/*
* 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 connected_handler(int fd, int events, void *arg)
{
int *counter = (int *)arg;
int i;
i = *counter;
if (events == IOBROKER_POLLIN) {
char buf[1024];
int len = read(fd, buf, sizeof(buf));
buf[len] = 0;
test(len == (int)strlen(msg[i]), "len match for message %d", i);
test(!memcmp(buf, msg[i], len), "data match for message %d", i);
}
i++;
if (i < 0 || i >= (int)ARRAY_SIZE(msg)) {
fprintf(stderr, "i = %d in connected_handler(). What's up with that?\n", i);
return 0;
}
if (!msg[i]) {
iobroker_close(iobs, fd);
return 0;
}
if (write(fd, msg[i], strlen(msg[i])) < 0)
t_fail("write returned failure: %s", strerror(errno));
*counter = i;
return 0;
}
static int nerd_deinit(void)
{
unsigned int i;
if(host_parent_path_cache) {
for(i = 0; i < num_objects.hosts; i++) {
my_free(host_parent_path_cache[i]);
}
my_free(host_parent_path_cache);
}
for(i = 0; i < num_channels; i++) {
struct nerd_channel *chan = channels[i];
objectlist *list, *next;
for(list = chan->subscriptions; list; list = next) {
struct nerd_subscription *subscr = (struct nerd_subscription *)list->object_ptr;
iobroker_close(nagios_iobs, subscr->sd);
next = list->next;
free(list);
free(subscr);
}
chan->subscriptions = NULL;
my_free(chan);
}
my_free(channels);
num_channels = 0;
alloc_channels = 0;
return 0;
}
static int send_command(int sd, int events, void *discard)
{
char buf[8192];
int ret;
simple_worker *wp;
struct kvvec *kvv;
ret = read(sd, buf, sizeof(buf));
if (ret == 0) {
iobroker_close(iobs, sd);
return 0;
}
if (ret < 0) {
printf("main: Failed to read() from fd %d: %s",
sd, strerror(errno));
}
/* this happens when we're reading from stdin */
buf[--ret] = 0;
kvv = kvvec_create(5);
wp = wps[wp_index++ % NWPS];
kvvec_addkv(kvv, "job_id", (char *)mkstr("%d", wp->job_index++));
kvvec_addkv_wlen(kvv, "command", sizeof("command") - 1, buf, ret);
kvvec_addkv(kvv, "timeout", (char *)mkstr("%d", 10));
printf("Sending kvvec with %d pairs to worker %d\n", kvv->kv_pairs, wp->pid);
worker_send_kvvec(wp->sd, kvv);
kvvec_destroy(kvv, 0);
return 0;
}
/* removes a subscriber entirely and closes its socket */
int nerd_cancel_subscriber(int sd)
{
unsigned int i;
for(i = 0; i < num_channels; i++) {
cancel_channel_subscription(channels[i], sd);
}
iobroker_close(nagios_iobs, sd);
return 0;
}
int wproc_destroy(worker_process *wp, int flags)
{
int i = 0, destroyed = 0, force = 0, sd, self;
if (!wp)
return 0;
force = !!(flags & WPROC_FORCE);
self = getpid();
/* master retains workers through restarts */
if (self == nagios_pid && !force)
return 0;
/* free all memory when either forcing or a worker called us */
iocache_destroy(wp->ioc);
wp->ioc = NULL;
if (wp->jobs) {
for (i = 0; i < wp->max_jobs; i++) {
if (!wp->jobs[i])
continue;
