2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
36 #include <sys/types.h>
38 #include <urcu/uatomic.h>
42 #include <common/common.h>
43 #include <common/compat/socket.h>
44 #include <common/compat/getenv.h>
45 #include <common/defaults.h>
46 #include <common/kernel-consumer/kernel-consumer.h>
47 #include <common/futex.h>
48 #include <common/relayd/relayd.h>
49 #include <common/utils.h>
50 #include <common/daemonize.h>
51 #include <common/config/session-config.h>
53 #include "lttng-sessiond.h"
54 #include "buffer-registry.h"
61 #include "kernel-consumer.h"
65 #include "ust-consumer.h"
68 #include "health-sessiond.h"
69 #include "testpoint.h"
70 #include "ust-thread.h"
71 #include "agent-thread.h"
73 #include "load-session-thread.h"
76 #include "ht-cleanup.h"
78 #define CONSUMERD_FILE "lttng-consumerd"
81 static const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
82 static int tracing_group_name_override
;
83 static char *opt_pidfile
;
84 static int opt_sig_parent
;
85 static int opt_verbose_consumer
;
86 static int opt_daemon
, opt_background
;
87 static int opt_no_kernel
;
88 static char *opt_load_session_path
;
89 static pid_t ppid
; /* Parent PID for --sig-parent option */
90 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
92 static int lockfile_fd
= -1;
94 /* Set to 1 when a SIGUSR1 signal is received. */
95 static int recv_child_signal
;
98 * Consumer daemon specific control data. Every value not initialized here is
99 * set to 0 by the static definition.
101 static struct consumer_data kconsumer_data
= {
102 .type
= LTTNG_CONSUMER_KERNEL
,
103 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
104 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
107 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
108 .lock
= PTHREAD_MUTEX_INITIALIZER
,
109 .cond
= PTHREAD_COND_INITIALIZER
,
110 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
112 static struct consumer_data ustconsumer64_data
= {
113 .type
= LTTNG_CONSUMER64_UST
,
114 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
115 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
118 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
119 .lock
= PTHREAD_MUTEX_INITIALIZER
,
120 .cond
= PTHREAD_COND_INITIALIZER
,
121 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
123 static struct consumer_data ustconsumer32_data
= {
124 .type
= LTTNG_CONSUMER32_UST
,
125 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
126 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
129 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
130 .lock
= PTHREAD_MUTEX_INITIALIZER
,
131 .cond
= PTHREAD_COND_INITIALIZER
,
132 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
135 /* Command line options */
136 static const struct option long_options
[] = {
137 { "client-sock", required_argument
, 0, 'c' },
138 { "apps-sock", required_argument
, 0, 'a' },
139 { "kconsumerd-cmd-sock", required_argument
, 0, '\0' },
140 { "kconsumerd-err-sock", required_argument
, 0, '\0' },
141 { "ustconsumerd32-cmd-sock", required_argument
, 0, '\0' },
142 { "ustconsumerd32-err-sock", required_argument
, 0, '\0' },
143 { "ustconsumerd64-cmd-sock", required_argument
, 0, '\0' },
144 { "ustconsumerd64-err-sock", required_argument
, 0, '\0' },
145 { "consumerd32-path", required_argument
, 0, '\0' },
146 { "consumerd32-libdir", required_argument
, 0, '\0' },
147 { "consumerd64-path", required_argument
, 0, '\0' },
148 { "consumerd64-libdir", required_argument
, 0, '\0' },
149 { "daemonize", no_argument
, 0, 'd' },
150 { "background", no_argument
, 0, 'b' },
151 { "sig-parent", no_argument
, 0, 'S' },
152 { "help", no_argument
, 0, 'h' },
153 { "group", required_argument
, 0, 'g' },
154 { "version", no_argument
, 0, 'V' },
155 { "quiet", no_argument
, 0, 'q' },
156 { "verbose", no_argument
, 0, 'v' },
157 { "verbose-consumer", no_argument
, 0, '\0' },
158 { "no-kernel", no_argument
, 0, '\0' },
159 { "pidfile", required_argument
, 0, 'p' },
160 { "agent-tcp-port", required_argument
, 0, '\0' },
161 { "config", required_argument
, 0, 'f' },
162 { "load", required_argument
, 0, 'l' },
163 { "kmod-probes", required_argument
, 0, '\0' },
164 { "extra-kmod-probes", required_argument
, 0, '\0' },
168 /* Command line options to ignore from configuration file */
169 static const char *config_ignore_options
[] = { "help", "version", "config" };
171 /* Shared between threads */
172 static int dispatch_thread_exit
;
174 /* Global application Unix socket path */
175 static char apps_unix_sock_path
[PATH_MAX
];
176 /* Global client Unix socket path */
177 static char client_unix_sock_path
[PATH_MAX
];
178 /* global wait shm path for UST */
179 static char wait_shm_path
[PATH_MAX
];
180 /* Global health check unix path */
181 static char health_unix_sock_path
[PATH_MAX
];
183 /* Sockets and FDs */
184 static int client_sock
= -1;
185 static int apps_sock
= -1;
186 int kernel_tracer_fd
= -1;
187 static int kernel_poll_pipe
[2] = { -1, -1 };
190 * Quit pipe for all threads. This permits a single cancellation point
191 * for all threads when receiving an event on the pipe.
193 static int thread_quit_pipe
[2] = { -1, -1 };
196 * This pipe is used to inform the thread managing application communication
197 * that a command is queued and ready to be processed.
199 static int apps_cmd_pipe
[2] = { -1, -1 };
201 int apps_cmd_notify_pipe
[2] = { -1, -1 };
203 /* Pthread, Mutexes and Semaphores */
204 static pthread_t apps_thread
;
205 static pthread_t apps_notify_thread
;
206 static pthread_t reg_apps_thread
;
207 static pthread_t client_thread
;
208 static pthread_t kernel_thread
;
209 static pthread_t dispatch_thread
;
210 static pthread_t health_thread
;
211 static pthread_t ht_cleanup_thread
;
212 static pthread_t agent_reg_thread
;
213 static pthread_t load_session_thread
;
216 * UST registration command queue. This queue is tied with a futex and uses a N
217 * wakers / 1 waiter implemented and detailed in futex.c/.h
219 * The thread_registration_apps and thread_dispatch_ust_registration uses this
220 * queue along with the wait/wake scheme. The thread_manage_apps receives down
221 * the line new application socket and monitors it for any I/O error or clean
222 * close that triggers an unregistration of the application.
224 static struct ust_cmd_queue ust_cmd_queue
;
227 * Pointer initialized before thread creation.
229 * This points to the tracing session list containing the session count and a
230 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
231 * MUST NOT be taken if you call a public function in session.c.
233 * The lock is nested inside the structure: session_list_ptr->lock. Please use
234 * session_lock_list and session_unlock_list for lock acquisition.
236 static struct ltt_session_list
*session_list_ptr
;
238 int ust_consumerd64_fd
= -1;
239 int ust_consumerd32_fd
= -1;
241 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
242 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
243 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
244 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
245 static int consumerd32_bin_override
;
246 static int consumerd64_bin_override
;
247 static int consumerd32_libdir_override
;
248 static int consumerd64_libdir_override
;
250 static const char *module_proc_lttng
= "/proc/lttng";
253 * Consumer daemon state which is changed when spawning it, killing it or in
254 * case of a fatal error.
256 enum consumerd_state
{
257 CONSUMER_STARTED
= 1,
258 CONSUMER_STOPPED
= 2,
263 * This consumer daemon state is used to validate if a client command will be
264 * able to reach the consumer. If not, the client is informed. For instance,
265 * doing a "lttng start" when the consumer state is set to ERROR will return an
266 * error to the client.
268 * The following example shows a possible race condition of this scheme:
270 * consumer thread error happens
272 * client cmd checks state -> still OK
273 * consumer thread exit, sets error
274 * client cmd try to talk to consumer
277 * However, since the consumer is a different daemon, we have no way of making
278 * sure the command will reach it safely even with this state flag. This is why
279 * we consider that up to the state validation during command processing, the
280 * command is safe. After that, we can not guarantee the correctness of the
281 * client request vis-a-vis the consumer.
283 static enum consumerd_state ust_consumerd_state
;
284 static enum consumerd_state kernel_consumerd_state
;
287 * Socket timeout for receiving and sending in seconds.
289 static int app_socket_timeout
;
291 /* Set in main() with the current page size. */
294 /* Application health monitoring */
295 struct health_app
*health_sessiond
;
297 /* Agent TCP port for registration. Used by the agent thread. */
298 unsigned int agent_tcp_port
= DEFAULT_AGENT_TCP_PORT
;
300 /* Am I root or not. */
301 int is_root
; /* Set to 1 if the daemon is running as root */
303 const char * const config_section_name
= "sessiond";
305 /* Load session thread information to operate. */
306 struct load_session_thread_data
*load_info
;
308 /* Global hash tables */
309 struct lttng_ht
*agent_apps_ht_by_sock
= NULL
;
312 * Whether sessiond is ready for commands/health check requests.
313 * NR_LTTNG_SESSIOND_READY must match the number of calls to
314 * sessiond_notify_ready().
316 #define NR_LTTNG_SESSIOND_READY 3
317 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
319 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
321 return (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) ? 1 : 0;
324 /* Notify parents that we are ready for cmd and health check */
326 void sessiond_notify_ready(void)
328 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
330 * Notify parent pid that we are ready to accept command
331 * for client side. This ppid is the one from the
332 * external process that spawned us.
334 if (opt_sig_parent
) {
339 * Notify the parent of the fork() process that we are
342 if (opt_daemon
|| opt_background
) {
343 kill(child_ppid
, SIGUSR1
);
349 void setup_consumerd_path(void)
351 const char *bin
, *libdir
;
354 * Allow INSTALL_BIN_PATH to be used as a target path for the
355 * native architecture size consumer if CONFIG_CONSUMER*_PATH
356 * has not been defined.
358 #if (CAA_BITS_PER_LONG == 32)
359 if (!consumerd32_bin
[0]) {
360 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
362 if (!consumerd32_libdir
[0]) {
363 consumerd32_libdir
= INSTALL_LIB_PATH
;
365 #elif (CAA_BITS_PER_LONG == 64)
366 if (!consumerd64_bin
[0]) {
367 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
369 if (!consumerd64_libdir
[0]) {
370 consumerd64_libdir
= INSTALL_LIB_PATH
;
373 #error "Unknown bitness"
377 * runtime env. var. overrides the build default.
379 bin
= lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
381 consumerd32_bin
= bin
;
383 bin
= lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
385 consumerd64_bin
= bin
;
387 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
389 consumerd32_libdir
= libdir
;
391 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
393 consumerd64_libdir
= libdir
;
398 int __sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
,
405 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
411 ret
= lttng_poll_add(events
, a_pipe
[0], LPOLLIN
| LPOLLERR
);
423 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
425 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
427 return __sessiond_set_thread_pollset(events
, size
, thread_quit_pipe
);
431 * Init thread quit pipe.
433 * Return -1 on error or 0 if all pipes are created.
435 static int __init_thread_quit_pipe(int *a_pipe
)
441 PERROR("thread quit pipe");
445 for (i
= 0; i
< 2; i
++) {
446 ret
= fcntl(a_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
457 static int init_thread_quit_pipe(void)
459 return __init_thread_quit_pipe(thread_quit_pipe
);
463 * Stop all threads by closing the thread quit pipe.
465 static void stop_threads(void)
469 /* Stopping all threads */
470 DBG("Terminating all threads");
471 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
473 ERR("write error on thread quit pipe");
476 /* Dispatch thread */
477 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
478 futex_nto1_wake(&ust_cmd_queue
.futex
);
482 * Close every consumer sockets.
484 static void close_consumer_sockets(void)
488 if (kconsumer_data
.err_sock
>= 0) {
489 ret
= close(kconsumer_data
.err_sock
);
491 PERROR("kernel consumer err_sock close");
494 if (ustconsumer32_data
.err_sock
>= 0) {
495 ret
= close(ustconsumer32_data
.err_sock
);
497 PERROR("UST consumerd32 err_sock close");
500 if (ustconsumer64_data
.err_sock
>= 0) {
501 ret
= close(ustconsumer64_data
.err_sock
);
503 PERROR("UST consumerd64 err_sock close");
506 if (kconsumer_data
.cmd_sock
>= 0) {
507 ret
= close(kconsumer_data
.cmd_sock
);
509 PERROR("kernel consumer cmd_sock close");
512 if (ustconsumer32_data
.cmd_sock
>= 0) {
513 ret
= close(ustconsumer32_data
.cmd_sock
);
515 PERROR("UST consumerd32 cmd_sock close");
518 if (ustconsumer64_data
.cmd_sock
>= 0) {
519 ret
= close(ustconsumer64_data
.cmd_sock
);
521 PERROR("UST consumerd64 cmd_sock close");
527 * Generate the full lock file path using the rundir.
529 * Return the snprintf() return value thus a negative value is an error.
531 static int generate_lock_file_path(char *path
, size_t len
)
538 /* Build lockfile path from rundir. */
539 ret
= snprintf(path
, len
, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE
, rundir
);
541 PERROR("snprintf lockfile path");
548 * Wait on consumer process termination.
550 * Need to be called with the consumer data lock held or from a context
551 * ensuring no concurrent access to data (e.g: cleanup).
553 static void wait_consumer(struct consumer_data
*consumer_data
)
558 if (consumer_data
->pid
<= 0) {
562 DBG("Waiting for complete teardown of consumerd (PID: %d)",
564 ret
= waitpid(consumer_data
->pid
, &status
, 0);
566 PERROR("consumerd waitpid pid: %d", consumer_data
->pid
)
568 if (!WIFEXITED(status
)) {
569 ERR("consumerd termination with error: %d",
572 consumer_data
->pid
= 0;
576 * Cleanup the session daemon's data structures.
578 static void sessiond_cleanup(void)
581 struct ltt_session
*sess
, *stmp
;
584 DBG("Cleanup sessiond");
587 * Close the thread quit pipe. It has already done its job,
588 * since we are now called.
590 utils_close_pipe(thread_quit_pipe
);
593 * If opt_pidfile is undefined, the default file will be wiped when
594 * removing the rundir.
