[lttng-tools.git] / src / bin / lttng-consumerd / health-consumerd.cpp

/*
 * Copyright (C) 2013 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
 *
 * SPDX-License-Identifier: GPL-2.0-only
 *
 */

#define _LGPL_SOURCE
#include <fcntl.h>
#include <getopt.h>
#include <grp.h>
#include <limits.h>
#include <pthread.h>
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/ipc.h>
#include <sys/resource.h>
#include <sys/shm.h>
#include <sys/socket.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <urcu/list.h>
#include <poll.h>
#include <unistd.h>
#include <sys/mman.h>
#include <urcu/compiler.h>
#include <ulimit.h>
#include <inttypes.h>

#include <common/defaults.h>
#include <common/common.h>
#include <common/consumer/consumer.h>
#include <common/consumer/consumer-timer.h>
#include <common/compat/poll.h>
#include <common/sessiond-comm/sessiond-comm.h>
#include <common/utils.h>

#include "lttng-consumerd.h"
#include "health-consumerd.h"

/* Global health check unix path */
static char health_unix_sock_path[PATH_MAX];

int health_quit_pipe[2];

/*
 * Check if the thread quit pipe was triggered.
 *
 * Return 1 if it was triggered else 0;
 */
static
int check_health_quit_pipe(int fd, uint32_t events)
{
	if (fd == health_quit_pipe[0] && (events & LPOLLIN)) {
		return 1;
	}

	return 0;
}

/*
 * Send data on a unix socket using the liblttsessiondcomm API.
 *
 * Return lttcomm error code.
 */
static int send_unix_sock(int sock, void *buf, size_t len)
{
	/* Check valid length */
	if (len == 0) {
		return -1;
	}

	return lttcomm_send_unix_sock(sock, buf, len);
}

static
int setup_health_path(void)
{
	int is_root, ret = 0;
	enum lttng_consumer_type type;
	const char *home_path;

	type = lttng_consumer_get_type();
	is_root = !getuid();

	if (is_root) {
		if (strlen(health_unix_sock_path) != 0) {
			goto end;
		}
		switch (type) {
		case LTTNG_CONSUMER_KERNEL:
			snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
				DEFAULT_GLOBAL_KCONSUMER_HEALTH_UNIX_SOCK);
			break;
		case LTTNG_CONSUMER64_UST:
			snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
				DEFAULT_GLOBAL_USTCONSUMER64_HEALTH_UNIX_SOCK);
			break;
		case LTTNG_CONSUMER32_UST:
			snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
				DEFAULT_GLOBAL_USTCONSUMER32_HEALTH_UNIX_SOCK);
			break;
		default:
			ret = -EINVAL;
			goto end;
		}
	} else {
		home_path = utils_get_home_dir();
		if (home_path == NULL) {
			/* TODO: Add --socket PATH option */
			ERR("Can't get HOME directory for sockets creation.");
			ret = -EPERM;
			goto end;
		}

		/* Set health check Unix path */
		if (strlen(health_unix_sock_path) != 0) {
			goto end;
		}
		switch (type) {
		case LTTNG_CONSUMER_KERNEL:
			snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
				DEFAULT_HOME_KCONSUMER_HEALTH_UNIX_SOCK, home_path);
			break;
		case LTTNG_CONSUMER64_UST:
			snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
				DEFAULT_HOME_USTCONSUMER64_HEALTH_UNIX_SOCK, home_path);
			break;
		case LTTNG_CONSUMER32_UST:
			snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
				DEFAULT_HOME_USTCONSUMER32_HEALTH_UNIX_SOCK, home_path);
			break;
		default:
			ret = -EINVAL;
			goto end;
		}
	}
end:
	return ret;
}

/*
 * Thread managing health check socket.
 */
void *thread_manage_health_consumerd(void *data __attribute__((unused)))
{
	int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
	uint32_t revents, nb_fd;
	struct lttng_poll_event events;
	struct health_comm_msg msg;
	struct health_comm_reply reply;
	int is_root;

	DBG("[thread] Manage health check started");

	setup_health_path();

	rcu_register_thread();

