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"
77 #define CONSUMERD_FILE "lttng-consumerd"
80 static const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
81 static int tracing_group_name_override
;
82 static char *opt_pidfile
;
83 static int opt_sig_parent
;
84 static int opt_verbose_consumer
;
85 static int opt_daemon
, opt_background
;
86 static int opt_no_kernel
;
87 static char *opt_load_session_path
;
88 static pid_t ppid
; /* Parent PID for --sig-parent option */
89 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
91 static int lockfile_fd
= -1;
93 /* Set to 1 when a SIGUSR1 signal is received. */
94 static int recv_child_signal
;
97 * Consumer daemon specific control data. Every value not initialized here is
98 * set to 0 by the static definition.
100 static struct consumer_data kconsumer_data
= {
101 .type
= LTTNG_CONSUMER_KERNEL
,
102 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
103 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
106 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
107 .lock
= PTHREAD_MUTEX_INITIALIZER
,
108 .cond
= PTHREAD_COND_INITIALIZER
,
109 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
111 static struct consumer_data ustconsumer64_data
= {
112 .type
= LTTNG_CONSUMER64_UST
,
113 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
114 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
117 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
118 .lock
= PTHREAD_MUTEX_INITIALIZER
,
119 .cond
= PTHREAD_COND_INITIALIZER
,
120 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
122 static struct consumer_data ustconsumer32_data
= {
123 .type
= LTTNG_CONSUMER32_UST
,
124 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
125 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
128 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
129 .lock
= PTHREAD_MUTEX_INITIALIZER
,
130 .cond
= PTHREAD_COND_INITIALIZER
,
131 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
134 /* Command line options */
135 static const struct option long_options
[] = {
136 { "client-sock", required_argument
, 0, 'c' },
137 { "apps-sock", required_argument
, 0, 'a' },
138 { "kconsumerd-cmd-sock", required_argument
, 0, '\0' },
139 { "kconsumerd-err-sock", required_argument
, 0, '\0' },
140 { "ustconsumerd32-cmd-sock", required_argument
, 0, '\0' },
141 { "ustconsumerd32-err-sock", required_argument
, 0, '\0' },
142 { "ustconsumerd64-cmd-sock", required_argument
, 0, '\0' },
143 { "ustconsumerd64-err-sock", required_argument
, 0, '\0' },
144 { "consumerd32-path", required_argument
, 0, '\0' },
145 { "consumerd32-libdir", required_argument
, 0, '\0' },
146 { "consumerd64-path", required_argument
, 0, '\0' },
147 { "consumerd64-libdir", required_argument
, 0, '\0' },
148 { "daemonize", no_argument
, 0, 'd' },
149 { "background", no_argument
, 0, 'b' },
150 { "sig-parent", no_argument
, 0, 'S' },
151 { "help", no_argument
, 0, 'h' },
152 { "group", required_argument
, 0, 'g' },
153 { "version", no_argument
, 0, 'V' },
154 { "quiet", no_argument
, 0, 'q' },
155 { "verbose", no_argument
, 0, 'v' },
156 { "verbose-consumer", no_argument
, 0, '\0' },
157 { "no-kernel", no_argument
, 0, '\0' },
158 { "pidfile", required_argument
, 0, 'p' },
159 { "agent-tcp-port", required_argument
, 0, '\0' },
160 { "config", required_argument
, 0, 'f' },
161 { "load", required_argument
, 0, 'l' },
162 { "kmod-probes", required_argument
, 0, '\0' },
163 { "extra-kmod-probes", required_argument
, 0, '\0' },
167 /* Command line options to ignore from configuration file */
168 static const char *config_ignore_options
[] = { "help", "version", "config" };
170 /* Shared between threads */
171 static int dispatch_thread_exit
;
173 /* Global application Unix socket path */
174 static char apps_unix_sock_path
[PATH_MAX
];
175 /* Global client Unix socket path */
176 static char client_unix_sock_path
[PATH_MAX
];
177 /* global wait shm path for UST */
178 static char wait_shm_path
[PATH_MAX
];
179 /* Global health check unix path */
180 static char health_unix_sock_path
[PATH_MAX
];
182 /* Sockets and FDs */
183 static int client_sock
= -1;
184 static int apps_sock
= -1;
185 int kernel_tracer_fd
= -1;
186 static int kernel_poll_pipe
[2] = { -1, -1 };
189 * Quit pipe for all threads. This permits a single cancellation point
190 * for all threads when receiving an event on the pipe.
192 static int thread_quit_pipe
[2] = { -1, -1 };
193 static int ht_cleanup_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 /* Notify parents that we are ready for cmd and health check */
321 void sessiond_notify_ready(void)
323 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
325 * Notify parent pid that we are ready to accept command
326 * for client side. This ppid is the one from the
327 * external process that spawned us.
329 if (opt_sig_parent
) {
334 * Notify the parent of the fork() process that we are
337 if (opt_daemon
|| opt_background
) {
338 kill(child_ppid
, SIGUSR1
);
344 void setup_consumerd_path(void)
346 const char *bin
, *libdir
;
349 * Allow INSTALL_BIN_PATH to be used as a target path for the
350 * native architecture size consumer if CONFIG_CONSUMER*_PATH
351 * has not been defined.
353 #if (CAA_BITS_PER_LONG == 32)
354 if (!consumerd32_bin
[0]) {
355 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
357 if (!consumerd32_libdir
[0]) {
358 consumerd32_libdir
= INSTALL_LIB_PATH
;
360 #elif (CAA_BITS_PER_LONG == 64)
361 if (!consumerd64_bin
[0]) {
362 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
364 if (!consumerd64_libdir
[0]) {
365 consumerd64_libdir
= INSTALL_LIB_PATH
;
368 #error "Unknown bitness"
372 * runtime env. var. overrides the build default.
374 bin
= lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
376 consumerd32_bin
= bin
;
378 bin
= lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
380 consumerd64_bin
= bin
;
382 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
384 consumerd32_libdir
= libdir
;
386 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
388 consumerd64_libdir
= libdir
;
393 int __sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
,
400 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
406 ret
= lttng_poll_add(events
, a_pipe
[0], LPOLLIN
| LPOLLERR
);
418 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
420 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
422 return __sessiond_set_thread_pollset(events
, size
, thread_quit_pipe
);
426 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
428 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event
*events
,
431 return __sessiond_set_thread_pollset(events
, size
,
432 ht_cleanup_quit_pipe
);
436 int __sessiond_check_thread_quit_pipe(int fd
, uint32_t events
, int a_pipe
)
438 if (fd
== a_pipe
&& (events
& LPOLLIN
)) {
445 * Check if the thread quit pipe was triggered.
447 * Return 1 if it was triggered else 0;
449 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
451 return __sessiond_check_thread_quit_pipe(fd
, events
,
452 thread_quit_pipe
[0]);
456 * Check if the ht_cleanup thread quit pipe was triggered.
458 * Return 1 if it was triggered else 0;
460 int sessiond_check_ht_cleanup_quit(int fd
, uint32_t events
)
462 return __sessiond_check_thread_quit_pipe(fd
, events
,
463 ht_cleanup_quit_pipe
[0]);
467 * Init thread quit pipe.
469 * Return -1 on error or 0 if all pipes are created.
471 static int __init_thread_quit_pipe(int *a_pipe
)
477 PERROR("thread quit pipe");
481 for (i
= 0; i
< 2; i
++) {
482 ret
= fcntl(a_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
493 static int init_thread_quit_pipe(void)
495 return __init_thread_quit_pipe(thread_quit_pipe
);
498 static int init_ht_cleanup_quit_pipe(void)
500 return __init_thread_quit_pipe(ht_cleanup_quit_pipe
);
504 * Stop all threads by closing the thread quit pipe.
506 static void stop_threads(void)
510 /* Stopping all threads */
511 DBG("Terminating all threads");
512 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
514 ERR("write error on thread quit pipe");
517 /* Dispatch thread */
518 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
519 futex_nto1_wake(&ust_cmd_queue
.futex
);
523 * Close every consumer sockets.
525 static void close_consumer_sockets(void)
529 if (kconsumer_data
.err_sock
>= 0) {
530 ret
= close(kconsumer_data
.err_sock
);
532 PERROR("kernel consumer err_sock close");
535 if (ustconsumer32_data
.err_sock
>= 0) {
536 ret
= close(ustconsumer32_data
.err_sock
);
538 PERROR("UST consumerd32 err_sock close");
541 if (ustconsumer64_data
.err_sock
>= 0) {
542 ret
= close(ustconsumer64_data
.err_sock
);
544 PERROR("UST consumerd64 err_sock close");
547 if (kconsumer_data
.cmd_sock
>= 0) {
548 ret
= close(kconsumer_data
.cmd_sock
);
550 PERROR("kernel consumer cmd_sock close");
553 if (ustconsumer32_data
.cmd_sock
>= 0) {
554 ret
= close(ustconsumer32_data
.cmd_sock
);
556 PERROR("UST consumerd32 cmd_sock close");
559 if (ustconsumer64_data
.cmd_sock
>= 0) {
560 ret
= close(ustconsumer64_data
.cmd_sock
);
562 PERROR("UST consumerd64 cmd_sock close");
568 * Generate the full lock file path using the rundir.
570 * Return the snprintf() return value thus a negative value is an error.
572 static int generate_lock_file_path(char *path
, size_t len
)
579 /* Build lockfile path from rundir. */
580 ret
= snprintf(path
, len
, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE
, rundir
);
582 PERROR("snprintf lockfile path");
589 * Wait on consumer process termination.
591 * Need to be called with the consumer data lock held or from a context
592 * ensuring no concurrent access to data (e.g: cleanup).
594 static void wait_consumer(struct consumer_data
*consumer_data
)
599 if (consumer_data
->pid
<= 0) {
603 DBG("Waiting for complete teardown of consumerd (PID: %d)",
605 ret
= waitpid(consumer_data
->pid
, &status
, 0);
607 PERROR("consumerd waitpid pid: %d", consumer_data
->pid
)
608 } else if (!WIFEXITED(status
)) {
609 ERR("consumerd termination with error: %d",
612 consumer_data
->pid
= 0;
616 * Cleanup the session daemon's data structures.
618 static void sessiond_cleanup(void)
621 struct ltt_session
*sess
, *stmp
;
624 DBG("Cleanup sessiond");
627 * Close the thread quit pipe. It has already done its job,
628 * since we are now called.
630 utils_close_pipe(thread_quit_pipe
);
633 * If opt_pidfile is undefined, the default file will be wiped when
634 * removing the rundir.
637 ret
= remove(opt_pidfile
);
639 PERROR("remove pidfile %s", opt_pidfile
);
643 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
646 snprintf(path
, PATH_MAX
,
648 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
649 DBG("Removing %s", path
);
652 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
653 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
);
654 DBG("Removing %s", path
);
658 snprintf(path
, PATH_MAX
,
659 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
661 DBG("Removing %s", path
);
664 snprintf(path
, PATH_MAX
,
665 DEFAULT_KCONSUMERD_PATH
,
667 DBG("Removing directory %s", path
);
670 /* ust consumerd 32 */
671 snprintf(path
, PATH_MAX
,
672 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
674 DBG("Removing %s", path
);
677 snprintf(path
, PATH_MAX
,
678 DEFAULT_USTCONSUMERD32_PATH
,
680 DBG("Removing directory %s", path
);
683 /* ust consumerd 64 */
684 snprintf(path
, PATH_MAX
,
685 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
687 DBG("Removing %s", path
);
690 snprintf(path
, PATH_MAX
,
691 DEFAULT_USTCONSUMERD64_PATH
,
693 DBG("Removing directory %s", path
);
696 DBG("Cleaning up all sessions");
698 /* Destroy session list mutex */
699 if (session_list_ptr
!= NULL
) {
700 pthread_mutex_destroy(&session_list_ptr
->lock
);
702 /* Cleanup ALL session */
703 cds_list_for_each_entry_safe(sess
, stmp
,
704 &session_list_ptr
->head
, list
) {
705 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
709 wait_consumer(&kconsumer_data
);
710 wait_consumer(&ustconsumer64_data
);
711 wait_consumer(&ustconsumer32_data
);
713 DBG("Cleaning up all agent apps");
714 agent_app_ht_clean();
716 DBG("Closing all UST sockets");
717 ust_app_clean_list();
718 buffer_reg_destroy_registries();
720 if (is_root
&& !opt_no_kernel
) {
721 DBG2("Closing kernel fd");
722 if (kernel_tracer_fd
>= 0) {
723 ret
= close(kernel_tracer_fd
);
728 DBG("Unloading kernel modules");
729 modprobe_remove_lttng_all();
733 close_consumer_sockets();
736 load_session_destroy_data(load_info
);
741 * Cleanup lock file by deleting it and finaly closing it which will
742 * release the file system lock.
744 if (lockfile_fd
>= 0) {
745 char lockfile_path
[PATH_MAX
];
747 ret
= generate_lock_file_path(lockfile_path
,
748 sizeof(lockfile_path
));
750 ret
= remove(lockfile_path
);
752 PERROR("remove lock file");
754 ret
= close(lockfile_fd
);
756 PERROR("close lock file");
762 * We do NOT rmdir rundir because there are other processes
763 * using it, for instance lttng-relayd, which can start in
764 * parallel with this teardown.
771 * Cleanup the daemon's option data structures.
773 static void sessiond_cleanup_options(void)
775 DBG("Cleaning up options");
778 * If the override option is set, the pointer points to a *non* const
779 * thus freeing it even though the variable type is set to const.
781 if (tracing_group_name_override
) {
782 free((void *) tracing_group_name
);
784 if (consumerd32_bin_override
) {
785 free((void *) consumerd32_bin
);
787 if (consumerd64_bin_override
) {
788 free((void *) consumerd64_bin
);
790 if (consumerd32_libdir_override
) {
791 free((void *) consumerd32_libdir
);
793 if (consumerd64_libdir_override
) {
794 free((void *) consumerd64_libdir
);
798 free(opt_load_session_path
);
799 free(kmod_probes_list
);
800 free(kmod_extra_probes_list
);
802 run_as_destroy_worker();
805 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
806 "Matthew, BEET driven development works!%c[%dm",
807 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
812 * Send data on a unix socket using the liblttsessiondcomm API.
814 * Return lttcomm error code.
816 static int send_unix_sock(int sock
, void *buf
, size_t len
)
818 /* Check valid length */
823 return lttcomm_send_unix_sock(sock
, buf
, len
);
827 * Free memory of a command context structure.