597 ret
= remove(opt_pidfile
);
599 PERROR("remove pidfile %s", opt_pidfile
);
603 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
606 snprintf(path
, PATH_MAX
,
608 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
609 DBG("Removing %s", path
);
612 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
613 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
);
614 DBG("Removing %s", path
);
618 snprintf(path
, PATH_MAX
,
619 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
621 DBG("Removing %s", path
);
624 snprintf(path
, PATH_MAX
,
625 DEFAULT_KCONSUMERD_PATH
,
627 DBG("Removing directory %s", path
);
630 /* ust consumerd 32 */
631 snprintf(path
, PATH_MAX
,
632 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
634 DBG("Removing %s", path
);
637 snprintf(path
, PATH_MAX
,
638 DEFAULT_USTCONSUMERD32_PATH
,
640 DBG("Removing directory %s", path
);
643 /* ust consumerd 64 */
644 snprintf(path
, PATH_MAX
,
645 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
647 DBG("Removing %s", path
);
650 snprintf(path
, PATH_MAX
,
651 DEFAULT_USTCONSUMERD64_PATH
,
653 DBG("Removing directory %s", path
);
656 DBG("Cleaning up all sessions");
658 /* Destroy session list mutex */
659 if (session_list_ptr
!= NULL
) {
660 pthread_mutex_destroy(&session_list_ptr
->lock
);
662 /* Cleanup ALL session */
663 cds_list_for_each_entry_safe(sess
, stmp
,
664 &session_list_ptr
->head
, list
) {
665 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
669 wait_consumer(&kconsumer_data
);
670 wait_consumer(&ustconsumer64_data
);
671 wait_consumer(&ustconsumer32_data
);
673 DBG("Cleaning up all agent apps");
674 agent_app_ht_clean();
676 DBG("Closing all UST sockets");
677 ust_app_clean_list();
678 buffer_reg_destroy_registries();
680 if (is_root
&& !opt_no_kernel
) {
681 DBG2("Closing kernel fd");
682 if (kernel_tracer_fd
>= 0) {
683 ret
= close(kernel_tracer_fd
);
688 DBG("Unloading kernel modules");
689 modprobe_remove_lttng_all();
693 close_consumer_sockets();
696 load_session_destroy_data(load_info
);
701 * Cleanup lock file by deleting it and finaly closing it which will
702 * release the file system lock.
704 if (lockfile_fd
>= 0) {
705 char lockfile_path
[PATH_MAX
];
707 ret
= generate_lock_file_path(lockfile_path
,
708 sizeof(lockfile_path
));
710 ret
= remove(lockfile_path
);
712 PERROR("remove lock file");
714 ret
= close(lockfile_fd
);
716 PERROR("close lock file");
722 * We do NOT rmdir rundir because there are other processes
723 * using it, for instance lttng-relayd, which can start in
724 * parallel with this teardown.
731 * Cleanup the daemon's option data structures.
733 static void sessiond_cleanup_options(void)
735 DBG("Cleaning up options");
738 * If the override option is set, the pointer points to a *non* const
739 * thus freeing it even though the variable type is set to const.
741 if (tracing_group_name_override
) {
742 free((void *) tracing_group_name
);
744 if (consumerd32_bin_override
) {
745 free((void *) consumerd32_bin
);
747 if (consumerd64_bin_override
) {
748 free((void *) consumerd64_bin
);
750 if (consumerd32_libdir_override
) {
751 free((void *) consumerd32_libdir
);
753 if (consumerd64_libdir_override
) {
754 free((void *) consumerd64_libdir
);
758 free(opt_load_session_path
);
759 free(kmod_probes_list
);
760 free(kmod_extra_probes_list
);
762 run_as_destroy_worker();
765 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
766 "Matthew, BEET driven development works!%c[%dm",
767 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
772 * Send data on a unix socket using the liblttsessiondcomm API.
774 * Return lttcomm error code.
776 static int send_unix_sock(int sock
, void *buf
, size_t len
)
778 /* Check valid length */
783 return lttcomm_send_unix_sock(sock
, buf
, len
);
787 * Free memory of a command context structure.
789 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
791 DBG("Clean command context structure");
793 if ((*cmd_ctx
)->llm
) {
794 free((*cmd_ctx
)->llm
);
796 if ((*cmd_ctx
)->lsm
) {
797 free((*cmd_ctx
)->lsm
);
805 * Notify UST applications using the shm mmap futex.
807 static int notify_ust_apps(int active
)
811 DBG("Notifying applications of session daemon state: %d", active
);
813 /* See shm.c for this call implying mmap, shm and futex calls */
814 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
815 if (wait_shm_mmap
== NULL
) {
819 /* Wake waiting process */
820 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
822 /* Apps notified successfully */
830 * Setup the outgoing data buffer for the response (llm) by allocating the
831 * right amount of memory and copying the original information from the lsm
834 * Return 0 on success, negative value on error.
836 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
,
837 const void *payload_buf
, size_t payload_len
,
838 const void *cmd_header_buf
, size_t cmd_header_len
)
841 const size_t header_len
= sizeof(struct lttcomm_lttng_msg
);
842 const size_t cmd_header_offset
= header_len
;
843 const size_t payload_offset
= cmd_header_offset
+ cmd_header_len
;
844 const size_t total_msg_size
= header_len
+ cmd_header_len
+ payload_len
;
846 cmd_ctx
->llm
= zmalloc(total_msg_size
);
848 if (cmd_ctx
->llm
== NULL
) {
854 /* Copy common data */
855 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
856 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
857 cmd_ctx
->llm
->cmd_header_size
= cmd_header_len
;
858 cmd_ctx
->llm
->data_size
= payload_len
;
859 cmd_ctx
->lttng_msg_size
= total_msg_size
;
861 /* Copy command header */
862 if (cmd_header_len
) {
863 memcpy(((uint8_t *) cmd_ctx
->llm
) + cmd_header_offset
, cmd_header_buf
,
869 memcpy(((uint8_t *) cmd_ctx
->llm
) + payload_offset
, payload_buf
,
878 * Version of setup_lttng_msg() without command header.
880 static int setup_lttng_msg_no_cmd_header(struct command_ctx
*cmd_ctx
,
881 void *payload_buf
, size_t payload_len
)
883 return setup_lttng_msg(cmd_ctx
, payload_buf
, payload_len
, NULL
, 0);
886 * Update the kernel poll set of all channel fd available over all tracing
887 * session. Add the wakeup pipe at the end of the set.
889 static int update_kernel_poll(struct lttng_poll_event
*events
)
892 struct ltt_session
*session
;
893 struct ltt_kernel_channel
*channel
;
895 DBG("Updating kernel poll set");
898 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
899 session_lock(session
);
900 if (session
->kernel_session
== NULL
) {
901 session_unlock(session
);
905 cds_list_for_each_entry(channel
,
906 &session
->kernel_session
->channel_list
.head
, list
) {
907 /* Add channel fd to the kernel poll set */
908 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
910 session_unlock(session
);
913 DBG("Channel fd %d added to kernel set", channel
->fd
);
915 session_unlock(session
);
917 session_unlock_list();
922 session_unlock_list();
927 * Find the channel fd from 'fd' over all tracing session. When found, check
928 * for new channel stream and send those stream fds to the kernel consumer.
930 * Useful for CPU hotplug feature.
932 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
935 struct ltt_session
*session
;
936 struct ltt_kernel_session
*ksess
;
937 struct ltt_kernel_channel
*channel
;
939 DBG("Updating kernel streams for channel fd %d", fd
);
942 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
943 session_lock(session
);
944 if (session
->kernel_session
== NULL
) {
945 session_unlock(session
);
948 ksess
= session
->kernel_session
;
950 cds_list_for_each_entry(channel
,
951 &ksess
->channel_list
.head
, list
) {
952 struct lttng_ht_iter iter
;
953 struct consumer_socket
*socket
;
955 if (channel
->fd
!= fd
) {
958 DBG("Channel found, updating kernel streams");
959 ret
= kernel_open_channel_stream(channel
);
963 /* Update the stream global counter */
964 ksess
->stream_count_global
+= ret
;
967 * Have we already sent fds to the consumer? If yes, it
968 * means that tracing is started so it is safe to send
969 * our updated stream fds.
971 if (ksess
->consumer_fds_sent
!= 1
972 || ksess
->consumer
== NULL
) {
978 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
979 &iter
.iter
, socket
, node
.node
) {
980 pthread_mutex_lock(socket
->lock
);
981 ret
= kernel_consumer_send_channel_stream(socket
,
983 session
->output_traces
? 1 : 0);
984 pthread_mutex_unlock(socket
->lock
);
992 session_unlock(session
);
994 session_unlock_list();
998 session_unlock(session
);
999 session_unlock_list();
1004 * For each tracing session, update newly registered apps. The session list
1005 * lock MUST be acquired before calling this.
1007 static void update_ust_app(int app_sock
)
1009 struct ltt_session
*sess
, *stmp
;
1011 /* Consumer is in an ERROR state. Stop any application update. */
1012 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
1013 /* Stop the update process since the consumer is dead. */
1017 /* For all tracing session(s) */
1018 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
1019 struct ust_app
*app
;
1022 if (!sess
->ust_session
) {
1023 goto unlock_session
;
1027 assert(app_sock
>= 0);
1028 app
= ust_app_find_by_sock(app_sock
);
1031 * Application can be unregistered before so
1032 * this is possible hence simply stopping the
1035 DBG3("UST app update failed to find app sock %d",
1039 ust_app_global_update(sess
->ust_session
, app
);
1043 session_unlock(sess
);
1048 * This thread manage event coming from the kernel.
1050 * Features supported in this thread:
1053 static void *thread_manage_kernel(void *data
)
1055 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
1056 uint32_t revents
, nb_fd
;
1058 struct lttng_poll_event events
;
1060 DBG("[thread] Thread manage kernel started");
1062 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
1065 * This first step of the while is to clean this structure which could free
1066 * non NULL pointers so initialize it before the loop.
1068 lttng_poll_init(&events
);
1070 if (testpoint(sessiond_thread_manage_kernel
)) {
1071 goto error_testpoint
;
1074 health_code_update();
1076 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
1077 goto error_testpoint
;
1081 health_code_update();
1083 if (update_poll_flag
== 1) {
1084 /* Clean events object. We are about to populate it again. */
1085 lttng_poll_clean(&events
);
1087 ret
= sessiond_set_thread_pollset(&events
, 2);
1089 goto error_poll_create
;
1092 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1097 /* This will add the available kernel channel if any. */
1098 ret
= update_kernel_poll(&events
);
1102 update_poll_flag
= 0;
1105 DBG("Thread kernel polling");
1107 /* Poll infinite value of time */
1109 health_poll_entry();
1110 ret
= lttng_poll_wait(&events
, -1);
1111 DBG("Thread kernel return from poll on %d fds",
1112 LTTNG_POLL_GETNB(&events
));
1116 * Restart interrupted system call.
1118 if (errno
== EINTR
) {
1122 } else if (ret
== 0) {
1123 /* Should not happen since timeout is infinite */
1124 ERR("Return value of poll is 0 with an infinite timeout.\n"
1125 "This should not have happened! Continuing...");
1131 for (i
= 0; i
< nb_fd
; i
++) {
1132 /* Fetch once the poll data */
1133 revents
= LTTNG_POLL_GETEV(&events
, i
);
1134 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1136 health_code_update();
1139 /* No activity for this FD (poll implementation). */
1143 /* Thread quit pipe has been closed. Killing thread. */
1144 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1150 /* Check for data on kernel pipe */
1151 if (revents
& LPOLLIN
) {
1152 if (pollfd
== kernel_poll_pipe
[0]) {
1153 (void) lttng_read(kernel_poll_pipe
[0],
1156 * Ret value is useless here, if this pipe gets any actions an
1157 * update is required anyway.
1159 update_poll_flag
= 1;
1163 * New CPU detected by the kernel. Adding kernel stream to
1164 * kernel session and updating the kernel consumer
1166 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1172 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1173 update_poll_flag
= 1;
1176 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1184 lttng_poll_clean(&events
);
1187 utils_close_pipe(kernel_poll_pipe
);
1188 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1191 ERR("Health error occurred in %s", __func__
);
1192 WARN("Kernel thread died unexpectedly. "
1193 "Kernel tracing can continue but CPU hotplug is disabled.");
1195 health_unregister(health_sessiond
);
1196 DBG("Kernel thread dying");
1201 * Signal pthread condition of the consumer data that the thread.
1203 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1205 pthread_mutex_lock(&data
->cond_mutex
);
1208 * The state is set before signaling. It can be any value, it's the waiter
1209 * job to correctly interpret this condition variable associated to the
1210 * consumer pthread_cond.
1212 * A value of 0 means that the corresponding thread of the consumer data
1213 * was not started. 1 indicates that the thread has started and is ready
1214 * for action. A negative value means that there was an error during the
1217 data
->consumer_thread_is_ready
= state
;
1218 (void) pthread_cond_signal(&data
->cond
);
1220 pthread_mutex_unlock(&data
->cond_mutex
);
1224 * This thread manage the consumer error sent back to the session daemon.
1226 static void *thread_manage_consumer(void *data
)
1228 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1229 uint32_t revents
, nb_fd
;
1230 enum lttcomm_return_code code
;
1231 struct lttng_poll_event events
;
1232 struct consumer_data
*consumer_data
= data
;
1234 DBG("[thread] Manage consumer started");
1236 rcu_register_thread();
1237 rcu_thread_online();
1239 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1241 health_code_update();
1244 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1245 * metadata_sock. Nothing more will be added to this poll set.
1247 ret
= sessiond_set_thread_pollset(&events
, 3);
1253 * The error socket here is already in a listening state which was done
1254 * just before spawning this thread to avoid a race between the consumer
1255 * daemon exec trying to connect and the listen() call.
1257 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1262 health_code_update();
1264 /* Infinite blocking call, waiting for transmission */
1266 health_poll_entry();
1268 if (testpoint(sessiond_thread_manage_consumer
)) {
1272 ret
= lttng_poll_wait(&events
, -1);
1276 * Restart interrupted system call.
1278 if (errno
== EINTR
) {
1286 for (i
= 0; i
< nb_fd
; i
++) {
1287 /* Fetch once the poll data */
1288 revents
= LTTNG_POLL_GETEV(&events
, i
);
1289 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1291 health_code_update();
1294 /* No activity for this FD (poll implementation). */
1298 /* Thread quit pipe has been closed. Killing thread. */
1299 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1305 /* Event on the registration socket */
1306 if (pollfd
== consumer_data
->err_sock
) {
1307 if (revents
& LPOLLIN
) {
1309 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1310 ERR("consumer err socket poll error");
1313 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1319 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1325 * Set the CLOEXEC flag. Return code is useless because either way, the
1328 (void) utils_set_fd_cloexec(sock
);
1330 health_code_update();
1332 DBG2("Receiving code from consumer err_sock");
1334 /* Getting status code from kconsumerd */
1335 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1336 sizeof(enum lttcomm_return_code
));
1341 health_code_update();
1342 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1343 /* Connect both socket, command and metadata. */
1344 consumer_data
->cmd_sock
=
1345 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1346 consumer_data
->metadata_fd
=
1347 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1348 if (consumer_data
->cmd_sock
< 0
1349 || consumer_data
->metadata_fd
< 0) {
1350 PERROR("consumer connect cmd socket");
1351 /* On error, signal condition and quit. */
1352 signal_consumer_condition(consumer_data
, -1);
1355 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1356 /* Create metadata socket lock. */
1357 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1358 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1359 PERROR("zmalloc pthread mutex");
1363 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1365 signal_consumer_condition(consumer_data
, 1);
1366 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1367 DBG("Consumer metadata socket ready (fd: %d)",
1368 consumer_data
->metadata_fd
);
1370 ERR("consumer error when waiting for SOCK_READY : %s",
1371 lttcomm_get_readable_code(-code
));
1375 /* Remove the consumerd error sock since we've established a connexion */
1376 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1381 /* Add new accepted error socket. */
1382 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1387 /* Add metadata socket that is successfully connected. */
1388 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1389 LPOLLIN
| LPOLLRDHUP
);
1394 health_code_update();
1396 /* Infinite blocking call, waiting for transmission */
1399 health_code_update();
1401 /* Exit the thread because the thread quit pipe has been triggered. */
1403 /* Not a health error. */
1408 health_poll_entry();
1409 ret
= lttng_poll_wait(&events
, -1);
1413 * Restart interrupted system call.
