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

/*
 * 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 <assert.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(void *data)
{
	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 1 for the consumer_channel pipe */
	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();

		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>
	28	#include <assert.h>
5c635c72 MD	29	#include <urcu/compiler.h>
5c635c72 MD	30	#include <ulimit.h>
6c71277b	31	#include <inttypes.h>
5c635c72 MD	32
	33	#include <common/defaults.h>
	34	#include <common/common.h>
c8fea79c JR	35	#include <common/consumer/consumer.h>
c8fea79c JR	36	#include <common/consumer/consumer-timer.h>
5c635c72 MD	37	#include <common/compat/poll.h>
	38	#include <common/sessiond-comm/sessiond-comm.h>
	39	#include <common/utils.h>
	40
	41	#include "lttng-consumerd.h"
	42	#include "health-consumerd.h"
	43
	44	/* Global health check unix path */
	45	static char health_unix_sock_path[PATH_MAX];
	46
	47	int health_quit_pipe[2];
	48
	49	/*
	50	* Check if the thread quit pipe was triggered.
	51	*
	52	* Return 1 if it was triggered else 0;
	53	*/
	54	static
	55	int check_health_quit_pipe(int fd, uint32_t events)
	56	{
	57	if (fd == health_quit_pipe[0] && (events & LPOLLIN)) {
	58	return 1;
	59	}
	60
	61	return 0;
	62	}
	63
	64	/*
	65	* Send data on a unix socket using the liblttsessiondcomm API.
	66	*
	67	* Return lttcomm error code.
	68	*/
	69	static int send_unix_sock(int sock, void *buf, size_t len)
	70	{
	71	/* Check valid length */
	72	if (len == 0) {
	73	return -1;
	74	}
	75
	76	return lttcomm_send_unix_sock(sock, buf, len);
	77	}
	78
	79	static
	80	int setup_health_path(void)
	81	{
	82	int is_root, ret = 0;
	83	enum lttng_consumer_type type;
	84	const char *home_path;
	85
	86	type = lttng_consumer_get_type();
	87	is_root = !getuid();
	88
	89	if (is_root) {
	90	if (strlen(health_unix_sock_path) != 0) {
	91	goto end;
	92	}
	93	switch (type) {
	94	case LTTNG_CONSUMER_KERNEL:
	95	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
	96	DEFAULT_GLOBAL_KCONSUMER_HEALTH_UNIX_SOCK);
	97	break;
	98	case LTTNG_CONSUMER64_UST:
	99	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
	100	DEFAULT_GLOBAL_USTCONSUMER64_HEALTH_UNIX_SOCK);
101	break;
102	case LTTNG_CONSUMER32_UST:
103	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
104	DEFAULT_GLOBAL_USTCONSUMER32_HEALTH_UNIX_SOCK);
105	break;
106	default:
107	ret = -EINVAL;
108	goto end;
109	}
110	} else {
5c635c72 MD	111	home_path = utils_get_home_dir();
	112	if (home_path == NULL) {
	113	/* TODO: Add --socket PATH option */
	114	ERR("Can't get HOME directory for sockets creation.");
	115	ret = -EPERM;
	116	goto end;
	117	}
	118
5c635c72 MD	119	/* Set health check Unix path */
	120	if (strlen(health_unix_sock_path) != 0) {
	121	goto end;
	122	}
	123	switch (type) {
	124	case LTTNG_CONSUMER_KERNEL:
	125	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
dbc8403d	126	DEFAULT_HOME_KCONSUMER_HEALTH_UNIX_SOCK, home_path);
5c635c72 MD	127	break;
	128	case LTTNG_CONSUMER64_UST:
	129	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
dbc8403d	130	DEFAULT_HOME_USTCONSUMER64_HEALTH_UNIX_SOCK, home_path);
5c635c72 MD	131	break;
	132	case LTTNG_CONSUMER32_UST:
	133	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
dbc8403d	134	DEFAULT_HOME_USTCONSUMER32_HEALTH_UNIX_SOCK, home_path);
5c635c72 MD	135	break;
	136	default:
	137	ret = -EINVAL;
	138	goto end;
	139	}
	140	}
5c635c72 MD	141	end:
	142	return ret;
	143	}
	144
	145	/*
	146	* Thread managing health check socket.
	147	*/
	148	void thread_manage_health(void data)
	149	{
	150	int sock = -1, new_sock = -1, ret, i, pollfd, err = -1;
	151	uint32_t revents, nb_fd;
	152	struct lttng_poll_event events;
	153	struct health_comm_msg msg;
	154	struct health_comm_reply reply;
6c71277b	155	int is_root;
5c635c72 MD	156
	157	DBG("[thread] Manage health check started");
	158
	159	setup_health_path();
	160
	161	rcu_register_thread();
	162
	163	/* We might hit an error path before this is created. */
	164	lttng_poll_init(&events);
	165
	166	/* Create unix socket */
	167	sock = lttcomm_create_unix_sock(health_unix_sock_path);
	168	if (sock < 0) {
	169	ERR("Unable to create health check Unix socket");
67fe4075	170	err = -1;
5c635c72 MD	171	goto error;
	172	}
	173
6c71277b MD	174	is_root = !getuid();
	175	if (is_root) {
	176	/* lttng health client socket path permissions */
28ab59d0 JR	177	gid_t gid;
	178
	179	ret = utils_get_group_id(tracing_group_name, true, &gid);
	180	if (ret) {
	181	/* Default to root group. */
	182	gid = 0;
	183	}
	184
	185	ret = chown(health_unix_sock_path, 0, gid);
6c71277b MD	186	if (ret < 0) {
	187	ERR("Unable to set group on %s", health_unix_sock_path);
	188	PERROR("chown");
67fe4075	189	err = -1;
6c71277b MD	190	goto error;
	191	}
	192
	193	ret = chmod(health_unix_sock_path,
	194	S_IRUSR \| S_IWUSR \| S_IRGRP \| S_IWGRP);
	195	if (ret < 0) {
	196	ERR("Unable to set permissions on %s", health_unix_sock_path);
	197	PERROR("chmod");
67fe4075	198	err = -1;
6c71277b MD	199	goto error;
	200	}
	201	}
	202
5c635c72 MD	203	/*
	204	* Set the CLOEXEC flag. Return code is useless because either way, the
	205	* show must go on.
