[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.hpp>
#include <common/common.hpp>
#include <common/consumer/consumer.hpp>
#include <common/consumer/consumer-timer.hpp>
#include <common/compat/poll.hpp>
#include <common/sessiond-comm/sessiond-comm.hpp>
#include <common/utils.hpp>

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

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

int health_quit_pipe[2] = {-1, -1};

/*
 * 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, err = -1;
	uint32_t 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 */
			const auto revents = LTTNG_POLL_GETEV(&events, i);
			const auto pollfd = LTTNG_POLL_GETFD(&events, i);

			/* Activity on health quit pipe, exiting. */
			if (pollfd == health_quit_pipe[0]) {
				DBG("Activity on health quit pipe");
				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	31
c9e313bc SM	32	#include <common/defaults.hpp>
	33	#include <common/common.hpp>
	34	#include <common/consumer/consumer.hpp>
	35	#include <common/consumer/consumer-timer.hpp>
	36	#include <common/compat/poll.hpp>
	37	#include <common/sessiond-comm/sessiond-comm.hpp>
	38	#include <common/utils.hpp>
5c635c72	39
c9e313bc SM	40	#include "lttng-consumerd.hpp"
c9e313bc SM	41	#include "health-consumerd.hpp"
5c635c72 MD	42
	43	/* Global health check unix path */
	44	static char health_unix_sock_path[PATH_MAX];
	45
8a00688e	46	int health_quit_pipe[2] = {-1, -1};
5c635c72 MD	47
	48	/*
	49	* Send data on a unix socket using the liblttsessiondcomm API.
	50	*
	51	* Return lttcomm error code.
	52	*/
	53	static int send_unix_sock(int sock, void *buf, size_t len)
	54	{
	55	/* Check valid length */
	56	if (len == 0) {
	57	return -1;
	58	}
	59
	60	return lttcomm_send_unix_sock(sock, buf, len);
	61	}
	62
	63	static
	64	int setup_health_path(void)
	65	{
	66	int is_root, ret = 0;
	67	enum lttng_consumer_type type;
	68	const char *home_path;
	69
	70	type = lttng_consumer_get_type();
	71	is_root = !getuid();
	72
	73	if (is_root) {
	74	if (strlen(health_unix_sock_path) != 0) {
	75	goto end;
	76	}
	77	switch (type) {
	78	case LTTNG_CONSUMER_KERNEL:
	79	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
	80	DEFAULT_GLOBAL_KCONSUMER_HEALTH_UNIX_SOCK);
	81	break;
	82	case LTTNG_CONSUMER64_UST:
	83	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
	84	DEFAULT_GLOBAL_USTCONSUMER64_HEALTH_UNIX_SOCK);
	85	break;
	86	case LTTNG_CONSUMER32_UST:
	87	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
	88	DEFAULT_GLOBAL_USTCONSUMER32_HEALTH_UNIX_SOCK);
	89	break;
	90	default:
	91	ret = -EINVAL;
	92	goto end;
	93	}
	94	} else {
5c635c72 MD	95	home_path = utils_get_home_dir();
	96	if (home_path == NULL) {
	97	/* TODO: Add --socket PATH option */
	98	ERR("Can't get HOME directory for sockets creation.");
	99	ret = -EPERM;
	100	goto end;
	101	}
	102
5c635c72 MD	103	/* Set health check Unix path */
	104	if (strlen(health_unix_sock_path) != 0) {
	105	goto end;
	106	}
	107	switch (type) {
	108	case LTTNG_CONSUMER_KERNEL:
	109	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
dbc8403d	110	DEFAULT_HOME_KCONSUMER_HEALTH_UNIX_SOCK, home_path);
5c635c72 MD	111	break;
	112	case LTTNG_CONSUMER64_UST:
	113	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
dbc8403d	114	DEFAULT_HOME_USTCONSUMER64_HEALTH_UNIX_SOCK, home_path);
5c635c72 MD	115	break;
	116	case LTTNG_CONSUMER32_UST:
	117	snprintf(health_unix_sock_path, sizeof(health_unix_sock_path),
dbc8403d	118	DEFAULT_HOME_USTCONSUMER32_HEALTH_UNIX_SOCK, home_path);
5c635c72 MD	119	break;
	120	default:
	121	ret = -EINVAL;
	122	goto end;
	123	}
	124	}
5c635c72 MD	125	end:
	126	return ret;
	127	}
	128
	129	/*
	130	* Thread managing health check socket.
	131	*/
f46376a1	132	void thread_manage_health_consumerd(void data __attribute__((unused)))
5c635c72	133	{
8a00688e MJ	134	int sock = -1, new_sock = -1, ret, i, err = -1;
8a00688e MJ	135	uint32_t nb_fd;
5c635c72 MD	136	struct lttng_poll_event events;
	137	struct health_comm_msg msg;
	138	struct health_comm_reply reply;
6c71277b	139	int is_root;
5c635c72 MD	140
	141	DBG("[thread] Manage health check started");
	142
	143	setup_health_path();
	144
	145	rcu_register_thread();
	146
	147	/* We might hit an error path before this is created. */
	148	lttng_poll_init(&events);
	149
	150	/* Create unix socket */
	151	sock = lttcomm_create_unix_sock(health_unix_sock_path);
	152	if (sock < 0) {
	153	ERR("Unable to create health check Unix socket");
67fe4075	154	err = -1;
5c635c72 MD	155	goto error;
	156	}
	157