1415 if (errno
== EINTR
) {
1423 for (i
= 0; i
< nb_fd
; i
++) {
1424 /* Fetch once the poll data */
1425 revents
= LTTNG_POLL_GETEV(&events
, i
);
1426 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1428 health_code_update();
1431 /* No activity for this FD (poll implementation). */
1436 * Thread quit pipe has been triggered, flag that we should stop
1437 * but continue the current loop to handle potential data from
1440 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1442 if (pollfd
== sock
) {
1443 /* Event on the consumerd socket */
1444 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1445 && !(revents
& LPOLLIN
)) {
1446 ERR("consumer err socket second poll error");
1449 health_code_update();
1450 /* Wait for any kconsumerd error */
1451 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1452 sizeof(enum lttcomm_return_code
));
1454 ERR("consumer closed the command socket");
1458 ERR("consumer return code : %s",
1459 lttcomm_get_readable_code(-code
));
1462 } else if (pollfd
== consumer_data
->metadata_fd
) {
1463 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1464 && !(revents
& LPOLLIN
)) {
1465 ERR("consumer err metadata socket second poll error");
1468 /* UST metadata requests */
1469 ret
= ust_consumer_metadata_request(
1470 &consumer_data
->metadata_sock
);
1472 ERR("Handling metadata request");
1476 /* No need for an else branch all FDs are tested prior. */
1478 health_code_update();
1484 * We lock here because we are about to close the sockets and some other
1485 * thread might be using them so get exclusive access which will abort all
1486 * other consumer command by other threads.
1488 pthread_mutex_lock(&consumer_data
->lock
);
1490 /* Immediately set the consumerd state to stopped */
1491 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1492 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1493 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1494 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1495 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1497 /* Code flow error... */
1501 if (consumer_data
->err_sock
>= 0) {
1502 ret
= close(consumer_data
->err_sock
);
1506 consumer_data
->err_sock
= -1;
1508 if (consumer_data
->cmd_sock
>= 0) {
1509 ret
= close(consumer_data
->cmd_sock
);
1513 consumer_data
->cmd_sock
= -1;
1515 if (consumer_data
->metadata_sock
.fd_ptr
&&
1516 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1517 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1529 unlink(consumer_data
->err_unix_sock_path
);
1530 unlink(consumer_data
->cmd_unix_sock_path
);
1531 pthread_mutex_unlock(&consumer_data
->lock
);
1533 /* Cleanup metadata socket mutex. */
1534 if (consumer_data
->metadata_sock
.lock
) {
1535 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1536 free(consumer_data
->metadata_sock
.lock
);
1538 lttng_poll_clean(&events
);
1542 ERR("Health error occurred in %s", __func__
);
1544 health_unregister(health_sessiond
);
1545 DBG("consumer thread cleanup completed");
1547 rcu_thread_offline();
1548 rcu_unregister_thread();
1554 * This thread manage application communication.
1556 static void *thread_manage_apps(void *data
)
1558 int i
, ret
, pollfd
, err
= -1;
1560 uint32_t revents
, nb_fd
;
1561 struct lttng_poll_event events
;
1563 DBG("[thread] Manage application started");
1565 rcu_register_thread();
1566 rcu_thread_online();
1568 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1570 if (testpoint(sessiond_thread_manage_apps
)) {
1571 goto error_testpoint
;
1574 health_code_update();
1576 ret
= sessiond_set_thread_pollset(&events
, 2);
1578 goto error_poll_create
;
1581 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1586 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1590 health_code_update();
1593 DBG("Apps thread polling");
1595 /* Inifinite blocking call, waiting for transmission */
1597 health_poll_entry();
1598 ret
= lttng_poll_wait(&events
, -1);
1599 DBG("Apps thread return from poll on %d fds",
1600 LTTNG_POLL_GETNB(&events
));
1604 * Restart interrupted system call.
1606 if (errno
== EINTR
) {
1614 for (i
= 0; i
< nb_fd
; i
++) {
1615 /* Fetch once the poll data */
1616 revents
= LTTNG_POLL_GETEV(&events
, i
);
1617 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1619 health_code_update();
1622 /* No activity for this FD (poll implementation). */
1626 /* Thread quit pipe has been closed. Killing thread. */
1627 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1633 /* Inspect the apps cmd pipe */
1634 if (pollfd
== apps_cmd_pipe
[0]) {
1635 if (revents
& LPOLLIN
) {
1639 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1640 if (size_ret
< sizeof(sock
)) {
1641 PERROR("read apps cmd pipe");
1645 health_code_update();
1648 * Since this is a command socket (write then read),
1649 * we only monitor the error events of the socket.
1651 ret
= lttng_poll_add(&events
, sock
,
1652 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1657 DBG("Apps with sock %d added to poll set", sock
);
1658 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1659 ERR("Apps command pipe error");
1662 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
1667 * At this point, we know that a registered application made
1668 * the event at poll_wait.
1670 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1671 /* Removing from the poll set */
1672 ret
= lttng_poll_del(&events
, pollfd
);
1677 /* Socket closed on remote end. */
1678 ust_app_unregister(pollfd
);
1680 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1685 health_code_update();
1691 lttng_poll_clean(&events
);
1694 utils_close_pipe(apps_cmd_pipe
);
1695 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1698 * We don't clean the UST app hash table here since already registered
1699 * applications can still be controlled so let them be until the session
1700 * daemon dies or the applications stop.
1705 ERR("Health error occurred in %s", __func__
);
1707 health_unregister(health_sessiond
);
1708 DBG("Application communication apps thread cleanup complete");
1709 rcu_thread_offline();
1710 rcu_unregister_thread();
1715 * Send a socket to a thread This is called from the dispatch UST registration
1716 * thread once all sockets are set for the application.
1718 * The sock value can be invalid, we don't really care, the thread will handle
1719 * it and make the necessary cleanup if so.
1721 * On success, return 0 else a negative value being the errno message of the
1724 static int send_socket_to_thread(int fd
, int sock
)
1729 * It's possible that the FD is set as invalid with -1 concurrently just
1730 * before calling this function being a shutdown state of the thread.
1737 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1738 if (ret
< sizeof(sock
)) {
1739 PERROR("write apps pipe %d", fd
);
1746 /* All good. Don't send back the write positive ret value. */
1753 * Sanitize the wait queue of the dispatch registration thread meaning removing
1754 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1755 * notify socket is never received.
1757 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1759 int ret
, nb_fd
= 0, i
;
1760 unsigned int fd_added
= 0;
1761 struct lttng_poll_event events
;
1762 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1766 lttng_poll_init(&events
);
1768 /* Just skip everything for an empty queue. */
1769 if (!wait_queue
->count
) {
1773 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1778 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1779 &wait_queue
->head
, head
) {
1780 assert(wait_node
->app
);
1781 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1782 LPOLLHUP
| LPOLLERR
);
1795 * Poll but don't block so we can quickly identify the faulty events and
1796 * clean them afterwards from the wait queue.
1798 ret
= lttng_poll_wait(&events
, 0);
1804 for (i
= 0; i
< nb_fd
; i
++) {
1805 /* Get faulty FD. */
1806 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1807 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1810 /* No activity for this FD (poll implementation). */
1814 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1815 &wait_queue
->head
, head
) {
1816 if (pollfd
== wait_node
->app
->sock
&&
1817 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1818 cds_list_del(&wait_node
->head
);
1819 wait_queue
->count
--;
1820 ust_app_destroy(wait_node
->app
);
1823 * Silence warning of use-after-free in
1824 * cds_list_for_each_entry_safe which uses
1825 * __typeof__(*wait_node).
1830 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1837 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1841 lttng_poll_clean(&events
);
1845 lttng_poll_clean(&events
);
1847 ERR("Unable to sanitize wait queue");
1852 * Dispatch request from the registration threads to the application
1853 * communication thread.
1855 static void *thread_dispatch_ust_registration(void *data
)
1858 struct cds_wfcq_node
*node
;
1859 struct ust_command
*ust_cmd
= NULL
;
1860 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1861 struct ust_reg_wait_queue wait_queue
= {
1865 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1867 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1868 goto error_testpoint
;
1871 health_code_update();
1873 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1875 DBG("[thread] Dispatch UST command started");
1877 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1878 health_code_update();
1880 /* Atomically prepare the queue futex */
1881 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1884 struct ust_app
*app
= NULL
;
1888 * Make sure we don't have node(s) that have hung up before receiving
1889 * the notify socket. This is to clean the list in order to avoid
1890 * memory leaks from notify socket that are never seen.
1892 sanitize_wait_queue(&wait_queue
);
1894 health_code_update();
1895 /* Dequeue command for registration */
1896 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1898 DBG("Woken up but nothing in the UST command queue");
1899 /* Continue thread execution */
1903 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1905 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1906 " gid:%d sock:%d name:%s (version %d.%d)",
1907 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1908 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1909 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1910 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1912 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1913 wait_node
= zmalloc(sizeof(*wait_node
));
1915 PERROR("zmalloc wait_node dispatch");
1916 ret
= close(ust_cmd
->sock
);
1918 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1920 lttng_fd_put(LTTNG_FD_APPS
, 1);
1924 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1926 /* Create application object if socket is CMD. */
1927 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1929 if (!wait_node
->app
) {
1930 ret
= close(ust_cmd
->sock
);
1932 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1934 lttng_fd_put(LTTNG_FD_APPS
, 1);
1940 * Add application to the wait queue so we can set the notify
1941 * socket before putting this object in the global ht.
1943 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1948 * We have to continue here since we don't have the notify
1949 * socket and the application MUST be added to the hash table
1950 * only at that moment.
1955 * Look for the application in the local wait queue and set the
1956 * notify socket if found.
1958 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1959 &wait_queue
.head
, head
) {
1960 health_code_update();
1961 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1962 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1963 cds_list_del(&wait_node
->head
);
1965 app
= wait_node
->app
;
1967 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1973 * With no application at this stage the received socket is
1974 * basically useless so close it before we free the cmd data
1975 * structure for good.
1978 ret
= close(ust_cmd
->sock
);
1980 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1982 lttng_fd_put(LTTNG_FD_APPS
, 1);
1989 * @session_lock_list
1991 * Lock the global session list so from the register up to the
1992 * registration done message, no thread can see the application
1993 * and change its state.
1995 session_lock_list();
1999 * Add application to the global hash table. This needs to be
2000 * done before the update to the UST registry can locate the
2005 /* Set app version. This call will print an error if needed. */
2006 (void) ust_app_version(app
);
2008 /* Send notify socket through the notify pipe. */
2009 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
2013 session_unlock_list();
2015 * No notify thread, stop the UST tracing. However, this is
2016 * not an internal error of the this thread thus setting
2017 * the health error code to a normal exit.
2024 * Update newly registered application with the tracing
2025 * registry info already enabled information.
2027 update_ust_app(app
->sock
);
2030 * Don't care about return value. Let the manage apps threads
2031 * handle app unregistration upon socket close.
2033 (void) ust_app_register_done(app
);
2036 * Even if the application socket has been closed, send the app
2037 * to the thread and unregistration will take place at that
2040 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
2043 session_unlock_list();
2045 * No apps. thread, stop the UST tracing. However, this is
2046 * not an internal error of the this thread thus setting
2047 * the health error code to a normal exit.
2054 session_unlock_list();
2056 } while (node
!= NULL
);
2058 health_poll_entry();
2059 /* Futex wait on queue. Blocking call on futex() */
2060 futex_nto1_wait(&ust_cmd_queue
.futex
);
2063 /* Normal exit, no error */
2067 /* Clean up wait queue. */
2068 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2069 &wait_queue
.head
, head
) {
2070 cds_list_del(&wait_node
->head
);
2075 /* Empty command queue. */
2077 /* Dequeue command for registration */
2078 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2082 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2083 ret
= close(ust_cmd
->sock
);
2085 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2087 lttng_fd_put(LTTNG_FD_APPS
, 1);
2092 DBG("Dispatch thread dying");
2095 ERR("Health error occurred in %s", __func__
);
2097 health_unregister(health_sessiond
);
2102 * This thread manage application registration.
2104 static void *thread_registration_apps(void *data
)
2106 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2107 uint32_t revents
, nb_fd
;
2108 struct lttng_poll_event events
;
2110 * Get allocated in this thread, enqueued to a global queue, dequeued and
2111 * freed in the manage apps thread.
2113 struct ust_command
*ust_cmd
= NULL
;
2115 DBG("[thread] Manage application registration started");
2117 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2119 if (testpoint(sessiond_thread_registration_apps
)) {
2120 goto error_testpoint
;
2123 ret
= lttcomm_listen_unix_sock(apps_sock
);
2129 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2130 * more will be added to this poll set.
2132 ret
= sessiond_set_thread_pollset(&events
, 2);
2134 goto error_create_poll
;
2137 /* Add the application registration socket */
2138 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2140 goto error_poll_add
;
2143 /* Notify all applications to register */
2144 ret
= notify_ust_apps(1);
2146 ERR("Failed to notify applications or create the wait shared memory.\n"
2147 "Execution continues but there might be problem for already\n"
2148 "running applications that wishes to register.");
2152 DBG("Accepting application registration");
2154 /* Inifinite blocking call, waiting for transmission */
2156 health_poll_entry();
2157 ret
= lttng_poll_wait(&events
, -1);
2161 * Restart interrupted system call.
2163 if (errno
== EINTR
) {
2171 for (i
= 0; i
< nb_fd
; i
++) {
2172 health_code_update();
2174 /* Fetch once the poll data */
2175 revents
= LTTNG_POLL_GETEV(&events
, i
);
2176 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2179 /* No activity for this FD (poll implementation). */
2183 /* Thread quit pipe has been closed. Killing thread. */
2184 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2190 /* Event on the registration socket */
2191 if (pollfd
== apps_sock
) {
2192 if (revents
& LPOLLIN
) {
2193 sock
= lttcomm_accept_unix_sock(apps_sock
);
2199 * Set socket timeout for both receiving and ending.
2200 * app_socket_timeout is in seconds, whereas
2201 * lttcomm_setsockopt_rcv_timeout and
2202 * lttcomm_setsockopt_snd_timeout expect msec as
2205 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2206 app_socket_timeout
* 1000);
2207 (void) lttcomm_setsockopt_snd_timeout(sock
,
2208 app_socket_timeout
* 1000);
2211 * Set the CLOEXEC flag. Return code is useless because
2212 * either way, the show must go on.