	/* We might hit an error path before this is created. */
	lttng_poll_init(&events);

	/* Create unix socket */
	sock = lttcomm_create_unix_sock(health_unix_sock_path);
	if (sock < 0) {
		ERR("Unable to create health check Unix socket");
		err = -1;
		goto error;
	}

	is_root = !getuid();
	if (is_root) {
		/* lttng health client socket path permissions */
		gid_t gid;

		ret = utils_get_group_id(tracing_group_name, true, &gid);
		if (ret) {
			/* Default to root group. */
			gid = 0;
		}

		ret = chown(health_unix_sock_path, 0, gid);
		if (ret < 0) {
			ERR("Unable to set group on %s", health_unix_sock_path);
			PERROR("chown");
			err = -1;
			goto error;
		}

		ret = chmod(health_unix_sock_path,
				S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
		if (ret < 0) {
			ERR("Unable to set permissions on %s", health_unix_sock_path);
			PERROR("chmod");
			err = -1;
			goto error;
		}
	}

	/*
	 * Set the CLOEXEC flag. Return code is useless because either way, the
	 * show must go on.
	 */
	(void) utils_set_fd_cloexec(sock);

	ret = lttcomm_listen_unix_sock(sock);
	if (ret < 0) {
		goto error;
	}

	/* Size is set to 2 for the quit pipe and registration socket. */
	ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
	if (ret < 0) {
		ERR("Poll set creation failed");
		goto error;
	}

	ret = lttng_poll_add(&events, health_quit_pipe[0], LPOLLIN);
	if (ret < 0) {
		goto error;
	}

	/* Add the application registration socket */
	ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLPRI);
	if (ret < 0) {
		goto error;
	}

	/* Perform prior memory accesses before decrementing ready */
	cmm_smp_mb__before_uatomic_dec();
	uatomic_dec(&lttng_consumer_ready);

	while (1) {
		DBG("Health check ready");

		/* Inifinite blocking call, waiting for transmission */
restart:
		ret = lttng_poll_wait(&events, -1);
		if (ret < 0) {
			/*
			 * Restart interrupted system call.
			 */
			if (errno == EINTR) {
				goto restart;
			}
			goto error;
		}

		nb_fd = ret;

		for (i = 0; i < nb_fd; i++) {
			/* Fetch once the poll data */
			revents = LTTNG_POLL_GETEV(&events, i);
			pollfd = LTTNG_POLL_GETFD(&events, i);

			/* Thread quit pipe has been closed. Killing thread. */
			ret = check_health_quit_pipe(pollfd, revents);
			if (ret) {
				err = 0;
				goto exit;
			}

			/* Event on the registration socket */
			if (pollfd == sock) {
				if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)
						&& !(revents & LPOLLIN)) {
					ERR("Health socket poll error");
					goto error;
				}
			}
		}

		new_sock = lttcomm_accept_unix_sock(sock);
		if (new_sock < 0) {
			goto error;
		}

		/*
		 * Set the CLOEXEC flag. Return code is useless because either way, the
		 * show must go on.
		 */
		(void) utils_set_fd_cloexec(new_sock);

		DBG("Receiving data from client for health...");
		ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
		if (ret <= 0) {
			DBG("Nothing recv() from client... continuing");
			ret = close(new_sock);
			if (ret) {
				PERROR("close");
			}
			new_sock = -1;
			continue;
		}

		rcu_thread_online();

		LTTNG_ASSERT(msg.cmd == HEALTH_CMD_CHECK);

		memset(&reply, 0, sizeof(reply));
		for (i = 0; i < NR_HEALTH_CONSUMERD_TYPES; i++) {
			/*
			 * health_check_state return 0 if thread is in
			 * error.
			 */
			if (!health_check_state(health_consumerd, i)) {
				reply.ret_code |= 1ULL << i;
			}
		}

		DBG("Health check return value %" PRIx64, reply.ret_code);

		ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
		if (ret < 0) {
			ERR("Failed to send health data back to client");
		}

