int run_watchdog (int pids[], int num_pids, char* port, char* email_address, char* mail_server) {
signal (SIGPIPE, SIG_IGN);
pid_t pid = fork();
if (pid < 0) {
PERR ("fork");
exit (EXIT_FAILURE);
}
if (pid == 0) {
umask(0);
openlog ("SPROCKETS_WATCHDOG", LOG_CONS, LOG_DAEMON);
pid_t sid = setsid();
if (sid < 0) {
PMSG ("Sprockets watchdog failed to setsid. Exiting now.\n");
exit (EXIT_FAILURE);
}
if ((chdir("/")) < 0) {
PMSG ("Sprockets watchdog failed to chdir. Exiting now.\n");
exit (EXIT_FAILURE);
}
int sigs[] = { SIGINT, SIGQUIT, SIGHUP, SIGCHLD };
int sfd = setup_sighandlers (sigs, 3);
if (sfd == 0) {
PMSG ("Sprockets watchdog failed to set sighandlers. Exiting now.\n");
exit (EXIT_FAILURE);
}
int epfd = init_epoll (sfd);
if (epfd == 0) {
PMSG ("Sprockets watchdog failed to initialize epoll. Exiting now.\n");
exit (EXIT_FAILURE);
}
int server_fd = NULL;
if (port != NULL) {
server_fd = sprocket_tcp_server (port, NULL);
if (server_fd == 0) {
PMSG ("Couldn't start watchdog server\n");
exit (EXIT_FAILURE);
}
if (add_fd_to_epoll (epfd, server_fd) == 0) {
PERR ("Couldn't add watchdog server to epoll\n");
exit (EXIT_FAILURE);
}
}
close(STDIN_FILENO);
close(STDOUT_FILENO);
close(STDERR_FILENO);
syslog (LOG_NOTICE, "Sprockets watchdog successfully started\n");
int paused_for_signal = 0;
char* stats = NULL;
auto_string* stat_buffer = NULL;
pthread_mutex_t stats_mutex;
pthread_mutex_init(&stats_mutex, NULL);
auto_array* thread_array = auto_array_create (10);
if (thread_array == NULL) {
syslog (LOG_CRIT, "Unable to create thread array - out of memory");
exit (EXIT_FAILURE);
}
while (1) {
struct epoll_event events[MAX_EPOLL_EVENTS];
int num_fd = epoll_wait(epfd, events, MAX_EPOLL_EVENTS, 500);
if (num_fd == -1) {
if (errno == EINTR) {
syslog (LOG_NOTICE, "epoll_wait interrupted. Continuing\n");
continue;
}
syslog (LOG_CRIT, "epoll_wait error: %s\n", strerror (errno));
exit (EXIT_FAILURE);
}
if (num_fd != 0) { // no fds timeout occurred
for (int i = 0; i < num_fd; ++i) {
if (events[i].data.fd == sfd) { // caught signal
struct signalfd_siginfo fdsi;
int s = read (sfd, &fdsi, sizeof fdsi);
if (s != sizeof fdsi) {
syslog (LOG_CRIT, "Read signal error: %s\n", strerror (errno));
continue;
}
switch (fdsi.ssi_signo) {
case SIGINT:
syslog (LOG_NOTICE, "Caught SIGINT - pausing\n");
paused_for_signal = 1;
break;
case SIGQUIT:
syslog (LOG_NOTICE, "Caught SIGQUIT - exiting\n");
for (int ii = 0; ii < num_pids; ++ii) {
if (pids[ii] != 0) {
kill (pids[ii], SIGTERM);
}
}
for (int ii = 0; ii < thread_array->count; ++ii) {
watchdog_thread* wt = auto_array_get (thread_array, ii);
char ex[2] = "EX";
write (wt->pipe_write, ex, 2);
auto_array_delete (thread_array, free);
}
exit (EXIT_SUCCESS);
case SIGHUP:
syslog (LOG_NOTICE, "Caught SIGHUP\n");
paused_for_signal = 0;
break;
case SIGCHLD:
syslog (LOG_NOTICE, "Caught SIGCHLD\n");
break;
default:
syslog (LOG_NOTICE, "Caught unknown signal\n");
break;
}
} else if (events[i].data.fd == server_fd) {
struct sockaddr_in addr;
socklen_t addr_sz = 0;
int client_fd = accept (server_fd, (struct sockaddr*) &addr, &addr_sz);
watchdog_thread_args* thread_args = malloc (sizeof thread_args);
if (thread_args == NULL) {
syslog (LOG_CRIT, "Malloc returned NULL: %s", strerror (errno));
exit (EXIT_FAILURE);
