int main(int argc, char **argv)
{
simple_worker *wp;
int i;
#ifdef HAVE_SIGACTION
struct sigaction sig_action;
sig_action.sa_sigaction = NULL;
sigfillset(&sig_action.sa_mask);
sig_action.sa_flags=SA_NOCLDSTOP;
sig_action.sa_handler = child_exited;
sigaction(SIGCHLD, &sig_action, NULL);
sig_action.sa_flags = SA_NODEFER|SA_RESTART;
sig_action.sa_handler = sighandler;
sigfillset(&sig_action.sa_mask);
sigaction(SIGINT, &sig_action, NULL);
sigaction(SIGPIPE, &sig_action, NULL);
#else /* HAVE_SIGACTION */
signal(SIGINT, sighandler);
signal(SIGPIPE, sighandler);
signal(SIGCHLD, child_exited);
#endif /* HAVE_SIGACTION */
iobs = iobroker_create();
if (!iobs)
die("Failed to create io broker set");
for (i = 0; i < NWPS; i++) {
wp = spawn_worker(print_some_crap, "lalala");
if (!wp) {
die("Failed to spawn worker(s)\n");
}
wps[i] = wp;
printf("Registering worker sd %d with io broker\n", wp->sd);
iobroker_register(iobs, wp->sd, wp, print_input);
}
iobroker_register(iobs, fileno(stdin), NULL, send_command);
/* get to work */
while (!sigreceived && iobroker_get_num_fds(iobs)) {
iobroker_poll(iobs, -1);
}
for (i = 0; i < NWPS; i++) {
kill(wps[i]->pid, SIGKILL);
}
return 0;
}
/*
* This gets called when a connect() attempt has become writable.
* It's entirely possible that the node we're trying to connect
* to has connected to us while we were waiting for them, in
* which case we need to figure out which of the two connections
* we're supposed to use.
*/
static int conn_writable(int sd, int events, void *node_)
{
merlin_node *node = (merlin_node *)node_;
int result;
int sel_sd;
/* unregister so we don't peg one cpu at 100% */
ldebug("CONN: In conn_writable(): node=%s; sd=%d; node->conn_sock=%d", node->name, sd, node->conn_sock);
iobroker_unregister(nagios_iobs, sd);
if (node->sock < 0) {
/* no inbound connection accept()'ed yet */
node->sock = sd;
node->conn_sock = -1;
if (!net_is_connected(node)) {
node_disconnect(node, "Connection attempt failed: %s", strerror(errno));
close(sd);
return 0;
}
iobroker_register(nagios_iobs, sd, node, net_input);
node_set_state(node, STATE_NEGOTIATING, "Connect completed successfully. Negotiating protocol");
return 0;
}
sel_sd = net_negotiate_socket(node, node->conn_sock, node->sock);
if (sel_sd < 0) {
node_disconnect(node, "Failed to negotiate socket");
return 0;
}
if (sel_sd == node->conn_sock) {
iobroker_close(nagios_iobs, node->sock);
} else if (sel_sd == node->sock) {
iobroker_close(nagios_iobs, node->conn_sock);
}
node->sock = sel_sd;
node->conn_sock = -1;
node_set_state(node, STATE_NEGOTIATING, "polled for writability");
/* now re-register for input */
ldebug("IOB: registering %s(%d) for input events", node->name, node->sock);
result = iobroker_register(nagios_iobs, node->sock, node, net_input);
if (result < 0) {
lerr("IOB: Failed to register %s(%d) for input events: %s",
node->name, node->sock, iobroker_strerror(result));
}
return 0;
}
static int fd_start_cmd(child_process *cp)
{
int pfd[2], pfderr[2];
cp->outstd.fd = runcmd_open(cp->cmd, pfd, pfderr, NULL);
if (cp->outstd.fd == -1) {
return -1;
}
gettimeofday(&cp->start, NULL);
cp->outerr.fd = pfderr[0];
cp->pid = runcmd_pid(cp->outstd.fd);
iobroker_register(iobs, cp->outstd.fd, cp, stdout_handler);