destroy_job(wp, wp->jobs[i]);
/* we can (often) break out early */
if (++destroyed >= wp->jobs_running)
break;
}
/* this triggers a double-free() for some reason */
/* free(wp->jobs); */
wp->jobs = NULL;
}
sd = wp->sd;
free(wp);
/* workers must never control other workers, so they return early */
if (self != nagios_pid)
return 0;
/* kill(0, SIGKILL) equals suicide, so we avoid it */
if (wp->pid) {
kill(wp->pid, SIGKILL);
}
iobroker_close(nagios_iobs, sd);
/* reap our possibly lost children */
while (waitpid(-1, &i, WNOHANG) > 0)
; /* do nothing */
return 0;
}
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;
}
/* close all sockets and release the memory used by
* static global vars for networking purposes */
int net_deinit(void)
{
uint i;
for (i = 0; i < num_nodes; i++) {
node_disconnect(node_table[i], "Deinitializing networking");
}
iobroker_close(nagios_iobs, net_sock);
close(net_sock);
net_sock = -1;
return 0;
}
static int echo_service(int fd, int events, void *arg)
{
char buf[1024];
int len;
len = read(fd, buf, sizeof(buf));
if (len < 0) {
perror("read");
iobroker_close(iobs, fd);
ok_int(iobroker_is_registered(iobs, fd), 0, "Closing must deregister");
return 0;
}
/* zero read means we're disconnected */
if (!len) {
iobroker_close(iobs, fd);
ok_int(iobroker_is_registered(iobs, fd), 0, "Closing must deregister");
return 0;
}
if (write(fd, buf, len) < 0)
t_fail("write returned failure: %s", strerror(errno));
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 wproc_destroy(struct wproc_worker *wp, int flags)
{
int i = 0, force = 0, self;
if (!wp)
return 0;
force = !!(flags & WPROC_FORCE);
self = getpid();
/* master retains workers through restarts */
if (self == nagios_pid && !force)
return 0;
/* free all memory when either forcing or a worker called us */
iocache_destroy(wp->ioc);
wp->ioc = NULL;
my_free(wp->name);
fanout_destroy(wp->jobs, fo_destroy_job);
wp->jobs = NULL;
/* workers must never control other workers, so they return early */
if (self != nagios_pid)
return 0;
/* kill(0, SIGKILL) equals suicide, so we avoid it */
if (wp->pid) {
kill(wp->pid, SIGKILL);
}
iobroker_close(nagios_iobs, wp->sd);
/* reap our possibly lost children */
while (waitpid(-1, &i, WNOHANG) > 0)
; /* do nothing */
free(wp);
return 0;
}
static int receive_command(int sd, int events, void *discard)
{
int ioc_ret;
char *buf;
unsigned long size;
if (!ioc) {
ioc = iocache_create(512 * 1024);
}
ioc_ret = iocache_read(ioc, sd);
/* master closed the connection, so we exit */
if (ioc_ret == 0) {
iobroker_close(iobs, sd);
exit_worker();
}
if (ioc_ret < 0) {
/* XXX: handle this somehow */
}
#if 0
/* debug-volley */
buf = iocache_use_size(ioc, ioc_ret);
write(master_sd, buf, ioc_ret);
return 0;
#endif
/*
* now loop over all inbound messages in the iocache.