}
watchdog_thread* dog_thread = malloc (sizeof dog_thread);
if (dog_thread == NULL) {
syslog (LOG_CRIT, "Malloc returned NULL: %s", strerror (errno));
exit (EXIT_FAILURE);
}
int pipefd[2];
if (pipe (pipefd) < 0) {
syslog (LOG_CRIT, "watchdog pipe error: %s", strerror (errno));
continue;
}
dog_thread->pipe_write = pipefd[1];
thread_args->pipe_read = pipefd[0];
thread_args->client_fd = client_fd;
thread_args->stats = &stats;
thread_args->stats_mutex = &stats_mutex;
pthread_attr_t attr;
int trv = 0;
if ((trv = pthread_attr_init (&attr)) != 0) {
syslog (LOG_CRIT, "pthread_attr_init: %s\n", strerror (trv));
continue;
}
if ((trv = pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED)) != 0) {
syslog (LOG_CRIT, "pthread_attr_setdetachstate: %s\n", strerror (trv));
continue;
}
if ((trv = pthread_create(&dog_thread->th_id, &attr, client_thread, thread_args)) != 0) {
syslog (LOG_CRIT, "pthread_create: %s\n", strerror (trv));
continue;
}
auto_array_add (thread_array, dog_thread);
pthread_attr_destroy(&attr);
}
}
}
if (paused_for_signal) {
syslog (LOG_NOTICE, "Paused for signal\n");
} else {
auto_array* stat_files = auto_array_create (num_pids);
for (int i = 0; i < num_pids; ++i) {
if (pids[i] == 0) {
continue;
}
char pid[32];
sprintf (pid, "%d", pids[i]);
FILE* p = get_stat_filep (pid);
if (p == NULL) {
syslog (LOG_CRIT, "Process with pid %s has halted", pid);
if (email_address != NULL && mail_server != NULL) {
char msg[2048];
snprintf (msg, 2048, email_alert_msg_format, email_address, email_address, pid);
send_email (mail_server, email_address, email_address, msg);
pids[i] = 0;
}
continue;
}
auto_array_add (stat_files, p);
}
pthread_mutex_lock(&stats_mutex);
if (stat_buffer != NULL) {
auto_string_delete (stat_buffer);
}
stat_buffer = auto_string_create (1024);
if (stat_buffer == NULL) {
syslog (LOG_CRIT, "Unable to create stat buffer - out of memory");
exit (EXIT_FAILURE);
}
for (int i = 0; i < stat_files->count; ++i) {
FILE* f = auto_array_get (stat_files, i);
char* tmp = get_proc_string (f);
if (tmp != NULL) {
auto_string_append (stat_buffer, tmp);
free (tmp);
}
}
stats = stat_buffer->buf;
pthread_mutex_unlock (&stats_mutex);
for (int i = 0; i < thread_array->count; ++i) {
watchdog_thread* wt = auto_array_get (thread_array, i);
char ok[2] = "OK";
if (write (wt->pipe_write, ok, 2) <= 0) {
syslog (LOG_CRIT, "watchdog write client thread: %s", strerror (errno));
auto_array_remove (thread_array, i);
close (wt->pipe_write);
free (wt);
}
}
auto_array_delete (stat_files, close_file);
}
}
}
return pid;
}
static void
event_loop(int sockfd, int hidfd)
{
int efd, i, s;
struct epoll_event events[MAXEVENTS];
struct timeval last_recv, last_heartbeat;
int clientfds[MAXCLIENTS];
int client_count = 0;
efd = epoll_create1(0);
if (efd == -1) {
perror("epoll_create");
goto out;
}
if (add_fd_to_epoll(efd, hidfd) < 0 || add_fd_to_epoll(efd, sockfd) < 0) {
goto out;
}
memset(events, 0, sizeof(events));
memset(&last_recv, 0, sizeof(last_recv));
memset(&last_heartbeat, 0, sizeof(last_heartbeat));
/* The event loop */
while (1) {
int n, item;
n = epoll_wait (efd, events, MAXEVENTS, 2000);
if (n < 0) {
if (errno != EINTR) {
perror("epoll_wait");
}
goto out;
}
if (client_count > 0) {
struct timeval now;
gettimeofday(&now, NULL);
if (now.tv_sec - last_recv.tv_sec > RESET_TIMEOUT) {
send_hmr_reset(hidfd);
last_recv = now;
}
if (now.tv_sec - last_heartbeat.tv_sec > HEARTBEAT_INTERVAL) {
send_hmr_heartbeat(hidfd);
last_heartbeat = now;
}
}
for (item = 0; item < n; item++) {
struct epoll_event *ev = &events[item];
if (ev->events & (EPOLLERR | EPOLLHUP)) {
fprintf (stderr, "epoll error\n");
if (ev->data.fd == sockfd || ev->data.fd == hidfd) {
goto out;
} else {
remove_client(ev->data.fd, clientfds, &client_count);
continue;
}
} else if (sockfd == ev->data.fd) {
/* We have a notification on the listening socket, which
* means one or more incoming connections. */
while (1) {
struct sockaddr in_addr;
socklen_t in_len;
int infd;
char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV];
in_len = sizeof(in_addr);
infd = accept(sockfd, &in_addr, &in_len);
if (infd == -1) {
if (errno == EAGAIN || errno == EWOULDBLOCK) {
/* We have processed all incoming
* connections. */
break;
} else {
perror("accept");
break;
}
}
s = getnameinfo(&in_addr, in_len, hbuf, sizeof(hbuf),
sbuf, sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV);
if (s == 0) {
printf("Accepted connection on descriptor %d "
"(host=%s, port=%s)\n", infd, hbuf, sbuf);
}
/* Make the incoming socket non-blocking and add it to the
* list of fds to monitor. */
s = make_fd_non_blocking(infd);
if (s == -1) {
close(infd);
continue;
}
if (add_fd_to_epoll(efd, infd) < 0) {
close(infd);
continue;
}
if (client_count < MAXCLIENTS) {
if (client_count == 0) {
send_hmr_reset(hidfd);
send_hmr_heartbeat(hidfd);
gettimeofday(&last_recv, NULL);
last_heartbeat = last_recv;
}
clientfds[client_count++] = infd;
}
}
} else if (hidfd == ev->data.fd) {
char buf[8];
ssize_t count = read(hidfd, buf, sizeof(buf));
gettimeofday(&last_recv, NULL);
if (count == 8) {
int cl, length = buf[0];
if (length < 8) {
for (cl = 0; cl < client_count; ) {
if (write_exact(clientfds[cl], buf + 1, length) < 0) {
remove_client(clientfds[cl], clientfds, &client_count);
/* keep cl, as remove_client shifted the array */
} else {
cl++;
}
}
}
}
} else {
/* We have data on the fd waiting to be read. Read and
* discard it. We must read whatever data is available
* completely, as we are running in edge-triggered mode
* and won't get a notification again for the same
* data. */
int done = 0;
while (1) {
ssize_t count;
char buf[512];
count = read (ev->data.fd, buf, sizeof buf);
if (count == -1) {
/* If errno == EAGAIN, that means we have read all
* data. So go back to the main loop. */
if (errno != EAGAIN && errno != EWOULDBLOCK) {
perror ("read");
done = 1;
}
break;
} else if (count == 0) {
/* End of file. The remote has closed the connection */
done = 1;
break;
}
}
if (done) {
printf ("Closed connection on descriptor %d\n", ev->data.fd);
/* Closing the descriptor will make epoll remove it
* from the set of descriptors which are monitored. */
remove_client(ev->data.fd, clientfds, &client_count);
}
}
}
}
out:
for (i = 0; i < client_count; i++) {
close(clientfds[i]);
}
if (efd >= 0) {
close(efd);
}
}
static
void
worker_thread_handle_msgqueue_event(
struct worker_thread* state,
uint32_t event_flags
)
{
if (event_flags & EPOLLERR) {
D_FMTSTRING("Error on message queue!");
return;
}
for (;;) {
/*
* Drain all the bytes from the queue descriptor.
*/
struct message msg;
ssize_t bytes = HANDLE_EINTR_ON_SYSCALL(
mq_receive(state->wk_messagequeue.mq_queuefds,
(char*) &msg,
sizeof(msg),
NULL /* don't care about message priority */));
if (bytes == -1) {
if (errno == EAGAIN) {
return;
}
D_FUNCFAIL_ERRNO(mq_receive);
return;
}
BUGSTOP_IF((msg.msg_code != kITTMessageAddClient),
"Unknown message code");
/*
* The client object gets ownership of the socket descriptor.
* Should anything go wrong in the creation process it will
* close the socket descriptor.
*/
struct client* clnt = client_create(msg.msg_data.msg_fd,
state->wk_allocator);
if (!clnt) {
/*
* Client creation failed. Try to get next message, if any.
* Close client socket.
*/
HANDLE_EINTR_ON_SYSCALL(close(msg.msg_data.msg_fd));
continue;
}
if (add_fd_to_epoll(state->wk_epoll_fds, clnt->cl_sockfd,
EPOLLIN | EPOLLRDHUP, kDataTypePTR,
clnt) == -1) {
/*
* Failed to add client socket to epoll so release all resources and get
* the next message.
*/
client_destroy(clnt, state->wk_allocator);
continue;
}
/*
* Add client to list.
*/
dlist_push_tail(state->wk_clients, clnt);
}
}
static
int
server_init(
struct server* p_srv
)
{
assert(p_srv);
memset(p_srv, 0, sizeof(*p_srv));
p_srv->sv_allocator = allocator_handle;
p_srv->sv_quitflag = 0;
p_srv->sv_acceptfd = create_server_socket();
if (-1 == p_srv->sv_acceptfd) {
return -1;
}
/*
* Level 1 - accept socket created.
*/
++p_srv->sv_rollback;
p_srv->sv_epollfd = epoll_create(kMaxEpollCompletionEntries);
if (-1 == p_srv->sv_epollfd) {
return -1;
}
/*
* Level 2 - epoll descriptor allocated.
*/
++p_srv->sv_rollback;
/*
* Block SIGINT and create a signal descriptor to receive it via epoll.
*/
sigset_t sig_mask;
sigemptyset(&sig_mask);
if (-1 == sigaddset(&sig_mask, SIGINT)) {
return -1;
}
if (-1 == sigprocmask(SIG_BLOCK, &sig_mask, NULL)) {
return -1;
}
p_srv->sv_sigfds = signalfd(-1, &sig_mask, SFD_NONBLOCK);
if (-1 == p_srv->sv_sigfds) {
return -1;
}
/*
* Level 3 - signal descriptor for SIGINT allocated.
*/
++p_srv->sv_rollback;
/*
* Add termination signal and accept socket to epoll interface.
*/
if (-1 == add_fd_to_epoll(p_srv->sv_epollfd,
p_srv->sv_sigfds,
EPOLLIN | EPOLLET,
kDataTypeFD,
p_srv->sv_sigfds)) {
return -1;
}
if (-1 == add_fd_to_epoll(p_srv->sv_epollfd,
p_srv->sv_acceptfd,
EPOLLIN | EPOLLET | EPOLLRDHUP,
kDataTypeFD,
p_srv->sv_acceptfd)) {
return -1;
}
p_srv->sv_threadrdy_eventfds = eventfd(0, 0);
if (p_srv->sv_threadrdy_eventfds == -1) {
D_FUNCFAIL_ERRNO(eventfd);
return -1;
}
/*
* Level 4 - thread notification event created.