		/* End of transmission */
		ret = close(new_sock);
		if (ret) {
			PERROR("close");
		}
		new_sock = -1;
	}

exit:
error:
	if (err) {
		ERR("Health error occurred in %s", __func__);
	}
	DBG("Health check thread dying");
	unlink(health_unix_sock_path);
	if (sock >= 0) {
		ret = close(sock);
		if (ret) {
			PERROR("close");
		}
	}

	lttng_poll_clean(&events);

	rcu_unregister_thread();
	return NULL;
}
Commit	Line	Data
5c635c72	1	/*
ab5be9fa	2	* Copyright (C) 2013 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5c635c72	3	*
ab5be9fa	4	* SPDX-License-Identifier: GPL-2.0-only
5c635c72	5	*
5c635c72 MD	6	*/
5c635c72 MD	7
6c1c0768	8	#define _LGPL_SOURCE
5c635c72 MD	9	#include <fcntl.h>
	10	#include <getopt.h>
	11	#include <grp.h>
	12	#include <limits.h>
	13	#include <pthread.h>
	14	#include <signal.h>
	15	#include <stdio.h>
	16	#include <stdlib.h>
	17	#include <string.h>
	18	#include <sys/ipc.h>
	19	#include <sys/resource.h>
	20	#include <sys/shm.h>
	21	#include <sys/socket.h>
	22	#include <sys/stat.h>
	23	#include <sys/types.h>
	24	#include <urcu/list.h>
	25	#include <poll.h>
	26	#include <unistd.h>
	27	#include <sys/mman.h>
5c635c72 MD	28	#include <urcu/compiler.h>
5c635c72 MD	29	#include <ulimit.h>
6c71277b	30	#include <inttypes.h>
5c635c72 MD	31
	32	#include <common/defaults.h>
	33	#include <common/common.h>
c8fea79c JR	34	#include <common/consumer/consumer.h>
c8fea79c JR	35	#include <common/consumer/consumer-timer.h>
5c635c72 MD	36	#include <common/compat/poll.h>
	37	#include <common/sessiond-comm/sessiond-comm.h>
	38	#include <common/utils.h>
	39
	40	#include "lttng-consumerd.h"
	41	#include "health-consumerd.h"
	42
	43	/* Global health check unix path */
	44	static char health_unix_sock_path[PATH_MAX];
	45
	46	int health_quit_pipe[2];
	47
	48	/*
	49	* Check if the thread quit pipe was triggered.
	50	*
	51	* Return 1 if it was triggered else 0;
	52	*/
	53	static
	54	int check_health_quit_pipe(int fd, uint32_t events)
	55	{
	56	if (fd == health_quit_pipe[0] && (events & LPOLLIN)) {
	57	return 1;
	58	}
	59
	60	return 0;
	61	}
	62
	63	/*
	64	* Send data on a unix socket using the liblttsessiondcomm API.
	65	*
	66	* Return lttcomm error code.
	67	*/
	68	static int send_unix_sock(int sock, void *buf, size_t len)
	69	{
	70	/* Check valid length */
	71	if (len == 0) {
	72	return -1;
	73	}
	74
	75	return lttcomm_send_unix_sock(sock, buf, len);
	76	}
	77
	78	static
	79	int setup_health_path(void)
	80	{
	81	int is_root, ret = 0;
	82	enum lttng_consumer_type type;
	83	const char *home_path;
	84
	85	type = lttng_consumer_get_type();
	86	is_root = !getuid();
	87
	88	if (is_root) {
	89	if (strlen(health_unix_sock_path) != 0) {
	90	goto end;
	91	}
	92	switch (type) {
	93	case LTTNG_CONSUMER_KERNEL:
	94	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
	95	DEFAULT_GLOBAL_KCONSUMER_HEALTH_UNIX_SOCK);
	96	break;
	97	case LTTNG_CONSUMER64_UST:
	98	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
	99	DEFAULT_GLOBAL_USTCONSUMER64_HEALTH_UNIX_SOCK);
100	break;
101	case LTTNG_CONSUMER32_UST:
102	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
103	DEFAULT_GLOBAL_USTCONSUMER32_HEALTH_UNIX_SOCK);
104	break;
105	default:
106	ret = -EINVAL;
107	goto end;
108	}
109	} else {
5c635c72 MD	110	home_path = utils_get_home_dir();
	111	if (home_path == NULL) {
	112	/* TODO: Add --socket PATH option */
	113	ERR("Can't get HOME directory for sockets creation.");
	114	ret = -EPERM;
	115	goto end;
	116	}
	117
5c635c72 MD	118	/* Set health check Unix path */
	119	if (strlen(health_unix_sock_path) != 0) {
	120	goto end;
	121	}
	122	switch (type) {
	123	case LTTNG_CONSUMER_KERNEL:
	124	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
dbc8403d	125	DEFAULT_HOME_KCONSUMER_HEALTH_UNIX_SOCK, home_path);
5c635c72 MD	126	break;
	127	case LTTNG_CONSUMER64_UST:
	128	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
dbc8403d	129	DEFAULT_HOME_USTCONSUMER64_HEALTH_UNIX_SOCK, home_path);
5c635c72 MD	130	break;
	131	case LTTNG_CONSUMER32_UST:
	132	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
dbc8403d	133	DEFAULT_HOME_USTCONSUMER32_HEALTH_UNIX_SOCK, home_path);
5c635c72 MD	134	break;
	135	default:
	136	ret = -EINVAL;
	137	goto end;
	138	}
	139	}
5c635c72 MD	140	end:
	141	return ret;
	142	}
	143
	144	/*
	145	* Thread managing health check socket.
	146	*/
f46376a1	147	void thread_manage_health_consumerd(void data __attribute__((unused)))
5c635c72 MD	148	{
	149	int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
	150	uint32_t revents, nb_fd;
	151	struct lttng_poll_event events;
	152	struct health_comm_msg msg;
	153	struct health_comm_reply reply;
6c71277b	154	int is_root;
5c635c72 MD	155
	156	DBG("[thread] Manage health check started");
	157
	158	setup_health_path();
	159
	160	rcu_register_thread();
	161
	162	/* We might hit an error path before this is created. */
	163	lttng_poll_init(&events);
	164
	165	/* Create unix socket */
	166	sock = lttcomm_create_unix_sock(health_unix_sock_path);
	167	if (sock < 0) {
	168	ERR("Unable to create health check Unix socket");
67fe4075	169	err = -1;
5c635c72 MD	170	goto error;
	171	}
	172
6c71277b MD	173	is_root = !getuid();
	174	if (is_root) {
	175	/* lttng health client socket path permissions */
28ab59d0 JR	176	gid_t gid;
	177
	178	ret = utils_get_group_id(tracing_group_name, true, &gid);
	179	if (ret) {
	180	/* Default to root group. */
	181	gid = 0;
	182	}
	183
	184	ret = chown(health_unix_sock_path, 0, gid);
6c71277b MD	185	if (ret < 0) {
	186	ERR("Unable to set group on %s", health_unix_sock_path);
	187	PERROR("chown");
67fe4075	188	err = -1;
6c71277b MD	189	goto error;
	190	}
	191
	192	ret = chmod(health_unix_sock_path,
	193	S_IRUSR \| S_IWUSR \| S_IRGRP \| S_IWGRP);
	194	if (ret < 0) {
	195	ERR("Unable to set permissions on %s", health_unix_sock_path);
	196	PERROR("chmod");
67fe4075	197	err = -1;
6c71277b MD	198	goto error;
	199	}
	200	}
	201
5c635c72 MD	202	/*
	203	* Set the CLOEXEC flag. Return code is useless because either way, the
	204	* show must go on.
	205	*/
	206	(void) utils_set_fd_cloexec(sock);
	207
	208	ret = lttcomm_listen_unix_sock(sock);
	209	if (ret < 0) {
	210	goto error;
	211	}
	212
3069d754	213	/* Size is set to 2 for the quit pipe and registration socket. */
5c635c72 MD	214	ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
	215	if (ret < 0) {
	216	ERR("Poll set creation failed");