iobroker_register(iobs, cp->outerr.fd, cp, stderr_handler);
return 0;
}
int init_workers(int desired_workers)
{
worker_process **wps;
int i;
if (desired_workers <= 0) {
desired_workers = 4;
}
if (workers_alive() == desired_workers)
return 0;
/* can't shrink the number of workers (yet) */
if (desired_workers < num_workers)
return -1;
wps = calloc(desired_workers, sizeof(worker_process *));
if (!wps)
return -1;
if (workers) {
if (num_workers < desired_workers) {
for (i = 0; i < num_workers; i++) {
wps[i] = workers[i];
}
}
free(workers);
}
workers = wps;
for (i = 0; i < desired_workers; i++) {
int ret;
worker_process *wp;
if (wps[i])
continue;
wp = spawn_worker(worker_init_func, (void *)get_global_macros());
if (!wp) {
logit(NSLOG_RUNTIME_WARNING, TRUE, "Failed to spawn worker: %s\n", strerror(errno));
free_worker_memory(0);
return ERROR;
}
set_socket_options(wp->sd, 256 * 1024);
wps[i] = wp;
ret = iobroker_register(nagios_iobs, wp->sd, wp, handle_worker_result);
if (ret < 0) {
printf("Failed to register worker socket with iobroker %p\n", nagios_iobs);
exit(1);
}
}
num_workers = desired_workers;
logit(NSLOG_INFO_MESSAGE, TRUE, "Workers spawned: %d\n", num_workers);
return 0;
}
/*
* Accept an inbound connection from a remote host
* Returns 0 on success and -1 on errors
*/
static int net_accept_one(int sd, int events, void *discard)
{
int sock, result;
merlin_node *node;
struct sockaddr_in sain;
socklen_t slen = sizeof(struct sockaddr_in);
sock = accept(sd, (struct sockaddr *)&sain, &slen);
if (sock < 0) {
lerr("accept() failed: %s", strerror(errno));
return -1;
}
node = find_node(&sain);
linfo("NODESTATE: %s connected from %s:%d. Current state is %s",
node ? node->name : "An unregistered node",
inet_ntoa(sain.sin_addr), ntohs(sain.sin_port),
node ? node_state_name(node->state) : "unknown");
if (!node) {
close(sock);
return 0;
}
switch (node->state) {
case STATE_NEGOTIATING:
case STATE_CONNECTED: case STATE_PENDING:
/* if node->sock >= 0, we must negotiate which one to use */
if (node->sock >= 0) {
int sel_sd = net_negotiate_socket(node, node->sock, sock);
if (sel_sd != sock) {
close(sock);
}
}
break;
case STATE_NONE:
/*
* we must close it unconditionally or we'll leak fd's
* for reconnecting nodes that were previously connected
*/
node_disconnect(node, "fd leak prevention for connecting nodes");
node->sock = sock;
break;
default:
lerr("%s %s has an unknown status", node_type(node), node->name);
break;
}
node_set_state(node, STATE_NEGOTIATING, "Inbound connection accepted. Negotiating protocol version");
result = iobroker_register(nagios_iobs, node->sock, node, net_input);
if (result < 0) {
lerr("IOB: Failed to register %d for %s node %s for input events: %s",
node->sock, node_type(node), node->name, iobroker_strerror(result));
}
return sock;
}
int main(int argc, char **argv)
{
int listen_fd, flags, sockopt = 1;
struct sockaddr_in sain;
int error;
const char *err_msg;
t_set_colors(0);
t_start("iobroker ipc test");
error = iobroker_get_max_fds(NULL);
ok_int(error, IOBROKER_ENOSET, "test errors when passing null");
err_msg = iobroker_strerror(error);
test(err_msg && !strcmp(err_msg, iobroker_errors[(~error) + 1].string), "iobroker_strerror() returns the right string");