2214 (void) utils_set_fd_cloexec(sock
);
2216 /* Create UST registration command for enqueuing */
2217 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2218 if (ust_cmd
== NULL
) {
2219 PERROR("ust command zmalloc");
2228 * Using message-based transmissions to ensure we don't
2229 * have to deal with partially received messages.
2231 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2233 ERR("Exhausted file descriptors allowed for applications.");
2243 health_code_update();
2244 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2247 /* Close socket of the application. */
2252 lttng_fd_put(LTTNG_FD_APPS
, 1);
2256 health_code_update();
2258 ust_cmd
->sock
= sock
;
2261 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2262 " gid:%d sock:%d name:%s (version %d.%d)",
2263 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2264 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2265 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2266 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2269 * Lock free enqueue the registration request. The red pill
2270 * has been taken! This apps will be part of the *system*.
2272 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2275 * Wake the registration queue futex. Implicit memory
2276 * barrier with the exchange in cds_wfcq_enqueue.
2278 futex_nto1_wake(&ust_cmd_queue
.futex
);
2279 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2280 ERR("Register apps socket poll error");
2283 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2292 /* Notify that the registration thread is gone */
2295 if (apps_sock
>= 0) {
2296 ret
= close(apps_sock
);
2306 lttng_fd_put(LTTNG_FD_APPS
, 1);
2308 unlink(apps_unix_sock_path
);
2311 lttng_poll_clean(&events
);
2315 DBG("UST Registration thread cleanup complete");
2318 ERR("Health error occurred in %s", __func__
);
2320 health_unregister(health_sessiond
);
2326 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2327 * exec or it will fails.
2329 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2332 struct timespec timeout
;
2335 * Make sure we set the readiness flag to 0 because we are NOT ready.
2336 * This access to consumer_thread_is_ready does not need to be
2337 * protected by consumer_data.cond_mutex (yet) since the consumer
2338 * management thread has not been started at this point.
2340 consumer_data
->consumer_thread_is_ready
= 0;
2342 /* Setup pthread condition */
2343 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2346 PERROR("pthread_condattr_init consumer data");
2351 * Set the monotonic clock in order to make sure we DO NOT jump in time
2352 * between the clock_gettime() call and the timedwait call. See bug #324
2353 * for a more details and how we noticed it.
2355 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2358 PERROR("pthread_condattr_setclock consumer data");
2362 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2365 PERROR("pthread_cond_init consumer data");
2369 ret
= pthread_create(&consumer_data
->thread
, default_pthread_attr(),
2370 thread_manage_consumer
, consumer_data
);
2373 PERROR("pthread_create consumer");
2378 /* We are about to wait on a pthread condition */
2379 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2381 /* Get time for sem_timedwait absolute timeout */
2382 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2384 * Set the timeout for the condition timed wait even if the clock gettime
2385 * call fails since we might loop on that call and we want to avoid to
2386 * increment the timeout too many times.
2388 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2391 * The following loop COULD be skipped in some conditions so this is why we
2392 * set ret to 0 in order to make sure at least one round of the loop is
2398 * Loop until the condition is reached or when a timeout is reached. Note
2399 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2400 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2401 * possible. This loop does not take any chances and works with both of
2404 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2405 if (clock_ret
< 0) {
2406 PERROR("clock_gettime spawn consumer");
2407 /* Infinite wait for the consumerd thread to be ready */
2408 ret
= pthread_cond_wait(&consumer_data
->cond
,
2409 &consumer_data
->cond_mutex
);
2411 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2412 &consumer_data
->cond_mutex
, &timeout
);
2416 /* Release the pthread condition */
2417 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2421 if (ret
== ETIMEDOUT
) {
2425 * Call has timed out so we kill the kconsumerd_thread and return
2428 ERR("Condition timed out. The consumer thread was never ready."
2430 pth_ret
= pthread_cancel(consumer_data
->thread
);
2432 PERROR("pthread_cancel consumer thread");
2435 PERROR("pthread_cond_wait failed consumer thread");
2437 /* Caller is expecting a negative value on failure. */
2442 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2443 if (consumer_data
->pid
== 0) {
2444 ERR("Consumerd did not start");
2445 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2448 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2457 * Join consumer thread
2459 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2463 /* Consumer pid must be a real one. */
2464 if (consumer_data
->pid
> 0) {
2466 ret
= kill(consumer_data
->pid
, SIGTERM
);
2468 PERROR("Error killing consumer daemon");
2471 return pthread_join(consumer_data
->thread
, &status
);
2478 * Fork and exec a consumer daemon (consumerd).
2480 * Return pid if successful else -1.
2482 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2486 const char *consumer_to_use
;
2487 const char *verbosity
;
2490 DBG("Spawning consumerd");
2497 if (opt_verbose_consumer
) {
2498 verbosity
= "--verbose";
2499 } else if (lttng_opt_quiet
) {
2500 verbosity
= "--quiet";
2505 switch (consumer_data
->type
) {
2506 case LTTNG_CONSUMER_KERNEL
:
2508 * Find out which consumerd to execute. We will first try the
2509 * 64-bit path, then the sessiond's installation directory, and
2510 * fallback on the 32-bit one,
2512 DBG3("Looking for a kernel consumer at these locations:");
2513 DBG3(" 1) %s", consumerd64_bin
);
2514 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2515 DBG3(" 3) %s", consumerd32_bin
);
2516 if (stat(consumerd64_bin
, &st
) == 0) {
2517 DBG3("Found location #1");
2518 consumer_to_use
= consumerd64_bin
;
2519 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2520 DBG3("Found location #2");
2521 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2522 } else if (stat(consumerd32_bin
, &st
) == 0) {
2523 DBG3("Found location #3");
2524 consumer_to_use
= consumerd32_bin
;
2526 DBG("Could not find any valid consumerd executable");
2530 DBG("Using kernel consumer at: %s", consumer_to_use
);
2531 ret
= execl(consumer_to_use
,
2532 "lttng-consumerd", verbosity
, "-k",
2533 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2534 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2535 "--group", tracing_group_name
,
2538 case LTTNG_CONSUMER64_UST
:
2540 char *tmpnew
= NULL
;
2542 if (consumerd64_libdir
[0] != '\0') {
2546 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2550 tmplen
= strlen("LD_LIBRARY_PATH=")
2551 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2552 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2557 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2558 strcat(tmpnew
, consumerd64_libdir
);
2559 if (tmp
[0] != '\0') {
2560 strcat(tmpnew
, ":");
2561 strcat(tmpnew
, tmp
);
2563 ret
= putenv(tmpnew
);
2570 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2571 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2572 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2573 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2574 "--group", tracing_group_name
,
2576 if (consumerd64_libdir
[0] != '\0') {
2581 case LTTNG_CONSUMER32_UST
:
2583 char *tmpnew
= NULL
;
2585 if (consumerd32_libdir
[0] != '\0') {
2589 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2593 tmplen
= strlen("LD_LIBRARY_PATH=")
2594 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2595 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2600 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2601 strcat(tmpnew
, consumerd32_libdir
);
2602 if (tmp
[0] != '\0') {
2603 strcat(tmpnew
, ":");
2604 strcat(tmpnew
, tmp
);
2606 ret
= putenv(tmpnew
);
2613 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2614 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2615 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2616 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2617 "--group", tracing_group_name
,
2619 if (consumerd32_libdir
[0] != '\0') {
2625 PERROR("unknown consumer type");
2629 PERROR("Consumer execl()");
2631 /* Reaching this point, we got a failure on our execl(). */
2633 } else if (pid
> 0) {
2636 PERROR("start consumer fork");
2644 * Spawn the consumerd daemon and session daemon thread.
2646 static int start_consumerd(struct consumer_data
*consumer_data
)
2651 * Set the listen() state on the socket since there is a possible race
2652 * between the exec() of the consumer daemon and this call if place in the
2653 * consumer thread. See bug #366 for more details.
2655 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2660 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2661 if (consumer_data
->pid
!= 0) {
2662 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2666 ret
= spawn_consumerd(consumer_data
);
2668 ERR("Spawning consumerd failed");
2669 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2673 /* Setting up the consumer_data pid */
2674 consumer_data
->pid
= ret
;
2675 DBG2("Consumer pid %d", consumer_data
->pid
);
2676 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2678 DBG2("Spawning consumer control thread");
2679 ret
= spawn_consumer_thread(consumer_data
);
2681 ERR("Fatal error spawning consumer control thread");
2689 /* Cleanup already created sockets on error. */
2690 if (consumer_data
->err_sock
>= 0) {
2693 err
= close(consumer_data
->err_sock
);
2695 PERROR("close consumer data error socket");
2702 * Setup necessary data for kernel tracer action.
2704 static int init_kernel_tracer(void)
2708 /* Modprobe lttng kernel modules */
2709 ret
= modprobe_lttng_control();
2714 /* Open debugfs lttng */
2715 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2716 if (kernel_tracer_fd
< 0) {
2717 DBG("Failed to open %s", module_proc_lttng
);
2722 /* Validate kernel version */
2723 ret
= kernel_validate_version(kernel_tracer_fd
);
2728 ret
= modprobe_lttng_data();
2733 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2737 modprobe_remove_lttng_control();
2738 ret
= close(kernel_tracer_fd
);
2742 kernel_tracer_fd
= -1;
2743 return LTTNG_ERR_KERN_VERSION
;
2746 ret
= close(kernel_tracer_fd
);
2752 modprobe_remove_lttng_control();
2755 WARN("No kernel tracer available");
2756 kernel_tracer_fd
= -1;
2758 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2760 return LTTNG_ERR_KERN_NA
;
2766 * Copy consumer output from the tracing session to the domain session. The
2767 * function also applies the right modification on a per domain basis for the
2768 * trace files destination directory.
2770 * Should *NOT* be called with RCU read-side lock held.
2772 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2775 const char *dir_name
;
2776 struct consumer_output
*consumer
;
2779 assert(session
->consumer
);
2782 case LTTNG_DOMAIN_KERNEL
:
2783 DBG3("Copying tracing session consumer output in kernel session");
2785 * XXX: We should audit the session creation and what this function
2786 * does "extra" in order to avoid a destroy since this function is used
2787 * in the domain session creation (kernel and ust) only. Same for UST
2790 if (session
->kernel_session
->consumer
) {
2791 consumer_output_put(session
->kernel_session
->consumer
);
2793 session
->kernel_session
->consumer
=
2794 consumer_copy_output(session
->consumer
);
2795 /* Ease our life a bit for the next part */
2796 consumer
= session
->kernel_session
->consumer
;
2797 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2799 case LTTNG_DOMAIN_JUL
:
2800 case LTTNG_DOMAIN_LOG4J
:
2801 case LTTNG_DOMAIN_PYTHON
:
2802 case LTTNG_DOMAIN_UST
:
2803 DBG3("Copying tracing session consumer output in UST session");
2804 if (session
->ust_session
->consumer
) {
2805 consumer_output_put(session
->ust_session
->consumer
);
2807 session
->ust_session
->consumer
=
2808 consumer_copy_output(session
->consumer
);
2809 /* Ease our life a bit for the next part */
2810 consumer
= session
->ust_session
->consumer
;
2811 dir_name
= DEFAULT_UST_TRACE_DIR
;
2814 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2818 /* Append correct directory to subdir */
2819 strncat(consumer
->subdir
, dir_name
,
2820 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2821 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2830 * Create an UST session and add it to the session ust list.
2832 * Should *NOT* be called with RCU read-side lock held.
2834 static int create_ust_session(struct ltt_session
*session
,
2835 struct lttng_domain
*domain
)
2838 struct ltt_ust_session
*lus
= NULL
;
2842 assert(session
->consumer
);
2844 switch (domain
->type
) {
2845 case LTTNG_DOMAIN_JUL
:
2846 case LTTNG_DOMAIN_LOG4J
:
2847 case LTTNG_DOMAIN_PYTHON
:
2848 case LTTNG_DOMAIN_UST
:
2851 ERR("Unknown UST domain on create session %d", domain
->type
);
2852 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2856 DBG("Creating UST session");
2858 lus
= trace_ust_create_session(session
->id
);
2860 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2864 lus
->uid
= session
->uid
;
2865 lus
->gid
= session
->gid
;
2866 lus
->output_traces
= session
->output_traces
;
2867 lus
->snapshot_mode
= session
->snapshot_mode
;
2868 lus
->live_timer_interval
= session
->live_timer
;
2869 session
->ust_session
= lus
;
2870 if (session
->shm_path
[0]) {
2871 strncpy(lus
->root_shm_path
, session
->shm_path
,
2872 sizeof(lus
->root_shm_path
));
2873 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2874 strncpy(lus
->shm_path
, session
->shm_path
,
2875 sizeof(lus
->shm_path
));
2876 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2877 strncat(lus
->shm_path
, "/ust",
2878 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2880 /* Copy session output to the newly created UST session */
2881 ret
= copy_session_consumer(domain
->type
, session
);
2882 if (ret
!= LTTNG_OK
) {
2890 session
->ust_session
= NULL
;
2895 * Create a kernel tracer session then create the default channel.
2897 static int create_kernel_session(struct ltt_session
*session
)
2901 DBG("Creating kernel session");
2903 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2905 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2909 /* Code flow safety */
2910 assert(session
->kernel_session
);
2912 /* Copy session output to the newly created Kernel session */
2913 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2914 if (ret
!= LTTNG_OK
) {
2918 /* Create directory(ies) on local filesystem. */
2919 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2920 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2921 ret
= run_as_mkdir_recursive(
2922 session
->kernel_session
->consumer
->dst
.trace_path
,
2923 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2925 if (errno
!= EEXIST
) {
2926 ERR("Trace directory creation error");
2932 session
->kernel_session
->uid
= session
->uid
;
2933 session
->kernel_session
->gid
= session
->gid
;
2934 session
->kernel_session
->output_traces
= session
->output_traces
;
2935 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2940 trace_kernel_destroy_session(session
->kernel_session
);
2941 session
->kernel_session
= NULL
;
2946 * Count number of session permitted by uid/gid.
2948 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2951 struct ltt_session
*session
;
2953 DBG("Counting number of available session for UID %d GID %d",
2955 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2957 * Only list the sessions the user can control.
2959 if (!session_access_ok(session
, uid
, gid
)) {
2968 * Process the command requested by the lttng client within the command
2969 * context structure. This function make sure that the return structure (llm)
2970 * is set and ready for transmission before returning.
2972 * Return any error encountered or 0 for success.
2974 * "sock" is only used for special-case var. len data.
2976 * Should *NOT* be called with RCU read-side lock held.