* Since KV_TERMINATOR is a nul-byte, they're separated by 3 nuls
*/
while ((buf = iocache_use_delim(ioc, MSG_DELIM, MSG_DELIM_LEN_RECV, &size))) {
struct kvvec *kvv;
/* we must copy vars here, as we preserve them for the response */
kvv = buf2kvvec(buf, (unsigned int)size, KV_SEP, PAIR_SEP, KVVEC_COPY);
if (kvv)
spawn_job(kvv);
}
return 0;
}
static void gather_output(child_process *cp, iobuf *io)
{
iobuf *other_io;
other_io = io == &cp->outstd ? &cp->outerr : &cp->outstd;
for (;;) {
char buf[4096];
int rd;
rd = read(io->fd, buf, sizeof(buf));
if (rd < 0) {
if (errno == EINTR)
continue;
/* XXX: handle the error somehow */
check_completion(cp, WNOHANG);
}
if (rd) {
/* we read some data */
io->buf = realloc(io->buf, rd + io->len + 1);
memcpy(&io->buf[io->len], buf, rd);
io->len += rd;
io->buf[io->len] = '\0';
} else {
iobroker_close(iobs, io->fd);
io->fd = -1;
if (other_io->fd < 0) {
check_completion(cp, 0);
} else {
check_completion(cp, WNOHANG);
}
}
break;
}
}
int finish_job(child_process *cp, int reason)
{
static struct kvvec resp = KVVEC_INITIALIZER;
struct rusage *ru = &cp->ei->rusage;
int i, ret;
/* get rid of still open filedescriptors */
if (cp->outstd.fd != -1) {
gather_output(cp, &cp->outstd, 1);
iobroker_close(iobs, cp->outstd.fd);
}
if (cp->outerr.fd != -1) {
gather_output(cp, &cp->outerr, 1);
iobroker_close(iobs, cp->outerr.fd);
}
/* Make sure network-supplied data doesn't contain nul bytes */
strip_nul_bytes(cp->outstd);
strip_nul_bytes(cp->outerr);
/* how many key/value pairs do we need? */
if (kvvec_init(&resp, 12 + cp->request->kv_pairs) == NULL) {
/* what the hell do we do now? */
exit_worker(1, "Failed to init response key/value vector");
}
gettimeofday(&cp->ei->stop, NULL);
if (running_jobs != squeue_size(sq)) {
wlog("running_jobs(%d) != squeue_size(sq) (%d)\n",
running_jobs, squeue_size(sq));
wlog("started: %d; running: %d; finished: %d\n",
started, running_jobs, started - running_jobs);
}
cp->ei->runtime = tv_delta_f(&cp->ei->start, &cp->ei->stop);
/*
* Now build the return message.
* First comes the request, minus environment variables
*/
for (i = 0; i < cp->request->kv_pairs; i++) {
struct key_value *kv = &cp->request->kv[i];
/* skip environment macros */
if (kv->key_len == 3 && !strcmp(kv->key, "env")) {
continue;
}
kvvec_addkv_wlen(&resp, kv->key, kv->key_len, kv->value, kv->value_len);
}
kvvec_addkv(&resp, "wait_status", mkstr("%d", cp->ret));
kvvec_add_tv(&resp, "start", cp->ei->start);
kvvec_add_tv(&resp, "stop", cp->ei->stop);
kvvec_addkv(&resp, "runtime", mkstr("%f", cp->ei->runtime));
if (!reason) {
/* child exited nicely (or with a signal, so check wait_status) */
kvvec_addkv(&resp, "exited_ok", "1");
kvvec_add_tv(&resp, "ru_utime", ru->ru_utime);
kvvec_add_tv(&resp, "ru_stime", ru->ru_stime);
kvvec_add_long(&resp, "ru_minflt", ru->ru_minflt);
kvvec_add_long(&resp, "ru_majflt", ru->ru_majflt);
kvvec_add_long(&resp, "ru_inblock", ru->ru_inblock);
kvvec_add_long(&resp, "ru_oublock", ru->ru_oublock);
} else {
/* some error happened */
kvvec_addkv(&resp, "exited_ok", "0");
kvvec_addkv(&resp, "error_code", mkstr("%d", reason));
}
kvvec_addkv_wlen(&resp, "outerr", 6, cp->outerr.buf, cp->outerr.len);
kvvec_addkv_wlen(&resp, "outstd", 6, cp->outstd.buf, cp->outstd.len);
ret = worker_send_kvvec(master_sd, &resp);
if (ret < 0 && errno == EPIPE)
exit_worker(1, "Failed to send kvvec struct to master");
return 0;
}
static int qh_input(int sd, int events, void *bq_)
{
nm_bufferqueue *bq = (nm_bufferqueue *)bq_;
int result;
size_t len;
unsigned int query_len = 0;
char *buf, *space;
struct query_handler *qh;
char *handler = NULL, *query = NULL;
result = nm_bufferqueue_read(bq, sd);
/* disconnect? */
if (result == 0 || (result < 0 && errno == EPIPE)) {
nm_bufferqueue_destroy(bq);
iobroker_close(nagios_iobs, sd);
qh_running--;
return 0;
}
/*
* A request looks like this: '[@|#]<qh>[<SP>][<query>]\0'.