iobs = iobroker_create();
error = iobroker_get_max_fds(iobs);
test(iobs && error >= 0, "max fd's for real iobroker set must be > 0");
listen_fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
flags = fcntl(listen_fd, F_GETFD);
flags |= FD_CLOEXEC;
fcntl(listen_fd, F_SETFD, flags);
(void)setsockopt(listen_fd, SOL_SOCKET, SO_REUSEADDR, &sockopt, sizeof(sockopt));
memset(&sain, 0, sizeof(sain));
sain.sin_port = ntohs(9123);
sain.sin_family = AF_INET;
bind(listen_fd, (struct sockaddr *)&sain, sizeof(sain));
listen(listen_fd, 128);
iobroker_register(iobs, listen_fd, iobs, listen_handler);
if (argc == 1)
conn_spam(&sain);
for (;;) {
iobroker_poll(iobs, -1);
if (iobroker_get_num_fds(iobs) <= 1) {
break;
}
}
iobroker_close(iobs, listen_fd);
iobroker_destroy(iobs, 0);
t_end();
return 0;
}
int qh_init(const char *path)
{
int result, old_umask;
if (qh_listen_sock >= 0)
iobroker_close(nagios_iobs, qh_listen_sock);
if (!path) {
nm_log(NSLOG_RUNTIME_ERROR, "qh: query_socket is NULL. What voodoo is this?\n");
return ERROR;
}
old_umask = umask(0117);
errno = 0;
qh_listen_sock = nsock_unix(path, NSOCK_TCP | NSOCK_UNLINK);
umask(old_umask);
if (qh_listen_sock < 0) {
nm_log(NSLOG_RUNTIME_ERROR, "qh: Failed to init socket '%s'. %s: %s\n",
path, nsock_strerror(qh_listen_sock), strerror(errno));
return ERROR;
}
/* plugins shouldn't have this socket */
(void)fcntl(qh_listen_sock, F_SETFD, FD_CLOEXEC);
/* most likely overkill, but it's small, so... */
qh_table = g_hash_table_new_full(g_str_hash, g_str_equal,
free, (GDestroyNotify) qh_remove);
errno = 0;
result = iobroker_register(nagios_iobs, qh_listen_sock, NULL, qh_registration_input);
if (result < 0) {
g_hash_table_destroy(qh_table);
close(qh_listen_sock);
nm_log(NSLOG_RUNTIME_ERROR, "qh: Failed to register socket with io broker: %s\n", iobroker_strerror(result));
return ERROR;
}
nm_log(NSLOG_INFO_MESSAGE, "qh: Socket '%s' successfully initialized\n", path);
/* now register our the in-core handlers */
qh_register_handler("command", "Naemon external commands interface", 0, qh_command);
qh_register_handler("echo", "The Echo Service - What You Put Is What You Get", 0, qh_echo);
qh_register_handler("help", "Help for the query handler", 0, qh_help);
return 0;
}
static int qh_registration_input(int sd, int events, void *bq_)
{
nm_bufferqueue *bq = (nm_bufferqueue *)bq_;
struct sockaddr sa;
socklen_t slen = 0;
int nsd, result;
memset(&sa, 0, sizeof(sa)); /* shut valgrind up */
nsd = accept(sd, &sa, &slen);
if (qh_max_running && qh_running >= qh_max_running) {
nsock_printf(nsd, "503: Server full");
close(nsd);
return 0;
}
if (!(bq = nm_bufferqueue_create())) {
nm_log(NSLOG_RUNTIME_ERROR, "qh: Failed to create iocache for inbound request\n");
nsock_printf(nsd, "500: Internal server error");
close(nsd);
return 0;
}
/*
* @todo: Stash the iocache and the socket in some
* addressable list so we can release them on deinit
*/
result = iobroker_register(nagios_iobs, nsd, bq, qh_input);
if (result < 0) {
nm_log(NSLOG_RUNTIME_ERROR, "qh: Failed to register input socket %d with I/O broker: %s; errno=%d (%s)\n",
nsd, iobroker_strerror(result), errno, strerror(errno));
nm_bufferqueue_destroy(bq);
close(nsd);
return 0;
}