2978 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2982 int need_tracing_session
= 1;
2985 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2987 assert(!rcu_read_ongoing());
2991 switch (cmd_ctx
->lsm
->cmd_type
) {
2992 case LTTNG_CREATE_SESSION
:
2993 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2994 case LTTNG_CREATE_SESSION_LIVE
:
2995 case LTTNG_DESTROY_SESSION
:
2996 case LTTNG_LIST_SESSIONS
:
2997 case LTTNG_LIST_DOMAINS
:
2998 case LTTNG_START_TRACE
:
2999 case LTTNG_STOP_TRACE
:
3000 case LTTNG_DATA_PENDING
:
3001 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3002 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3003 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3004 case LTTNG_SNAPSHOT_RECORD
:
3005 case LTTNG_SAVE_SESSION
:
3006 case LTTNG_SET_SESSION_SHM_PATH
:
3007 case LTTNG_REGENERATE_METADATA
:
3008 case LTTNG_REGENERATE_STATEDUMP
:
3015 if (opt_no_kernel
&& need_domain
3016 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
3018 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3020 ret
= LTTNG_ERR_KERN_NA
;
3025 /* Deny register consumer if we already have a spawned consumer. */
3026 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
3027 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3028 if (kconsumer_data
.pid
> 0) {
3029 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3030 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3033 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3037 * Check for command that don't needs to allocate a returned payload. We do
3038 * this here so we don't have to make the call for no payload at each
3041 switch(cmd_ctx
->lsm
->cmd_type
) {
3042 case LTTNG_LIST_SESSIONS
:
3043 case LTTNG_LIST_TRACEPOINTS
:
3044 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3045 case LTTNG_LIST_DOMAINS
:
3046 case LTTNG_LIST_CHANNELS
:
3047 case LTTNG_LIST_EVENTS
:
3048 case LTTNG_LIST_SYSCALLS
:
3049 case LTTNG_LIST_TRACKER_PIDS
:
3050 case LTTNG_DATA_PENDING
:
3053 /* Setup lttng message with no payload */
3054 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0);
3056 /* This label does not try to unlock the session */
3057 goto init_setup_error
;
3061 /* Commands that DO NOT need a session. */
3062 switch (cmd_ctx
->lsm
->cmd_type
) {
3063 case LTTNG_CREATE_SESSION
:
3064 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3065 case LTTNG_CREATE_SESSION_LIVE
:
3066 case LTTNG_CALIBRATE
:
3067 case LTTNG_LIST_SESSIONS
:
3068 case LTTNG_LIST_TRACEPOINTS
:
3069 case LTTNG_LIST_SYSCALLS
:
3070 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3071 case LTTNG_SAVE_SESSION
:
3072 need_tracing_session
= 0;
3075 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3077 * We keep the session list lock across _all_ commands
3078 * for now, because the per-session lock does not
3079 * handle teardown properly.
3081 session_lock_list();
3082 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3083 if (cmd_ctx
->session
== NULL
) {
3084 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3087 /* Acquire lock for the session */
3088 session_lock(cmd_ctx
->session
);
3094 * Commands that need a valid session but should NOT create one if none
3095 * exists. Instead of creating one and destroying it when the command is
3096 * handled, process that right before so we save some round trip in useless
3099 switch (cmd_ctx
->lsm
->cmd_type
) {
3100 case LTTNG_DISABLE_CHANNEL
:
3101 case LTTNG_DISABLE_EVENT
:
3102 switch (cmd_ctx
->lsm
->domain
.type
) {
3103 case LTTNG_DOMAIN_KERNEL
:
3104 if (!cmd_ctx
->session
->kernel_session
) {
3105 ret
= LTTNG_ERR_NO_CHANNEL
;
3109 case LTTNG_DOMAIN_JUL
:
3110 case LTTNG_DOMAIN_LOG4J
:
3111 case LTTNG_DOMAIN_PYTHON
:
3112 case LTTNG_DOMAIN_UST
:
3113 if (!cmd_ctx
->session
->ust_session
) {
3114 ret
= LTTNG_ERR_NO_CHANNEL
;
3119 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3131 * Check domain type for specific "pre-action".
3133 switch (cmd_ctx
->lsm
->domain
.type
) {
3134 case LTTNG_DOMAIN_KERNEL
:
3136 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3140 /* Kernel tracer check */
3141 if (kernel_tracer_fd
== -1) {
3142 /* Basically, load kernel tracer modules */
3143 ret
= init_kernel_tracer();
3149 /* Consumer is in an ERROR state. Report back to client */
3150 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3151 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3155 /* Need a session for kernel command */
3156 if (need_tracing_session
) {
3157 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3158 ret
= create_kernel_session(cmd_ctx
->session
);
3160 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3165 /* Start the kernel consumer daemon */
3166 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3167 if (kconsumer_data
.pid
== 0 &&
3168 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3169 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3170 ret
= start_consumerd(&kconsumer_data
);
3172 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3175 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3177 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3181 * The consumer was just spawned so we need to add the socket to
3182 * the consumer output of the session if exist.
3184 ret
= consumer_create_socket(&kconsumer_data
,
3185 cmd_ctx
->session
->kernel_session
->consumer
);
3192 case LTTNG_DOMAIN_JUL
:
3193 case LTTNG_DOMAIN_LOG4J
:
3194 case LTTNG_DOMAIN_PYTHON
:
3195 case LTTNG_DOMAIN_UST
:
3197 if (!ust_app_supported()) {
3198 ret
= LTTNG_ERR_NO_UST
;
3201 /* Consumer is in an ERROR state. Report back to client */
3202 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3203 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3207 if (need_tracing_session
) {
3208 /* Create UST session if none exist. */
3209 if (cmd_ctx
->session
->ust_session
== NULL
) {
3210 ret
= create_ust_session(cmd_ctx
->session
,
3211 &cmd_ctx
->lsm
->domain
);
3212 if (ret
!= LTTNG_OK
) {
3217 /* Start the UST consumer daemons */
3219 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3220 if (consumerd64_bin
[0] != '\0' &&
3221 ustconsumer64_data
.pid
== 0 &&
3222 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3223 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3224 ret
= start_consumerd(&ustconsumer64_data
);
3226 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3227 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3231 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3232 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3234 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3238 * Setup socket for consumer 64 bit. No need for atomic access
3239 * since it was set above and can ONLY be set in this thread.
3241 ret
= consumer_create_socket(&ustconsumer64_data
,
3242 cmd_ctx
->session
->ust_session
->consumer
);
3248 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3249 if (consumerd32_bin
[0] != '\0' &&
3250 ustconsumer32_data
.pid
== 0 &&
3251 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3252 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3253 ret
= start_consumerd(&ustconsumer32_data
);
3255 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3256 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3260 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3261 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3263 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3267 * Setup socket for consumer 64 bit. No need for atomic access
3268 * since it was set above and can ONLY be set in this thread.
3270 ret
= consumer_create_socket(&ustconsumer32_data
,
3271 cmd_ctx
->session
->ust_session
->consumer
);
3283 /* Validate consumer daemon state when start/stop trace command */
3284 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3285 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3286 switch (cmd_ctx
->lsm
->domain
.type
) {
3287 case LTTNG_DOMAIN_NONE
:
3289 case LTTNG_DOMAIN_JUL
:
3290 case LTTNG_DOMAIN_LOG4J
:
3291 case LTTNG_DOMAIN_PYTHON
:
3292 case LTTNG_DOMAIN_UST
:
3293 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3294 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3298 case LTTNG_DOMAIN_KERNEL
:
3299 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3300 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3305 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3311 * Check that the UID or GID match that of the tracing session.
3312 * The root user can interact with all sessions.
3314 if (need_tracing_session
) {
3315 if (!session_access_ok(cmd_ctx
->session
,
3316 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3317 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3318 ret
= LTTNG_ERR_EPERM
;
3324 * Send relayd information to consumer as soon as we have a domain and a
3327 if (cmd_ctx
->session
&& need_domain
) {
3329 * Setup relayd if not done yet. If the relayd information was already
3330 * sent to the consumer, this call will gracefully return.
3332 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3333 if (ret
!= LTTNG_OK
) {
3338 /* Process by command type */
3339 switch (cmd_ctx
->lsm
->cmd_type
) {
3340 case LTTNG_ADD_CONTEXT
:
3343 * An LTTNG_ADD_CONTEXT command might have a supplementary
3344 * payload if the context being added is an application context.
3346 if (cmd_ctx
->lsm
->u
.context
.ctx
.ctx
==
3347 LTTNG_EVENT_CONTEXT_APP_CONTEXT
) {
3348 char *provider_name
= NULL
, *context_name
= NULL
;
3349 size_t provider_name_len
=
3350 cmd_ctx
->lsm
->u
.context
.provider_name_len
;
3351 size_t context_name_len
=
3352 cmd_ctx
->lsm
->u
.context
.context_name_len
;
3354 if (provider_name_len
== 0 || context_name_len
== 0) {
3356 * Application provider and context names MUST
3359 ret
= -LTTNG_ERR_INVALID
;
3363 provider_name
= zmalloc(provider_name_len
+ 1);
3364 if (!provider_name
) {
3365 ret
= -LTTNG_ERR_NOMEM
;
3368 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
=
3371 context_name
= zmalloc(context_name_len
+ 1);
3372 if (!context_name
) {
3373 ret
= -LTTNG_ERR_NOMEM
;
3374 goto error_add_context
;
3376 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
=
3379 ret
= lttcomm_recv_unix_sock(sock
, provider_name
,
3382 goto error_add_context
;
3385 ret
= lttcomm_recv_unix_sock(sock
, context_name
,
3388 goto error_add_context
;
3393 * cmd_add_context assumes ownership of the provider and context
3396 ret
= cmd_add_context(cmd_ctx
->session
,
3397 cmd_ctx
->lsm
->domain
.type
,
3398 cmd_ctx
->lsm
->u
.context
.channel_name
,
3399 &cmd_ctx
->lsm
->u
.context
.ctx
,
3400 kernel_poll_pipe
[1]);
3402 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
= NULL
;
3403 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
= NULL
;
3405 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
);
3406 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
);
3412 case LTTNG_DISABLE_CHANNEL
:
3414 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3415 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3418 case LTTNG_DISABLE_EVENT
:
3422 * FIXME: handle filter; for now we just receive the filter's
3423 * bytecode along with the filter expression which are sent by
3424 * liblttng-ctl and discard them.
3426 * This fixes an issue where the client may block while sending
3427 * the filter payload and encounter an error because the session
3428 * daemon closes the socket without ever handling this data.
3430 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3431 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3434 char data
[LTTNG_FILTER_MAX_LEN
];
3436 DBG("Discarding disable event command payload of size %zu", count
);
3438 ret
= lttcomm_recv_unix_sock(sock
, data
,
3439 count
> sizeof(data
) ? sizeof(data
) : count
);
3444 count
-= (size_t) ret
;
3447 /* FIXME: passing packed structure to non-packed pointer */
3448 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3449 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3450 &cmd_ctx
->lsm
->u
.disable
.event
);
3453 case LTTNG_ENABLE_CHANNEL
:
3455 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3456 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3459 case LTTNG_TRACK_PID
:
3461 ret
= cmd_track_pid(cmd_ctx
->session
,
3462 cmd_ctx
->lsm
->domain
.type
,
3463 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3466 case LTTNG_UNTRACK_PID
:
3468 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3469 cmd_ctx
->lsm
->domain
.type
,
3470 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3473 case LTTNG_ENABLE_EVENT
:
3475 struct lttng_event_exclusion
*exclusion
= NULL
;
3476 struct lttng_filter_bytecode
*bytecode
= NULL
;
3477 char *filter_expression
= NULL
;
3479 /* Handle exclusion events and receive it from the client. */
3480 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3481 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3483 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3484 (count
* LTTNG_SYMBOL_NAME_LEN
));
3486 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3490 DBG("Receiving var len exclusion event list from client ...");
3491 exclusion
->count
= count
;
3492 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3493 count
* LTTNG_SYMBOL_NAME_LEN
);
3495 DBG("Nothing recv() from client var len data... continuing");
3498 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3503 /* Get filter expression from client. */
3504 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3505 size_t expression_len
=
3506 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3508 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3509 ret
= LTTNG_ERR_FILTER_INVAL
;
3514 filter_expression
= zmalloc(expression_len
);
3515 if (!filter_expression
) {
3517 ret
= LTTNG_ERR_FILTER_NOMEM
;
3521 /* Receive var. len. data */
3522 DBG("Receiving var len filter's expression from client ...");
3523 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3526 DBG("Nothing recv() from client car len data... continuing");
3528 free(filter_expression
);
3530 ret
= LTTNG_ERR_FILTER_INVAL
;
3535 /* Handle filter and get bytecode from client. */
3536 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3537 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3539 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3540 ret
= LTTNG_ERR_FILTER_INVAL
;
3541 free(filter_expression
);
3546 bytecode
= zmalloc(bytecode_len
);
3548 free(filter_expression
);
3550 ret
= LTTNG_ERR_FILTER_NOMEM
;
3554 /* Receive var. len. data */
3555 DBG("Receiving var len filter's bytecode from client ...");
3556 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3558 DBG("Nothing recv() from client car len data... continuing");
3560 free(filter_expression
);
3563 ret
= LTTNG_ERR_FILTER_INVAL
;
3567 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3568 free(filter_expression
);
3571 ret
= LTTNG_ERR_FILTER_INVAL
;
3576 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3577 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3578 &cmd_ctx
->lsm
->u
.enable
.event
,
3579 filter_expression
, bytecode
, exclusion
,
3580 kernel_poll_pipe
[1]);
3583 case LTTNG_LIST_TRACEPOINTS
:
3585 struct lttng_event
*events
;
3588 session_lock_list();
3589 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3590 session_unlock_list();
3591 if (nb_events
< 0) {
3592 /* Return value is a negative lttng_error_code. */
3598 * Setup lttng message with payload size set to the event list size in
3599 * bytes and then copy list into the llm payload.
3601 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3602 sizeof(struct lttng_event
) * nb_events
);
3612 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3614 struct lttng_event_field
*fields
;
3617 session_lock_list();
3618 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3620 session_unlock_list();
3621 if (nb_fields
< 0) {
3622 /* Return value is a negative lttng_error_code. */
3628 * Setup lttng message with payload size set to the event list size in
3629 * bytes and then copy list into the llm payload.
3631 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, fields
,
3632 sizeof(struct lttng_event_field
) * nb_fields
);
3642 case LTTNG_LIST_SYSCALLS
:
3644 struct lttng_event
*events
;
3647 nb_events
= cmd_list_syscalls(&events
);
3648 if (nb_events
< 0) {
3649 /* Return value is a negative lttng_error_code. */
3655 * Setup lttng message with payload size set to the event list size in
3656 * bytes and then copy list into the llm payload.
3658 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3659 sizeof(struct lttng_event
) * nb_events
);
3669 case LTTNG_LIST_TRACKER_PIDS
:
3671 int32_t *pids
= NULL
;
3674 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3675 cmd_ctx
->lsm
->domain
.type
, &pids
);
3677 /* Return value is a negative lttng_error_code. */
3683 * Setup lttng message with payload size set to the event list size in
3684 * bytes and then copy list into the llm payload.