	217	goto error;
	218	}
	219
	220	ret = lttng_poll_add(&events, health_quit_pipe[0], LPOLLIN);
	221	if (ret < 0) {
	222	goto error;
	223	}
	224
	225	/* Add the application registration socket */
	226	ret = lttng_poll_add(&events, sock, LPOLLIN \| LPOLLPRI);
	227	if (ret < 0) {
	228	goto error;
	229	}
	230
748b7b07 MD	231	/* Perform prior memory accesses before decrementing ready */
	232	cmm_smp_mb__before_uatomic_dec();
	233	uatomic_dec(&lttng_consumer_ready);
	234
5c635c72 MD	235	while (1) {
	236	DBG("Health check ready");
	237
	238	/* Inifinite blocking call, waiting for transmission */
	239	restart:
	240	ret = lttng_poll_wait(&events, -1);
	241	if (ret < 0) {
	242	/*
	243	* Restart interrupted system call.
	244	*/
	245	if (errno == EINTR) {
	246	goto restart;
	247	}
	248	goto error;
	249	}
	250
	251	nb_fd = ret;
	252
	253	for (i = 0; i < nb_fd; i++) {
	254	/* Fetch once the poll data */
	255	revents = LTTNG_POLL_GETEV(&events, i);
	256	pollfd = LTTNG_POLL_GETFD(&events, i);
	257
	258	/* Thread quit pipe has been closed. Killing thread. */
	259	ret = check_health_quit_pipe(pollfd, revents);
	260	if (ret) {
	261	err = 0;
	262	goto exit;
	263	}
	264
	265	/* Event on the registration socket */
	266	if (pollfd == sock) {
03e43155 MD	267	if (revents & (LPOLLERR \| LPOLLHUP \| LPOLLRDHUP)
03e43155 MD	268	&& !(revents & LPOLLIN)) {
5c635c72 MD	269	ERR("Health socket poll error");
	270	goto error;
	271	}
	272	}
	273	}
	274
	275	new_sock = lttcomm_accept_unix_sock(sock);
	276	if (new_sock < 0) {
	277	goto error;
	278	}
	279
	280	/*
	281	* Set the CLOEXEC flag. Return code is useless because either way, the
	282	* show must go on.
	283	*/
	284	(void) utils_set_fd_cloexec(new_sock);
	285
	286	DBG("Receiving data from client for health...");
	287	ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
	288	if (ret <= 0) {
	289	DBG("Nothing recv() from client... continuing");
	290	ret = close(new_sock);
	291	if (ret) {
	292	PERROR("close");
	293	}
	294	new_sock = -1;
	295	continue;
	296	}
	297
	298	rcu_thread_online();
	299
a0377dfe	300	LTTNG_ASSERT(msg.cmd == HEALTH_CMD_CHECK);
5c635c72	301
53efb85a	302	memset(&reply, 0, sizeof(reply));
6c71277b MD	303	for (i = 0; i < NR_HEALTH_CONSUMERD_TYPES; i++) {
	304	/*
	305	* health_check_state return 0 if thread is in
	306	* error.
	307	*/
	308	if (!health_check_state(health_consumerd, i)) {
	309	reply.ret_code \|= 1ULL << i;
	310	}
5c635c72 MD	311	}
5c635c72 MD	312
6137f630	313	DBG("Health check return value %" PRIx64, reply.ret_code);
5c635c72 MD	314
	315	ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
	316	if (ret < 0) {
	317	ERR("Failed to send health data back to client");
	318	}
	319
	320	/* End of transmission */
	321	ret = close(new_sock);
	322	if (ret) {
	323	PERROR("close");
	324	}
	325	new_sock = -1;
	326	}
	327
	328	exit:
	329	error:
	330	if (err) {
	331	ERR("Health error occurred in %s", __func__);
	332	}
	333	DBG("Health check thread dying");
	334	unlink(health_unix_sock_path);
	335	if (sock >= 0) {
	336	ret = close(sock);
	337	if (ret) {
	338	PERROR("close");
	339	}
	340	}
	341
	342	lttng_poll_clean(&events);
	343
	344	rcu_unregister_thread();
	345	return NULL;
	346	}