829 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
831 DBG("Clean command context structure");
833 if ((*cmd_ctx
)->llm
) {
834 free((*cmd_ctx
)->llm
);
836 if ((*cmd_ctx
)->lsm
) {
837 free((*cmd_ctx
)->lsm
);
845 * Notify UST applications using the shm mmap futex.
847 static int notify_ust_apps(int active
)
851 DBG("Notifying applications of session daemon state: %d", active
);
853 /* See shm.c for this call implying mmap, shm and futex calls */
854 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
855 if (wait_shm_mmap
== NULL
) {
859 /* Wake waiting process */
860 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
862 /* Apps notified successfully */
870 * Setup the outgoing data buffer for the response (llm) by allocating the
871 * right amount of memory and copying the original information from the lsm
874 * Return 0 on success, negative value on error.
876 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
,
877 const void *payload_buf
, size_t payload_len
,
878 const void *cmd_header_buf
, size_t cmd_header_len
)
881 const size_t header_len
= sizeof(struct lttcomm_lttng_msg
);
882 const size_t cmd_header_offset
= header_len
;
883 const size_t payload_offset
= cmd_header_offset
+ cmd_header_len
;
884 const size_t total_msg_size
= header_len
+ cmd_header_len
+ payload_len
;
886 cmd_ctx
->llm
= zmalloc(total_msg_size
);
888 if (cmd_ctx
->llm
== NULL
) {
894 /* Copy common data */
895 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
896 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
897 cmd_ctx
->llm
->cmd_header_size
= cmd_header_len
;
898 cmd_ctx
->llm
->data_size
= payload_len
;
899 cmd_ctx
->lttng_msg_size
= total_msg_size
;
901 /* Copy command header */
902 if (cmd_header_len
) {
903 memcpy(((uint8_t *) cmd_ctx
->llm
) + cmd_header_offset
, cmd_header_buf
,
909 memcpy(((uint8_t *) cmd_ctx
->llm
) + payload_offset
, payload_buf
,
918 * Version of setup_lttng_msg() without command header.
920 static int setup_lttng_msg_no_cmd_header(struct command_ctx
*cmd_ctx
,
921 void *payload_buf
, size_t payload_len
)
923 return setup_lttng_msg(cmd_ctx
, payload_buf
, payload_len
, NULL
, 0);
926 * Update the kernel poll set of all channel fd available over all tracing
927 * session. Add the wakeup pipe at the end of the set.
929 static int update_kernel_poll(struct lttng_poll_event
*events
)
932 struct ltt_session
*session
;
933 struct ltt_kernel_channel
*channel
;
935 DBG("Updating kernel poll set");
938 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
939 session_lock(session
);
940 if (session
->kernel_session
== NULL
) {
941 session_unlock(session
);
945 cds_list_for_each_entry(channel
,
946 &session
->kernel_session
->channel_list
.head
, list
) {
947 /* Add channel fd to the kernel poll set */
948 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
950 session_unlock(session
);
953 DBG("Channel fd %d added to kernel set", channel
->fd
);
955 session_unlock(session
);
957 session_unlock_list();
962 session_unlock_list();
967 * Find the channel fd from 'fd' over all tracing session. When found, check
968 * for new channel stream and send those stream fds to the kernel consumer.
970 * Useful for CPU hotplug feature.
972 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
975 struct ltt_session
*session
;
976 struct ltt_kernel_session
*ksess
;
977 struct ltt_kernel_channel
*channel
;
979 DBG("Updating kernel streams for channel fd %d", fd
);
982 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
983 session_lock(session
);
984 if (session
->kernel_session
== NULL
) {
985 session_unlock(session
);
988 ksess
= session
->kernel_session
;
990 cds_list_for_each_entry(channel
,
991 &ksess
->channel_list
.head
, list
) {
992 struct lttng_ht_iter iter
;
993 struct consumer_socket
*socket
;
995 if (channel
->fd
!= fd
) {
998 DBG("Channel found, updating kernel streams");
999 ret
= kernel_open_channel_stream(channel
);
1003 /* Update the stream global counter */
1004 ksess
->stream_count_global
+= ret
;
1007 * Have we already sent fds to the consumer? If yes, it
1008 * means that tracing is started so it is safe to send
1009 * our updated stream fds.
1011 if (ksess
->consumer_fds_sent
!= 1
1012 || ksess
->consumer
== NULL
) {
1018 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
1019 &iter
.iter
, socket
, node
.node
) {
1020 pthread_mutex_lock(socket
->lock
);
1021 ret
= kernel_consumer_send_channel_stream(socket
,
1023 session
->output_traces
? 1 : 0);
1024 pthread_mutex_unlock(socket
->lock
);
1032 session_unlock(session
);
1034 session_unlock_list();
1038 session_unlock(session
);
1039 session_unlock_list();
1044 * For each tracing session, update newly registered apps. The session list
1045 * lock MUST be acquired before calling this.
1047 static void update_ust_app(int app_sock
)
1049 struct ltt_session
*sess
, *stmp
;
1051 /* Consumer is in an ERROR state. Stop any application update. */
1052 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
1053 /* Stop the update process since the consumer is dead. */
1057 /* For all tracing session(s) */
1058 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
1059 struct ust_app
*app
;
1062 if (!sess
->ust_session
) {
1063 goto unlock_session
;
1067 assert(app_sock
>= 0);
1068 app
= ust_app_find_by_sock(app_sock
);
1071 * Application can be unregistered before so
1072 * this is possible hence simply stopping the
1075 DBG3("UST app update failed to find app sock %d",
1079 ust_app_global_update(sess
->ust_session
, app
);
1083 session_unlock(sess
);
1088 * This thread manage event coming from the kernel.
1090 * Features supported in this thread:
1093 static void *thread_manage_kernel(void *data
)
1095 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
1096 uint32_t revents
, nb_fd
;
1098 struct lttng_poll_event events
;
1100 DBG("[thread] Thread manage kernel started");
1102 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
1105 * This first step of the while is to clean this structure which could free
1106 * non NULL pointers so initialize it before the loop.
1108 lttng_poll_init(&events
);
1110 if (testpoint(sessiond_thread_manage_kernel
)) {
1111 goto error_testpoint
;
1114 health_code_update();
1116 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
1117 goto error_testpoint
;
1121 health_code_update();
1123 if (update_poll_flag
== 1) {
1124 /* Clean events object. We are about to populate it again. */
1125 lttng_poll_clean(&events
);
1127 ret
= sessiond_set_thread_pollset(&events
, 2);
1129 goto error_poll_create
;
1132 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1137 /* This will add the available kernel channel if any. */
1138 ret
= update_kernel_poll(&events
);
1142 update_poll_flag
= 0;
1145 DBG("Thread kernel polling");
1147 /* Poll infinite value of time */
1149 health_poll_entry();
1150 ret
= lttng_poll_wait(&events
, -1);
1151 DBG("Thread kernel return from poll on %d fds",
1152 LTTNG_POLL_GETNB(&events
));
1156 * Restart interrupted system call.
1158 if (errno
== EINTR
) {
1162 } else if (ret
== 0) {
1163 /* Should not happen since timeout is infinite */
1164 ERR("Return value of poll is 0 with an infinite timeout.\n"
1165 "This should not have happened! Continuing...");
1171 for (i
= 0; i
< nb_fd
; i
++) {
1172 /* Fetch once the poll data */
1173 revents
= LTTNG_POLL_GETEV(&events
, i
);
1174 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1176 health_code_update();
1179 /* No activity for this FD (poll implementation). */
1183 /* Thread quit pipe has been closed. Killing thread. */
1184 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1190 /* Check for data on kernel pipe */
1191 if (revents
& LPOLLIN
) {
1192 if (pollfd
== kernel_poll_pipe
[0]) {
1193 (void) lttng_read(kernel_poll_pipe
[0],
1196 * Ret value is useless here, if this pipe gets any actions an
1197 * update is required anyway.
1199 update_poll_flag
= 1;
1203 * New CPU detected by the kernel. Adding kernel stream to
1204 * kernel session and updating the kernel consumer
1206 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1212 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1213 update_poll_flag
= 1;
1216 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1224 lttng_poll_clean(&events
);
1227 utils_close_pipe(kernel_poll_pipe
);
1228 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1231 ERR("Health error occurred in %s", __func__
);
1232 WARN("Kernel thread died unexpectedly. "
1233 "Kernel tracing can continue but CPU hotplug is disabled.");
1235 health_unregister(health_sessiond
);
1236 DBG("Kernel thread dying");
1241 * Signal pthread condition of the consumer data that the thread.
1243 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1245 pthread_mutex_lock(&data
->cond_mutex
);
1248 * The state is set before signaling. It can be any value, it's the waiter
1249 * job to correctly interpret this condition variable associated to the
1250 * consumer pthread_cond.
1252 * A value of 0 means that the corresponding thread of the consumer data
1253 * was not started. 1 indicates that the thread has started and is ready
1254 * for action. A negative value means that there was an error during the
1257 data
->consumer_thread_is_ready
= state
;
1258 (void) pthread_cond_signal(&data
->cond
);
1260 pthread_mutex_unlock(&data
->cond_mutex
);
1264 * This thread manage the consumer error sent back to the session daemon.
1266 static void *thread_manage_consumer(void *data
)
1268 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1269 uint32_t revents
, nb_fd
;
1270 enum lttcomm_return_code code
;
1271 struct lttng_poll_event events
;
1272 struct consumer_data
*consumer_data
= data
;
1274 DBG("[thread] Manage consumer started");
1276 rcu_register_thread();
1277 rcu_thread_online();
1279 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1281 health_code_update();
1284 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1285 * metadata_sock. Nothing more will be added to this poll set.
1287 ret
= sessiond_set_thread_pollset(&events
, 3);
1293 * The error socket here is already in a listening state which was done
1294 * just before spawning this thread to avoid a race between the consumer
1295 * daemon exec trying to connect and the listen() call.
1297 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1302 health_code_update();
1304 /* Infinite blocking call, waiting for transmission */
1306 health_poll_entry();
1308 if (testpoint(sessiond_thread_manage_consumer
)) {
1312 ret
= lttng_poll_wait(&events
, -1);
1316 * Restart interrupted system call.
1318 if (errno
== EINTR
) {
1326 for (i
= 0; i
< nb_fd
; i
++) {
1327 /* Fetch once the poll data */
1328 revents
= LTTNG_POLL_GETEV(&events
, i
);
1329 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1331 health_code_update();
1334 /* No activity for this FD (poll implementation). */
1338 /* Thread quit pipe has been closed. Killing thread. */
1339 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1345 /* Event on the registration socket */
1346 if (pollfd
== consumer_data
->err_sock
) {
1347 if (revents
& LPOLLIN
) {
1349 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1350 ERR("consumer err socket poll error");
1353 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1359 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1365 * Set the CLOEXEC flag. Return code is useless because either way, the
1368 (void) utils_set_fd_cloexec(sock
);
1370 health_code_update();
1372 DBG2("Receiving code from consumer err_sock");
1374 /* Getting status code from kconsumerd */
1375 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1376 sizeof(enum lttcomm_return_code
));
1381 health_code_update();
1382 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1383 /* Connect both socket, command and metadata. */
1384 consumer_data
->cmd_sock
=
1385 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1386 consumer_data
->metadata_fd
=
1387 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1388 if (consumer_data
->cmd_sock
< 0
1389 || consumer_data
->metadata_fd
< 0) {
1390 PERROR("consumer connect cmd socket");
1391 /* On error, signal condition and quit. */
1392 signal_consumer_condition(consumer_data
, -1);
1395 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1396 /* Create metadata socket lock. */
1397 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1398 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1399 PERROR("zmalloc pthread mutex");
1403 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1405 signal_consumer_condition(consumer_data
, 1);
1406 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1407 DBG("Consumer metadata socket ready (fd: %d)",
1408 consumer_data
->metadata_fd
);
1410 ERR("consumer error when waiting for SOCK_READY : %s",
1411 lttcomm_get_readable_code(-code
));
1415 /* Remove the consumerd error sock since we've established a connexion */
1416 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1421 /* Add new accepted error socket. */
1422 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1427 /* Add metadata socket that is successfully connected. */
1428 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1429 LPOLLIN
| LPOLLRDHUP
);
1434 health_code_update();
1436 /* Infinite blocking call, waiting for transmission */
1439 health_code_update();
1441 /* Exit the thread because the thread quit pipe has been triggered. */
1443 /* Not a health error. */
1448 health_poll_entry();
1449 ret
= lttng_poll_wait(&events
, -1);
1453 * Restart interrupted system call.
1455 if (errno
== EINTR
) {
1463 for (i
= 0; i
< nb_fd
; i
++) {
1464 /* Fetch once the poll data */
1465 revents
= LTTNG_POLL_GETEV(&events
, i
);
1466 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1468 health_code_update();
1471 /* No activity for this FD (poll implementation). */
1476 * Thread quit pipe has been triggered, flag that we should stop
1477 * but continue the current loop to handle potential data from
1480 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1482 if (pollfd
== sock
) {
1483 /* Event on the consumerd socket */
1484 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1485 && !(revents
& LPOLLIN
)) {
1486 ERR("consumer err socket second poll error");
1489 health_code_update();
1490 /* Wait for any kconsumerd error */
1491 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1492 sizeof(enum lttcomm_return_code
));
1494 ERR("consumer closed the command socket");
1498 ERR("consumer return code : %s",
1499 lttcomm_get_readable_code(-code
));
1502 } else if (pollfd
== consumer_data
->metadata_fd
) {
1503 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1504 && !(revents
& LPOLLIN
)) {
1505 ERR("consumer err metadata socket second poll error");
1508 /* UST metadata requests */
1509 ret
= ust_consumer_metadata_request(
1510 &consumer_data
->metadata_sock
);
1512 ERR("Handling metadata request");
1516 /* No need for an else branch all FDs are tested prior. */
1518 health_code_update();
1524 * We lock here because we are about to close the sockets and some other
1525 * thread might be using them so get exclusive access which will abort all
1526 * other consumer command by other threads.
1528 pthread_mutex_lock(&consumer_data
->lock
);
1530 /* Immediately set the consumerd state to stopped */
1531 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1532 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1533 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1534 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1535 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1537 /* Code flow error... */
1541 if (consumer_data
->err_sock
>= 0) {
1542 ret
= close(consumer_data
->err_sock
);
1546 consumer_data
->err_sock
= -1;
1548 if (consumer_data
->cmd_sock
>= 0) {
1549 ret
= close(consumer_data
->cmd_sock
);
1553 consumer_data
->cmd_sock
= -1;
1555 if (consumer_data
->metadata_sock
.fd_ptr
&&
1556 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1557 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1569 unlink(consumer_data
->err_unix_sock_path
);
1570 unlink(consumer_data
->cmd_unix_sock_path
);
1571 pthread_mutex_unlock(&consumer_data
->lock
);
1573 /* Cleanup metadata socket mutex. */
1574 if (consumer_data
->metadata_sock
.lock
) {
1575 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1576 free(consumer_data
->metadata_sock
.lock
);
1578 lttng_poll_clean(&events
);
1582 ERR("Health error occurred in %s", __func__
);
1584 health_unregister(health_sessiond
);
1585 DBG("consumer thread cleanup completed");
1587 rcu_thread_offline();
1588 rcu_unregister_thread();
1594 * This thread manage application communication.