3686 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, pids
,
3687 sizeof(int32_t) * nr_pids
);
3697 case LTTNG_SET_CONSUMER_URI
:
3700 struct lttng_uri
*uris
;
3702 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3703 len
= nb_uri
* sizeof(struct lttng_uri
);
3706 ret
= LTTNG_ERR_INVALID
;
3710 uris
= zmalloc(len
);
3712 ret
= LTTNG_ERR_FATAL
;
3716 /* Receive variable len data */
3717 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3718 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3720 DBG("No URIs received from client... continuing");
3722 ret
= LTTNG_ERR_SESSION_FAIL
;
3727 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3729 if (ret
!= LTTNG_OK
) {
3736 case LTTNG_START_TRACE
:
3738 ret
= cmd_start_trace(cmd_ctx
->session
);
3741 case LTTNG_STOP_TRACE
:
3743 ret
= cmd_stop_trace(cmd_ctx
->session
);
3746 case LTTNG_CREATE_SESSION
:
3749 struct lttng_uri
*uris
= NULL
;
3751 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3752 len
= nb_uri
* sizeof(struct lttng_uri
);
3755 uris
= zmalloc(len
);
3757 ret
= LTTNG_ERR_FATAL
;
3761 /* Receive variable len data */
3762 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3763 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3765 DBG("No URIs received from client... continuing");
3767 ret
= LTTNG_ERR_SESSION_FAIL
;
3772 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3773 DBG("Creating session with ONE network URI is a bad call");
3774 ret
= LTTNG_ERR_SESSION_FAIL
;
3780 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3781 &cmd_ctx
->creds
, 0);
3787 case LTTNG_DESTROY_SESSION
:
3789 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3791 /* Set session to NULL so we do not unlock it after free. */
3792 cmd_ctx
->session
= NULL
;
3795 case LTTNG_LIST_DOMAINS
:
3798 struct lttng_domain
*domains
= NULL
;
3800 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3802 /* Return value is a negative lttng_error_code. */
3807 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, domains
,
3808 nb_dom
* sizeof(struct lttng_domain
));
3818 case LTTNG_LIST_CHANNELS
:
3820 ssize_t payload_size
;
3821 struct lttng_channel
*channels
= NULL
;
3823 payload_size
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3824 cmd_ctx
->session
, &channels
);
3825 if (payload_size
< 0) {
3826 /* Return value is a negative lttng_error_code. */
3827 ret
= -payload_size
;
3831 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, channels
,
3842 case LTTNG_LIST_EVENTS
:
3845 struct lttng_event
*events
= NULL
;
3846 struct lttcomm_event_command_header cmd_header
;
3849 memset(&cmd_header
, 0, sizeof(cmd_header
));
3850 /* Extended infos are included at the end of events */
3851 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
,
3852 cmd_ctx
->session
, cmd_ctx
->lsm
->u
.list
.channel_name
,
3853 &events
, &total_size
);
3856 /* Return value is a negative lttng_error_code. */
3861 cmd_header
.nb_events
= nb_event
;
3862 ret
= setup_lttng_msg(cmd_ctx
, events
, total_size
,
3863 &cmd_header
, sizeof(cmd_header
));
3873 case LTTNG_LIST_SESSIONS
:
3875 unsigned int nr_sessions
;
3876 void *sessions_payload
;
3879 session_lock_list();
3880 nr_sessions
= lttng_sessions_count(
3881 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3882 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3883 payload_len
= sizeof(struct lttng_session
) * nr_sessions
;
3884 sessions_payload
= zmalloc(payload_len
);
3886 if (!sessions_payload
) {
3887 session_unlock_list();
3892 cmd_list_lttng_sessions(sessions_payload
,
3893 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3894 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3895 session_unlock_list();
3897 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, sessions_payload
,
3899 free(sessions_payload
);
3908 case LTTNG_CALIBRATE
:
3910 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3911 &cmd_ctx
->lsm
->u
.calibrate
);
3914 case LTTNG_REGISTER_CONSUMER
:
3916 struct consumer_data
*cdata
;
3918 switch (cmd_ctx
->lsm
->domain
.type
) {
3919 case LTTNG_DOMAIN_KERNEL
:
3920 cdata
= &kconsumer_data
;
3923 ret
= LTTNG_ERR_UND
;
3927 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3928 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3931 case LTTNG_DATA_PENDING
:
3934 uint8_t pending_ret_byte
;
3936 pending_ret
= cmd_data_pending(cmd_ctx
->session
);
3941 * This function may returns 0 or 1 to indicate whether or not
3942 * there is data pending. In case of error, it should return an
3943 * LTTNG_ERR code. However, some code paths may still return
3944 * a nondescript error code, which we handle by returning an
3947 if (pending_ret
== 0 || pending_ret
== 1) {
3949 * ret will be set to LTTNG_OK at the end of
3952 } else if (pending_ret
< 0) {
3953 ret
= LTTNG_ERR_UNK
;
3960 pending_ret_byte
= (uint8_t) pending_ret
;
3962 /* 1 byte to return whether or not data is pending */
3963 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
,
3964 &pending_ret_byte
, 1);
3973 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3975 struct lttcomm_lttng_output_id reply
;
3977 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3978 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3979 if (ret
!= LTTNG_OK
) {
3983 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &reply
,
3989 /* Copy output list into message payload */
3993 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3995 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3996 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3999 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
4002 struct lttng_snapshot_output
*outputs
= NULL
;
4004 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
4005 if (nb_output
< 0) {
4010 assert((nb_output
> 0 && outputs
) || nb_output
== 0);
4011 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, outputs
,
4012 nb_output
* sizeof(struct lttng_snapshot_output
));
4022 case LTTNG_SNAPSHOT_RECORD
:
4024 ret
= cmd_snapshot_record(cmd_ctx
->session
,
4025 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
4026 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
4029 case LTTNG_CREATE_SESSION_SNAPSHOT
:
4032 struct lttng_uri
*uris
= NULL
;
4034 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4035 len
= nb_uri
* sizeof(struct lttng_uri
);
4038 uris
= zmalloc(len
);
4040 ret
= LTTNG_ERR_FATAL
;
4044 /* Receive variable len data */
4045 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4046 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4048 DBG("No URIs received from client... continuing");
4050 ret
= LTTNG_ERR_SESSION_FAIL
;
4055 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4056 DBG("Creating session with ONE network URI is a bad call");
4057 ret
= LTTNG_ERR_SESSION_FAIL
;
4063 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
4064 nb_uri
, &cmd_ctx
->creds
);
4068 case LTTNG_CREATE_SESSION_LIVE
:
4071 struct lttng_uri
*uris
= NULL
;
4073 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4074 len
= nb_uri
* sizeof(struct lttng_uri
);
4077 uris
= zmalloc(len
);
4079 ret
= LTTNG_ERR_FATAL
;
4083 /* Receive variable len data */
4084 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4085 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4087 DBG("No URIs received from client... continuing");
4089 ret
= LTTNG_ERR_SESSION_FAIL
;
4094 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4095 DBG("Creating session with ONE network URI is a bad call");
4096 ret
= LTTNG_ERR_SESSION_FAIL
;
4102 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4103 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4107 case LTTNG_SAVE_SESSION
:
4109 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4113 case LTTNG_SET_SESSION_SHM_PATH
:
4115 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4116 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4119 case LTTNG_REGENERATE_METADATA
:
4121 ret
= cmd_regenerate_metadata(cmd_ctx
->session
);
4124 case LTTNG_REGENERATE_STATEDUMP
:
4126 ret
= cmd_regenerate_statedump(cmd_ctx
->session
);
4130 ret
= LTTNG_ERR_UND
;
4135 if (cmd_ctx
->llm
== NULL
) {
4136 DBG("Missing llm structure. Allocating one.");
4137 if (setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0) < 0) {
4141 /* Set return code */
4142 cmd_ctx
->llm
->ret_code
= ret
;
4144 if (cmd_ctx
->session
) {
4145 session_unlock(cmd_ctx
->session
);
4147 if (need_tracing_session
) {
4148 session_unlock_list();
4151 assert(!rcu_read_ongoing());
4156 * Thread managing health check socket.
4158 static void *thread_manage_health(void *data
)
4160 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4161 uint32_t revents
, nb_fd
;
4162 struct lttng_poll_event events
;
4163 struct health_comm_msg msg
;
4164 struct health_comm_reply reply
;
4166 DBG("[thread] Manage health check started");
4168 rcu_register_thread();
4170 /* We might hit an error path before this is created. */
4171 lttng_poll_init(&events
);
4173 /* Create unix socket */
4174 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
4176 ERR("Unable to create health check Unix socket");
4181 /* lttng health client socket path permissions */
4182 ret
= chown(health_unix_sock_path
, 0,
4183 utils_get_group_id(tracing_group_name
));
4185 ERR("Unable to set group on %s", health_unix_sock_path
);
4190 ret
= chmod(health_unix_sock_path
,
4191 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4193 ERR("Unable to set permissions on %s", health_unix_sock_path
);
4200 * Set the CLOEXEC flag. Return code is useless because either way, the
4203 (void) utils_set_fd_cloexec(sock
);
4205 ret
= lttcomm_listen_unix_sock(sock
);
4211 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4212 * more will be added to this poll set.
4214 ret
= sessiond_set_thread_pollset(&events
, 2);
4219 /* Add the application registration socket */
4220 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4225 sessiond_notify_ready();
4228 DBG("Health check ready");
4230 /* Inifinite blocking call, waiting for transmission */
4232 ret
= lttng_poll_wait(&events
, -1);
4235 * Restart interrupted system call.
4237 if (errno
== EINTR
) {
4245 for (i
= 0; i
< nb_fd
; i
++) {
4246 /* Fetch once the poll data */
4247 revents
= LTTNG_POLL_GETEV(&events
, i
);
4248 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4251 /* No activity for this FD (poll implementation). */
4255 /* Thread quit pipe has been closed. Killing thread. */
4256 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4262 /* Event on the registration socket */
4263 if (pollfd
== sock
) {
4264 if (revents
& LPOLLIN
) {
4266 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4267 ERR("Health socket poll error");
4270 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4276 new_sock
= lttcomm_accept_unix_sock(sock
);
4282 * Set the CLOEXEC flag. Return code is useless because either way, the
4285 (void) utils_set_fd_cloexec(new_sock
);
4287 DBG("Receiving data from client for health...");
4288 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4290 DBG("Nothing recv() from client... continuing");
4291 ret
= close(new_sock
);
4299 rcu_thread_online();
4301 memset(&reply
, 0, sizeof(reply
));
4302 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4304 * health_check_state returns 0 if health is
4307 if (!health_check_state(health_sessiond
, i
)) {
4308 reply
.ret_code
|= 1ULL << i
;
4312 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4314 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4316 ERR("Failed to send health data back to client");
4319 /* End of transmission */
4320 ret
= close(new_sock
);
4330 ERR("Health error occurred in %s", __func__
);
4332 DBG("Health check thread dying");
4333 unlink(health_unix_sock_path
);
4341 lttng_poll_clean(&events
);
4343 rcu_unregister_thread();
4348 * This thread manage all clients request using the unix client socket for
4351 static void *thread_manage_clients(void *data
)
4353 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4355 uint32_t revents
, nb_fd
;
4356 struct command_ctx
*cmd_ctx
= NULL
;
4357 struct lttng_poll_event events
;
4359 DBG("[thread] Manage client started");
4361 rcu_register_thread();
4363 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4365 health_code_update();
4367 ret
= lttcomm_listen_unix_sock(client_sock
);
4373 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4374 * more will be added to this poll set.
4376 ret
= sessiond_set_thread_pollset(&events
, 2);
4378 goto error_create_poll
;
4381 /* Add the application registration socket */
4382 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4387 sessiond_notify_ready();
4388 ret
= sem_post(&load_info
->message_thread_ready
);
4390 PERROR("sem_post message_thread_ready");
4394 /* This testpoint is after we signal readiness to the parent. */
4395 if (testpoint(sessiond_thread_manage_clients
)) {
4399 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4403 health_code_update();
4406 DBG("Accepting client command ...");
4408 /* Inifinite blocking call, waiting for transmission */
4410 health_poll_entry();
4411 ret
= lttng_poll_wait(&events
, -1);
4415 * Restart interrupted system call.
4417 if (errno
== EINTR
) {
4425 for (i
= 0; i
< nb_fd
; i
++) {
4426 /* Fetch once the poll data */
4427 revents
= LTTNG_POLL_GETEV(&events
, i
);
4428 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4430 health_code_update();
4433 /* No activity for this FD (poll implementation). */
4437 /* Thread quit pipe has been closed. Killing thread. */
4438 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4444 /* Event on the registration socket */
4445 if (pollfd
== client_sock
) {
4446 if (revents
& LPOLLIN
) {
4448 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4449 ERR("Client socket poll error");
4452 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4458 DBG("Wait for client response");
4460 health_code_update();
4462 sock
= lttcomm_accept_unix_sock(client_sock
);
4468 * Set the CLOEXEC flag. Return code is useless because either way, the
4471 (void) utils_set_fd_cloexec(sock
);
4473 /* Set socket option for credentials retrieval */
4474 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4479 /* Allocate context command to process the client request */
4480 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4481 if (cmd_ctx
== NULL
) {
4482 PERROR("zmalloc cmd_ctx");
4486 /* Allocate data buffer for reception */
4487 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4488 if (cmd_ctx
->lsm
== NULL
) {
4489 PERROR("zmalloc cmd_ctx->lsm");
4493 cmd_ctx
->llm
= NULL
;
4494 cmd_ctx
->session
= NULL
;
4496 health_code_update();
4499 * Data is received from the lttng client. The struct
4500 * lttcomm_session_msg (lsm) contains the command and data request of
4503 DBG("Receiving data from client ...");
4504 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4505 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4507 DBG("Nothing recv() from client... continuing");
4513 clean_command_ctx(&cmd_ctx
);
4517 health_code_update();
4519 // TODO: Validate cmd_ctx including sanity check for
4520 // security purpose.
4522 rcu_thread_online();
4524 * This function dispatch the work to the kernel or userspace tracer
4525 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4526 * informations for the client. The command context struct contains
4527 * everything this function may needs.
4529 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4530 rcu_thread_offline();
4538 * TODO: Inform client somehow of the fatal error. At
4539 * this point, ret < 0 means that a zmalloc failed
4540 * (ENOMEM). Error detected but still accept
4541 * command, unless a socket error has been
4544 clean_command_ctx(&cmd_ctx
);
4548 health_code_update();
4550 DBG("Sending response (size: %d, retcode: %s (%d))",
4551 cmd_ctx
->lttng_msg_size
,
4552 lttng_strerror(-cmd_ctx
->llm
->ret_code
),
4553 cmd_ctx
->llm
->ret_code
);
4554 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4556 ERR("Failed to send data back to client");
4559 /* End of transmission */
4566 clean_command_ctx(&cmd_ctx
);
4568 health_code_update();
4580 lttng_poll_clean(&events
);
4581 clean_command_ctx(&cmd_ctx
);
4585 unlink(client_unix_sock_path
);
4586 if (client_sock
>= 0) {
4587 ret
= close(client_sock
);
4595 ERR("Health error occurred in %s", __func__
);
4598 health_unregister(health_sessiond
);
4600 DBG("Client thread dying");
4602 rcu_unregister_thread();
4605 * Since we are creating the consumer threads, we own them, so we need
4606 * to join them before our thread exits.