* That is, optional '#' (oneshot) or '@' (keepalive),
* followed by the name of a registered handler, followed by
* an optional space and an optional query. If the handler
* has no "default" handler, a query is required or an error
* will be thrown.
*/
/* Use data up to the first nul byte */
nm_bufferqueue_unshift_to_delim(bq, "\0", 1, &len, (void **)&buf);
if (!buf)
return 0;
/* Identify handler part and any magic query bytes */
if (*buf == '@' || *buf == '#') {
handler = buf + 1;
} else {
handler = buf;
}
/* Locate query (if any) */
if ((space = strchr(buf, ' '))) {
*space = 0;
query = space + 1;
query_len = len - ((unsigned long)query - (unsigned long)buf);
} else {
query = "";
query_len = 0;
}
/* locate the handler */
if (!(qh = qh_find_handler(handler))) {
/* not found. that's a 404 */
nsock_printf(sd, "404: %s: No such handler", handler);
nm_free(buf);
iobroker_close(nagios_iobs, sd);
nm_bufferqueue_destroy(bq);
return 0;
}
/* strip trailing newlines */
while (query_len > 0 && (query[query_len - 1] == 0 || query[query_len - 1] == '\n'))
query[--query_len] = 0;
/* now pass the query to the handler */
if ((result = qh->handler(sd, query, query_len)) >= 100) {
nsock_printf_nul(sd, "%d: %s", result, qh_strerror(result));
}
if (result >= 300 || *buf != '@') {
/* error code or one-shot query */
nm_free(buf);
iobroker_close(nagios_iobs, sd);
nm_bufferqueue_destroy(bq);
return 0;
}
nm_free(buf);
/* check for magic handler codes */
switch (result) {
case QH_CLOSE: /* oneshot handler */
case -1: /* general error */
iobroker_close(nagios_iobs, sd);
/* fallthrough */
case QH_TAKEOVER: /* handler takes over */
case 101: /* switch protocol (takeover + message) */
nm_bufferqueue_destroy(bq);
break;
}
return 0;
}
/*
* Negotiate which socket to use for communication when the remote
* host has accepted a connection attempt from us while we have
* accepted one from the remote host. We must make sure both ends
* agree on one socket to use.
*/
static int net_negotiate_socket(merlin_node *node, int con, int lis)
{
struct sockaddr_in lissain, consain;
socklen_t slen = sizeof(struct sockaddr_in);
linfo("negotiate: Choosing socket for %s %s (%d or %d)", node_type(node), node->name, con, lis);
if (con < 0)
return lis;
if (lis < 0)
return con;
/* we prefer the socket with the lowest ip-address */
if (getsockname(lis, (struct sockaddr *)&lissain, &slen) < 0) {
lerr("negotiate: getsockname(%d, ...) failed: %s",
lis, strerror(errno));
return con;
}
if (getpeername(con, (struct sockaddr *)&consain, &slen) < 0) {
lerr("negotiate: getpeername(%d, ...) failed: %s",
con, strerror(errno));
return lis;
}
ldebug("negotiate: lis(%d): %s:%d", lis,
inet_ntoa(lissain.sin_addr), ntohs(lissain.sin_port));
ldebug("negotiate: con(%d): %s:%d", con,
inet_ntoa(consain.sin_addr), ntohs(consain.sin_port));
if (lissain.sin_addr.s_addr > consain.sin_addr.s_addr) {
ldebug("negotiate: con has lowest ip. using that");
return con;
}
if (consain.sin_addr.s_addr > lissain.sin_addr.s_addr) {
ldebug("negotiate: lis has lowest ip. using that");
return lis;
}
/*
* this will happen if multiple merlin instances run