1596 static void *thread_manage_apps(void *data
)
1598 int i
, ret
, pollfd
, err
= -1;
1600 uint32_t revents
, nb_fd
;
1601 struct lttng_poll_event events
;
1603 DBG("[thread] Manage application started");
1605 rcu_register_thread();
1606 rcu_thread_online();
1608 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1610 if (testpoint(sessiond_thread_manage_apps
)) {
1611 goto error_testpoint
;
1614 health_code_update();
1616 ret
= sessiond_set_thread_pollset(&events
, 2);
1618 goto error_poll_create
;
1621 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1626 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1630 health_code_update();
1633 DBG("Apps thread polling");
1635 /* Inifinite blocking call, waiting for transmission */
1637 health_poll_entry();
1638 ret
= lttng_poll_wait(&events
, -1);
1639 DBG("Apps thread return from poll on %d fds",
1640 LTTNG_POLL_GETNB(&events
));
1644 * Restart interrupted system call.
1646 if (errno
== EINTR
) {
1654 for (i
= 0; i
< nb_fd
; i
++) {
1655 /* Fetch once the poll data */
1656 revents
= LTTNG_POLL_GETEV(&events
, i
);
1657 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1659 health_code_update();
1662 /* No activity for this FD (poll implementation). */
1666 /* Thread quit pipe has been closed. Killing thread. */
1667 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1673 /* Inspect the apps cmd pipe */
1674 if (pollfd
== apps_cmd_pipe
[0]) {
1675 if (revents
& LPOLLIN
) {
1679 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1680 if (size_ret
< sizeof(sock
)) {
1681 PERROR("read apps cmd pipe");
1685 health_code_update();
1688 * Since this is a command socket (write then read),
1689 * we only monitor the error events of the socket.
1691 ret
= lttng_poll_add(&events
, sock
,
1692 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1697 DBG("Apps with sock %d added to poll set", sock
);
1698 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1699 ERR("Apps command pipe error");
1702 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
1707 * At this point, we know that a registered application made
1708 * the event at poll_wait.
1710 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1711 /* Removing from the poll set */
1712 ret
= lttng_poll_del(&events
, pollfd
);
1717 /* Socket closed on remote end. */
1718 ust_app_unregister(pollfd
);
1720 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1725 health_code_update();
1731 lttng_poll_clean(&events
);
1734 utils_close_pipe(apps_cmd_pipe
);
1735 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1738 * We don't clean the UST app hash table here since already registered
1739 * applications can still be controlled so let them be until the session
1740 * daemon dies or the applications stop.
1745 ERR("Health error occurred in %s", __func__
);
1747 health_unregister(health_sessiond
);
1748 DBG("Application communication apps thread cleanup complete");
1749 rcu_thread_offline();
1750 rcu_unregister_thread();
1755 * Send a socket to a thread This is called from the dispatch UST registration
1756 * thread once all sockets are set for the application.
1758 * The sock value can be invalid, we don't really care, the thread will handle
1759 * it and make the necessary cleanup if so.
1761 * On success, return 0 else a negative value being the errno message of the
1764 static int send_socket_to_thread(int fd
, int sock
)
1769 * It's possible that the FD is set as invalid with -1 concurrently just
1770 * before calling this function being a shutdown state of the thread.
1777 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1778 if (ret
< sizeof(sock
)) {
1779 PERROR("write apps pipe %d", fd
);
1786 /* All good. Don't send back the write positive ret value. */
1793 * Sanitize the wait queue of the dispatch registration thread meaning removing
1794 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1795 * notify socket is never received.
1797 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1799 int ret
, nb_fd
= 0, i
;
1800 unsigned int fd_added
= 0;
1801 struct lttng_poll_event events
;
1802 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1806 lttng_poll_init(&events
);
1808 /* Just skip everything for an empty queue. */
1809 if (!wait_queue
->count
) {
1813 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1818 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1819 &wait_queue
->head
, head
) {
1820 assert(wait_node
->app
);
1821 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1822 LPOLLHUP
| LPOLLERR
);
1835 * Poll but don't block so we can quickly identify the faulty events and
1836 * clean them afterwards from the wait queue.
1838 ret
= lttng_poll_wait(&events
, 0);
1844 for (i
= 0; i
< nb_fd
; i
++) {
1845 /* Get faulty FD. */
1846 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1847 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1850 /* No activity for this FD (poll implementation). */
1854 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1855 &wait_queue
->head
, head
) {
1856 if (pollfd
== wait_node
->app
->sock
&&
1857 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1858 cds_list_del(&wait_node
->head
);
1859 wait_queue
->count
--;
1860 ust_app_destroy(wait_node
->app
);
1863 * Silence warning of use-after-free in
1864 * cds_list_for_each_entry_safe which uses
1865 * __typeof__(*wait_node).
1870 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1877 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1881 lttng_poll_clean(&events
);
1885 lttng_poll_clean(&events
);
1887 ERR("Unable to sanitize wait queue");
1892 * Dispatch request from the registration threads to the application
1893 * communication thread.
1895 static void *thread_dispatch_ust_registration(void *data
)
1898 struct cds_wfcq_node
*node
;
1899 struct ust_command
*ust_cmd
= NULL
;
1900 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1901 struct ust_reg_wait_queue wait_queue
= {
1905 rcu_register_thread();
1907 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1909 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1910 goto error_testpoint
;
1913 health_code_update();
1915 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1917 DBG("[thread] Dispatch UST command started");
1920 health_code_update();
1922 /* Atomically prepare the queue futex */
1923 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1925 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1930 struct ust_app
*app
= NULL
;
1934 * Make sure we don't have node(s) that have hung up before receiving
1935 * the notify socket. This is to clean the list in order to avoid
1936 * memory leaks from notify socket that are never seen.
1938 sanitize_wait_queue(&wait_queue
);
1940 health_code_update();
1941 /* Dequeue command for registration */
1942 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1944 DBG("Woken up but nothing in the UST command queue");
1945 /* Continue thread execution */
1949 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1951 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1952 " gid:%d sock:%d name:%s (version %d.%d)",
1953 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1954 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1955 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1956 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1958 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1959 wait_node
= zmalloc(sizeof(*wait_node
));
1961 PERROR("zmalloc wait_node dispatch");
1962 ret
= close(ust_cmd
->sock
);
1964 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1966 lttng_fd_put(LTTNG_FD_APPS
, 1);
1970 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1972 /* Create application object if socket is CMD. */
1973 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1975 if (!wait_node
->app
) {
1976 ret
= close(ust_cmd
->sock
);
1978 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1980 lttng_fd_put(LTTNG_FD_APPS
, 1);
1986 * Add application to the wait queue so we can set the notify
1987 * socket before putting this object in the global ht.
1989 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1994 * We have to continue here since we don't have the notify
1995 * socket and the application MUST be added to the hash table
1996 * only at that moment.
2001 * Look for the application in the local wait queue and set the
2002 * notify socket if found.
2004 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2005 &wait_queue
.head
, head
) {
2006 health_code_update();
2007 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
2008 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
2009 cds_list_del(&wait_node
->head
);
2011 app
= wait_node
->app
;
2013 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
2019 * With no application at this stage the received socket is
2020 * basically useless so close it before we free the cmd data
2021 * structure for good.
2024 ret
= close(ust_cmd
->sock
);
2026 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
2028 lttng_fd_put(LTTNG_FD_APPS
, 1);
2035 * @session_lock_list
2037 * Lock the global session list so from the register up to the
2038 * registration done message, no thread can see the application
2039 * and change its state.
2041 session_lock_list();
2045 * Add application to the global hash table. This needs to be
2046 * done before the update to the UST registry can locate the
2051 /* Set app version. This call will print an error if needed. */
2052 (void) ust_app_version(app
);
2054 /* Send notify socket through the notify pipe. */
2055 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
2059 session_unlock_list();
2061 * No notify thread, stop the UST tracing. However, this is
2062 * not an internal error of the this thread thus setting
2063 * the health error code to a normal exit.
2070 * Update newly registered application with the tracing
2071 * registry info already enabled information.
2073 update_ust_app(app
->sock
);
2076 * Don't care about return value. Let the manage apps threads
2077 * handle app unregistration upon socket close.
2079 (void) ust_app_register_done(app
);
2082 * Even if the application socket has been closed, send the app
2083 * to the thread and unregistration will take place at that
2086 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
2089 session_unlock_list();
2091 * No apps. thread, stop the UST tracing. However, this is
2092 * not an internal error of the this thread thus setting
2093 * the health error code to a normal exit.
2100 session_unlock_list();
2102 } while (node
!= NULL
);
2104 health_poll_entry();
2105 /* Futex wait on queue. Blocking call on futex() */
2106 futex_nto1_wait(&ust_cmd_queue
.futex
);
2109 /* Normal exit, no error */
2113 /* Clean up wait queue. */
2114 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2115 &wait_queue
.head
, head
) {
2116 cds_list_del(&wait_node
->head
);
2121 /* Empty command queue. */
2123 /* Dequeue command for registration */
2124 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2128 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2129 ret
= close(ust_cmd
->sock
);
2131 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2133 lttng_fd_put(LTTNG_FD_APPS
, 1);
2138 DBG("Dispatch thread dying");
2141 ERR("Health error occurred in %s", __func__
);
2143 health_unregister(health_sessiond
);
2144 rcu_unregister_thread();
2149 * This thread manage application registration.
2151 static void *thread_registration_apps(void *data
)
2153 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2154 uint32_t revents
, nb_fd
;
2155 struct lttng_poll_event events
;
2157 * Get allocated in this thread, enqueued to a global queue, dequeued and
2158 * freed in the manage apps thread.
2160 struct ust_command
*ust_cmd
= NULL
;
2162 DBG("[thread] Manage application registration started");
2164 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2166 if (testpoint(sessiond_thread_registration_apps
)) {
2167 goto error_testpoint
;
2170 ret
= lttcomm_listen_unix_sock(apps_sock
);
2176 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2177 * more will be added to this poll set.
2179 ret
= sessiond_set_thread_pollset(&events
, 2);
2181 goto error_create_poll
;
2184 /* Add the application registration socket */
2185 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2187 goto error_poll_add
;
2190 /* Notify all applications to register */
2191 ret
= notify_ust_apps(1);
2193 ERR("Failed to notify applications or create the wait shared memory.\n"
2194 "Execution continues but there might be problem for already\n"
2195 "running applications that wishes to register.");
2199 DBG("Accepting application registration");
2201 /* Inifinite blocking call, waiting for transmission */
2203 health_poll_entry();
2204 ret
= lttng_poll_wait(&events
, -1);
2208 * Restart interrupted system call.
2210 if (errno
== EINTR
) {
2218 for (i
= 0; i
< nb_fd
; i
++) {
2219 health_code_update();
2221 /* Fetch once the poll data */
2222 revents
= LTTNG_POLL_GETEV(&events
, i
);
2223 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2226 /* No activity for this FD (poll implementation). */
2230 /* Thread quit pipe has been closed. Killing thread. */
2231 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2237 /* Event on the registration socket */
2238 if (pollfd
== apps_sock
) {
2239 if (revents
& LPOLLIN
) {
2240 sock
= lttcomm_accept_unix_sock(apps_sock
);
2246 * Set socket timeout for both receiving and ending.
2247 * app_socket_timeout is in seconds, whereas
2248 * lttcomm_setsockopt_rcv_timeout and
2249 * lttcomm_setsockopt_snd_timeout expect msec as
2252 if (app_socket_timeout
>= 0) {
2253 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2254 app_socket_timeout
* 1000);
2255 (void) lttcomm_setsockopt_snd_timeout(sock
,
2256 app_socket_timeout
* 1000);
2260 * Set the CLOEXEC flag. Return code is useless because
2261 * either way, the show must go on.
2263 (void) utils_set_fd_cloexec(sock
);
2265 /* Create UST registration command for enqueuing */
2266 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2267 if (ust_cmd
== NULL
) {
2268 PERROR("ust command zmalloc");
2277 * Using message-based transmissions to ensure we don't
2278 * have to deal with partially received messages.
2280 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2282 ERR("Exhausted file descriptors allowed for applications.");
2292 health_code_update();
2293 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2296 /* Close socket of the application. */
2301 lttng_fd_put(LTTNG_FD_APPS
, 1);
2305 health_code_update();
2307 ust_cmd
->sock
= sock
;
2310 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2311 " gid:%d sock:%d name:%s (version %d.%d)",
2312 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2313 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2314 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2315 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2318 * Lock free enqueue the registration request. The red pill
2319 * has been taken! This apps will be part of the *system*.
2321 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2324 * Wake the registration queue futex. Implicit memory
2325 * barrier with the exchange in cds_wfcq_enqueue.
2327 futex_nto1_wake(&ust_cmd_queue
.futex
);
2328 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2329 ERR("Register apps socket poll error");
2332 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2341 /* Notify that the registration thread is gone */
2344 if (apps_sock
>= 0) {
2345 ret
= close(apps_sock
);
2355 lttng_fd_put(LTTNG_FD_APPS
, 1);
2357 unlink(apps_unix_sock_path
);
2360 lttng_poll_clean(&events
);
2364 DBG("UST Registration thread cleanup complete");
2367 ERR("Health error occurred in %s", __func__
);
2369 health_unregister(health_sessiond
);
2375 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2376 * exec or it will fails.
2378 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2381 struct timespec timeout
;
2383 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2384 consumer_data
->consumer_thread_is_ready
= 0;
2386 /* Setup pthread condition */
2387 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2390 PERROR("pthread_condattr_init consumer data");
2395 * Set the monotonic clock in order to make sure we DO NOT jump in time
2396 * between the clock_gettime() call and the timedwait call. See bug #324
2397 * for a more details and how we noticed it.
2399 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2402 PERROR("pthread_condattr_setclock consumer data");
2406 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2409 PERROR("pthread_cond_init consumer data");
2413 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2417 PERROR("pthread_create consumer");
2422 /* We are about to wait on a pthread condition */
2423 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2425 /* Get time for sem_timedwait absolute timeout */
2426 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2428 * Set the timeout for the condition timed wait even if the clock gettime
2429 * call fails since we might loop on that call and we want to avoid to
2430 * increment the timeout too many times.