/* make it non-blocking, but leave kernel buffers unchanged */
worker_set_sockopts(nsd, 0);
qh_running++;
return 0;
}
/*
* set up the listening socket (if applicable)
*/
int net_init(void)
{
int result, sockopt = 1;
struct sockaddr_in sain, inbound;
struct sockaddr *sa = (struct sockaddr *)&sain;
socklen_t addrlen = sizeof(inbound);
if (!num_nodes)
return 0;
sain.sin_addr.s_addr = default_addr;
sain.sin_port = htons(default_port);
sain.sin_family = AF_INET;
net_sock = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
if (net_sock < 0)
return -1;
merlin_set_socket_options(net_sock, 0);
/* if this fails we can do nothing but try anyway */
(void)setsockopt(net_sock, SOL_SOCKET, SO_REUSEADDR, &sockopt, sizeof(int));
result = bind(net_sock, sa, addrlen);
if (result < 0)
return -1;
result = listen(net_sock, SOMAXCONN);
if (result < 0)
return -1;
result = iobroker_register(nagios_iobs, net_sock, NULL, net_accept_one);
if (result < 0) {
lerr("IOB: Failed to register network socket with I/O broker: %s", iobroker_strerror(result));
return -1;
}
return 0;
}
static int conn_spam(struct sockaddr_in *sain)
{
int i;
signal(SIGALRM, (void (*)(int))sighandler);
signal(SIGINT, (void (*)(int))sighandler);
signal(SIGPIPE, SIG_IGN);
alarm(20);
for (i = 0; i < NUM_PROCS; i++) {
int fd, sockopt = 1;
fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
(void)setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &sockopt, sizeof(sockopt));
proc_counter[i] = 0;
iobroker_register(iobs, fd, (void *)&proc_counter[i], connected_handler);
if (connect(fd, (struct sockaddr *)sain, sizeof(*sain))) {
perror("connect");
}
iobroker_poll(iobs, -1);
}
return 0;
}
static int listen_handler(int fd, int events, void *arg)
{
int sock;
struct sockaddr_in sain;
socklen_t addrlen;
if (!arg || arg != iobs) {
printf("Argument passing seems to fail spectacularly\n");
}
addrlen = sizeof(sain);
//printf("listen_handler(%d, %d, %p) called\n", fd, events, arg);
sock = accept(fd, (struct sockaddr *)&sain, &addrlen);
if (sock < 0) {
perror("accept");
return -1;
}
if (write(sock, msg[0], strlen(msg[0])) < 0)
t_fail("write returned failure: %s", strerror(errno));
iobroker_register(iobs, sock, iobs, echo_service);
ok_int(iobroker_is_registered(iobs, sock), 1, "is_registered must be true");
return 0;
}
/* a service for registering workers */
static int register_worker(int sd, char *buf, unsigned int len)
{
int i, is_global = 1;
struct kvvec *info;
struct wproc_worker *worker;
logit(NSLOG_INFO_MESSAGE, TRUE, "wproc: Registry request: %s\n", buf);
if (!(worker = calloc(1, sizeof(*worker)))) {
logit(NSLOG_RUNTIME_ERROR, TRUE, "wproc: Failed to allocate worker: %s\n", strerror(errno));
return 500;
}
info = buf2kvvec(buf, len, '=', ';', 0);
if (info == NULL) {
free(worker);
logit(NSLOG_RUNTIME_ERROR, TRUE, "wproc: Failed to parse registration request\n");
return 500;
}
worker->sd = sd;
worker->ioc = iocache_create(1 * 1024 * 1024);
iobroker_unregister(nagios_iobs, sd);
iobroker_register(nagios_iobs, sd, worker, handle_worker_result);
for(i = 0; i < info->kv_pairs; i++) {
struct key_value *kv = &info->kv[i];
if (!strcmp(kv->key, "name")) {
worker->name = strdup(kv->value);
}
else if (!strcmp(kv->key, "pid")) {
worker->pid = atoi(kv->value);
}
else if (!strcmp(kv->key, "max_jobs")) {
worker->max_jobs = atoi(kv->value);
}
else if (!strcmp(kv->key, "plugin")) {
struct wproc_list *command_handlers;
is_global = 0;
if (!(command_handlers = dkhash_get(specialized_workers, kv->value, NULL))) {
command_handlers = calloc(1, sizeof(struct wproc_list));
command_handlers->wps = calloc(1, sizeof(struct wproc_worker**));
command_handlers->len = 1;
command_handlers->wps[0] = worker;
dkhash_insert(specialized_workers, strdup(kv->value), NULL, command_handlers);
}
else {
command_handlers->len++;
command_handlers->wps = realloc(command_handlers->wps, command_handlers->len * sizeof(struct wproc_worker**));
command_handlers->wps[command_handlers->len - 1] = worker;
}
worker->wp_list = command_handlers;
}
}
if (!worker->max_jobs) {
/*
* each worker uses two filedescriptors per job, one to
* connect to the master and about 13 to handle libraries
* and memory allocation, so this guesstimate shouldn't
* be too far off (for local workers, at least).
*/
worker->max_jobs = (iobroker_max_usable_fds() / 2) - 50;
}
worker->jobs = fanout_create(worker->max_jobs);
if (is_global) {
workers.len++;
workers.wps = realloc(workers.wps, workers.len * sizeof(struct wproc_worker *));
workers.wps[workers.len - 1] = worker;
worker->wp_list = &workers;
}
wproc_num_workers_online++;
kvvec_destroy(info, 0);
nsock_printf_nul(sd, "OK");
/* signal query handler to release its iocache for this one */
return QH_TAKEOVER;
}
static void enter_worker(int sd)
{
/* created with socketpair(), usually */
master_sd = sd;
parent_pid = getppid();
(void)chdir("/tmp");
(void)chdir("nagios-workers");
if (setpgid(0, 0)) {
/* XXX: handle error somehow, or maybe just ignore it */
}
/* we need to catch child signals the default way */
signal(SIGCHLD, SIG_DFL);
fcntl(fileno(stdout), F_SETFD, FD_CLOEXEC);
fcntl(fileno(stderr), F_SETFD, FD_CLOEXEC);
fcntl(master_sd, F_SETFD, FD_CLOEXEC);
iobs = iobroker_create();
if (!iobs) {
/* XXX: handle this a bit better */
worker_die("Worker failed to create io broker socket set");
}
/*
* Create a modest scheduling queue that will be
* more than enough for our needs
*/
sq = squeue_create(1024);
set_socket_options(master_sd, 256 * 1024);
iobroker_register(iobs, master_sd, NULL, receive_command);
while (iobroker_get_num_fds(iobs) > 0) {
int poll_time = -1;
/* check for timed out jobs */
for (;;) {
child_process *cp;
struct timeval now, tmo;
/* stop when scheduling queue is empty */
cp = (child_process *)squeue_peek(sq);
if (!cp)
break;
tmo.tv_usec = cp->start.tv_usec;
tmo.tv_sec = cp->start.tv_sec + cp->timeout;
gettimeofday(&now, NULL);
poll_time = tv_delta_msec(&now, &tmo);
/*
* A little extra takes care of rounding errors and
* ensures we never kill a job before it times out.
* 5 milliseconds is enough to take care of that.
*/
poll_time += 5;
if (poll_time > 0)
break;
/* this job timed out, so kill it */
wlog("job with pid %d timed out. Killing it", cp->pid);
kill_job(cp, ETIME);
}
iobroker_poll(iobs, poll_time);
/*
* if our parent goes away we can't really do anything
* sensible at all, so let's just break out and exit
*/
if (kill(parent_pid, 0) < 0 && errno == ESRCH) {
break;
}
}
/* we exit when the master shuts us down */
exit(EXIT_SUCCESS);
}