4608 ret
= join_consumer_thread(&kconsumer_data
);
4611 PERROR("join_consumer");
4614 ret
= join_consumer_thread(&ustconsumer32_data
);
4617 PERROR("join_consumer ust32");
4620 ret
= join_consumer_thread(&ustconsumer64_data
);
4623 PERROR("join_consumer ust64");
4628 static int string_match(const char *str1
, const char *str2
)
4630 return (str1
&& str2
) && !strcmp(str1
, str2
);
4634 * Take an option from the getopt output and set it in the right variable to be
4637 * Return 0 on success else a negative value.
4639 static int set_option(int opt
, const char *arg
, const char *optname
)
4643 if (string_match(optname
, "client-sock") || opt
== 'c') {
4644 if (!arg
|| *arg
== '\0') {
4648 if (lttng_is_setuid_setgid()) {
4649 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4650 "-c, --client-sock");
4652 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4654 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4655 if (!arg
|| *arg
== '\0') {
4659 if (lttng_is_setuid_setgid()) {
4660 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4663 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4665 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4667 } else if (string_match(optname
, "background") || opt
== 'b') {
4669 } else if (string_match(optname
, "group") || opt
== 'g') {
4670 if (!arg
|| *arg
== '\0') {
4674 if (lttng_is_setuid_setgid()) {
4675 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4679 * If the override option is set, the pointer points to a
4680 * *non* const thus freeing it even though the variable type is
4683 if (tracing_group_name_override
) {
4684 free((void *) tracing_group_name
);
4686 tracing_group_name
= strdup(arg
);
4687 if (!tracing_group_name
) {
4691 tracing_group_name_override
= 1;
4693 } else if (string_match(optname
, "help") || opt
== 'h') {
4694 ret
= utils_show_man_page(8, "lttng-sessiond");
4696 ERR("Cannot view man page lttng-sessiond(8)");
4699 exit(ret
? EXIT_FAILURE
: EXIT_SUCCESS
);
4700 } else if (string_match(optname
, "version") || opt
== 'V') {
4701 fprintf(stdout
, "%s\n", VERSION
);
4703 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4705 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4706 if (!arg
|| *arg
== '\0') {
4710 if (lttng_is_setuid_setgid()) {
4711 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4712 "--kconsumerd-err-sock");
4714 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4716 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4717 if (!arg
|| *arg
== '\0') {
4721 if (lttng_is_setuid_setgid()) {
4722 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4723 "--kconsumerd-cmd-sock");
4725 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4727 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4728 if (!arg
|| *arg
== '\0') {
4732 if (lttng_is_setuid_setgid()) {
4733 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4734 "--ustconsumerd64-err-sock");
4736 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4738 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4739 if (!arg
|| *arg
== '\0') {
4743 if (lttng_is_setuid_setgid()) {
4744 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4745 "--ustconsumerd64-cmd-sock");
4747 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4749 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4750 if (!arg
|| *arg
== '\0') {
4754 if (lttng_is_setuid_setgid()) {
4755 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4756 "--ustconsumerd32-err-sock");
4758 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4760 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4761 if (!arg
|| *arg
== '\0') {
4765 if (lttng_is_setuid_setgid()) {
4766 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4767 "--ustconsumerd32-cmd-sock");
4769 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4771 } else if (string_match(optname
, "no-kernel")) {
4773 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4774 lttng_opt_quiet
= 1;
4775 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4776 /* Verbose level can increase using multiple -v */
4778 /* Value obtained from config file */
4779 lttng_opt_verbose
= config_parse_value(arg
);
4781 /* -v used on command line */
4782 lttng_opt_verbose
++;
4784 /* Clamp value to [0, 3] */
4785 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4786 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4787 } else if (string_match(optname
, "verbose-consumer")) {
4789 opt_verbose_consumer
= config_parse_value(arg
);
4791 opt_verbose_consumer
++;
4793 } else if (string_match(optname
, "consumerd32-path")) {
4794 if (!arg
|| *arg
== '\0') {
4798 if (lttng_is_setuid_setgid()) {
4799 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4800 "--consumerd32-path");
4802 if (consumerd32_bin_override
) {
4803 free((void *) consumerd32_bin
);
4805 consumerd32_bin
= strdup(arg
);
4806 if (!consumerd32_bin
) {
4810 consumerd32_bin_override
= 1;
4812 } else if (string_match(optname
, "consumerd32-libdir")) {
4813 if (!arg
|| *arg
== '\0') {
4817 if (lttng_is_setuid_setgid()) {
4818 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4819 "--consumerd32-libdir");
4821 if (consumerd32_libdir_override
) {
4822 free((void *) consumerd32_libdir
);
4824 consumerd32_libdir
= strdup(arg
);
4825 if (!consumerd32_libdir
) {
4829 consumerd32_libdir_override
= 1;
4831 } else if (string_match(optname
, "consumerd64-path")) {
4832 if (!arg
|| *arg
== '\0') {
4836 if (lttng_is_setuid_setgid()) {
4837 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4838 "--consumerd64-path");
4840 if (consumerd64_bin_override
) {
4841 free((void *) consumerd64_bin
);
4843 consumerd64_bin
= strdup(arg
);
4844 if (!consumerd64_bin
) {
4848 consumerd64_bin_override
= 1;
4850 } else if (string_match(optname
, "consumerd64-libdir")) {
4851 if (!arg
|| *arg
== '\0') {
4855 if (lttng_is_setuid_setgid()) {
4856 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4857 "--consumerd64-libdir");
4859 if (consumerd64_libdir_override
) {
4860 free((void *) consumerd64_libdir
);
4862 consumerd64_libdir
= strdup(arg
);
4863 if (!consumerd64_libdir
) {
4867 consumerd64_libdir_override
= 1;
4869 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4870 if (!arg
|| *arg
== '\0') {
4874 if (lttng_is_setuid_setgid()) {
4875 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4879 opt_pidfile
= strdup(arg
);
4885 } else if (string_match(optname
, "agent-tcp-port")) {
4886 if (!arg
|| *arg
== '\0') {
4890 if (lttng_is_setuid_setgid()) {
4891 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4892 "--agent-tcp-port");
4897 v
= strtoul(arg
, NULL
, 0);
4898 if (errno
!= 0 || !isdigit(arg
[0])) {
4899 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4902 if (v
== 0 || v
>= 65535) {
4903 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4906 agent_tcp_port
= (uint32_t) v
;
4907 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4909 } else if (string_match(optname
, "load") || opt
== 'l') {
4910 if (!arg
|| *arg
== '\0') {
4914 if (lttng_is_setuid_setgid()) {
4915 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4918 free(opt_load_session_path
);
4919 opt_load_session_path
= strdup(arg
);
4920 if (!opt_load_session_path
) {
4925 } else if (string_match(optname
, "kmod-probes")) {
4926 if (!arg
|| *arg
== '\0') {
4930 if (lttng_is_setuid_setgid()) {
4931 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4934 free(kmod_probes_list
);
4935 kmod_probes_list
= strdup(arg
);
4936 if (!kmod_probes_list
) {
4941 } else if (string_match(optname
, "extra-kmod-probes")) {
4942 if (!arg
|| *arg
== '\0') {
4946 if (lttng_is_setuid_setgid()) {
4947 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4948 "--extra-kmod-probes");
4950 free(kmod_extra_probes_list
);
4951 kmod_extra_probes_list
= strdup(arg
);
4952 if (!kmod_extra_probes_list
) {
4957 } else if (string_match(optname
, "config") || opt
== 'f') {
4958 /* This is handled in set_options() thus silent skip. */
4961 /* Unknown option or other error.
4962 * Error is printed by getopt, just return */
4967 if (ret
== -EINVAL
) {
4968 const char *opt_name
= "unknown";
4971 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
4973 if (opt
== long_options
[i
].val
) {
4974 opt_name
= long_options
[i
].name
;
4979 WARN("Invalid argument provided for option \"%s\", using default value.",
4987 * config_entry_handler_cb used to handle options read from a config file.
4988 * See config_entry_handler_cb comment in common/config/session-config.h for the
4989 * return value conventions.
4991 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4995 if (!entry
|| !entry
->name
|| !entry
->value
) {
5000 /* Check if the option is to be ignored */
5001 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
5002 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
5007 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
5010 /* Ignore if not fully matched. */
5011 if (strcmp(entry
->name
, long_options
[i
].name
)) {
5016 * If the option takes no argument on the command line, we have to
5017 * check if the value is "true". We support non-zero numeric values,
5020 if (!long_options
[i
].has_arg
) {
5021 ret
= config_parse_value(entry
->value
);
5024 WARN("Invalid configuration value \"%s\" for option %s",
5025 entry
->value
, entry
->name
);
5027 /* False, skip boolean config option. */
5032 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
5036 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
5043 * daemon configuration loading and argument parsing
5045 static int set_options(int argc
, char **argv
)
5047 int ret
= 0, c
= 0, option_index
= 0;
5048 int orig_optopt
= optopt
, orig_optind
= optind
;
5050 const char *config_path
= NULL
;
5052 optstring
= utils_generate_optstring(long_options
,
5053 sizeof(long_options
) / sizeof(struct option
));
5059 /* Check for the --config option */
5060 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
5061 &option_index
)) != -1) {
5065 } else if (c
!= 'f') {
5066 /* if not equal to --config option. */
5070 if (lttng_is_setuid_setgid()) {
5071 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5074 config_path
= utils_expand_path(optarg
);
5076 ERR("Failed to resolve path: %s", optarg
);
5081 ret
= config_get_section_entries(config_path
, config_section_name
,
5082 config_entry_handler
, NULL
);
5085 ERR("Invalid configuration option at line %i", ret
);
5091 /* Reset getopt's global state */
5092 optopt
= orig_optopt
;
5093 optind
= orig_optind
;
5097 * getopt_long() will not set option_index if it encounters a
5100 c
= getopt_long(argc
, argv
, optstring
, long_options
,
5107 * Pass NULL as the long option name if popt left the index
5110 ret
= set_option(c
, optarg
,
5111 option_index
< 0 ? NULL
:
5112 long_options
[option_index
].name
);
5124 * Creates the two needed socket by the daemon.
5125 * apps_sock - The communication socket for all UST apps.
5126 * client_sock - The communication of the cli tool (lttng).
5128 static int init_daemon_socket(void)
5133 old_umask
= umask(0);
5135 /* Create client tool unix socket */
5136 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
5137 if (client_sock
< 0) {
5138 ERR("Create unix sock failed: %s", client_unix_sock_path
);
5143 /* Set the cloexec flag */
5144 ret
= utils_set_fd_cloexec(client_sock
);
5146 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5147 "Continuing but note that the consumer daemon will have a "
5148 "reference to this socket on exec()", client_sock
);
5151 /* File permission MUST be 660 */
5152 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5154 ERR("Set file permissions failed: %s", client_unix_sock_path
);
5159 /* Create the application unix socket */
5160 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
5161 if (apps_sock
< 0) {
5162 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
5167 /* Set the cloexec flag */
5168 ret
= utils_set_fd_cloexec(apps_sock
);
5170 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5171 "Continuing but note that the consumer daemon will have a "
5172 "reference to this socket on exec()", apps_sock
);
5175 /* File permission MUST be 666 */
5176 ret
= chmod(apps_unix_sock_path
,
5177 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5179 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
5184 DBG3("Session daemon client socket %d and application socket %d created",
5185 client_sock
, apps_sock
);
5193 * Check if the global socket is available, and if a daemon is answering at the
5194 * other side. If yes, error is returned.
5196 static int check_existing_daemon(void)
5198 /* Is there anybody out there ? */
5199 if (lttng_session_daemon_alive()) {
5207 * Set the tracing group gid onto the client socket.
5209 * Race window between mkdir and chown is OK because we are going from more
5210 * permissive (root.root) to less permissive (root.tracing).
5212 static int set_permissions(char *rundir
)
5217 gid
= utils_get_group_id(tracing_group_name
);
5219 /* Set lttng run dir */
5220 ret
= chown(rundir
, 0, gid
);
5222 ERR("Unable to set group on %s", rundir
);
5227 * Ensure all applications and tracing group can search the run
5228 * dir. Allow everyone to read the directory, since it does not
5229 * buy us anything to hide its content.
5231 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5233 ERR("Unable to set permissions on %s", rundir
);
5237 /* lttng client socket path */
5238 ret
= chown(client_unix_sock_path
, 0, gid
);
5240 ERR("Unable to set group on %s", client_unix_sock_path
);
5244 /* kconsumer error socket path */
5245 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5247 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5251 /* 64-bit ustconsumer error socket path */
5252 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5254 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5258 /* 32-bit ustconsumer compat32 error socket path */
5259 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5261 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5265 DBG("All permissions are set");
5271 * Create the lttng run directory needed for all global sockets and pipe.
5273 static int create_lttng_rundir(const char *rundir
)
5277 DBG3("Creating LTTng run directory: %s", rundir
);
5279 ret
= mkdir(rundir
, S_IRWXU
);
5281 if (errno
!= EEXIST
) {
5282 ERR("Unable to create %s", rundir
);
5294 * Setup sockets and directory needed by the kconsumerd communication with the
5297 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5301 char path
[PATH_MAX
];
5303 switch (consumer_data
->type
) {
5304 case LTTNG_CONSUMER_KERNEL
:
5305 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5307 case LTTNG_CONSUMER64_UST
:
5308 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5310 case LTTNG_CONSUMER32_UST
:
5311 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5314 ERR("Consumer type unknown");
5319 DBG2("Creating consumer directory: %s", path
);
5321 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5323 if (errno
!= EEXIST
) {
5325 ERR("Failed to create %s", path
);
5331 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5333 ERR("Unable to set group on %s", path
);
5339 /* Create the kconsumerd error unix socket */
5340 consumer_data
->err_sock
=
5341 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5342 if (consumer_data
->err_sock
< 0) {
5343 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5349 * Set the CLOEXEC flag. Return code is useless because either way, the
5352 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5354 PERROR("utils_set_fd_cloexec");
5355 /* continue anyway */
5358 /* File permission MUST be 660 */
5359 ret
= chmod(consumer_data
->err_unix_sock_path
,
5360 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5362 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5372 * Signal handler for the daemon
5374 * Simply stop all worker threads, leaving main() return gracefully after
5375 * joining all threads and calling cleanup().