6c71277b MD	158	is_root = !getuid();
	159	if (is_root) {
	160	/* lttng health client socket path permissions */
28ab59d0 JR	161	gid_t gid;
	162
	163	ret = utils_get_group_id(tracing_group_name, true, &gid);
	164	if (ret) {
	165	/* Default to root group. */
	166	gid = 0;
	167	}
	168
	169	ret = chown(health_unix_sock_path, 0, gid);
6c71277b MD	170	if (ret < 0) {
	171	ERR("Unable to set group on %s", health_unix_sock_path);
	172	PERROR("chown");
67fe4075	173	err = -1;
6c71277b MD	174	goto error;
	175	}
	176
	177	ret = chmod(health_unix_sock_path,
	178	S_IRUSR \| S_IWUSR \| S_IRGRP \| S_IWGRP);
	179	if (ret < 0) {
	180	ERR("Unable to set permissions on %s", health_unix_sock_path);
	181	PERROR("chmod");
67fe4075	182	err = -1;
6c71277b MD	183	goto error;
	184	}
	185	}
	186
5c635c72 MD	187	/*
	188	* Set the CLOEXEC flag. Return code is useless because either way, the
	189	* show must go on.
	190	*/
	191	(void) utils_set_fd_cloexec(sock);
	192
	193	ret = lttcomm_listen_unix_sock(sock);
	194	if (ret < 0) {
	195	goto error;
	196	}
	197
3069d754	198	/* Size is set to 2 for the quit pipe and registration socket. */
5c635c72 MD	199	ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
	200	if (ret < 0) {
	201	ERR("Poll set creation failed");
	202	goto error;
	203	}
	204
	205	ret = lttng_poll_add(&events, health_quit_pipe[0], LPOLLIN);
	206	if (ret < 0) {
	207	goto error;
	208	}
	209
	210	/* Add the application registration socket */
	211	ret = lttng_poll_add(&events, sock, LPOLLIN \| LPOLLPRI);
	212	if (ret < 0) {
	213	goto error;
	214	}
	215
748b7b07 MD	216	/* Perform prior memory accesses before decrementing ready */
	217	cmm_smp_mb__before_uatomic_dec();
	218	uatomic_dec(&lttng_consumer_ready);
	219
5c635c72 MD	220	while (1) {
	221	DBG("Health check ready");
	222
	223	/* Inifinite blocking call, waiting for transmission */
	224	restart:
	225	ret = lttng_poll_wait(&events, -1);
	226	if (ret < 0) {
	227	/*
	228	* Restart interrupted system call.
	229	*/
	230	if (errno == EINTR) {
	231	goto restart;
	232	}
	233	goto error;
	234	}
	235
	236	nb_fd = ret;
	237
	238	for (i = 0; i < nb_fd; i++) {
	239	/* Fetch once the poll data */
8a00688e MJ	240	const auto revents = LTTNG_POLL_GETEV(&events, i);
8a00688e MJ	241	const auto pollfd = LTTNG_POLL_GETFD(&events, i);
5c635c72	242
8a00688e MJ	243	/* Activity on health quit pipe, exiting. */
	244	if (pollfd == health_quit_pipe[0]) {
	245	DBG("Activity on health quit pipe");
5c635c72 MD	246	err = 0;
	247	goto exit;
	248	}
	249
	250	/* Event on the registration socket */
	251	if (pollfd == sock) {
03e43155 MD	252	if (revents & (LPOLLERR \| LPOLLHUP \| LPOLLRDHUP)
03e43155 MD	253	&& !(revents & LPOLLIN)) {
5c635c72 MD	254	ERR("Health socket poll error");
	255	goto error;
	256	}
	257	}
	258	}
	259
	260	new_sock = lttcomm_accept_unix_sock(sock);
	261	if (new_sock < 0) {
	262	goto error;
	263	}
	264
	265	/*
	266	* Set the CLOEXEC flag. Return code is useless because either way, the
	267	* show must go on.
	268	*/
	269	(void) utils_set_fd_cloexec(new_sock);
	270
	271	DBG("Receiving data from client for health...");
	272	ret = lttcomm_recv_unix_sock(new_sock, (void *)&msg, sizeof(msg));
	273	if (ret <= 0) {
	274	DBG("Nothing recv() from client... continuing");
	275	ret = close(new_sock);
	276	if (ret) {
	277	PERROR("close");
	278	}
	279	new_sock = -1;
	280	continue;
	281	}
	282
	283	rcu_thread_online();
	284
a0377dfe	285	LTTNG_ASSERT(msg.cmd == HEALTH_CMD_CHECK);
5c635c72	286
53efb85a	287	memset(&reply, 0, sizeof(reply));
6c71277b MD	288	for (i = 0; i < NR_HEALTH_CONSUMERD_TYPES; i++) {
	289	/*
	290	* health_check_state return 0 if thread is in
	291	* error.
	292	*/
	293	if (!health_check_state(health_consumerd, i)) {
	294	reply.ret_code \|= 1ULL << i;
	295	}
5c635c72 MD	296	}
5c635c72 MD	297
6137f630	298	DBG("Health check return value %" PRIx64, reply.ret_code);
5c635c72 MD	299
	300	ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply));
	301	if (ret < 0) {
	302	ERR("Failed to send health data back to client");
	303	}
	304
	305	/* End of transmission */
	306	ret = close(new_sock);
	307	if (ret) {
	308	PERROR("close");
	309	}
	310	new_sock = -1;
	311	}
	312
	313	exit:
	314	error:
	315	if (err) {
	316	ERR("Health error occurred in %s", __func__);
	317	}
	318	DBG("Health check thread dying");
	319	unlink(health_unix_sock_path);
	320	if (sock >= 0) {
	321	ret = close(sock);
	322	if (ret) {
	323	PERROR("close");
	324	}
	325	}
	326
	327	lttng_poll_clean(&events);
	328
	329	rcu_unregister_thread();
	330	return NULL;
	331	}