2432 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2435 * The following loop COULD be skipped in some conditions so this is why we
2436 * set ret to 0 in order to make sure at least one round of the loop is
2442 * Loop until the condition is reached or when a timeout is reached. Note
2443 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2444 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2445 * possible. This loop does not take any chances and works with both of
2448 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2449 if (clock_ret
< 0) {
2450 PERROR("clock_gettime spawn consumer");
2451 /* Infinite wait for the consumerd thread to be ready */
2452 ret
= pthread_cond_wait(&consumer_data
->cond
,
2453 &consumer_data
->cond_mutex
);
2455 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2456 &consumer_data
->cond_mutex
, &timeout
);
2460 /* Release the pthread condition */
2461 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2465 if (ret
== ETIMEDOUT
) {
2469 * Call has timed out so we kill the kconsumerd_thread and return
2472 ERR("Condition timed out. The consumer thread was never ready."
2474 pth_ret
= pthread_cancel(consumer_data
->thread
);
2476 PERROR("pthread_cancel consumer thread");
2479 PERROR("pthread_cond_wait failed consumer thread");
2481 /* Caller is expecting a negative value on failure. */
2486 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2487 if (consumer_data
->pid
== 0) {
2488 ERR("Consumerd did not start");
2489 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2492 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2501 * Join consumer thread
2503 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2507 /* Consumer pid must be a real one. */
2508 if (consumer_data
->pid
> 0) {
2510 ret
= kill(consumer_data
->pid
, SIGTERM
);
2512 PERROR("Error killing consumer daemon");
2515 return pthread_join(consumer_data
->thread
, &status
);
2522 * Fork and exec a consumer daemon (consumerd).
2524 * Return pid if successful else -1.
2526 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2530 const char *consumer_to_use
;
2531 const char *verbosity
;
2534 DBG("Spawning consumerd");
2541 if (opt_verbose_consumer
) {
2542 verbosity
= "--verbose";
2543 } else if (lttng_opt_quiet
) {
2544 verbosity
= "--quiet";
2549 switch (consumer_data
->type
) {
2550 case LTTNG_CONSUMER_KERNEL
:
2552 * Find out which consumerd to execute. We will first try the
2553 * 64-bit path, then the sessiond's installation directory, and
2554 * fallback on the 32-bit one,
2556 DBG3("Looking for a kernel consumer at these locations:");
2557 DBG3(" 1) %s", consumerd64_bin
);
2558 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2559 DBG3(" 3) %s", consumerd32_bin
);
2560 if (stat(consumerd64_bin
, &st
) == 0) {
2561 DBG3("Found location #1");
2562 consumer_to_use
= consumerd64_bin
;
2563 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2564 DBG3("Found location #2");
2565 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2566 } else if (stat(consumerd32_bin
, &st
) == 0) {
2567 DBG3("Found location #3");
2568 consumer_to_use
= consumerd32_bin
;
2570 DBG("Could not find any valid consumerd executable");
2574 DBG("Using kernel consumer at: %s", consumer_to_use
);
2575 ret
= execl(consumer_to_use
,
2576 "lttng-consumerd", verbosity
, "-k",
2577 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2578 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2579 "--group", tracing_group_name
,
2582 case LTTNG_CONSUMER64_UST
:
2584 char *tmpnew
= NULL
;
2586 if (consumerd64_libdir
[0] != '\0') {
2590 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2594 tmplen
= strlen("LD_LIBRARY_PATH=")
2595 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2596 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2601 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2602 strcat(tmpnew
, consumerd64_libdir
);
2603 if (tmp
[0] != '\0') {
2604 strcat(tmpnew
, ":");
2605 strcat(tmpnew
, tmp
);
2607 ret
= putenv(tmpnew
);
2614 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2615 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2616 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2617 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2618 "--group", tracing_group_name
,
2620 if (consumerd64_libdir
[0] != '\0') {
2625 case LTTNG_CONSUMER32_UST
:
2627 char *tmpnew
= NULL
;
2629 if (consumerd32_libdir
[0] != '\0') {
2633 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2637 tmplen
= strlen("LD_LIBRARY_PATH=")
2638 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2639 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2644 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2645 strcat(tmpnew
, consumerd32_libdir
);
2646 if (tmp
[0] != '\0') {
2647 strcat(tmpnew
, ":");
2648 strcat(tmpnew
, tmp
);
2650 ret
= putenv(tmpnew
);
2657 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2658 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2659 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2660 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2661 "--group", tracing_group_name
,
2663 if (consumerd32_libdir
[0] != '\0') {
2669 PERROR("unknown consumer type");
2673 PERROR("Consumer execl()");
2675 /* Reaching this point, we got a failure on our execl(). */
2677 } else if (pid
> 0) {
2680 PERROR("start consumer fork");
2688 * Spawn the consumerd daemon and session daemon thread.
2690 static int start_consumerd(struct consumer_data
*consumer_data
)
2695 * Set the listen() state on the socket since there is a possible race
2696 * between the exec() of the consumer daemon and this call if place in the
2697 * consumer thread. See bug #366 for more details.
2699 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2704 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2705 if (consumer_data
->pid
!= 0) {
2706 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2710 ret
= spawn_consumerd(consumer_data
);
2712 ERR("Spawning consumerd failed");
2713 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2717 /* Setting up the consumer_data pid */
2718 consumer_data
->pid
= ret
;
2719 DBG2("Consumer pid %d", consumer_data
->pid
);
2720 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2722 DBG2("Spawning consumer control thread");
2723 ret
= spawn_consumer_thread(consumer_data
);
2725 ERR("Fatal error spawning consumer control thread");
2733 /* Cleanup already created sockets on error. */
2734 if (consumer_data
->err_sock
>= 0) {
2737 err
= close(consumer_data
->err_sock
);
2739 PERROR("close consumer data error socket");
2746 * Setup necessary data for kernel tracer action.
2748 static int init_kernel_tracer(void)
2752 /* Modprobe lttng kernel modules */
2753 ret
= modprobe_lttng_control();
2758 /* Open debugfs lttng */
2759 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2760 if (kernel_tracer_fd
< 0) {
2761 DBG("Failed to open %s", module_proc_lttng
);
2766 /* Validate kernel version */
2767 ret
= kernel_validate_version(kernel_tracer_fd
);
2772 ret
= modprobe_lttng_data();
2777 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2781 modprobe_remove_lttng_control();
2782 ret
= close(kernel_tracer_fd
);
2786 kernel_tracer_fd
= -1;
2787 return LTTNG_ERR_KERN_VERSION
;
2790 ret
= close(kernel_tracer_fd
);
2796 modprobe_remove_lttng_control();
2799 WARN("No kernel tracer available");
2800 kernel_tracer_fd
= -1;
2802 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2804 return LTTNG_ERR_KERN_NA
;
2810 * Copy consumer output from the tracing session to the domain session. The
2811 * function also applies the right modification on a per domain basis for the
2812 * trace files destination directory.
2814 * Should *NOT* be called with RCU read-side lock held.
2816 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2819 const char *dir_name
;
2820 struct consumer_output
*consumer
;
2823 assert(session
->consumer
);
2826 case LTTNG_DOMAIN_KERNEL
:
2827 DBG3("Copying tracing session consumer output in kernel session");
2829 * XXX: We should audit the session creation and what this function
2830 * does "extra" in order to avoid a destroy since this function is used
2831 * in the domain session creation (kernel and ust) only. Same for UST
2834 if (session
->kernel_session
->consumer
) {
2835 consumer_output_put(session
->kernel_session
->consumer
);
2837 session
->kernel_session
->consumer
=
2838 consumer_copy_output(session
->consumer
);
2839 /* Ease our life a bit for the next part */
2840 consumer
= session
->kernel_session
->consumer
;
2841 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2843 case LTTNG_DOMAIN_JUL
:
2844 case LTTNG_DOMAIN_LOG4J
:
2845 case LTTNG_DOMAIN_PYTHON
:
2846 case LTTNG_DOMAIN_UST
:
2847 DBG3("Copying tracing session consumer output in UST session");
2848 if (session
->ust_session
->consumer
) {
2849 consumer_output_put(session
->ust_session
->consumer
);
2851 session
->ust_session
->consumer
=
2852 consumer_copy_output(session
->consumer
);
2853 /* Ease our life a bit for the next part */
2854 consumer
= session
->ust_session
->consumer
;
2855 dir_name
= DEFAULT_UST_TRACE_DIR
;
2858 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2862 /* Append correct directory to subdir */
2863 strncat(consumer
->subdir
, dir_name
,
2864 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2865 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2874 * Create an UST session and add it to the session ust list.
2876 * Should *NOT* be called with RCU read-side lock held.
2878 static int create_ust_session(struct ltt_session
*session
,
2879 struct lttng_domain
*domain
)
2882 struct ltt_ust_session
*lus
= NULL
;
2886 assert(session
->consumer
);
2888 switch (domain
->type
) {
2889 case LTTNG_DOMAIN_JUL
:
2890 case LTTNG_DOMAIN_LOG4J
:
2891 case LTTNG_DOMAIN_PYTHON
:
2892 case LTTNG_DOMAIN_UST
:
2895 ERR("Unknown UST domain on create session %d", domain
->type
);
2896 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2900 DBG("Creating UST session");
2902 lus
= trace_ust_create_session(session
->id
);
2904 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2908 lus
->uid
= session
->uid
;
2909 lus
->gid
= session
->gid
;
2910 lus
->output_traces
= session
->output_traces
;
2911 lus
->snapshot_mode
= session
->snapshot_mode
;
2912 lus
->live_timer_interval
= session
->live_timer
;
2913 session
->ust_session
= lus
;
2914 if (session
->shm_path
[0]) {
2915 strncpy(lus
->root_shm_path
, session
->shm_path
,
2916 sizeof(lus
->root_shm_path
));
2917 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2918 strncpy(lus
->shm_path
, session
->shm_path
,
2919 sizeof(lus
->shm_path
));
2920 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2921 strncat(lus
->shm_path
, "/ust",
2922 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2924 /* Copy session output to the newly created UST session */
2925 ret
= copy_session_consumer(domain
->type
, session
);
2926 if (ret
!= LTTNG_OK
) {
2934 session
->ust_session
= NULL
;
2939 * Create a kernel tracer session then create the default channel.
2941 static int create_kernel_session(struct ltt_session
*session
)
2945 DBG("Creating kernel session");
2947 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2949 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2953 /* Code flow safety */
2954 assert(session
->kernel_session
);
2956 /* Copy session output to the newly created Kernel session */
2957 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2958 if (ret
!= LTTNG_OK
) {
2962 /* Create directory(ies) on local filesystem. */
2963 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2964 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2965 ret
= run_as_mkdir_recursive(
2966 session
->kernel_session
->consumer
->dst
.trace_path
,
2967 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2969 if (errno
!= EEXIST
) {
2970 ERR("Trace directory creation error");
2976 session
->kernel_session
->uid
= session
->uid
;
2977 session
->kernel_session
->gid
= session
->gid
;
2978 session
->kernel_session
->output_traces
= session
->output_traces
;
2979 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2984 trace_kernel_destroy_session(session
->kernel_session
);
2985 session
->kernel_session
= NULL
;
2990 * Count number of session permitted by uid/gid.
2992 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2995 struct ltt_session
*session
;
2997 DBG("Counting number of available session for UID %d GID %d",
2999 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
3001 * Only list the sessions the user can control.
3003 if (!session_access_ok(session
, uid
, gid
)) {
3012 * Process the command requested by the lttng client within the command
3013 * context structure. This function make sure that the return structure (llm)
3014 * is set and ready for transmission before returning.
3016 * Return any error encountered or 0 for success.
3018 * "sock" is only used for special-case var. len data.
3020 * Should *NOT* be called with RCU read-side lock held.
3022 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
3026 int need_tracing_session
= 1;
3029 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
3031 assert(!rcu_read_ongoing());
3035 switch (cmd_ctx
->lsm
->cmd_type
) {
3036 case LTTNG_CREATE_SESSION
:
3037 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3038 case LTTNG_CREATE_SESSION_LIVE
:
3039 case LTTNG_DESTROY_SESSION
:
3040 case LTTNG_LIST_SESSIONS
:
3041 case LTTNG_LIST_DOMAINS
:
3042 case LTTNG_START_TRACE
:
3043 case LTTNG_STOP_TRACE
:
3044 case LTTNG_DATA_PENDING
:
3045 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3046 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3047 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3048 case LTTNG_SNAPSHOT_RECORD
:
3049 case LTTNG_SAVE_SESSION
:
3050 case LTTNG_SET_SESSION_SHM_PATH
:
3051 case LTTNG_METADATA_REGENERATE
:
3058 if (opt_no_kernel
&& need_domain
3059 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
3061 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3063 ret
= LTTNG_ERR_KERN_NA
;
3068 /* Deny register consumer if we already have a spawned consumer. */
3069 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
3070 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3071 if (kconsumer_data
.pid
> 0) {
3072 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3073 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3076 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3080 * Check for command that don't needs to allocate a returned payload. We do
3081 * this here so we don't have to make the call for no payload at each
3084 switch(cmd_ctx
->lsm
->cmd_type
) {
3085 case LTTNG_LIST_SESSIONS
:
3086 case LTTNG_LIST_TRACEPOINTS
:
3087 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3088 case LTTNG_LIST_DOMAINS
:
3089 case LTTNG_LIST_CHANNELS
:
3090 case LTTNG_LIST_EVENTS
:
3091 case LTTNG_LIST_SYSCALLS
:
3092 case LTTNG_LIST_TRACKER_PIDS
:
3093 case LTTNG_DATA_PENDING
:
3096 /* Setup lttng message with no payload */
3097 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0);
3099 /* This label does not try to unlock the session */
3100 goto init_setup_error
;
3104 /* Commands that DO NOT need a session. */
3105 switch (cmd_ctx
->lsm
->cmd_type
) {
3106 case LTTNG_CREATE_SESSION
:
3107 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3108 case LTTNG_CREATE_SESSION_LIVE
:
3109 case LTTNG_CALIBRATE
:
3110 case LTTNG_LIST_SESSIONS
:
3111 case LTTNG_LIST_TRACEPOINTS
:
3112 case LTTNG_LIST_SYSCALLS
:
3113 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3114 case LTTNG_SAVE_SESSION
:
3115 need_tracing_session
= 0;
3118 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3120 * We keep the session list lock across _all_ commands
3121 * for now, because the per-session lock does not
3122 * handle teardown properly.
3124 session_lock_list();
3125 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3126 if (cmd_ctx
->session
== NULL
) {
3127 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3130 /* Acquire lock for the session */
3131 session_lock(cmd_ctx
->session
);
3137 * Commands that need a valid session but should NOT create one if none
3138 * exists. Instead of creating one and destroying it when the command is
3139 * handled, process that right before so we save some round trip in useless
3142 switch (cmd_ctx
->lsm
->cmd_type
) {
3143 case LTTNG_DISABLE_CHANNEL
:
3144 case LTTNG_DISABLE_EVENT
:
3145 switch (cmd_ctx
->lsm
->domain
.type
) {
3146 case LTTNG_DOMAIN_KERNEL
:
3147 if (!cmd_ctx
->session
->kernel_session
) {
3148 ret
= LTTNG_ERR_NO_CHANNEL
;
3152 case LTTNG_DOMAIN_JUL
:
3153 case LTTNG_DOMAIN_LOG4J
:
3154 case LTTNG_DOMAIN_PYTHON
:
3155 case LTTNG_DOMAIN_UST
:
3156 if (!cmd_ctx
->session
->ust_session
) {
3157 ret
= LTTNG_ERR_NO_CHANNEL
;
3162 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3174 * Check domain type for specific "pre-action".
3176 switch (cmd_ctx
->lsm
->domain
.type
) {
3177 case LTTNG_DOMAIN_KERNEL
:
3179 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3183 /* Kernel tracer check */
3184 if (kernel_tracer_fd
== -1) {
3185 /* Basically, load kernel tracer modules */
3186 ret
= init_kernel_tracer();
3192 /* Consumer is in an ERROR state. Report back to client */
3193 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3194 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3198 /* Need a session for kernel command */
3199 if (need_tracing_session
) {
3200 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3201 ret
= create_kernel_session(cmd_ctx
->session
);
3203 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3208 /* Start the kernel consumer daemon */
3209 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3210 if (kconsumer_data
.pid
== 0 &&
3211 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3212 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3213 ret
= start_consumerd(&kconsumer_data
);
3215 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3218 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3220 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3224 * The consumer was just spawned so we need to add the socket to
3225 * the consumer output of the session if exist.
3227 ret
= consumer_create_socket(&kconsumer_data
,
3228 cmd_ctx
->session
->kernel_session
->consumer
);
3235 case LTTNG_DOMAIN_JUL
:
3236 case LTTNG_DOMAIN_LOG4J
:
3237 case LTTNG_DOMAIN_PYTHON
:
3238 case LTTNG_DOMAIN_UST
:
3240 if (!ust_app_supported()) {
3241 ret
= LTTNG_ERR_NO_UST
;
3244 /* Consumer is in an ERROR state. Report back to client */
3245 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3246 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3250 if (need_tracing_session
) {
3251 /* Create UST session if none exist. */
3252 if (cmd_ctx
->session
->ust_session
== NULL
) {
3253 ret
= create_ust_session(cmd_ctx
->session
,
3254 &cmd_ctx
->lsm
->domain
);
3255 if (ret
!= LTTNG_OK
) {
3260 /* Start the UST consumer daemons */
3262 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3263 if (consumerd64_bin
[0] != '\0' &&
3264 ustconsumer64_data
.pid
== 0 &&
3265 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3266 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3267 ret
= start_consumerd(&ustconsumer64_data
);
3269 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3270 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3274 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3275 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3277 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3281 * Setup socket for consumer 64 bit. No need for atomic access
3282 * since it was set above and can ONLY be set in this thread.
3284 ret
= consumer_create_socket(&ustconsumer64_data
,
3285 cmd_ctx
->session
->ust_session
->consumer
);
3291 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3292 if (consumerd32_bin
[0] != '\0' &&
3293 ustconsumer32_data
.pid
== 0 &&
3294 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3295 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3296 ret
= start_consumerd(&ustconsumer32_data
);
3298 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3299 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3303 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3304 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3306 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3310 * Setup socket for consumer 64 bit. No need for atomic access
3311 * since it was set above and can ONLY be set in this thread.
3313 ret
= consumer_create_socket(&ustconsumer32_data
,
3314 cmd_ctx
->session
->ust_session
->consumer
);
3326 /* Validate consumer daemon state when start/stop trace command */
3327 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3328 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3329 switch (cmd_ctx
->lsm
->domain
.type
) {
3330 case LTTNG_DOMAIN_NONE
:
3332 case LTTNG_DOMAIN_JUL
:
3333 case LTTNG_DOMAIN_LOG4J
:
3334 case LTTNG_DOMAIN_PYTHON
:
3335 case LTTNG_DOMAIN_UST
:
3336 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3337 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3341 case LTTNG_DOMAIN_KERNEL
:
3342 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3343 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3348 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3354 * Check that the UID or GID match that of the tracing session.
3355 * The root user can interact with all sessions.
3357 if (need_tracing_session
) {
3358 if (!session_access_ok(cmd_ctx
->session
,
3359 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3360 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3361 ret
= LTTNG_ERR_EPERM
;
3367 * Send relayd information to consumer as soon as we have a domain and a
3370 if (cmd_ctx
->session
&& need_domain
) {
3372 * Setup relayd if not done yet. If the relayd information was already
3373 * sent to the consumer, this call will gracefully return.
3375 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3376 if (ret
!= LTTNG_OK
) {
3381 /* Process by command type */
3382 switch (cmd_ctx
->lsm
->cmd_type
) {
3383 case LTTNG_ADD_CONTEXT
:
3386 * An LTTNG_ADD_CONTEXT command might have a supplementary
3387 * payload if the context being added is an application context.
3389 if (cmd_ctx
->lsm
->u
.context
.ctx
.ctx
==
3390 LTTNG_EVENT_CONTEXT_APP_CONTEXT
) {
3391 char *provider_name
= NULL
, *context_name
= NULL
;
3392 size_t provider_name_len
=
3393 cmd_ctx
->lsm
->u
.context
.provider_name_len
;
3394 size_t context_name_len
=
3395 cmd_ctx
->lsm
->u
.context
.context_name_len
;
3397 if (provider_name_len
== 0 || context_name_len
== 0) {
3399 * Application provider and context names MUST
3402 ret
= -LTTNG_ERR_INVALID
;
3406 provider_name
= zmalloc(provider_name_len
+ 1);
3407 if (!provider_name
) {
3408 ret
= -LTTNG_ERR_NOMEM
;
3411 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
=
3414 context_name
= zmalloc(context_name_len
+ 1);
3415 if (!context_name
) {
3416 ret
= -LTTNG_ERR_NOMEM
;
3417 goto error_add_context
;
3419 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
=
3422 ret
= lttcomm_recv_unix_sock(sock
, provider_name
,
3425 goto error_add_context
;
3428 ret
= lttcomm_recv_unix_sock(sock
, context_name
,
3431 goto error_add_context
;
3436 * cmd_add_context assumes ownership of the provider and context
3439 ret
= cmd_add_context(cmd_ctx
->session
,
3440 cmd_ctx
->lsm
->domain
.type
,
3441 cmd_ctx
->lsm
->u
.context
.channel_name
,
3442 &cmd_ctx
->lsm
->u
.context
.ctx
,
3443 kernel_poll_pipe
[1]);
3445 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
= NULL
;
3446 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
= NULL
;
3448 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
);
3449 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
);
3455 case LTTNG_DISABLE_CHANNEL
:
3457 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3458 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3461 case LTTNG_DISABLE_EVENT
:
3465 * FIXME: handle filter; for now we just receive the filter's
3466 * bytecode along with the filter expression which are sent by
3467 * liblttng-ctl and discard them.
3469 * This fixes an issue where the client may block while sending
3470 * the filter payload and encounter an error because the session
3471 * daemon closes the socket without ever handling this data.
3473 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3474 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3477 char data
[LTTNG_FILTER_MAX_LEN
];
3479 DBG("Discarding disable event command payload of size %zu", count
);
3481 ret
= lttcomm_recv_unix_sock(sock
, data
,
3482 count
> sizeof(data
) ? sizeof(data
) : count
);
3487 count
-= (size_t) ret
;
3490 /* FIXME: passing packed structure to non-packed pointer */
3491 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3492 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3493 &cmd_ctx
->lsm
->u
.disable
.event
);
3496 case LTTNG_ENABLE_CHANNEL
:
3498 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3499 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3502 case LTTNG_TRACK_PID
:
3504 ret
= cmd_track_pid(cmd_ctx
->session
,
3505 cmd_ctx
->lsm
->domain
.type
,
3506 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3509 case LTTNG_UNTRACK_PID
:
3511 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3512 cmd_ctx
->lsm
->domain
.type
,
3513 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3516 case LTTNG_ENABLE_EVENT
:
3518 struct lttng_event_exclusion
*exclusion
= NULL
;
3519 struct lttng_filter_bytecode
*bytecode
= NULL
;
3520 char *filter_expression
= NULL
;
3522 /* Handle exclusion events and receive it from the client. */
3523 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3524 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3526 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3527 (count
* LTTNG_SYMBOL_NAME_LEN
));
3529 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3533 DBG("Receiving var len exclusion event list from client ...");
3534 exclusion
->count
= count
;
3535 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3536 count
* LTTNG_SYMBOL_NAME_LEN
);
3538 DBG("Nothing recv() from client var len data... continuing");
3541 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3546 /* Get filter expression from client. */
3547 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3548 size_t expression_len
=
3549 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3551 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3552 ret
= LTTNG_ERR_FILTER_INVAL
;
3557 filter_expression
= zmalloc(expression_len
);
3558 if (!filter_expression
) {
3560 ret
= LTTNG_ERR_FILTER_NOMEM
;
3564 /* Receive var. len. data */
3565 DBG("Receiving var len filter's expression from client ...");
3566 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3569 DBG("Nothing recv() from client car len data... continuing");
3571 free(filter_expression
);
3573 ret
= LTTNG_ERR_FILTER_INVAL
;
3578 /* Handle filter and get bytecode from client. */
3579 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3580 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3582 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3583 ret
= LTTNG_ERR_FILTER_INVAL
;
3584 free(filter_expression
);
3589 bytecode
= zmalloc(bytecode_len
);
3591 free(filter_expression
);
3593 ret
= LTTNG_ERR_FILTER_NOMEM
;
3597 /* Receive var. len. data */
3598 DBG("Receiving var len filter's bytecode from client ...");
3599 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3601 DBG("Nothing recv() from client car len data... continuing");
3603 free(filter_expression
);
3606 ret
= LTTNG_ERR_FILTER_INVAL
;
3610 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3611 free(filter_expression
);
3614 ret
= LTTNG_ERR_FILTER_INVAL
;
3619 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3620 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3621 &cmd_ctx
->lsm
->u
.enable
.event
,
3622 filter_expression
, bytecode
, exclusion
,
3623 kernel_poll_pipe
[1]);
3626 case LTTNG_LIST_TRACEPOINTS
:
3628 struct lttng_event
*events
;
3631 session_lock_list();
3632 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3633 session_unlock_list();
3634 if (nb_events
< 0) {
3635 /* Return value is a negative lttng_error_code. */
3641 * Setup lttng message with payload size set to the event list size in
3642 * bytes and then copy list into the llm payload.
3644 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3645 sizeof(struct lttng_event
) * nb_events
);
3655 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3657 struct lttng_event_field
*fields
;
3660 session_lock_list();
3661 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3663 session_unlock_list();
3664 if (nb_fields
< 0) {
3665 /* Return value is a negative lttng_error_code. */
3671 * Setup lttng message with payload size set to the event list size in
3672 * bytes and then copy list into the llm payload.
3674 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, fields
,
3675 sizeof(struct lttng_event_field
) * nb_fields
);
3685 case LTTNG_LIST_SYSCALLS
:
3687 struct lttng_event
*events
;
3690 nb_events
= cmd_list_syscalls(&events
);
3691 if (nb_events
< 0) {
3692 /* Return value is a negative lttng_error_code. */
3698 * Setup lttng message with payload size set to the event list size in
3699 * bytes and then copy list into the llm payload.
3701 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3702 sizeof(struct lttng_event
) * nb_events
);
3712 case LTTNG_LIST_TRACKER_PIDS
:
3714 int32_t *pids
= NULL
;
3717 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3718 cmd_ctx
->lsm
->domain
.type
, &pids
);
3720 /* Return value is a negative lttng_error_code. */
3726 * Setup lttng message with payload size set to the event list size in
3727 * bytes and then copy list into the llm payload.
3729 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, pids
,
3730 sizeof(int32_t) * nr_pids
);
3740 case LTTNG_SET_CONSUMER_URI
:
3743 struct lttng_uri
*uris
;
3745 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3746 len
= nb_uri
* sizeof(struct lttng_uri
);
3749 ret
= LTTNG_ERR_INVALID
;
3753 uris
= zmalloc(len
);
3755 ret
= LTTNG_ERR_FATAL
;
3759 /* Receive variable len data */
3760 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3761 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3763 DBG("No URIs received from client... continuing");
3765 ret
= LTTNG_ERR_SESSION_FAIL
;
3770 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3772 if (ret
!= LTTNG_OK
) {
3779 case LTTNG_START_TRACE
:
3781 ret
= cmd_start_trace(cmd_ctx
->session
);
3784 case LTTNG_STOP_TRACE
:
3786 ret
= cmd_stop_trace(cmd_ctx
->session
);
3789 case LTTNG_CREATE_SESSION
:
3792 struct lttng_uri
*uris
= NULL
;
3794 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3795 len
= nb_uri
* sizeof(struct lttng_uri
);
3798 uris
= zmalloc(len
);
3800 ret
= LTTNG_ERR_FATAL
;
3804 /* Receive variable len data */
3805 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3806 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3808 DBG("No URIs received from client... continuing");
3810 ret
= LTTNG_ERR_SESSION_FAIL
;
3815 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3816 DBG("Creating session with ONE network URI is a bad call");
3817 ret
= LTTNG_ERR_SESSION_FAIL
;
3823 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3824 &cmd_ctx
->creds
, 0);
3830 case LTTNG_DESTROY_SESSION
:
3832 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3834 /* Set session to NULL so we do not unlock it after free. */
3835 cmd_ctx
->session
= NULL
;
3838 case LTTNG_LIST_DOMAINS
:
3841 struct lttng_domain
*domains
= NULL
;
3843 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3845 /* Return value is a negative lttng_error_code. */
3850 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, domains
,
3851 nb_dom
* sizeof(struct lttng_domain
));
3861 case LTTNG_LIST_CHANNELS
:
3863 ssize_t payload_size
;
3864 struct lttng_channel
*channels
= NULL
;
3866 payload_size
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3867 cmd_ctx
->session
, &channels
);
3868 if (payload_size
< 0) {
3869 /* Return value is a negative lttng_error_code. */
3870 ret
= -payload_size
;
3874 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, channels
,
3885 case LTTNG_LIST_EVENTS
:
3888 struct lttng_event
*events
= NULL
;
3889 struct lttcomm_event_command_header cmd_header
;
3892 memset(&cmd_header
, 0, sizeof(cmd_header
));
3893 /* Extended infos are included at the end of events */
3894 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
,
3895 cmd_ctx
->session
, cmd_ctx
->lsm
->u
.list
.channel_name
,
3896 &events
, &total_size
);
3899 /* Return value is a negative lttng_error_code. */
3904 cmd_header
.nb_events
= nb_event
;
3905 ret
= setup_lttng_msg(cmd_ctx
, events
, total_size
,
3906 &cmd_header
, sizeof(cmd_header
));
3916 case LTTNG_LIST_SESSIONS
:
3918 unsigned int nr_sessions
;
3919 void *sessions_payload
;
3922 session_lock_list();
3923 nr_sessions
= lttng_sessions_count(
3924 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3925 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3926 payload_len
= sizeof(struct lttng_session
) * nr_sessions
;
3927 sessions_payload
= zmalloc(payload_len
);
3929 if (!sessions_payload
) {
3930 session_unlock_list();
3935 cmd_list_lttng_sessions(sessions_payload
,
3936 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3937 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3938 session_unlock_list();
3940 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, sessions_payload
,
3942 free(sessions_payload
);
3951 case LTTNG_CALIBRATE
:
3953 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3954 &cmd_ctx
->lsm
->u
.calibrate
);
3957 case LTTNG_REGISTER_CONSUMER
:
3959 struct consumer_data
*cdata
;
3961 switch (cmd_ctx
->lsm
->domain
.type
) {
3962 case LTTNG_DOMAIN_KERNEL
:
3963 cdata
= &kconsumer_data
;
3966 ret
= LTTNG_ERR_UND
;
3970 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3971 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3974 case LTTNG_DATA_PENDING
:
3977 uint8_t pending_ret_byte
;
3979 pending_ret
= cmd_data_pending(cmd_ctx
->session
);
3984 * This function may returns 0 or 1 to indicate whether or not
3985 * there is data pending. In case of error, it should return an
3986 * LTTNG_ERR code. However, some code paths may still return
3987 * a nondescript error code, which we handle by returning an
3990 if (pending_ret
== 0 || pending_ret
== 1) {
3992 * ret will be set to LTTNG_OK at the end of
3995 } else if (pending_ret
< 0) {
3996 ret
= LTTNG_ERR_UNK
;
4003 pending_ret_byte
= (uint8_t) pending_ret
;
4005 /* 1 byte to return whether or not data is pending */
4006 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
,
4007 &pending_ret_byte
, 1);
4016 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
4018 struct lttcomm_lttng_output_id reply
;
4020 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
4021 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
4022 if (ret
!= LTTNG_OK
) {
4026 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &reply
,
4032 /* Copy output list into message payload */
4036 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
4038 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
4039 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
4042 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
4045 struct lttng_snapshot_output
*outputs
= NULL
;
4047 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
4048 if (nb_output
< 0) {
4053 assert((nb_output
> 0 && outputs
) || nb_output
== 0);
4054 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, outputs
,
4055 nb_output
* sizeof(struct lttng_snapshot_output
));
4065 case LTTNG_SNAPSHOT_RECORD
:
4067 ret
= cmd_snapshot_record(cmd_ctx
->session
,
4068 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
4069 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
4072 case LTTNG_CREATE_SESSION_SNAPSHOT
:
4075 struct lttng_uri
*uris
= NULL
;
4077 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4078 len
= nb_uri
* sizeof(struct lttng_uri
);
4081 uris
= zmalloc(len
);
4083 ret
= LTTNG_ERR_FATAL
;
4087 /* Receive variable len data */
4088 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4089 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4091 DBG("No URIs received from client... continuing");
4093 ret
= LTTNG_ERR_SESSION_FAIL
;
4098 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4099 DBG("Creating session with ONE network URI is a bad call");
4100 ret
= LTTNG_ERR_SESSION_FAIL
;
4106 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
4107 nb_uri
, &cmd_ctx
->creds
);
4111 case LTTNG_CREATE_SESSION_LIVE
:
4114 struct lttng_uri
*uris
= NULL
;
4116 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4117 len
= nb_uri
* sizeof(struct lttng_uri
);
4120 uris
= zmalloc(len
);
4122 ret
= LTTNG_ERR_FATAL
;
4126 /* Receive variable len data */
4127 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4128 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4130 DBG("No URIs received from client... continuing");
4132 ret
= LTTNG_ERR_SESSION_FAIL
;
4137 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4138 DBG("Creating session with ONE network URI is a bad call");
4139 ret
= LTTNG_ERR_SESSION_FAIL
;
4145 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4146 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4150 case LTTNG_SAVE_SESSION
:
4152 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4156 case LTTNG_SET_SESSION_SHM_PATH
:
4158 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4159 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4162 case LTTNG_METADATA_REGENERATE
:
4164 ret
= cmd_metadata_regenerate(cmd_ctx
->session
);
4168 ret
= LTTNG_ERR_UND
;
4173 if (cmd_ctx
->llm
== NULL
) {
4174 DBG("Missing llm structure. Allocating one.");
4175 if (setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0) < 0) {
4179 /* Set return code */
4180 cmd_ctx
->llm
->ret_code
= ret
;
4182 if (cmd_ctx
->session
) {
4183 session_unlock(cmd_ctx
->session
);
4185 if (need_tracing_session
) {
4186 session_unlock_list();
4189 assert(!rcu_read_ongoing());
4194 * Thread managing health check socket.
4196 static void *thread_manage_health(void *data
)
4198 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4199 uint32_t revents
, nb_fd
;
4200 struct lttng_poll_event events
;
4201 struct health_comm_msg msg
;
4202 struct health_comm_reply reply
;
4204 DBG("[thread] Manage health check started");
4206 rcu_register_thread();
4208 /* We might hit an error path before this is created. */
4209 lttng_poll_init(&events
);
4211 /* Create unix socket */
4212 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
4214 ERR("Unable to create health check Unix socket");
4220 /* lttng health client socket path permissions */
4221 ret
= chown(health_unix_sock_path
, 0,
4222 utils_get_group_id(tracing_group_name
));
4224 ERR("Unable to set group on %s", health_unix_sock_path
);
4230 ret
= chmod(health_unix_sock_path
,
4231 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4233 ERR("Unable to set permissions on %s", health_unix_sock_path
);
4241 * Set the CLOEXEC flag. Return code is useless because either way, the
4244 (void) utils_set_fd_cloexec(sock
);
4246 ret
= lttcomm_listen_unix_sock(sock
);
4252 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4253 * more will be added to this poll set.
4255 ret
= sessiond_set_thread_pollset(&events
, 2);
4260 /* Add the application registration socket */
4261 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4266 sessiond_notify_ready();
4269 DBG("Health check ready");
4271 /* Inifinite blocking call, waiting for transmission */
4273 ret
= lttng_poll_wait(&events
, -1);
4276 * Restart interrupted system call.
4278 if (errno
== EINTR
) {
4286 for (i
= 0; i
< nb_fd
; i
++) {
4287 /* Fetch once the poll data */
4288 revents
= LTTNG_POLL_GETEV(&events
, i
);
4289 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4292 /* No activity for this FD (poll implementation). */
4296 /* Thread quit pipe has been closed. Killing thread. */
4297 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4303 /* Event on the registration socket */
4304 if (pollfd
== sock
) {
4305 if (revents
& LPOLLIN
) {
4307 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4308 ERR("Health socket poll error");
4311 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4317 new_sock
= lttcomm_accept_unix_sock(sock
);
4323 * Set the CLOEXEC flag. Return code is useless because either way, the
4326 (void) utils_set_fd_cloexec(new_sock
);
4328 DBG("Receiving data from client for health...");
4329 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4331 DBG("Nothing recv() from client... continuing");
4332 ret
= close(new_sock
);
4340 rcu_thread_online();
4342 memset(&reply
, 0, sizeof(reply
));
4343 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4345 * health_check_state returns 0 if health is
4348 if (!health_check_state(health_sessiond
, i
)) {
4349 reply
.ret_code
|= 1ULL << i
;
4353 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4355 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4357 ERR("Failed to send health data back to client");
4360 /* End of transmission */
4361 ret
= close(new_sock
);
4371 ERR("Health error occurred in %s", __func__
);
4373 DBG("Health check thread dying");
4374 unlink(health_unix_sock_path
);
4382 lttng_poll_clean(&events
);
4384 rcu_unregister_thread();
4389 * This thread manage all clients request using the unix client socket for
4392 static void *thread_manage_clients(void *data
)
4394 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4396 uint32_t revents
, nb_fd
;
4397 struct command_ctx
*cmd_ctx
= NULL
;
4398 struct lttng_poll_event events
;
4400 DBG("[thread] Manage client started");
4402 rcu_register_thread();
4404 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4406 health_code_update();
4408 ret
= lttcomm_listen_unix_sock(client_sock
);
4414 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4415 * more will be added to this poll set.
4417 ret
= sessiond_set_thread_pollset(&events
, 2);
4419 goto error_create_poll
;
4422 /* Add the application registration socket */
4423 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4428 sessiond_notify_ready();
4429 ret
= sem_post(&load_info
->message_thread_ready
);
4431 PERROR("sem_post message_thread_ready");
4435 /* This testpoint is after we signal readiness to the parent. */
4436 if (testpoint(sessiond_thread_manage_clients
)) {
4440 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4444 health_code_update();
4447 DBG("Accepting client command ...");
4449 /* Inifinite blocking call, waiting for transmission */
4451 health_poll_entry();
4452 ret
= lttng_poll_wait(&events
, -1);
4456 * Restart interrupted system call.
4458 if (errno
== EINTR
) {
4466 for (i
= 0; i
< nb_fd
; i
++) {
4467 /* Fetch once the poll data */
4468 revents
= LTTNG_POLL_GETEV(&events
, i
);
4469 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4471 health_code_update();
4474 /* No activity for this FD (poll implementation). */
4478 /* Thread quit pipe has been closed. Killing thread. */
4479 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4485 /* Event on the registration socket */
4486 if (pollfd
== client_sock
) {
4487 if (revents
& LPOLLIN
) {
4489 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4490 ERR("Client socket poll error");
4493 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4499 DBG("Wait for client response");
4501 health_code_update();
4503 sock
= lttcomm_accept_unix_sock(client_sock
);
4509 * Set the CLOEXEC flag. Return code is useless because either way, the
4512 (void) utils_set_fd_cloexec(sock
);
4514 /* Set socket option for credentials retrieval */
4515 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4520 /* Allocate context command to process the client request */
4521 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4522 if (cmd_ctx
== NULL
) {
4523 PERROR("zmalloc cmd_ctx");
4527 /* Allocate data buffer for reception */
4528 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4529 if (cmd_ctx
->lsm
== NULL
) {
4530 PERROR("zmalloc cmd_ctx->lsm");
4534 cmd_ctx
->llm
= NULL
;
4535 cmd_ctx
->session
= NULL
;
4537 health_code_update();
4540 * Data is received from the lttng client. The struct
4541 * lttcomm_session_msg (lsm) contains the command and data request of
4544 DBG("Receiving data from client ...");
4545 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4546 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4548 DBG("Nothing recv() from client... continuing");
4554 clean_command_ctx(&cmd_ctx
);
4558 health_code_update();
4560 // TODO: Validate cmd_ctx including sanity check for
4561 // security purpose.
4563 rcu_thread_online();
4565 * This function dispatch the work to the kernel or userspace tracer
4566 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4567 * informations for the client. The command context struct contains
4568 * everything this function may needs.
4570 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4571 rcu_thread_offline();
4579 * TODO: Inform client somehow of the fatal error. At
4580 * this point, ret < 0 means that a zmalloc failed
4581 * (ENOMEM). Error detected but still accept
4582 * command, unless a socket error has been
4585 clean_command_ctx(&cmd_ctx
);
4589 health_code_update();
4591 DBG("Sending response (size: %d, retcode: %s (%d))",
4592 cmd_ctx
->lttng_msg_size
,
4593 lttng_strerror(-cmd_ctx
->llm
->ret_code
),
4594 cmd_ctx
->llm
->ret_code
);
4595 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4597 ERR("Failed to send data back to client");
4600 /* End of transmission */
4607 clean_command_ctx(&cmd_ctx
);
4609 health_code_update();
4621 lttng_poll_clean(&events
);
4622 clean_command_ctx(&cmd_ctx
);
4626 unlink(client_unix_sock_path
);
4627 if (client_sock
>= 0) {
4628 ret
= close(client_sock
);
4636 ERR("Health error occurred in %s", __func__
);
4639 health_unregister(health_sessiond
);
4641 DBG("Client thread dying");
4643 rcu_unregister_thread();
4646 * Since we are creating the consumer threads, we own them, so we need
4647 * to join them before our thread exits.
4649 ret
= join_consumer_thread(&kconsumer_data
);
4652 PERROR("join_consumer");
4655 ret
= join_consumer_thread(&ustconsumer32_data
);
4658 PERROR("join_consumer ust32");
4661 ret
= join_consumer_thread(&ustconsumer64_data
);
4664 PERROR("join_consumer ust64");
4669 static int string_match(const char *str1
, const char *str2
)
4671 return (str1
&& str2
) && !strcmp(str1
, str2
);
4675 * Take an option from the getopt output and set it in the right variable to be
4678 * Return 0 on success else a negative value.