* on the same server, such as when we're testing
* things. In that case, let the portnumber decide
* the tiebreak
*/
if (lissain.sin_port > consain.sin_port) {
ldebug("negotiate: con has lowest port. using that");
return con;
}
if (consain.sin_port > lissain.sin_port) {
ldebug("negotiate: lis has lowest port. using that");
return lis;
}
ldebug("negotiate: con and lis are equal. killing both");
node->last_conn_attempt_logged = 0;
node_disconnect(node, "socket negotiation failed");
iobroker_close(nagios_iobs, lis);
node->sock = -1;
return -1;
}
static int finish_job(child_process *cp, int reason)
{
static struct kvvec resp = KVVEC_INITIALIZER;
struct rusage *ru = &cp->rusage;
int i, ret;
/* how many key/value pairs do we need? */
if (kvvec_init(&resp, 12 + cp->request->kv_pairs) == NULL) {
/* what the hell do we do now? */
exit_worker();
}
gettimeofday(&cp->stop, NULL);
if (running_jobs != squeue_size(sq)) {
wlog("running_jobs(%d) != squeue_size(sq) (%d)\n",
running_jobs, squeue_size(sq));
wlog("started: %d; running: %d; finished: %d\n",
started, running_jobs, started - running_jobs);
}
/*
* we must remove the job's timeout ticker,
* or we'll end up accessing an already free()'d
* pointer, or the pointer to a different child.
*/
squeue_remove(sq, cp->sq_event);
/* get rid of still open filedescriptors */
if (cp->outstd.fd != -1)
iobroker_close(iobs, cp->outstd.fd);
if (cp->outerr.fd != -1)
iobroker_close(iobs, cp->outerr.fd);
cp->runtime = tv_delta_f(&cp->start, &cp->stop);
/*
* Now build the return message.
* First comes the request, minus environment variables
*/
for (i = 0; i < cp->request->kv_pairs; i++) {
struct key_value *kv = &cp->request->kv[i];
/* skip environment macros */
if (kv->key_len == 3 && !strcmp(kv->key, "env")) {
continue;
}
kvvec_addkv_wlen(&resp, kv->key, kv->key_len, kv->value, kv->value_len);
}
kvvec_addkv(&resp, "wait_status", (char *)mkstr("%d", cp->ret));
kvvec_addkv_wlen(&resp, "outstd", 6, cp->outstd.buf, cp->outstd.len);
kvvec_addkv_wlen(&resp, "outerr", 6, cp->outerr.buf, cp->outerr.len);
kvvec_add_tv(&resp, "start", cp->start);
kvvec_add_tv(&resp, "stop", cp->stop);
kvvec_addkv(&resp, "runtime", (char *)mkstr("%f", cp->runtime));
if (!reason) {
/* child exited nicely */
kvvec_addkv(&resp, "exited_ok", "1");
kvvec_add_tv(&resp, "ru_utime", ru->ru_utime);
kvvec_add_tv(&resp, "ru_stime", ru->ru_stime);
kvvec_add_long(&resp, "ru_minflt", ru->ru_minflt);
kvvec_add_long(&resp, "ru_majflt", ru->ru_majflt);
kvvec_add_long(&resp, "ru_nswap", ru->ru_nswap);
kvvec_add_long(&resp, "ru_inblock", ru->ru_inblock);
kvvec_add_long(&resp, "ru_oublock", ru->ru_oublock);
kvvec_add_long(&resp, "ru_nsignals", ru->ru_nsignals);
} else {
/* some error happened */
kvvec_addkv(&resp, "exited_ok", "0");
kvvec_addkv(&resp, "error_code", (char *)mkstr("%d", reason));
}
ret = send_kvvec(master_sd, &resp);
if (ret < 0 && errno == EPIPE)
exit_worker();
running_jobs--;
if (cp->outstd.buf) {
free(cp->outstd.buf);
cp->outstd.buf = NULL;
}
if (cp->outerr.buf) {
free(cp->outerr.buf);
cp->outerr.buf = NULL;
}
kvvec_destroy(cp->request, KVVEC_FREE_ALL);
free(cp->cmd);
free(cp);
return 0;
}