5377 static void sighandler(int sig
)
5381 DBG("SIGPIPE caught");
5384 DBG("SIGINT caught");
5388 DBG("SIGTERM caught");
5392 CMM_STORE_SHARED(recv_child_signal
, 1);
5400 * Setup signal handler for :
5401 * SIGINT, SIGTERM, SIGPIPE
5403 static int set_signal_handler(void)
5406 struct sigaction sa
;
5409 if ((ret
= sigemptyset(&sigset
)) < 0) {
5410 PERROR("sigemptyset");
5414 sa
.sa_handler
= sighandler
;
5415 sa
.sa_mask
= sigset
;
5417 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5418 PERROR("sigaction");
5422 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5423 PERROR("sigaction");
5427 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5428 PERROR("sigaction");
5432 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5433 PERROR("sigaction");
5437 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5443 * Set open files limit to unlimited. This daemon can open a large number of
5444 * file descriptors in order to consume multiple kernel traces.
5446 static void set_ulimit(void)
5451 /* The kernel does not allow an infinite limit for open files */
5452 lim
.rlim_cur
= 65535;
5453 lim
.rlim_max
= 65535;
5455 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5457 PERROR("failed to set open files limit");
5462 * Write pidfile using the rundir and opt_pidfile.
5464 static int write_pidfile(void)
5467 char pidfile_path
[PATH_MAX
];
5472 if (lttng_strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
))) {
5477 /* Build pidfile path from rundir and opt_pidfile. */
5478 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5479 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5481 PERROR("snprintf pidfile path");
5487 * Create pid file in rundir.
5489 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5495 * Create lockfile using the rundir and return its fd.
5497 static int create_lockfile(void)
5500 char lockfile_path
[PATH_MAX
];
5502 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5507 ret
= utils_create_lock_file(lockfile_path
);
5513 * Write agent TCP port using the rundir.
5515 static int write_agent_port(void)
5518 char path
[PATH_MAX
];
5522 ret
= snprintf(path
, sizeof(path
), "%s/"
5523 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5525 PERROR("snprintf agent port path");
5530 * Create TCP agent port file in rundir.
5532 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5541 int main(int argc
, char **argv
)
5543 int ret
= 0, retval
= 0;
5545 const char *home_path
, *env_app_timeout
;
5547 init_kernel_workarounds();
5549 rcu_register_thread();
5551 if (set_signal_handler()) {
5553 goto exit_set_signal_handler
;
5556 setup_consumerd_path();
5558 page_size
= sysconf(_SC_PAGESIZE
);
5559 if (page_size
< 0) {
5560 PERROR("sysconf _SC_PAGESIZE");
5561 page_size
= LONG_MAX
;
5562 WARN("Fallback page size to %ld", page_size
);
5566 * Parse arguments and load the daemon configuration file.
5568 * We have an exit_options exit path to free memory reserved by
5569 * set_options. This is needed because the rest of sessiond_cleanup()
5570 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5571 * depends on set_options.
5574 if (set_options(argc
, argv
)) {
5580 if (opt_daemon
|| opt_background
) {
5583 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5591 * We are in the child. Make sure all other file descriptors are
5592 * closed, in case we are called with more opened file
5593 * descriptors than the standard ones.
5595 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5600 if (run_as_create_worker(argv
[0]) < 0) {
5601 goto exit_create_run_as_worker_cleanup
;
5605 * Starting from here, we can create threads. This needs to be after
5606 * lttng_daemonize due to RCU.
5610 * Initialize the health check subsystem. This call should set the
5611 * appropriate time values.
5613 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5614 if (!health_sessiond
) {
5615 PERROR("health_app_create error");
5617 goto exit_health_sessiond_cleanup
;
5620 /* Create thread to clean up RCU hash tables */
5621 if (init_ht_cleanup_thread(&ht_cleanup_thread
)) {
5623 goto exit_ht_cleanup
;
5626 /* Create thread quit pipe */
5627 if (init_thread_quit_pipe()) {
5629 goto exit_init_data
;
5632 /* Check if daemon is UID = 0 */
5633 is_root
= !getuid();
5636 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5639 goto exit_init_data
;
5642 /* Create global run dir with root access */
5643 if (create_lttng_rundir(rundir
)) {
5645 goto exit_init_data
;
5648 if (strlen(apps_unix_sock_path
) == 0) {
5649 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5650 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5653 goto exit_init_data
;
5657 if (strlen(client_unix_sock_path
) == 0) {
5658 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5659 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5662 goto exit_init_data
;
5666 /* Set global SHM for ust */
5667 if (strlen(wait_shm_path
) == 0) {
5668 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5669 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5672 goto exit_init_data
;
5676 if (strlen(health_unix_sock_path
) == 0) {
5677 ret
= snprintf(health_unix_sock_path
,
5678 sizeof(health_unix_sock_path
),
5679 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5682 goto exit_init_data
;
5686 /* Setup kernel consumerd path */
5687 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5688 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5691 goto exit_init_data
;
5693 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5694 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5697 goto exit_init_data
;
5700 DBG2("Kernel consumer err path: %s",
5701 kconsumer_data
.err_unix_sock_path
);
5702 DBG2("Kernel consumer cmd path: %s",
5703 kconsumer_data
.cmd_unix_sock_path
);
5705 home_path
= utils_get_home_dir();
5706 if (home_path
== NULL
) {
5707 /* TODO: Add --socket PATH option */
5708 ERR("Can't get HOME directory for sockets creation.");
5710 goto exit_init_data
;
5714 * Create rundir from home path. This will create something like
5717 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5720 goto exit_init_data
;
5723 if (create_lttng_rundir(rundir
)) {
5725 goto exit_init_data
;
5728 if (strlen(apps_unix_sock_path
) == 0) {
5729 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5730 DEFAULT_HOME_APPS_UNIX_SOCK
,
5734 goto exit_init_data
;
5738 /* Set the cli tool unix socket path */
5739 if (strlen(client_unix_sock_path
) == 0) {
5740 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5741 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5745 goto exit_init_data
;
5749 /* Set global SHM for ust */
5750 if (strlen(wait_shm_path
) == 0) {
5751 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5752 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5756 goto exit_init_data
;
5760 /* Set health check Unix path */
5761 if (strlen(health_unix_sock_path
) == 0) {
5762 ret
= snprintf(health_unix_sock_path
,
5763 sizeof(health_unix_sock_path
),
5764 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5768 goto exit_init_data
;
5773 lockfile_fd
= create_lockfile();
5774 if (lockfile_fd
< 0) {
5776 goto exit_init_data
;
5779 /* Set consumer initial state */
5780 kernel_consumerd_state
= CONSUMER_STOPPED
;
5781 ust_consumerd_state
= CONSUMER_STOPPED
;
5783 DBG("Client socket path %s", client_unix_sock_path
);
5784 DBG("Application socket path %s", apps_unix_sock_path
);
5785 DBG("Application wait path %s", wait_shm_path
);
5786 DBG("LTTng run directory path: %s", rundir
);
5788 /* 32 bits consumerd path setup */
5789 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5790 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5792 PERROR("snprintf 32-bit consumer error socket path");
5794 goto exit_init_data
;
5796 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5797 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5799 PERROR("snprintf 32-bit consumer command socket path");
5801 goto exit_init_data
;
5804 DBG2("UST consumer 32 bits err path: %s",
5805 ustconsumer32_data
.err_unix_sock_path
);
5806 DBG2("UST consumer 32 bits cmd path: %s",
5807 ustconsumer32_data
.cmd_unix_sock_path
);
5809 /* 64 bits consumerd path setup */
5810 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5811 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5813 PERROR("snprintf 64-bit consumer error socket path");
5815 goto exit_init_data
;
5817 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5818 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5820 PERROR("snprintf 64-bit consumer command socket path");
5822 goto exit_init_data
;
5825 DBG2("UST consumer 64 bits err path: %s",
5826 ustconsumer64_data
.err_unix_sock_path
);
5827 DBG2("UST consumer 64 bits cmd path: %s",
5828 ustconsumer64_data
.cmd_unix_sock_path
);
5831 * See if daemon already exist.
5833 if (check_existing_daemon()) {
5834 ERR("Already running daemon.\n");
5836 * We do not goto exit because we must not cleanup()
5837 * because a daemon is already running.
5840 goto exit_init_data
;
5844 * Init UST app hash table. Alloc hash table before this point since
5845 * cleanup() can get called after that point.
5847 if (ust_app_ht_alloc()) {
5848 ERR("Failed to allocate UST app hash table");
5850 goto exit_init_data
;
5854 * Initialize agent app hash table. We allocate the hash table here
5855 * since cleanup() can get called after this point.
5857 if (agent_app_ht_alloc()) {
5858 ERR("Failed to allocate Agent app hash table");
5860 goto exit_init_data
;
5864 * These actions must be executed as root. We do that *after* setting up
5865 * the sockets path because we MUST make the check for another daemon using
5866 * those paths *before* trying to set the kernel consumer sockets and init
5870 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
5872 goto exit_init_data
;
5875 /* Setup kernel tracer */
5876 if (!opt_no_kernel
) {
5877 init_kernel_tracer();
5878 if (kernel_tracer_fd
>= 0) {
5879 ret
= syscall_init_table();
5881 ERR("Unable to populate syscall table. "
5882 "Syscall tracing won't work "
5883 "for this session daemon.");
5888 /* Set ulimit for open files */
5891 /* init lttng_fd tracking must be done after set_ulimit. */
5894 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
5896 goto exit_init_data
;
5899 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
5901 goto exit_init_data
;
5904 /* Setup the needed unix socket */
5905 if (init_daemon_socket()) {
5907 goto exit_init_data
;
5910 /* Set credentials to socket */
5911 if (is_root
&& set_permissions(rundir
)) {
5913 goto exit_init_data
;
5916 /* Get parent pid if -S, --sig-parent is specified. */
5917 if (opt_sig_parent
) {
5921 /* Setup the kernel pipe for waking up the kernel thread */
5922 if (is_root
&& !opt_no_kernel
) {
5923 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5925 goto exit_init_data
;
5929 /* Setup the thread apps communication pipe. */
5930 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
5932 goto exit_init_data
;
5935 /* Setup the thread apps notify communication pipe. */
5936 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
5938 goto exit_init_data
;
5941 /* Initialize global buffer per UID and PID registry. */
5942 buffer_reg_init_uid_registry();
5943 buffer_reg_init_pid_registry();
5945 /* Init UST command queue. */
5946 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
5949 * Get session list pointer. This pointer MUST NOT be free'd. This list
5950 * is statically declared in session.c
5952 session_list_ptr
= session_get_list();
5956 /* Check for the application socket timeout env variable. */
5957 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5958 if (env_app_timeout
) {
5959 app_socket_timeout
= atoi(env_app_timeout
);
5961 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5964 ret
= write_pidfile();
5966 ERR("Error in write_pidfile");
5968 goto exit_init_data
;
5970 ret
= write_agent_port();
5972 ERR("Error in write_agent_port");
5974 goto exit_init_data
;
5977 /* Initialize communication library */
5979 /* Initialize TCP timeout values */
5980 lttcomm_inet_init();
5982 if (load_session_init_data(&load_info
) < 0) {
5984 goto exit_init_data
;
5986 load_info
->path
= opt_load_session_path
;
5988 /* Create health-check thread */
5989 ret
= pthread_create(&health_thread
, default_pthread_attr(),
5990 thread_manage_health
, (void *) NULL
);
5993 PERROR("pthread_create health");
5998 /* Create thread to manage the client socket */
5999 ret
= pthread_create(&client_thread
, default_pthread_attr(),
6000 thread_manage_clients
, (void *) NULL
);
6003 PERROR("pthread_create clients");
6008 /* Create thread to dispatch registration */
6009 ret
= pthread_create(&dispatch_thread
, default_pthread_attr(),
6010 thread_dispatch_ust_registration
, (void *) NULL
);
6013 PERROR("pthread_create dispatch");
6018 /* Create thread to manage application registration. */
6019 ret
= pthread_create(®_apps_thread
, default_pthread_attr(),
6020 thread_registration_apps
, (void *) NULL
);
6023 PERROR("pthread_create registration");
6028 /* Create thread to manage application socket */
6029 ret
= pthread_create(&apps_thread
, default_pthread_attr(),
6030 thread_manage_apps
, (void *) NULL
);
6033 PERROR("pthread_create apps");
6038 /* Create thread to manage application notify socket */
6039 ret
= pthread_create(&apps_notify_thread
, default_pthread_attr(),
6040 ust_thread_manage_notify
, (void *) NULL
);
6043 PERROR("pthread_create notify");
6045 goto exit_apps_notify
;
6048 /* Create agent registration thread. */
6049 ret
= pthread_create(&agent_reg_thread
, default_pthread_attr(),
6050 agent_thread_manage_registration
, (void *) NULL
);
6053 PERROR("pthread_create agent");
6055 goto exit_agent_reg
;
6058 /* Don't start this thread if kernel tracing is not requested nor root */
6059 if (is_root
&& !opt_no_kernel
) {
6060 /* Create kernel thread to manage kernel event */
6061 ret
= pthread_create(&kernel_thread
, default_pthread_attr(),
6062 thread_manage_kernel
, (void *) NULL
);
6065 PERROR("pthread_create kernel");
6071 /* Create session loading thread. */
6072 ret
= pthread_create(&load_session_thread
, default_pthread_attr(),
6073 thread_load_session
, load_info
);
6076 PERROR("pthread_create load_session_thread");
6078 goto exit_load_session
;
6082 * This is where we start awaiting program completion (e.g. through
6083 * signal that asks threads to teardown).
6086 ret
= pthread_join(load_session_thread
, &status
);
6089 PERROR("pthread_join load_session_thread");
6094 if (is_root
&& !opt_no_kernel
) {
6095 ret
= pthread_join(kernel_thread
, &status
);
6098 PERROR("pthread_join");
6104 ret
= pthread_join(agent_reg_thread
, &status
);
6107 PERROR("pthread_join agent");
6112 ret
= pthread_join(apps_notify_thread
, &status
);
6115 PERROR("pthread_join apps notify");
6120 ret
= pthread_join(apps_thread
, &status
);
6123 PERROR("pthread_join apps");
6128 ret
= pthread_join(reg_apps_thread
, &status
);
6131 PERROR("pthread_join");
6137 * Join dispatch thread after joining reg_apps_thread to ensure
6138 * we don't leak applications in the queue.
6140 ret
= pthread_join(dispatch_thread
, &status
);
6143 PERROR("pthread_join");
6148 ret
= pthread_join(client_thread
, &status
);
6151 PERROR("pthread_join");
6156 ret
= pthread_join(health_thread
, &status
);
6159 PERROR("pthread_join health thread");
6166 * sessiond_cleanup() is called when no other thread is running, except
6167 * the ht_cleanup thread, which is needed to destroy the hash tables.
6169 rcu_thread_online();
6171 rcu_thread_offline();
6172 rcu_unregister_thread();
6175 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6176 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6177 * the queue is empty before shutting down the clean-up thread.
6181 ret
= fini_ht_cleanup_thread(&ht_cleanup_thread
);
6187 health_app_destroy(health_sessiond
);
6188 exit_health_sessiond_cleanup
:
6189 exit_create_run_as_worker_cleanup
:
6192 sessiond_cleanup_options();
6194 exit_set_signal_handler
: