN;
	225	ev.data.ptr = alloc_client(sigfd, 0);
	226	if (epoll_ctl(epollfd, EPOLL_CTL_ADD, sigfd, &ev) == -1) {
	227	perror("epoll_ctl add sigfd");
	228	goto err;
	229	}
	230
	231	for (;;) {
	232	nfds = epoll_wait(epollfd, events, MAX_EVENTS, 1000);
	233	if (nfds == -1) {
	234	if (errno == EINTR)
	235	continue;
	236
	237	perror("epoll_pwait");
	238	goto err;
	239	}
	240
	241	if (nfds == 0) { // timeout
	242
	243	// check for timeouts
	244	if (update_watchdog) {
	245	int i;
	246	time_t ctime = time(NULL);
	247	for (i = 0; i < MAX_CLIENTS; i++) {
	248	if (client_list[i].fd != 0 && client_list[i].time != 0 &&
	249	((ctime - client_list[i].time) > client_watchdog_timeout)) {
	250	update_watchdog = 0;
	251	fprintf(stderr, "client watchdog expired - disable watchdog updates\n");
	252	}
	253	}
	254	}
	255
	256	if (update_watchdog) {
	257	if (ioctl(watchdog_fd, WDIOC_KEEPALIVE, 0) == -1) {
	258	perror("watchdog update failed");
	259	}
	260	}
	261
	262	continue;
	263	}
	264
	265	if (!update_watchdog)
	266	break;
	267
	268	int terminate = 0;
	269
	270	int n;
	271	for (n = 0; n < nfds; ++n) {
	272	wd_client_t *wd_client = events[n].data.ptr;
	273	if (wd_client->fd == listen_sock) {
	274	int conn_sock = accept(listen_sock, (struct sockaddr *) &peer_addr, &peer_addr_size);
	275	if (conn_sock == -1) {
	276	perror("accept");
	277	goto err; // fixme
	278	}
	279	if (fcntl(conn_sock, F_SETFL, O_NONBLOCK) == -1) {
	280	perror("setnonblocking");
	281	goto err; // fixme
	282	}
	283
	284	wd_client_t *new_client = alloc_client(conn_sock, time(NULL));
	285	if (new_client == NULL) {
	286	fprintf(stderr, "unable to alloc wd_client structure\n");
	287	goto err; // fixme;
	288	}
	289
	290	mkdir(WD_ACTIVE_MARKER, 0600);
	291
	292	ev.events = EPOLLIN;
	293	ev.data.ptr = new_client;
	294	if (epoll_ctl(epollfd, EPOLL_CTL_ADD, conn_sock, &ev) == -1) {
	295	perror("epoll_ctl: add conn_sock");
	296	goto err; // fixme
	297	}
	298	} else if (wd_client->fd == sigfd) {
	299
	300	/* signal handling */
	301
	302	int rv = 0;
	303	struct signalfd_siginfo si;
	304
	305	if ((rv = read(sigfd, &si, sizeof(si))) && rv >= 0) {
	306	if (si.ssi_signo == SIGHUP) {
	307	perror("got SIGHUP - ignored");
	308	} else {
	309	terminate = 1;
	310	fprintf(stderr, "got terminate request\n");
	311	}
	312	}
	313
	314	} else {
	315	char buf[4096];
	316	int cfd = wd_client->fd;
	317
	318	ssize_t bytes = read(cfd, buf, sizeof(buf));
	319	if (bytes == -1) {
	320	perror("read");
	321	goto err; // fixme
	322	} else if (bytes > 0) {
	323	int i;
	324	for (i = 0; i < bytes; i++) {
	325	if (buf[i] == 'V') {
	326	wd_client->magic_close = 1;
	327	} else {
	328	wd_client->magic_close = 0;
	329	}
	330	}
	331	wd_client->time = time(NULL);
	332	} else {
	333	if (events[n].events & EPOLLHUP \|\| events[n].events & EPOLLERR) {
	334	//printf("GOT %016x event\n", events[n].events);
	335	if (epoll_ctl(epollfd, EPOLL_CTL_DEL, cfd, NULL) == -1) {
	336	perror("epoll_ctl: del conn_sock");
	337	goto err; // fixme
	338	}
	339	if (close(cfd) == -1) {
	340	perror("close conn_sock");
	341	goto err; // fixme
	342	}
	343
	344	if (!wd_client->magic_close) {
	345	fprintf(stderr, "client did not stop watchdog - disable watchdog updates\n");
	346	sync_journal_unsafe();
	347	update_watchdog = 0;
	348	} else {
	349	free_client(wd_client);
	350	}
	351
	352	if (!active_client_count()) {
	353	rmdir(WD_ACTIVE_MARKER);
	354	}
	355	}
	356	}
	357	}
	358	}
	359	if (terminate)
	360	break;
	361	}
	362
	363	int active_count = active_client_count();
	364	if (active_count > 0) {
	365	fprintf(stderr, "exit watchdog-mux with active connections\n");
	366	sync_journal_unsafe();
	367	} else {
	368	fprintf(stderr, "clean exit\n");
	369	watchdog_close();
	370	}
	371
	372	unlink(WD_SOCK_PATH);
	373
	374	exit(EXIT_SUCCESS);
	375
	376	err:
	377	unlink(WD_SOCK_PATH);
	378
	379	exit(EXIT_FAILURE);
	380	}