*/
++p_srv->sv_rollback;
/*
* Get number of available processors. The number of spawned threads is
* nr_processors * thread_to_proc_ratio.
*/
long nr_procs = sysconf(_SC_NPROCESSORS_ONLN);
if (nr_procs == -1) {
D_FUNCFAIL_ERRNO(sysconf);
return -1;
}
D_FMTSTRING("Online processors %d, will spawn %d threads.",
nr_procs, nr_procs);
p_srv->sv_workers = p_srv->sv_allocator->al_mem_alloc(
p_srv->sv_allocator, (sizeof(struct worker_thread*) * nr_procs));
if (!p_srv->sv_workers) {
D_FMTSTRING("Out of memory!");
return -1;
}
/*
* Level 5 - memory for worker thread data allocated.
*/
++p_srv->sv_rollback;
memset(p_srv->sv_workers, 0, sizeof(struct worker_thread*) * nr_procs);
/*
* Initialize data and start worker threads.
*/
for (long l = 0; l < nr_procs; ++l) {
char thread_msgqueue[NAME_MAX];
snprintf(thread_msgqueue, sizeof(thread_msgqueue) - 1,
"/__msgqueue_thread_%d__", (int) l);
struct worker_thread* current = worker_thread_create(thread_msgqueue,
p_srv->sv_allocator);
if (current) {
if (worker_thread_start(p_srv, current, NULL) == 0) {
/*
* Thread successfully initialized, add it to list.
*/
p_srv->sv_workers[p_srv->sv_worker_count++] = current;
} else {
/*
* Cleanup thread data since pthread_create() failed.
*/
worker_thread_destroy(current);
}
}
}
if (!p_srv->sv_worker_count) {
D_FMTSTRING("Fatal : failed to initialize at least one worker thread!");
return -1;
}
D_FMTSTRING("Started a total of %d worker threads", p_srv->sv_worker_count);
/*
* Server is up and running.
*/
return 0;
}
/*
* @@ Not implemented. @@
*/
static
void*
worker_thread_proc(
void* args
) {
struct worker_thread* state = (struct worker_thread*) args;
BUGSTOP_IF((!state), "Invalid thread state specified!");
D_FMTSTRING("Client thread (%u) starting\n", syscall(SYS_gettid));
/*
* Add the message queue and the termination event to epoll.
*/
int result = 0;
if (add_fd_to_epoll(state->wk_epoll_fds,
state->wk_termsig.so_sigfds,
EPOLLIN | EPOLLET,
kDataTypePTR,
(void*) &state->wk_termsig) == 0) {
++result;
}
if (add_fd_to_epoll(state->wk_epoll_fds,
state->wk_messagequeue.mq_queuefds,
EPOLLIN | EPOLLET,
kDataTypePTR,
(void*) &state->wk_messagequeue) == 0) {
++result;
}
/*
* Notify waiter with our initialize status.
*/
uint64_t init_status = (result == 2 ? kThreadInitOk : kThreadInitFail);
HANDLE_EINTR_ON_SYSCALL(write(state->wk_readyevent, &init_status,
sizeof(init_status)));
if (result != 2) {
/*
* Failed to init so return.
*/
return NULL;
}
/*
* Loop forever waiting for events.
*/
for (; !state->wk_quitflag;) {
struct epoll_event rec_events[kMaxEpollCompletionEntries];
int ev_count = HANDLE_EINTR_ON_SYSCALL(epoll_wait(state->wk_epoll_fds,
rec_events,
kMaxEpollCompletionEntries,
-1 /* don;t timeout */));
if (ev_count == -1) {
D_FUNCFAIL_ERRNO(epoll_wait);
break;
}
for (int i = 0; i < ev_count && !state->wk_quitflag; ++i) {
worker_thread_handle_event(state, rec_events + i);
}
}
return NULL;
}