4680 static int set_option(int opt
, const char *arg
, const char *optname
)
4684 if (string_match(optname
, "client-sock") || opt
== 'c') {
4685 if (!arg
|| *arg
== '\0') {
4689 if (lttng_is_setuid_setgid()) {
4690 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4691 "-c, --client-sock");
4693 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4695 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4696 if (!arg
|| *arg
== '\0') {
4700 if (lttng_is_setuid_setgid()) {
4701 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4704 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4706 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4708 } else if (string_match(optname
, "background") || opt
== 'b') {
4710 } else if (string_match(optname
, "group") || opt
== 'g') {
4711 if (!arg
|| *arg
== '\0') {
4715 if (lttng_is_setuid_setgid()) {
4716 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4720 * If the override option is set, the pointer points to a
4721 * *non* const thus freeing it even though the variable type is
4724 if (tracing_group_name_override
) {
4725 free((void *) tracing_group_name
);
4727 tracing_group_name
= strdup(arg
);
4728 if (!tracing_group_name
) {
4732 tracing_group_name_override
= 1;
4734 } else if (string_match(optname
, "help") || opt
== 'h') {
4735 ret
= utils_show_man_page(8, "lttng-sessiond");
4737 ERR("Cannot view man page lttng-sessiond(8)");
4740 exit(ret
? EXIT_FAILURE
: EXIT_SUCCESS
);
4741 } else if (string_match(optname
, "version") || opt
== 'V') {
4742 fprintf(stdout
, "%s\n", VERSION
);
4744 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4746 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4747 if (!arg
|| *arg
== '\0') {
4751 if (lttng_is_setuid_setgid()) {
4752 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4753 "--kconsumerd-err-sock");
4755 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4757 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4758 if (!arg
|| *arg
== '\0') {
4762 if (lttng_is_setuid_setgid()) {
4763 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4764 "--kconsumerd-cmd-sock");
4766 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4768 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4769 if (!arg
|| *arg
== '\0') {
4773 if (lttng_is_setuid_setgid()) {
4774 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4775 "--ustconsumerd64-err-sock");
4777 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4779 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4780 if (!arg
|| *arg
== '\0') {
4784 if (lttng_is_setuid_setgid()) {
4785 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4786 "--ustconsumerd64-cmd-sock");
4788 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4790 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4791 if (!arg
|| *arg
== '\0') {
4795 if (lttng_is_setuid_setgid()) {
4796 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4797 "--ustconsumerd32-err-sock");
4799 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4801 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4802 if (!arg
|| *arg
== '\0') {
4806 if (lttng_is_setuid_setgid()) {
4807 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4808 "--ustconsumerd32-cmd-sock");
4810 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4812 } else if (string_match(optname
, "no-kernel")) {
4814 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4815 lttng_opt_quiet
= 1;
4816 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4817 /* Verbose level can increase using multiple -v */
4819 /* Value obtained from config file */
4820 lttng_opt_verbose
= config_parse_value(arg
);
4822 /* -v used on command line */
4823 lttng_opt_verbose
++;
4825 /* Clamp value to [0, 3] */
4826 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4827 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4828 } else if (string_match(optname
, "verbose-consumer")) {
4830 opt_verbose_consumer
= config_parse_value(arg
);
4832 opt_verbose_consumer
++;
4834 } else if (string_match(optname
, "consumerd32-path")) {
4835 if (!arg
|| *arg
== '\0') {
4839 if (lttng_is_setuid_setgid()) {
4840 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4841 "--consumerd32-path");
4843 if (consumerd32_bin_override
) {
4844 free((void *) consumerd32_bin
);
4846 consumerd32_bin
= strdup(arg
);
4847 if (!consumerd32_bin
) {
4851 consumerd32_bin_override
= 1;
4853 } else if (string_match(optname
, "consumerd32-libdir")) {
4854 if (!arg
|| *arg
== '\0') {
4858 if (lttng_is_setuid_setgid()) {
4859 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4860 "--consumerd32-libdir");
4862 if (consumerd32_libdir_override
) {
4863 free((void *) consumerd32_libdir
);
4865 consumerd32_libdir
= strdup(arg
);
4866 if (!consumerd32_libdir
) {
4870 consumerd32_libdir_override
= 1;
4872 } else if (string_match(optname
, "consumerd64-path")) {
4873 if (!arg
|| *arg
== '\0') {
4877 if (lttng_is_setuid_setgid()) {
4878 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4879 "--consumerd64-path");
4881 if (consumerd64_bin_override
) {
4882 free((void *) consumerd64_bin
);
4884 consumerd64_bin
= strdup(arg
);
4885 if (!consumerd64_bin
) {
4889 consumerd64_bin_override
= 1;
4891 } else if (string_match(optname
, "consumerd64-libdir")) {
4892 if (!arg
|| *arg
== '\0') {
4896 if (lttng_is_setuid_setgid()) {
4897 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4898 "--consumerd64-libdir");
4900 if (consumerd64_libdir_override
) {
4901 free((void *) consumerd64_libdir
);
4903 consumerd64_libdir
= strdup(arg
);
4904 if (!consumerd64_libdir
) {
4908 consumerd64_libdir_override
= 1;
4910 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4911 if (!arg
|| *arg
== '\0') {
4915 if (lttng_is_setuid_setgid()) {
4916 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4920 opt_pidfile
= strdup(arg
);
4926 } else if (string_match(optname
, "agent-tcp-port")) {
4927 if (!arg
|| *arg
== '\0') {
4931 if (lttng_is_setuid_setgid()) {
4932 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4933 "--agent-tcp-port");
4942 v
= strtoul(arg
, NULL
, 0);
4943 if (errno
!= 0 || !isdigit(arg
[0])) {
4944 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4947 if (v
== 0 || v
>= 65535) {
4948 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4951 agent_tcp_port
= (uint32_t) v
;
4952 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4954 } else if (string_match(optname
, "load") || opt
== 'l') {
4955 if (!arg
|| *arg
== '\0') {
4959 if (lttng_is_setuid_setgid()) {
4960 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4963 free(opt_load_session_path
);
4964 opt_load_session_path
= strdup(arg
);
4965 if (!opt_load_session_path
) {
4970 } else if (string_match(optname
, "kmod-probes")) {
4971 if (!arg
|| *arg
== '\0') {
4975 if (lttng_is_setuid_setgid()) {
4976 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4979 free(kmod_probes_list
);
4980 kmod_probes_list
= strdup(arg
);
4981 if (!kmod_probes_list
) {
4986 } else if (string_match(optname
, "extra-kmod-probes")) {
4987 if (!arg
|| *arg
== '\0') {
4991 if (lttng_is_setuid_setgid()) {
4992 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4993 "--extra-kmod-probes");
4995 free(kmod_extra_probes_list
);
4996 kmod_extra_probes_list
= strdup(arg
);
4997 if (!kmod_extra_probes_list
) {
5002 } else if (string_match(optname
, "config") || opt
== 'f') {
5003 /* This is handled in set_options() thus silent skip. */
5006 /* Unknown option or other error.
5007 * Error is printed by getopt, just return */
5012 if (ret
== -EINVAL
) {
5013 const char *opt_name
= "unknown";
5016 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
5018 if (opt
== long_options
[i
].val
) {
5019 opt_name
= long_options
[i
].name
;
5024 WARN("Invalid argument provided for option \"%s\", using default value.",
5032 * config_entry_handler_cb used to handle options read from a config file.
5033 * See config_entry_handler_cb comment in common/config/session-config.h for the
5034 * return value conventions.
5036 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
5040 if (!entry
|| !entry
->name
|| !entry
->value
) {
5045 /* Check if the option is to be ignored */
5046 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
5047 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
5052 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
5055 /* Ignore if not fully matched. */
5056 if (strcmp(entry
->name
, long_options
[i
].name
)) {
5061 * If the option takes no argument on the command line, we have to
5062 * check if the value is "true". We support non-zero numeric values,
5065 if (!long_options
[i
].has_arg
) {
5066 ret
= config_parse_value(entry
->value
);
5069 WARN("Invalid configuration value \"%s\" for option %s",
5070 entry
->value
, entry
->name
);
5072 /* False, skip boolean config option. */
5077 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
5081 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
5088 * daemon configuration loading and argument parsing
5090 static int set_options(int argc
, char **argv
)
5092 int ret
= 0, c
= 0, option_index
= 0;
5093 int orig_optopt
= optopt
, orig_optind
= optind
;
5095 const char *config_path
= NULL
;
5097 optstring
= utils_generate_optstring(long_options
,
5098 sizeof(long_options
) / sizeof(struct option
));
5104 /* Check for the --config option */
5105 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
5106 &option_index
)) != -1) {
5110 } else if (c
!= 'f') {
5111 /* if not equal to --config option. */
5115 if (lttng_is_setuid_setgid()) {
5116 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5119 config_path
= utils_expand_path(optarg
);
5121 ERR("Failed to resolve path: %s", optarg
);
5126 ret
= config_get_section_entries(config_path
, config_section_name
,
5127 config_entry_handler
, NULL
);
5130 ERR("Invalid configuration option at line %i", ret
);
5136 /* Reset getopt's global state */
5137 optopt
= orig_optopt
;
5138 optind
= orig_optind
;
5142 * getopt_long() will not set option_index if it encounters a
5145 c
= getopt_long(argc
, argv
, optstring
, long_options
,
5152 * Pass NULL as the long option name if popt left the index
5155 ret
= set_option(c
, optarg
,
5156 option_index
< 0 ? NULL
:
5157 long_options
[option_index
].name
);
5169 * Creates the two needed socket by the daemon.
5170 * apps_sock - The communication socket for all UST apps.
5171 * client_sock - The communication of the cli tool (lttng).
5173 static int init_daemon_socket(void)
5178 old_umask
= umask(0);
5180 /* Create client tool unix socket */
5181 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
5182 if (client_sock
< 0) {
5183 ERR("Create unix sock failed: %s", client_unix_sock_path
);
5188 /* Set the cloexec flag */
5189 ret
= utils_set_fd_cloexec(client_sock
);
5191 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5192 "Continuing but note that the consumer daemon will have a "
5193 "reference to this socket on exec()", client_sock
);
5196 /* File permission MUST be 660 */
5197 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5199 ERR("Set file permissions failed: %s", client_unix_sock_path
);
5204 /* Create the application unix socket */
5205 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
5206 if (apps_sock
< 0) {
5207 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
5212 /* Set the cloexec flag */
5213 ret
= utils_set_fd_cloexec(apps_sock
);
5215 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5216 "Continuing but note that the consumer daemon will have a "
5217 "reference to this socket on exec()", apps_sock
);
5220 /* File permission MUST be 666 */
5221 ret
= chmod(apps_unix_sock_path
,
5222 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5224 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
5229 DBG3("Session daemon client socket %d and application socket %d created",
5230 client_sock
, apps_sock
);
5238 * Check if the global socket is available, and if a daemon is answering at the
5239 * other side. If yes, error is returned.
5241 static int check_existing_daemon(void)
5243 /* Is there anybody out there ? */
5244 if (lttng_session_daemon_alive()) {
5252 * Set the tracing group gid onto the client socket.
5254 * Race window between mkdir and chown is OK because we are going from more
5255 * permissive (root.root) to less permissive (root.tracing).
5257 static int set_permissions(char *rundir
)
5262 gid
= utils_get_group_id(tracing_group_name
);
5264 /* Set lttng run dir */
5265 ret
= chown(rundir
, 0, gid
);
5267 ERR("Unable to set group on %s", rundir
);
5272 * Ensure all applications and tracing group can search the run
5273 * dir. Allow everyone to read the directory, since it does not
5274 * buy us anything to hide its content.
5276 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5278 ERR("Unable to set permissions on %s", rundir
);
5282 /* lttng client socket path */
5283 ret
= chown(client_unix_sock_path
, 0, gid
);
5285 ERR("Unable to set group on %s", client_unix_sock_path
);
5289 /* kconsumer error socket path */
5290 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5292 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5296 /* 64-bit ustconsumer error socket path */
5297 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5299 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5303 /* 32-bit ustconsumer compat32 error socket path */
5304 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5306 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5310 DBG("All permissions are set");
5316 * Create the lttng run directory needed for all global sockets and pipe.
5318 static int create_lttng_rundir(const char *rundir
)
5322 DBG3("Creating LTTng run directory: %s", rundir
);
5324 ret
= mkdir(rundir
, S_IRWXU
);
5326 if (errno
!= EEXIST
) {
5327 ERR("Unable to create %s", rundir
);
5339 * Setup sockets and directory needed by the kconsumerd communication with the
5342 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5346 char path
[PATH_MAX
];
5348 switch (consumer_data
->type
) {
5349 case LTTNG_CONSUMER_KERNEL
:
5350 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5352 case LTTNG_CONSUMER64_UST
:
5353 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5355 case LTTNG_CONSUMER32_UST
:
5356 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5359 ERR("Consumer type unknown");
5364 DBG2("Creating consumer directory: %s", path
);
5366 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5368 if (errno
!= EEXIST
) {
5370 ERR("Failed to create %s", path
);
5376 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5378 ERR("Unable to set group on %s", path
);
5384 /* Create the kconsumerd error unix socket */
5385 consumer_data
->err_sock
=
5386 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5387 if (consumer_data
->err_sock
< 0) {
5388 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5394 * Set the CLOEXEC flag. Return code is useless because either way, the
5397 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5399 PERROR("utils_set_fd_cloexec");
5400 /* continue anyway */
5403 /* File permission MUST be 660 */
5404 ret
= chmod(consumer_data
->err_unix_sock_path
,
5405 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5407 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5417 * Signal handler for the daemon
5419 * Simply stop all worker threads, leaving main() return gracefully after
5420 * joining all threads and calling cleanup().
5422 static void sighandler(int sig
)
5426 DBG("SIGINT caught");
5430 DBG("SIGTERM caught");
5434 CMM_STORE_SHARED(recv_child_signal
, 1);
5442 * Setup signal handler for :
5443 * SIGINT, SIGTERM, SIGPIPE
5445 static int set_signal_handler(void)
5448 struct sigaction sa
;
5451 if ((ret
= sigemptyset(&sigset
)) < 0) {
5452 PERROR("sigemptyset");
5456 sa
.sa_mask
= sigset
;
5459 sa
.sa_handler
= sighandler
;
5460 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5461 PERROR("sigaction");
5465 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5466 PERROR("sigaction");
5470 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5471 PERROR("sigaction");
5475 sa
.sa_handler
= SIG_IGN
;
5476 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5477 PERROR("sigaction");
5481 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5487 * Set open files limit to unlimited. This daemon can open a large number of
5488 * file descriptors in order to consumer multiple kernel traces.
5490 static void set_ulimit(void)
5495 /* The kernel does not allowed an infinite limit for open files */
5496 lim
.rlim_cur
= 65535;
5497 lim
.rlim_max
= 65535;
5499 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5501 PERROR("failed to set open files limit");
5506 * Write pidfile using the rundir and opt_pidfile.
5508 static int write_pidfile(void)
5511 char pidfile_path
[PATH_MAX
];
5516 if (lttng_strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
))) {
5521 /* Build pidfile path from rundir and opt_pidfile. */
5522 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5523 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5525 PERROR("snprintf pidfile path");
5531 * Create pid file in rundir.
5533 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5539 * Create lockfile using the rundir and return its fd.