	206	*/
	207	(void) utils_set_fd_cloexec(sock);
	208
	209	ret = lttcomm_listen_unix_sock(sock);
	210	if (ret < 0) {
	211	goto error;
	212	}
	213
	214	/* Size is set to 1 for the consumer_channel pipe */
	215	ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
	216	if (ret < 0) {
	217	ERR("Poll set creation failed");
	218	goto error;
	219	}
	220
	221	ret = lttng_poll_add(&events, health_quit_pipe[0], LPOLLIN);
	222	if (ret < 0) {
	223	goto error;
	224	}
	225
	226	/* Add the application registration socket */
	227	ret = lttng_poll_add(&events, sock, LPOLLIN \| LPOLLPRI);
	228	if (ret < 0) {
	229	goto error;
	230	}
	231
748b7b07 MD	232	/* Perform prior memory accesses before decrementing ready */
	233	cmm_smp_mb__before_uatomic_dec();
	234	uatomic_dec(&lttng_consumer_ready);
	235
5c635c72 MD	236	while (1) {
	237	DBG("Health check ready");
	238
	239	/* Inifinite blocking call, waiting for transmission */
	240	restart:
	241	ret = lttng_poll_wait(&events, -1);
	242	if (ret < 0) {
	243	/*
	244	* Restart interrupted system call.
	245	*/
	246	if (errno == EINTR) {
	247	goto restart;
	248	}
	249	goto error;
	250	}
	251
	252	nb_fd = ret;
	253
	254	for (i = 0; i < nb_fd; i++) {
	255	/* Fetch once the poll data */
	256	revents = LTTNG_POLL_GETEV(&events, i);
	257	pollfd = LTTNG_POLL_GETFD(&events, i);
	258
	259	/* Thread quit pipe has been closed. Killing thread. */
	260	ret = check_health_quit_pipe(pollfd, revents);
	261	if (ret) {
	262	err = 0;
	263	goto exit;
	264	}
	265
	266	/* Event on the registration socket */
	267	if (pollfd == sock) {
03e43155 MD	268	if (revents & (LPOLLERR \| LPOLLHUP \| LPOLLRDHUP)
03e43155 MD	269	&& !(revents & LPOLLIN)) {
5c635c72 MD	270	ERR("Health socket poll error");
	271	goto error;
	272	}
	273	}
	274	}
	275
	276	new_sock = lttcomm_accept_unix_sock(sock);
	277	if (new_sock < 0) {
	278	goto error;
	279	}
	280
	281	/*
	282	* Set the CLOEXEC flag. Return code is useless because either way, the
	283	* show must go on.
	284	*/
	285	(void) utils_set_fd_cloexec(new_sock);
	286
	287	DBG("Receiving data from client for health...");
	288	ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
	289	if (ret <= 0) {
	290	DBG("Nothing recv() from client... continuing");
	291	ret = close(new_sock);
	292	if (ret) {
	293	PERROR("close");
	294	}
	295	new_sock = -1;
	296	continue;
	297	}
	298
	299	rcu_thread_online();
	300
	301	assert(msg.cmd == HEALTH_CMD_CHECK);
	302
53efb85a	303	memset(&reply, 0, sizeof(reply));
6c71277b MD	304	for (i = 0; i < NR_HEALTH_CONSUMERD_TYPES; i++) {
	305	/*
	306	* health_check_state return 0 if thread is in
	307	* error.
	308	*/
	309	if (!health_check_state(health_consumerd, i)) {
	310	reply.ret_code \|= 1ULL << i;
	311	}
5c635c72 MD	312	}
5c635c72 MD	313
6137f630	314	DBG("Health check return value %" PRIx64, reply.ret_code);
5c635c72 MD	315
	316	ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
	317	if (ret < 0) {
	318	ERR("Failed to send health data back to client");
	319	}
	320
	321	/* End of transmission */
	322	ret = close(new_sock);
	323	if (ret) {
	324	PERROR("close");
	325	}
	326	new_sock = -1;
	327	}
	328
	329	exit:
	330	error:
	331	if (err) {
	332	ERR("Health error occurred in %s", __func__);
	333	}
	334	DBG("Health check thread dying");
	335	unlink(health_unix_sock_path);
	336	if (sock >= 0) {
	337	ret = close(sock);
	338	if (ret) {
	339	PERROR("close");
	340	}
	341	}
	342
	343	lttng_poll_clean(&events);
	344
	345	rcu_unregister_thread();
	346	return NULL;
	347	}