5541 static int create_lockfile(void)
5544 char lockfile_path
[PATH_MAX
];
5546 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5551 ret
= utils_create_lock_file(lockfile_path
);
5557 * Write agent TCP port using the rundir.
5559 static int write_agent_port(void)
5562 char path
[PATH_MAX
];
5566 ret
= snprintf(path
, sizeof(path
), "%s/"
5567 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5569 PERROR("snprintf agent port path");
5574 * Create TCP agent port file in rundir.
5576 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5585 int main(int argc
, char **argv
)
5587 int ret
= 0, retval
= 0;
5589 const char *home_path
, *env_app_timeout
;
5591 init_kernel_workarounds();
5593 rcu_register_thread();
5595 if (set_signal_handler()) {
5597 goto exit_set_signal_handler
;
5600 setup_consumerd_path();
5602 page_size
= sysconf(_SC_PAGESIZE
);
5603 if (page_size
< 0) {
5604 PERROR("sysconf _SC_PAGESIZE");
5605 page_size
= LONG_MAX
;
5606 WARN("Fallback page size to %ld", page_size
);
5610 * Parse arguments and load the daemon configuration file.
5612 * We have an exit_options exit path to free memory reserved by
5613 * set_options. This is needed because the rest of sessiond_cleanup()
5614 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5615 * depends on set_options.
5618 if (set_options(argc
, argv
)) {
5624 if (opt_daemon
|| opt_background
) {
5627 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5635 * We are in the child. Make sure all other file descriptors are
5636 * closed, in case we are called with more opened file
5637 * descriptors than the standard ones.
5639 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5644 if (run_as_create_worker(argv
[0]) < 0) {
5645 goto exit_create_run_as_worker_cleanup
;
5649 * Starting from here, we can create threads. This needs to be after
5650 * lttng_daemonize due to RCU.
5654 * Initialize the health check subsystem. This call should set the
5655 * appropriate time values.
5657 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5658 if (!health_sessiond
) {
5659 PERROR("health_app_create error");
5661 goto exit_health_sessiond_cleanup
;
5664 if (init_ht_cleanup_quit_pipe()) {
5666 goto exit_ht_cleanup_quit_pipe
;
5669 /* Setup the thread ht_cleanup communication pipe. */
5670 if (utils_create_pipe_cloexec(ht_cleanup_pipe
)) {
5672 goto exit_ht_cleanup_pipe
;
5675 /* Set up max poll set size */
5676 if (lttng_poll_set_max_size()) {
5678 goto exit_set_max_size
;
5681 /* Create thread to clean up RCU hash tables */
5682 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5683 thread_ht_cleanup
, (void *) NULL
);
5686 PERROR("pthread_create ht_cleanup");
5688 goto exit_ht_cleanup
;
5691 /* Create thread quit pipe */
5692 if (init_thread_quit_pipe()) {
5694 goto exit_init_data
;
5697 /* Check if daemon is UID = 0 */
5698 is_root
= !getuid();
5701 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5704 goto exit_init_data
;
5707 /* Create global run dir with root access */
5708 if (create_lttng_rundir(rundir
)) {
5710 goto exit_init_data
;
5713 if (strlen(apps_unix_sock_path
) == 0) {
5714 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5715 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5718 goto exit_init_data
;
5722 if (strlen(client_unix_sock_path
) == 0) {
5723 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5724 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5727 goto exit_init_data
;
5731 /* Set global SHM for ust */
5732 if (strlen(wait_shm_path
) == 0) {
5733 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5734 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5737 goto exit_init_data
;
5741 if (strlen(health_unix_sock_path
) == 0) {
5742 ret
= snprintf(health_unix_sock_path
,
5743 sizeof(health_unix_sock_path
),
5744 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5747 goto exit_init_data
;
5751 /* Setup kernel consumerd path */
5752 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5753 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5756 goto exit_init_data
;
5758 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5759 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5762 goto exit_init_data
;
5765 DBG2("Kernel consumer err path: %s",
5766 kconsumer_data
.err_unix_sock_path
);
5767 DBG2("Kernel consumer cmd path: %s",
5768 kconsumer_data
.cmd_unix_sock_path
);
5770 home_path
= utils_get_home_dir();
5771 if (home_path
== NULL
) {
5772 /* TODO: Add --socket PATH option */
5773 ERR("Can't get HOME directory for sockets creation.");
5775 goto exit_init_data
;
5779 * Create rundir from home path. This will create something like
5782 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5785 goto exit_init_data
;
5788 if (create_lttng_rundir(rundir
)) {
5790 goto exit_init_data
;
5793 if (strlen(apps_unix_sock_path
) == 0) {
5794 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5795 DEFAULT_HOME_APPS_UNIX_SOCK
,
5799 goto exit_init_data
;
5803 /* Set the cli tool unix socket path */
5804 if (strlen(client_unix_sock_path
) == 0) {
5805 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5806 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5810 goto exit_init_data
;
5814 /* Set global SHM for ust */
5815 if (strlen(wait_shm_path
) == 0) {
5816 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5817 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5821 goto exit_init_data
;
5825 /* Set health check Unix path */
5826 if (strlen(health_unix_sock_path
) == 0) {
5827 ret
= snprintf(health_unix_sock_path
,
5828 sizeof(health_unix_sock_path
),
5829 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5833 goto exit_init_data
;
5838 lockfile_fd
= create_lockfile();
5839 if (lockfile_fd
< 0) {
5841 goto exit_init_data
;
5844 /* Set consumer initial state */
5845 kernel_consumerd_state
= CONSUMER_STOPPED
;
5846 ust_consumerd_state
= CONSUMER_STOPPED
;
5848 DBG("Client socket path %s", client_unix_sock_path
);
5849 DBG("Application socket path %s", apps_unix_sock_path
);
5850 DBG("Application wait path %s", wait_shm_path
);
5851 DBG("LTTng run directory path: %s", rundir
);
5853 /* 32 bits consumerd path setup */
5854 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5855 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5857 PERROR("snprintf 32-bit consumer error socket path");
5859 goto exit_init_data
;
5861 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5862 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5864 PERROR("snprintf 32-bit consumer command socket path");
5866 goto exit_init_data
;
5869 DBG2("UST consumer 32 bits err path: %s",
5870 ustconsumer32_data
.err_unix_sock_path
);
5871 DBG2("UST consumer 32 bits cmd path: %s",
5872 ustconsumer32_data
.cmd_unix_sock_path
);
5874 /* 64 bits consumerd path setup */
5875 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5876 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5878 PERROR("snprintf 64-bit consumer error socket path");
5880 goto exit_init_data
;
5882 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5883 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5885 PERROR("snprintf 64-bit consumer command socket path");
5887 goto exit_init_data
;
5890 DBG2("UST consumer 64 bits err path: %s",
5891 ustconsumer64_data
.err_unix_sock_path
);
5892 DBG2("UST consumer 64 bits cmd path: %s",
5893 ustconsumer64_data
.cmd_unix_sock_path
);
5896 * See if daemon already exist.
5898 if (check_existing_daemon()) {
5899 ERR("Already running daemon.\n");
5901 * We do not goto exit because we must not cleanup()
5902 * because a daemon is already running.
5905 goto exit_init_data
;
5909 * Init UST app hash table. Alloc hash table before this point since
5910 * cleanup() can get called after that point.
5912 if (ust_app_ht_alloc()) {
5913 ERR("Failed to allocate UST app hash table");
5915 goto exit_init_data
;
5919 * Initialize agent app hash table. We allocate the hash table here
5920 * since cleanup() can get called after this point.
5922 if (agent_app_ht_alloc()) {
5923 ERR("Failed to allocate Agent app hash table");
5925 goto exit_init_data
;
5929 * These actions must be executed as root. We do that *after* setting up
5930 * the sockets path because we MUST make the check for another daemon using
5931 * those paths *before* trying to set the kernel consumer sockets and init
5935 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
5937 goto exit_init_data
;
5940 /* Setup kernel tracer */
5941 if (!opt_no_kernel
) {
5942 init_kernel_tracer();
5943 if (kernel_tracer_fd
>= 0) {
5944 ret
= syscall_init_table();
5946 ERR("Unable to populate syscall table. "
5947 "Syscall tracing won't work "
5948 "for this session daemon.");
5953 /* Set ulimit for open files */
5956 /* init lttng_fd tracking must be done after set_ulimit. */
5959 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
5961 goto exit_init_data
;
5964 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
5966 goto exit_init_data
;
5969 /* Setup the needed unix socket */
5970 if (init_daemon_socket()) {
5972 goto exit_init_data
;
5975 /* Set credentials to socket */
5976 if (is_root
&& set_permissions(rundir
)) {
5978 goto exit_init_data
;
5981 /* Get parent pid if -S, --sig-parent is specified. */
5982 if (opt_sig_parent
) {
5986 /* Setup the kernel pipe for waking up the kernel thread */
5987 if (is_root
&& !opt_no_kernel
) {
5988 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5990 goto exit_init_data
;
5994 /* Setup the thread apps communication pipe. */
5995 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
5997 goto exit_init_data
;
6000 /* Setup the thread apps notify communication pipe. */
6001 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
6003 goto exit_init_data
;
6006 /* Initialize global buffer per UID and PID registry. */
6007 buffer_reg_init_uid_registry();
6008 buffer_reg_init_pid_registry();
6010 /* Init UST command queue. */
6011 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
6014 * Get session list pointer. This pointer MUST NOT be free'd. This list
6015 * is statically declared in session.c
6017 session_list_ptr
= session_get_list();
6021 /* Check for the application socket timeout env variable. */
6022 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
6023 if (env_app_timeout
) {
6024 app_socket_timeout
= atoi(env_app_timeout
);
6026 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
6029 ret
= write_pidfile();
6031 ERR("Error in write_pidfile");
6033 goto exit_init_data
;
6035 ret
= write_agent_port();
6037 ERR("Error in write_agent_port");
6039 goto exit_init_data
;
6042 /* Initialize communication library */
6044 /* Initialize TCP timeout values */
6045 lttcomm_inet_init();
6047 if (load_session_init_data(&load_info
) < 0) {
6049 goto exit_init_data
;
6051 load_info
->path
= opt_load_session_path
;
6053 /* Create health-check thread */
6054 ret
= pthread_create(&health_thread
, NULL
,
6055 thread_manage_health
, (void *) NULL
);
6058 PERROR("pthread_create health");
6063 /* Create thread to manage the client socket */
6064 ret
= pthread_create(&client_thread
, NULL
,
6065 thread_manage_clients
, (void *) NULL
);
6068 PERROR("pthread_create clients");
6073 /* Create thread to dispatch registration */
6074 ret
= pthread_create(&dispatch_thread
, NULL
,
6075 thread_dispatch_ust_registration
, (void *) NULL
);
6078 PERROR("pthread_create dispatch");
6083 /* Create thread to manage application registration. */
6084 ret
= pthread_create(®_apps_thread
, NULL
,
6085 thread_registration_apps
, (void *) NULL
);
6088 PERROR("pthread_create registration");
6093 /* Create thread to manage application socket */
6094 ret
= pthread_create(&apps_thread
, NULL
,
6095 thread_manage_apps
, (void *) NULL
);
6098 PERROR("pthread_create apps");
6103 /* Create thread to manage application notify socket */
6104 ret
= pthread_create(&apps_notify_thread
, NULL
,
6105 ust_thread_manage_notify
, (void *) NULL
);
6108 PERROR("pthread_create notify");
6110 goto exit_apps_notify
;
6113 /* Create agent registration thread. */
6114 ret
= pthread_create(&agent_reg_thread
, NULL
,
6115 agent_thread_manage_registration
, (void *) NULL
);
6118 PERROR("pthread_create agent");
6120 goto exit_agent_reg
;
6123 /* Don't start this thread if kernel tracing is not requested nor root */
6124 if (is_root
&& !opt_no_kernel
) {
6125 /* Create kernel thread to manage kernel event */
6126 ret
= pthread_create(&kernel_thread
, NULL
,
6127 thread_manage_kernel
, (void *) NULL
);
6130 PERROR("pthread_create kernel");
6136 /* Create session loading thread. */
6137 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
6141 PERROR("pthread_create load_session_thread");
6143 goto exit_load_session
;
6147 * This is where we start awaiting program completion (e.g. through
6148 * signal that asks threads to teardown).
6151 ret
= pthread_join(load_session_thread
, &status
);
6154 PERROR("pthread_join load_session_thread");
6159 if (is_root
&& !opt_no_kernel
) {
6160 ret
= pthread_join(kernel_thread
, &status
);
6163 PERROR("pthread_join");
6169 ret
= pthread_join(agent_reg_thread
, &status
);
6172 PERROR("pthread_join agent");
6177 ret
= pthread_join(apps_notify_thread
, &status
);
6180 PERROR("pthread_join apps notify");
6185 ret
= pthread_join(apps_thread
, &status
);
6188 PERROR("pthread_join apps");
6193 ret
= pthread_join(reg_apps_thread
, &status
);
6196 PERROR("pthread_join");
6202 * Join dispatch thread after joining reg_apps_thread to ensure
6203 * we don't leak applications in the queue.
6205 ret
= pthread_join(dispatch_thread
, &status
);
6208 PERROR("pthread_join");
6213 ret
= pthread_join(client_thread
, &status
);
6216 PERROR("pthread_join");
6221 ret
= pthread_join(health_thread
, &status
);
6224 PERROR("pthread_join health thread");
6231 * Wait for all pending call_rcu work to complete before tearing
6232 * down data structures. call_rcu worker may be trying to
6233 * perform lookups in those structures.
6237 * sessiond_cleanup() is called when no other thread is running, except
6238 * the ht_cleanup thread, which is needed to destroy the hash tables.
6240 rcu_thread_online();
6242 rcu_thread_offline();
6243 rcu_unregister_thread();
6246 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6247 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6248 * the queue is empty before shutting down the clean-up thread.
6252 ret
= notify_thread_pipe(ht_cleanup_quit_pipe
[1]);
6254 ERR("write error on ht_cleanup quit pipe");
6258 ret
= pthread_join(ht_cleanup_thread
, &status
);
6261 PERROR("pthread_join ht cleanup thread");
6267 utils_close_pipe(ht_cleanup_pipe
);
6268 exit_ht_cleanup_pipe
:
6271 * Close the ht_cleanup quit pipe.
6273 utils_close_pipe(ht_cleanup_quit_pipe
);
6274 exit_ht_cleanup_quit_pipe
:
6276 health_app_destroy(health_sessiond
);
6277 exit_health_sessiond_cleanup
:
6278 exit_create_run_as_worker_cleanup
:
6281 sessiond_cleanup_options();
6283 exit_set_signal_handler
: