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>
41 #include <common/common.h>
42 #include <common/compat/socket.h>
43 #include <common/compat/getenv.h>
44 #include <common/defaults.h>
45 #include <common/kernel-consumer/kernel-consumer.h>
46 #include <common/futex.h>
47 #include <common/relayd/relayd.h>
48 #include <common/utils.h>
49 #include <common/daemonize.h>
50 #include <common/config/session-config.h>
52 #include "lttng-sessiond.h"
53 #include "buffer-registry.h"
60 #include "kernel-consumer.h"
64 #include "ust-consumer.h"
67 #include "health-sessiond.h"
68 #include "testpoint.h"
69 #include "ust-thread.h"
70 #include "agent-thread.h"
72 #include "load-session-thread.h"
76 #define CONSUMERD_FILE "lttng-consumerd"
79 static const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
80 static int tracing_group_name_override
;
81 static char *opt_pidfile
;
82 static int opt_sig_parent
;
83 static int opt_verbose_consumer
;
84 static int opt_daemon
, opt_background
;
85 static int opt_no_kernel
;
86 static char *opt_load_session_path
;
87 static pid_t ppid
; /* Parent PID for --sig-parent option */
88 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
90 static int lockfile_fd
= -1;
92 /* Set to 1 when a SIGUSR1 signal is received. */
93 static int recv_child_signal
;
96 * Consumer daemon specific control data. Every value not initialized here is
97 * set to 0 by the static definition.
99 static struct consumer_data kconsumer_data
= {
100 .type
= LTTNG_CONSUMER_KERNEL
,
101 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
102 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
105 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
106 .lock
= PTHREAD_MUTEX_INITIALIZER
,
107 .cond
= PTHREAD_COND_INITIALIZER
,
108 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
110 static struct consumer_data ustconsumer64_data
= {
111 .type
= LTTNG_CONSUMER64_UST
,
112 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
113 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
116 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
117 .lock
= PTHREAD_MUTEX_INITIALIZER
,
118 .cond
= PTHREAD_COND_INITIALIZER
,
119 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
121 static struct consumer_data ustconsumer32_data
= {
122 .type
= LTTNG_CONSUMER32_UST
,
123 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
124 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
127 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
128 .lock
= PTHREAD_MUTEX_INITIALIZER
,
129 .cond
= PTHREAD_COND_INITIALIZER
,
130 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
133 /* Command line options */
134 static const struct option long_options
[] = {
135 { "client-sock", required_argument
, 0, 'c' },
136 { "apps-sock", required_argument
, 0, 'a' },
137 { "kconsumerd-cmd-sock", required_argument
, 0, '\0' },
138 { "kconsumerd-err-sock", required_argument
, 0, '\0' },
139 { "ustconsumerd32-cmd-sock", required_argument
, 0, '\0' },
140 { "ustconsumerd32-err-sock", required_argument
, 0, '\0' },
141 { "ustconsumerd64-cmd-sock", required_argument
, 0, '\0' },
142 { "ustconsumerd64-err-sock", required_argument
, 0, '\0' },
143 { "consumerd32-path", required_argument
, 0, '\0' },
144 { "consumerd32-libdir", required_argument
, 0, '\0' },
145 { "consumerd64-path", required_argument
, 0, '\0' },
146 { "consumerd64-libdir", required_argument
, 0, '\0' },
147 { "daemonize", no_argument
, 0, 'd' },
148 { "background", no_argument
, 0, 'b' },
149 { "sig-parent", no_argument
, 0, 'S' },
150 { "help", no_argument
, 0, 'h' },
151 { "group", required_argument
, 0, 'g' },
152 { "version", no_argument
, 0, 'V' },
153 { "quiet", no_argument
, 0, 'q' },
154 { "verbose", no_argument
, 0, 'v' },
155 { "verbose-consumer", no_argument
, 0, '\0' },
156 { "no-kernel", no_argument
, 0, '\0' },
157 { "pidfile", required_argument
, 0, 'p' },
158 { "agent-tcp-port", required_argument
, 0, '\0' },
159 { "config", required_argument
, 0, 'f' },
160 { "load", required_argument
, 0, 'l' },
161 { "kmod-probes", required_argument
, 0, '\0' },
162 { "extra-kmod-probes", required_argument
, 0, '\0' },
166 /* Command line options to ignore from configuration file */
167 static const char *config_ignore_options
[] = { "help", "version", "config" };
169 /* Shared between threads */
170 static int dispatch_thread_exit
;
172 /* Global application Unix socket path */
173 static char apps_unix_sock_path
[PATH_MAX
];
174 /* Global client Unix socket path */
175 static char client_unix_sock_path
[PATH_MAX
];
176 /* global wait shm path for UST */
177 static char wait_shm_path
[PATH_MAX
];
178 /* Global health check unix path */
179 static char health_unix_sock_path
[PATH_MAX
];
181 /* Sockets and FDs */
182 static int client_sock
= -1;
183 static int apps_sock
= -1;
184 int kernel_tracer_fd
= -1;
185 static int kernel_poll_pipe
[2] = { -1, -1 };
188 * Quit pipe for all threads. This permits a single cancellation point
189 * for all threads when receiving an event on the pipe.
191 static int thread_quit_pipe
[2] = { -1, -1 };
192 static int ht_cleanup_quit_pipe
[2] = { -1, -1 };
195 * This pipe is used to inform the thread managing application communication
196 * that a command is queued and ready to be processed.
198 static int apps_cmd_pipe
[2] = { -1, -1 };
200 int apps_cmd_notify_pipe
[2] = { -1, -1 };
202 /* Pthread, Mutexes and Semaphores */
203 static pthread_t apps_thread
;
204 static pthread_t apps_notify_thread
;
205 static pthread_t reg_apps_thread
;
206 static pthread_t client_thread
;
207 static pthread_t kernel_thread
;
208 static pthread_t dispatch_thread
;
209 static pthread_t health_thread
;
210 static pthread_t ht_cleanup_thread
;
211 static pthread_t agent_reg_thread
;
212 static pthread_t load_session_thread
;
215 * UST registration command queue. This queue is tied with a futex and uses a N
216 * wakers / 1 waiter implemented and detailed in futex.c/.h
218 * The thread_registration_apps and thread_dispatch_ust_registration uses this
219 * queue along with the wait/wake scheme. The thread_manage_apps receives down
220 * the line new application socket and monitors it for any I/O error or clean
221 * close that triggers an unregistration of the application.
223 static struct ust_cmd_queue ust_cmd_queue
;
226 * Pointer initialized before thread creation.
228 * This points to the tracing session list containing the session count and a
229 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
230 * MUST NOT be taken if you call a public function in session.c.
232 * The lock is nested inside the structure: session_list_ptr->lock. Please use
233 * session_lock_list and session_unlock_list for lock acquisition.
235 static struct ltt_session_list
*session_list_ptr
;
237 int ust_consumerd64_fd
= -1;
238 int ust_consumerd32_fd
= -1;
240 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
241 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
242 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
243 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
244 static int consumerd32_bin_override
;
245 static int consumerd64_bin_override
;
246 static int consumerd32_libdir_override
;
247 static int consumerd64_libdir_override
;
249 static const char *module_proc_lttng
= "/proc/lttng";
252 * Consumer daemon state which is changed when spawning it, killing it or in
253 * case of a fatal error.
255 enum consumerd_state
{
256 CONSUMER_STARTED
= 1,
257 CONSUMER_STOPPED
= 2,
262 * This consumer daemon state is used to validate if a client command will be
263 * able to reach the consumer. If not, the client is informed. For instance,
264 * doing a "lttng start" when the consumer state is set to ERROR will return an
265 * error to the client.
267 * The following example shows a possible race condition of this scheme:
269 * consumer thread error happens
271 * client cmd checks state -> still OK
272 * consumer thread exit, sets error
273 * client cmd try to talk to consumer
276 * However, since the consumer is a different daemon, we have no way of making
277 * sure the command will reach it safely even with this state flag. This is why
278 * we consider that up to the state validation during command processing, the
279 * command is safe. After that, we can not guarantee the correctness of the
280 * client request vis-a-vis the consumer.
282 static enum consumerd_state ust_consumerd_state
;
283 static enum consumerd_state kernel_consumerd_state
;
286 * Socket timeout for receiving and sending in seconds.
288 static int app_socket_timeout
;
290 /* Set in main() with the current page size. */
293 /* Application health monitoring */
294 struct health_app
*health_sessiond
;
296 /* Agent TCP port for registration. Used by the agent thread. */
297 unsigned int agent_tcp_port
= DEFAULT_AGENT_TCP_PORT
;
299 /* Am I root or not. */
300 int is_root
; /* Set to 1 if the daemon is running as root */
302 const char * const config_section_name
= "sessiond";
304 /* Load session thread information to operate. */
305 struct load_session_thread_data
*load_info
;
307 /* Global hash tables */
308 struct lttng_ht
*agent_apps_ht_by_sock
= NULL
;
311 * Whether sessiond is ready for commands/health check requests.
312 * NR_LTTNG_SESSIOND_READY must match the number of calls to
313 * sessiond_notify_ready().
315 #define NR_LTTNG_SESSIOND_READY 3
316 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
318 /* Notify parents that we are ready for cmd and health check */
320 void sessiond_notify_ready(void)
322 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
324 * Notify parent pid that we are ready to accept command
325 * for client side. This ppid is the one from the
326 * external process that spawned us.
328 if (opt_sig_parent
) {
333 * Notify the parent of the fork() process that we are
336 if (opt_daemon
|| opt_background
) {
337 kill(child_ppid
, SIGUSR1
);
343 void setup_consumerd_path(void)
345 const char *bin
, *libdir
;
348 * Allow INSTALL_BIN_PATH to be used as a target path for the
349 * native architecture size consumer if CONFIG_CONSUMER*_PATH
350 * has not been defined.
352 #if (CAA_BITS_PER_LONG == 32)
353 if (!consumerd32_bin
[0]) {
354 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
356 if (!consumerd32_libdir
[0]) {
357 consumerd32_libdir
= INSTALL_LIB_PATH
;
359 #elif (CAA_BITS_PER_LONG == 64)
360 if (!consumerd64_bin
[0]) {
361 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
363 if (!consumerd64_libdir
[0]) {
364 consumerd64_libdir
= INSTALL_LIB_PATH
;
367 #error "Unknown bitness"
371 * runtime env. var. overrides the build default.
373 bin
= lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
375 consumerd32_bin
= bin
;
377 bin
= lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
379 consumerd64_bin
= bin
;
381 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
383 consumerd32_libdir
= libdir
;
385 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
387 consumerd64_libdir
= libdir
;
392 int __sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
,
399 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
405 ret
= lttng_poll_add(events
, a_pipe
[0], LPOLLIN
| LPOLLERR
);
417 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
419 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
421 return __sessiond_set_thread_pollset(events
, size
, thread_quit_pipe
);
425 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
427 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event
*events
,
430 return __sessiond_set_thread_pollset(events
, size
,
431 ht_cleanup_quit_pipe
);
435 int __sessiond_check_thread_quit_pipe(int fd
, uint32_t events
, int a_pipe
)
437 if (fd
== a_pipe
&& (events
& LPOLLIN
)) {
444 * Check if the thread quit pipe was triggered.
446 * Return 1 if it was triggered else 0;
448 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
450 return __sessiond_check_thread_quit_pipe(fd
, events
,
451 thread_quit_pipe
[0]);
455 * Check if the ht_cleanup thread quit pipe was triggered.
457 * Return 1 if it was triggered else 0;
459 int sessiond_check_ht_cleanup_quit(int fd
, uint32_t events
)
461 return __sessiond_check_thread_quit_pipe(fd
, events
,
462 ht_cleanup_quit_pipe
[0]);
466 * Init thread quit pipe.
468 * Return -1 on error or 0 if all pipes are created.
470 static int __init_thread_quit_pipe(int *a_pipe
)
476 PERROR("thread quit pipe");
480 for (i
= 0; i
< 2; i
++) {
481 ret
= fcntl(a_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
492 static int init_thread_quit_pipe(void)
494 return __init_thread_quit_pipe(thread_quit_pipe
);
497 static int init_ht_cleanup_quit_pipe(void)
499 return __init_thread_quit_pipe(ht_cleanup_quit_pipe
);
503 * Stop all threads by closing the thread quit pipe.
505 static void stop_threads(void)
509 /* Stopping all threads */
510 DBG("Terminating all threads");
511 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
513 ERR("write error on thread quit pipe");
516 /* Dispatch thread */
517 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
518 futex_nto1_wake(&ust_cmd_queue
.futex
);
522 * Close every consumer sockets.
524 static void close_consumer_sockets(void)
528 if (kconsumer_data
.err_sock
>= 0) {
529 ret
= close(kconsumer_data
.err_sock
);
531 PERROR("kernel consumer err_sock close");
534 if (ustconsumer32_data
.err_sock
>= 0) {
535 ret
= close(ustconsumer32_data
.err_sock
);
537 PERROR("UST consumerd32 err_sock close");
540 if (ustconsumer64_data
.err_sock
>= 0) {
541 ret
= close(ustconsumer64_data
.err_sock
);
543 PERROR("UST consumerd64 err_sock close");
546 if (kconsumer_data
.cmd_sock
>= 0) {
547 ret
= close(kconsumer_data
.cmd_sock
);
549 PERROR("kernel consumer cmd_sock close");
552 if (ustconsumer32_data
.cmd_sock
>= 0) {
553 ret
= close(ustconsumer32_data
.cmd_sock
);
555 PERROR("UST consumerd32 cmd_sock close");
558 if (ustconsumer64_data
.cmd_sock
>= 0) {
559 ret
= close(ustconsumer64_data
.cmd_sock
);
561 PERROR("UST consumerd64 cmd_sock close");
567 * Generate the full lock file path using the rundir.
569 * Return the snprintf() return value thus a negative value is an error.
571 static int generate_lock_file_path(char *path
, size_t len
)
578 /* Build lockfile path from rundir. */
579 ret
= snprintf(path
, len
, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE
, rundir
);
581 PERROR("snprintf lockfile path");
588 * Wait on consumer process termination.
590 * Need to be called with the consumer data lock held or from a context
591 * ensuring no concurrent access to data (e.g: cleanup).
593 static void wait_consumer(struct consumer_data
*consumer_data
)
598 if (consumer_data
->pid
<= 0) {
602 DBG("Waiting for complete teardown of consumerd (PID: %d)",
604 ret
= waitpid(consumer_data
->pid
, &status
, 0);
606 PERROR("consumerd waitpid pid: %d", consumer_data
->pid
)
608 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 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1907 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1908 goto error_testpoint
;
1911 health_code_update();
1913 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1915 DBG("[thread] Dispatch UST command started");
1917 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1918 health_code_update();
1920 /* Atomically prepare the queue futex */
1921 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1924 struct ust_app
*app
= NULL
;
1928 * Make sure we don't have node(s) that have hung up before receiving
1929 * the notify socket. This is to clean the list in order to avoid
1930 * memory leaks from notify socket that are never seen.
1932 sanitize_wait_queue(&wait_queue
);
1934 health_code_update();
1935 /* Dequeue command for registration */
1936 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1938 DBG("Woken up but nothing in the UST command queue");
1939 /* Continue thread execution */
1943 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1945 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1946 " gid:%d sock:%d name:%s (version %d.%d)",
1947 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1948 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1949 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1950 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1952 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1953 wait_node
= zmalloc(sizeof(*wait_node
));
1955 PERROR("zmalloc wait_node dispatch");
1956 ret
= close(ust_cmd
->sock
);
1958 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1960 lttng_fd_put(LTTNG_FD_APPS
, 1);
1964 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1966 /* Create application object if socket is CMD. */
1967 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1969 if (!wait_node
->app
) {
1970 ret
= close(ust_cmd
->sock
);
1972 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1974 lttng_fd_put(LTTNG_FD_APPS
, 1);
1980 * Add application to the wait queue so we can set the notify
1981 * socket before putting this object in the global ht.
1983 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1988 * We have to continue here since we don't have the notify
1989 * socket and the application MUST be added to the hash table
1990 * only at that moment.
1995 * Look for the application in the local wait queue and set the
1996 * notify socket if found.
1998 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1999 &wait_queue
.head
, head
) {
2000 health_code_update();
2001 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
2002 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
2003 cds_list_del(&wait_node
->head
);
2005 app
= wait_node
->app
;
2007 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
2013 * With no application at this stage the received socket is
2014 * basically useless so close it before we free the cmd data
2015 * structure for good.
2018 ret
= close(ust_cmd
->sock
);
2020 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
2022 lttng_fd_put(LTTNG_FD_APPS
, 1);
2029 * @session_lock_list
2031 * Lock the global session list so from the register up to the
2032 * registration done message, no thread can see the application
2033 * and change its state.
2035 session_lock_list();
2039 * Add application to the global hash table. This needs to be
2040 * done before the update to the UST registry can locate the
2045 /* Set app version. This call will print an error if needed. */
2046 (void) ust_app_version(app
);
2048 /* Send notify socket through the notify pipe. */
2049 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
2053 session_unlock_list();
2055 * No notify thread, stop the UST tracing. However, this is
2056 * not an internal error of the this thread thus setting
2057 * the health error code to a normal exit.
2064 * Update newly registered application with the tracing
2065 * registry info already enabled information.
2067 update_ust_app(app
->sock
);
2070 * Don't care about return value. Let the manage apps threads
2071 * handle app unregistration upon socket close.
2073 (void) ust_app_register_done(app
);
2076 * Even if the application socket has been closed, send the app
2077 * to the thread and unregistration will take place at that
2080 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
2083 session_unlock_list();
2085 * No apps. thread, stop the UST tracing. However, this is
2086 * not an internal error of the this thread thus setting
2087 * the health error code to a normal exit.
2094 session_unlock_list();
2096 } while (node
!= NULL
);
2098 health_poll_entry();
2099 /* Futex wait on queue. Blocking call on futex() */
2100 futex_nto1_wait(&ust_cmd_queue
.futex
);
2103 /* Normal exit, no error */
2107 /* Clean up wait queue. */
2108 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2109 &wait_queue
.head
, head
) {
2110 cds_list_del(&wait_node
->head
);
2115 /* Empty command queue. */
2117 /* Dequeue command for registration */
2118 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2122 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2123 ret
= close(ust_cmd
->sock
);
2125 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2127 lttng_fd_put(LTTNG_FD_APPS
, 1);
2132 DBG("Dispatch thread dying");
2135 ERR("Health error occurred in %s", __func__
);
2137 health_unregister(health_sessiond
);
2142 * This thread manage application registration.
2144 static void *thread_registration_apps(void *data
)
2146 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2147 uint32_t revents
, nb_fd
;
2148 struct lttng_poll_event events
;
2150 * Get allocated in this thread, enqueued to a global queue, dequeued and
2151 * freed in the manage apps thread.
2153 struct ust_command
*ust_cmd
= NULL
;
2155 DBG("[thread] Manage application registration started");
2157 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2159 if (testpoint(sessiond_thread_registration_apps
)) {
2160 goto error_testpoint
;
2163 ret
= lttcomm_listen_unix_sock(apps_sock
);
2169 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2170 * more will be added to this poll set.
2172 ret
= sessiond_set_thread_pollset(&events
, 2);
2174 goto error_create_poll
;
2177 /* Add the application registration socket */
2178 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2180 goto error_poll_add
;
2183 /* Notify all applications to register */
2184 ret
= notify_ust_apps(1);
2186 ERR("Failed to notify applications or create the wait shared memory.\n"
2187 "Execution continues but there might be problem for already\n"
2188 "running applications that wishes to register.");
2192 DBG("Accepting application registration");
2194 /* Inifinite blocking call, waiting for transmission */
2196 health_poll_entry();
2197 ret
= lttng_poll_wait(&events
, -1);
2201 * Restart interrupted system call.
2203 if (errno
== EINTR
) {
2211 for (i
= 0; i
< nb_fd
; i
++) {
2212 health_code_update();
2214 /* Fetch once the poll data */
2215 revents
= LTTNG_POLL_GETEV(&events
, i
);
2216 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2219 /* No activity for this FD (poll implementation). */
2223 /* Thread quit pipe has been closed. Killing thread. */
2224 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2230 /* Event on the registration socket */
2231 if (pollfd
== apps_sock
) {
2232 if (revents
& LPOLLIN
) {
2233 sock
= lttcomm_accept_unix_sock(apps_sock
);
2239 * Set socket timeout for both receiving and ending.
2240 * app_socket_timeout is in seconds, whereas
2241 * lttcomm_setsockopt_rcv_timeout and
2242 * lttcomm_setsockopt_snd_timeout expect msec as
2245 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2246 app_socket_timeout
* 1000);
2247 (void) lttcomm_setsockopt_snd_timeout(sock
,
2248 app_socket_timeout
* 1000);
2251 * Set the CLOEXEC flag. Return code is useless because
2252 * either way, the show must go on.
2254 (void) utils_set_fd_cloexec(sock
);
2256 /* Create UST registration command for enqueuing */
2257 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2258 if (ust_cmd
== NULL
) {
2259 PERROR("ust command zmalloc");
2268 * Using message-based transmissions to ensure we don't
2269 * have to deal with partially received messages.
2271 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2273 ERR("Exhausted file descriptors allowed for applications.");
2283 health_code_update();
2284 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2287 /* Close socket of the application. */
2292 lttng_fd_put(LTTNG_FD_APPS
, 1);
2296 health_code_update();
2298 ust_cmd
->sock
= sock
;
2301 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2302 " gid:%d sock:%d name:%s (version %d.%d)",
2303 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2304 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2305 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2306 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2309 * Lock free enqueue the registration request. The red pill
2310 * has been taken! This apps will be part of the *system*.
2312 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2315 * Wake the registration queue futex. Implicit memory
2316 * barrier with the exchange in cds_wfcq_enqueue.
2318 futex_nto1_wake(&ust_cmd_queue
.futex
);
2319 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2320 ERR("Register apps socket poll error");
2323 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2332 /* Notify that the registration thread is gone */
2335 if (apps_sock
>= 0) {
2336 ret
= close(apps_sock
);
2346 lttng_fd_put(LTTNG_FD_APPS
, 1);
2348 unlink(apps_unix_sock_path
);
2351 lttng_poll_clean(&events
);
2355 DBG("UST Registration thread cleanup complete");
2358 ERR("Health error occurred in %s", __func__
);
2360 health_unregister(health_sessiond
);
2366 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2367 * exec or it will fails.
2369 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2372 struct timespec timeout
;
2375 * Make sure we set the readiness flag to 0 because we are NOT ready.
2376 * This access to consumer_thread_is_ready does not need to be
2377 * protected by consumer_data.cond_mutex (yet) since the consumer
2378 * management thread has not been started at this point.
2380 consumer_data
->consumer_thread_is_ready
= 0;
2382 /* Setup pthread condition */
2383 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2386 PERROR("pthread_condattr_init consumer data");
2391 * Set the monotonic clock in order to make sure we DO NOT jump in time
2392 * between the clock_gettime() call and the timedwait call. See bug #324
2393 * for a more details and how we noticed it.
2395 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2398 PERROR("pthread_condattr_setclock consumer data");
2402 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2405 PERROR("pthread_cond_init consumer data");
2409 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2413 PERROR("pthread_create consumer");
2418 /* We are about to wait on a pthread condition */
2419 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2421 /* Get time for sem_timedwait absolute timeout */
2422 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2424 * Set the timeout for the condition timed wait even if the clock gettime
2425 * call fails since we might loop on that call and we want to avoid to
2426 * increment the timeout too many times.
2428 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2431 * The following loop COULD be skipped in some conditions so this is why we
2432 * set ret to 0 in order to make sure at least one round of the loop is
2438 * Loop until the condition is reached or when a timeout is reached. Note
2439 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2440 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2441 * possible. This loop does not take any chances and works with both of
2444 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2445 if (clock_ret
< 0) {
2446 PERROR("clock_gettime spawn consumer");
2447 /* Infinite wait for the consumerd thread to be ready */
2448 ret
= pthread_cond_wait(&consumer_data
->cond
,
2449 &consumer_data
->cond_mutex
);
2451 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2452 &consumer_data
->cond_mutex
, &timeout
);
2456 /* Release the pthread condition */
2457 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2461 if (ret
== ETIMEDOUT
) {
2465 * Call has timed out so we kill the kconsumerd_thread and return
2468 ERR("Condition timed out. The consumer thread was never ready."
2470 pth_ret
= pthread_cancel(consumer_data
->thread
);
2472 PERROR("pthread_cancel consumer thread");
2475 PERROR("pthread_cond_wait failed consumer thread");
2477 /* Caller is expecting a negative value on failure. */
2482 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2483 if (consumer_data
->pid
== 0) {
2484 ERR("Consumerd did not start");
2485 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2488 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2497 * Join consumer thread
2499 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2503 /* Consumer pid must be a real one. */
2504 if (consumer_data
->pid
> 0) {
2506 ret
= kill(consumer_data
->pid
, SIGTERM
);
2508 PERROR("Error killing consumer daemon");
2511 return pthread_join(consumer_data
->thread
, &status
);
2518 * Fork and exec a consumer daemon (consumerd).
2520 * Return pid if successful else -1.
2522 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2526 const char *consumer_to_use
;
2527 const char *verbosity
;
2530 DBG("Spawning consumerd");
2537 if (opt_verbose_consumer
) {
2538 verbosity
= "--verbose";
2539 } else if (lttng_opt_quiet
) {
2540 verbosity
= "--quiet";
2545 switch (consumer_data
->type
) {
2546 case LTTNG_CONSUMER_KERNEL
:
2548 * Find out which consumerd to execute. We will first try the
2549 * 64-bit path, then the sessiond's installation directory, and
2550 * fallback on the 32-bit one,
2552 DBG3("Looking for a kernel consumer at these locations:");
2553 DBG3(" 1) %s", consumerd64_bin
);
2554 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2555 DBG3(" 3) %s", consumerd32_bin
);
2556 if (stat(consumerd64_bin
, &st
) == 0) {
2557 DBG3("Found location #1");
2558 consumer_to_use
= consumerd64_bin
;
2559 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2560 DBG3("Found location #2");
2561 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2562 } else if (stat(consumerd32_bin
, &st
) == 0) {
2563 DBG3("Found location #3");
2564 consumer_to_use
= consumerd32_bin
;
2566 DBG("Could not find any valid consumerd executable");
2570 DBG("Using kernel consumer at: %s", consumer_to_use
);
2571 ret
= execl(consumer_to_use
,
2572 "lttng-consumerd", verbosity
, "-k",
2573 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2574 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2575 "--group", tracing_group_name
,
2578 case LTTNG_CONSUMER64_UST
:
2580 char *tmpnew
= NULL
;
2582 if (consumerd64_libdir
[0] != '\0') {
2586 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2590 tmplen
= strlen("LD_LIBRARY_PATH=")
2591 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2592 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2597 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2598 strcat(tmpnew
, consumerd64_libdir
);
2599 if (tmp
[0] != '\0') {
2600 strcat(tmpnew
, ":");
2601 strcat(tmpnew
, tmp
);
2603 ret
= putenv(tmpnew
);
2610 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2611 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2612 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2613 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2614 "--group", tracing_group_name
,
2616 if (consumerd64_libdir
[0] != '\0') {
2621 case LTTNG_CONSUMER32_UST
:
2623 char *tmpnew
= NULL
;
2625 if (consumerd32_libdir
[0] != '\0') {
2629 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2633 tmplen
= strlen("LD_LIBRARY_PATH=")
2634 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2635 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2640 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2641 strcat(tmpnew
, consumerd32_libdir
);
2642 if (tmp
[0] != '\0') {
2643 strcat(tmpnew
, ":");
2644 strcat(tmpnew
, tmp
);
2646 ret
= putenv(tmpnew
);
2653 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2654 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2655 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2656 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2657 "--group", tracing_group_name
,
2659 if (consumerd32_libdir
[0] != '\0') {
2665 PERROR("unknown consumer type");
2669 PERROR("Consumer execl()");
2671 /* Reaching this point, we got a failure on our execl(). */
2673 } else if (pid
> 0) {
2676 PERROR("start consumer fork");
2684 * Spawn the consumerd daemon and session daemon thread.
2686 static int start_consumerd(struct consumer_data
*consumer_data
)
2691 * Set the listen() state on the socket since there is a possible race
2692 * between the exec() of the consumer daemon and this call if place in the
2693 * consumer thread. See bug #366 for more details.
2695 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2700 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2701 if (consumer_data
->pid
!= 0) {
2702 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2706 ret
= spawn_consumerd(consumer_data
);
2708 ERR("Spawning consumerd failed");
2709 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2713 /* Setting up the consumer_data pid */
2714 consumer_data
->pid
= ret
;
2715 DBG2("Consumer pid %d", consumer_data
->pid
);
2716 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2718 DBG2("Spawning consumer control thread");
2719 ret
= spawn_consumer_thread(consumer_data
);
2721 ERR("Fatal error spawning consumer control thread");
2729 /* Cleanup already created sockets on error. */
2730 if (consumer_data
->err_sock
>= 0) {
2733 err
= close(consumer_data
->err_sock
);
2735 PERROR("close consumer data error socket");
2742 * Setup necessary data for kernel tracer action.
2744 static int init_kernel_tracer(void)
2748 /* Modprobe lttng kernel modules */
2749 ret
= modprobe_lttng_control();
2754 /* Open debugfs lttng */
2755 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2756 if (kernel_tracer_fd
< 0) {
2757 DBG("Failed to open %s", module_proc_lttng
);
2762 /* Validate kernel version */
2763 ret
= kernel_validate_version(kernel_tracer_fd
);
2768 ret
= modprobe_lttng_data();
2773 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2777 modprobe_remove_lttng_control();
2778 ret
= close(kernel_tracer_fd
);
2782 kernel_tracer_fd
= -1;
2783 return LTTNG_ERR_KERN_VERSION
;
2786 ret
= close(kernel_tracer_fd
);
2792 modprobe_remove_lttng_control();
2795 WARN("No kernel tracer available");
2796 kernel_tracer_fd
= -1;
2798 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2800 return LTTNG_ERR_KERN_NA
;
2806 * Copy consumer output from the tracing session to the domain session. The
2807 * function also applies the right modification on a per domain basis for the
2808 * trace files destination directory.
2810 * Should *NOT* be called with RCU read-side lock held.
2812 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2815 const char *dir_name
;
2816 struct consumer_output
*consumer
;
2819 assert(session
->consumer
);
2822 case LTTNG_DOMAIN_KERNEL
:
2823 DBG3("Copying tracing session consumer output in kernel session");
2825 * XXX: We should audit the session creation and what this function
2826 * does "extra" in order to avoid a destroy since this function is used
2827 * in the domain session creation (kernel and ust) only. Same for UST
2830 if (session
->kernel_session
->consumer
) {
2831 consumer_output_put(session
->kernel_session
->consumer
);
2833 session
->kernel_session
->consumer
=
2834 consumer_copy_output(session
->consumer
);
2835 /* Ease our life a bit for the next part */
2836 consumer
= session
->kernel_session
->consumer
;
2837 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2839 case LTTNG_DOMAIN_JUL
:
2840 case LTTNG_DOMAIN_LOG4J
:
2841 case LTTNG_DOMAIN_PYTHON
:
2842 case LTTNG_DOMAIN_UST
:
2843 DBG3("Copying tracing session consumer output in UST session");
2844 if (session
->ust_session
->consumer
) {
2845 consumer_output_put(session
->ust_session
->consumer
);
2847 session
->ust_session
->consumer
=
2848 consumer_copy_output(session
->consumer
);
2849 /* Ease our life a bit for the next part */
2850 consumer
= session
->ust_session
->consumer
;
2851 dir_name
= DEFAULT_UST_TRACE_DIR
;
2854 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2858 /* Append correct directory to subdir */
2859 strncat(consumer
->subdir
, dir_name
,
2860 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2861 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2870 * Create an UST session and add it to the session ust list.
2872 * Should *NOT* be called with RCU read-side lock held.
2874 static int create_ust_session(struct ltt_session
*session
,
2875 struct lttng_domain
*domain
)
2878 struct ltt_ust_session
*lus
= NULL
;
2882 assert(session
->consumer
);
2884 switch (domain
->type
) {
2885 case LTTNG_DOMAIN_JUL
:
2886 case LTTNG_DOMAIN_LOG4J
:
2887 case LTTNG_DOMAIN_PYTHON
:
2888 case LTTNG_DOMAIN_UST
:
2891 ERR("Unknown UST domain on create session %d", domain
->type
);
2892 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2896 DBG("Creating UST session");
2898 lus
= trace_ust_create_session(session
->id
);
2900 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2904 lus
->uid
= session
->uid
;
2905 lus
->gid
= session
->gid
;
2906 lus
->output_traces
= session
->output_traces
;
2907 lus
->snapshot_mode
= session
->snapshot_mode
;
2908 lus
->live_timer_interval
= session
->live_timer
;
2909 session
->ust_session
= lus
;
2910 if (session
->shm_path
[0]) {
2911 strncpy(lus
->root_shm_path
, session
->shm_path
,
2912 sizeof(lus
->root_shm_path
));
2913 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2914 strncpy(lus
->shm_path
, session
->shm_path
,
2915 sizeof(lus
->shm_path
));
2916 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2917 strncat(lus
->shm_path
, "/ust",
2918 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2920 /* Copy session output to the newly created UST session */
2921 ret
= copy_session_consumer(domain
->type
, session
);
2922 if (ret
!= LTTNG_OK
) {
2930 session
->ust_session
= NULL
;
2935 * Create a kernel tracer session then create the default channel.
2937 static int create_kernel_session(struct ltt_session
*session
)
2941 DBG("Creating kernel session");
2943 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2945 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2949 /* Code flow safety */
2950 assert(session
->kernel_session
);
2952 /* Copy session output to the newly created Kernel session */
2953 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2954 if (ret
!= LTTNG_OK
) {
2958 /* Create directory(ies) on local filesystem. */
2959 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2960 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2961 ret
= run_as_mkdir_recursive(
2962 session
->kernel_session
->consumer
->dst
.trace_path
,
2963 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2965 if (errno
!= EEXIST
) {
2966 ERR("Trace directory creation error");
2972 session
->kernel_session
->uid
= session
->uid
;
2973 session
->kernel_session
->gid
= session
->gid
;
2974 session
->kernel_session
->output_traces
= session
->output_traces
;
2975 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2980 trace_kernel_destroy_session(session
->kernel_session
);
2981 session
->kernel_session
= NULL
;
2986 * Count number of session permitted by uid/gid.
2988 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2991 struct ltt_session
*session
;
2993 DBG("Counting number of available session for UID %d GID %d",
2995 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2997 * Only list the sessions the user can control.
2999 if (!session_access_ok(session
, uid
, gid
)) {
3008 * Process the command requested by the lttng client within the command
3009 * context structure. This function make sure that the return structure (llm)
3010 * is set and ready for transmission before returning.
3012 * Return any error encountered or 0 for success.
3014 * "sock" is only used for special-case var. len data.
3016 * Should *NOT* be called with RCU read-side lock held.
3018 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
3022 int need_tracing_session
= 1;
3025 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
3027 assert(!rcu_read_ongoing());
3031 switch (cmd_ctx
->lsm
->cmd_type
) {
3032 case LTTNG_CREATE_SESSION
:
3033 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3034 case LTTNG_CREATE_SESSION_LIVE
:
3035 case LTTNG_DESTROY_SESSION
:
3036 case LTTNG_LIST_SESSIONS
:
3037 case LTTNG_LIST_DOMAINS
:
3038 case LTTNG_START_TRACE
:
3039 case LTTNG_STOP_TRACE
:
3040 case LTTNG_DATA_PENDING
:
3041 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3042 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3043 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3044 case LTTNG_SNAPSHOT_RECORD
:
3045 case LTTNG_SAVE_SESSION
:
3046 case LTTNG_SET_SESSION_SHM_PATH
:
3047 case LTTNG_METADATA_REGENERATE
:
3054 if (opt_no_kernel
&& need_domain
3055 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
3057 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3059 ret
= LTTNG_ERR_KERN_NA
;
3064 /* Deny register consumer if we already have a spawned consumer. */
3065 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
3066 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3067 if (kconsumer_data
.pid
> 0) {
3068 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3069 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3072 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3076 * Check for command that don't needs to allocate a returned payload. We do
3077 * this here so we don't have to make the call for no payload at each
3080 switch(cmd_ctx
->lsm
->cmd_type
) {
3081 case LTTNG_LIST_SESSIONS
:
3082 case LTTNG_LIST_TRACEPOINTS
:
3083 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3084 case LTTNG_LIST_DOMAINS
:
3085 case LTTNG_LIST_CHANNELS
:
3086 case LTTNG_LIST_EVENTS
:
3087 case LTTNG_LIST_SYSCALLS
:
3088 case LTTNG_LIST_TRACKER_PIDS
:
3091 /* Setup lttng message with no payload */
3092 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0);
3094 /* This label does not try to unlock the session */
3095 goto init_setup_error
;
3099 /* Commands that DO NOT need a session. */
3100 switch (cmd_ctx
->lsm
->cmd_type
) {
3101 case LTTNG_CREATE_SESSION
:
3102 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3103 case LTTNG_CREATE_SESSION_LIVE
:
3104 case LTTNG_CALIBRATE
:
3105 case LTTNG_LIST_SESSIONS
:
3106 case LTTNG_LIST_TRACEPOINTS
:
3107 case LTTNG_LIST_SYSCALLS
:
3108 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3109 case LTTNG_SAVE_SESSION
:
3110 need_tracing_session
= 0;
3113 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3115 * We keep the session list lock across _all_ commands
3116 * for now, because the per-session lock does not
3117 * handle teardown properly.
3119 session_lock_list();
3120 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3121 if (cmd_ctx
->session
== NULL
) {
3122 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3125 /* Acquire lock for the session */
3126 session_lock(cmd_ctx
->session
);
3132 * Commands that need a valid session but should NOT create one if none
3133 * exists. Instead of creating one and destroying it when the command is
3134 * handled, process that right before so we save some round trip in useless
3137 switch (cmd_ctx
->lsm
->cmd_type
) {
3138 case LTTNG_DISABLE_CHANNEL
:
3139 case LTTNG_DISABLE_EVENT
:
3140 switch (cmd_ctx
->lsm
->domain
.type
) {
3141 case LTTNG_DOMAIN_KERNEL
:
3142 if (!cmd_ctx
->session
->kernel_session
) {
3143 ret
= LTTNG_ERR_NO_CHANNEL
;
3147 case LTTNG_DOMAIN_JUL
:
3148 case LTTNG_DOMAIN_LOG4J
:
3149 case LTTNG_DOMAIN_PYTHON
:
3150 case LTTNG_DOMAIN_UST
:
3151 if (!cmd_ctx
->session
->ust_session
) {
3152 ret
= LTTNG_ERR_NO_CHANNEL
;
3157 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3169 * Check domain type for specific "pre-action".
3171 switch (cmd_ctx
->lsm
->domain
.type
) {
3172 case LTTNG_DOMAIN_KERNEL
:
3174 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3178 /* Kernel tracer check */
3179 if (kernel_tracer_fd
== -1) {
3180 /* Basically, load kernel tracer modules */
3181 ret
= init_kernel_tracer();
3187 /* Consumer is in an ERROR state. Report back to client */
3188 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3189 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3193 /* Need a session for kernel command */
3194 if (need_tracing_session
) {
3195 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3196 ret
= create_kernel_session(cmd_ctx
->session
);
3198 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3203 /* Start the kernel consumer daemon */
3204 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3205 if (kconsumer_data
.pid
== 0 &&
3206 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3207 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3208 ret
= start_consumerd(&kconsumer_data
);
3210 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3213 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3215 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3219 * The consumer was just spawned so we need to add the socket to
3220 * the consumer output of the session if exist.
3222 ret
= consumer_create_socket(&kconsumer_data
,
3223 cmd_ctx
->session
->kernel_session
->consumer
);
3230 case LTTNG_DOMAIN_JUL
:
3231 case LTTNG_DOMAIN_LOG4J
:
3232 case LTTNG_DOMAIN_PYTHON
:
3233 case LTTNG_DOMAIN_UST
:
3235 if (!ust_app_supported()) {
3236 ret
= LTTNG_ERR_NO_UST
;
3239 /* Consumer is in an ERROR state. Report back to client */
3240 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3241 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3245 if (need_tracing_session
) {
3246 /* Create UST session if none exist. */
3247 if (cmd_ctx
->session
->ust_session
== NULL
) {
3248 ret
= create_ust_session(cmd_ctx
->session
,
3249 &cmd_ctx
->lsm
->domain
);
3250 if (ret
!= LTTNG_OK
) {
3255 /* Start the UST consumer daemons */
3257 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3258 if (consumerd64_bin
[0] != '\0' &&
3259 ustconsumer64_data
.pid
== 0 &&
3260 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3261 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3262 ret
= start_consumerd(&ustconsumer64_data
);
3264 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3265 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3269 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3270 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3272 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3276 * Setup socket for consumer 64 bit. No need for atomic access
3277 * since it was set above and can ONLY be set in this thread.
3279 ret
= consumer_create_socket(&ustconsumer64_data
,
3280 cmd_ctx
->session
->ust_session
->consumer
);
3286 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3287 if (consumerd32_bin
[0] != '\0' &&
3288 ustconsumer32_data
.pid
== 0 &&
3289 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3290 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3291 ret
= start_consumerd(&ustconsumer32_data
);
3293 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3294 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3298 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3299 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3301 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3305 * Setup socket for consumer 64 bit. No need for atomic access
3306 * since it was set above and can ONLY be set in this thread.
3308 ret
= consumer_create_socket(&ustconsumer32_data
,
3309 cmd_ctx
->session
->ust_session
->consumer
);
3321 /* Validate consumer daemon state when start/stop trace command */
3322 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3323 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3324 switch (cmd_ctx
->lsm
->domain
.type
) {
3325 case LTTNG_DOMAIN_NONE
:
3327 case LTTNG_DOMAIN_JUL
:
3328 case LTTNG_DOMAIN_LOG4J
:
3329 case LTTNG_DOMAIN_PYTHON
:
3330 case LTTNG_DOMAIN_UST
:
3331 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3332 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3336 case LTTNG_DOMAIN_KERNEL
:
3337 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3338 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3343 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3349 * Check that the UID or GID match that of the tracing session.
3350 * The root user can interact with all sessions.
3352 if (need_tracing_session
) {
3353 if (!session_access_ok(cmd_ctx
->session
,
3354 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3355 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3356 ret
= LTTNG_ERR_EPERM
;
3362 * Send relayd information to consumer as soon as we have a domain and a
3365 if (cmd_ctx
->session
&& need_domain
) {
3367 * Setup relayd if not done yet. If the relayd information was already
3368 * sent to the consumer, this call will gracefully return.
3370 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3371 if (ret
!= LTTNG_OK
) {
3376 /* Process by command type */
3377 switch (cmd_ctx
->lsm
->cmd_type
) {
3378 case LTTNG_ADD_CONTEXT
:
3381 * An LTTNG_ADD_CONTEXT command might have a supplementary
3382 * payload if the context being added is an application context.
3384 if (cmd_ctx
->lsm
->u
.context
.ctx
.ctx
==
3385 LTTNG_EVENT_CONTEXT_APP_CONTEXT
) {
3386 char *provider_name
= NULL
, *context_name
= NULL
;
3387 size_t provider_name_len
=
3388 cmd_ctx
->lsm
->u
.context
.provider_name_len
;
3389 size_t context_name_len
=
3390 cmd_ctx
->lsm
->u
.context
.context_name_len
;
3392 if (provider_name_len
== 0 || context_name_len
== 0) {
3394 * Application provider and context names MUST
3397 ret
= -LTTNG_ERR_INVALID
;
3401 provider_name
= zmalloc(provider_name_len
+ 1);
3402 if (!provider_name
) {
3403 ret
= -LTTNG_ERR_NOMEM
;
3406 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
=
3409 context_name
= zmalloc(context_name_len
+ 1);
3410 if (!context_name
) {
3411 ret
= -LTTNG_ERR_NOMEM
;
3412 goto error_add_context
;
3414 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
=
3417 ret
= lttcomm_recv_unix_sock(sock
, provider_name
,
3420 goto error_add_context
;
3423 ret
= lttcomm_recv_unix_sock(sock
, context_name
,
3426 goto error_add_context
;
3431 * cmd_add_context assumes ownership of the provider and context
3434 ret
= cmd_add_context(cmd_ctx
->session
,
3435 cmd_ctx
->lsm
->domain
.type
,
3436 cmd_ctx
->lsm
->u
.context
.channel_name
,
3437 &cmd_ctx
->lsm
->u
.context
.ctx
,
3438 kernel_poll_pipe
[1]);
3440 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
= NULL
;
3441 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
= NULL
;
3443 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
);
3444 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
);
3450 case LTTNG_DISABLE_CHANNEL
:
3452 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3453 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3456 case LTTNG_DISABLE_EVENT
:
3460 * FIXME: handle filter; for now we just receive the filter's
3461 * bytecode along with the filter expression which are sent by
3462 * liblttng-ctl and discard them.
3464 * This fixes an issue where the client may block while sending
3465 * the filter payload and encounter an error because the session
3466 * daemon closes the socket without ever handling this data.
3468 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3469 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3472 char data
[LTTNG_FILTER_MAX_LEN
];
3474 DBG("Discarding disable event command payload of size %zu", count
);
3476 ret
= lttcomm_recv_unix_sock(sock
, data
,
3477 count
> sizeof(data
) ? sizeof(data
) : count
);
3482 count
-= (size_t) ret
;
3485 /* FIXME: passing packed structure to non-packed pointer */
3486 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3487 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3488 &cmd_ctx
->lsm
->u
.disable
.event
);
3491 case LTTNG_ENABLE_CHANNEL
:
3493 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3494 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3497 case LTTNG_TRACK_PID
:
3499 ret
= cmd_track_pid(cmd_ctx
->session
,
3500 cmd_ctx
->lsm
->domain
.type
,
3501 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3504 case LTTNG_UNTRACK_PID
:
3506 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3507 cmd_ctx
->lsm
->domain
.type
,
3508 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3511 case LTTNG_ENABLE_EVENT
:
3513 struct lttng_event_exclusion
*exclusion
= NULL
;
3514 struct lttng_filter_bytecode
*bytecode
= NULL
;
3515 char *filter_expression
= NULL
;
3517 /* Handle exclusion events and receive it from the client. */
3518 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3519 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3521 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3522 (count
* LTTNG_SYMBOL_NAME_LEN
));
3524 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3528 DBG("Receiving var len exclusion event list from client ...");
3529 exclusion
->count
= count
;
3530 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3531 count
* LTTNG_SYMBOL_NAME_LEN
);
3533 DBG("Nothing recv() from client var len data... continuing");
3536 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3541 /* Get filter expression from client. */
3542 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3543 size_t expression_len
=
3544 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3546 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3547 ret
= LTTNG_ERR_FILTER_INVAL
;
3552 filter_expression
= zmalloc(expression_len
);
3553 if (!filter_expression
) {
3555 ret
= LTTNG_ERR_FILTER_NOMEM
;
3559 /* Receive var. len. data */
3560 DBG("Receiving var len filter's expression from client ...");
3561 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3564 DBG("Nothing recv() from client car len data... continuing");
3566 free(filter_expression
);
3568 ret
= LTTNG_ERR_FILTER_INVAL
;
3573 /* Handle filter and get bytecode from client. */
3574 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3575 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3577 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3578 ret
= LTTNG_ERR_FILTER_INVAL
;
3579 free(filter_expression
);
3584 bytecode
= zmalloc(bytecode_len
);
3586 free(filter_expression
);
3588 ret
= LTTNG_ERR_FILTER_NOMEM
;
3592 /* Receive var. len. data */
3593 DBG("Receiving var len filter's bytecode from client ...");
3594 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3596 DBG("Nothing recv() from client car len data... continuing");
3598 free(filter_expression
);
3601 ret
= LTTNG_ERR_FILTER_INVAL
;
3605 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3606 free(filter_expression
);
3609 ret
= LTTNG_ERR_FILTER_INVAL
;
3614 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3615 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3616 &cmd_ctx
->lsm
->u
.enable
.event
,
3617 filter_expression
, bytecode
, exclusion
,
3618 kernel_poll_pipe
[1]);
3621 case LTTNG_LIST_TRACEPOINTS
:
3623 struct lttng_event
*events
;
3626 session_lock_list();
3627 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3628 session_unlock_list();
3629 if (nb_events
< 0) {
3630 /* Return value is a negative lttng_error_code. */
3636 * Setup lttng message with payload size set to the event list size in
3637 * bytes and then copy list into the llm payload.
3639 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3640 sizeof(struct lttng_event
) * nb_events
);
3650 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3652 struct lttng_event_field
*fields
;
3655 session_lock_list();
3656 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3658 session_unlock_list();
3659 if (nb_fields
< 0) {
3660 /* Return value is a negative lttng_error_code. */
3666 * Setup lttng message with payload size set to the event list size in
3667 * bytes and then copy list into the llm payload.
3669 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, fields
,
3670 sizeof(struct lttng_event_field
) * nb_fields
);
3680 case LTTNG_LIST_SYSCALLS
:
3682 struct lttng_event
*events
;
3685 nb_events
= cmd_list_syscalls(&events
);
3686 if (nb_events
< 0) {
3687 /* Return value is a negative lttng_error_code. */
3693 * Setup lttng message with payload size set to the event list size in
3694 * bytes and then copy list into the llm payload.
3696 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3697 sizeof(struct lttng_event
) * nb_events
);
3707 case LTTNG_LIST_TRACKER_PIDS
:
3709 int32_t *pids
= NULL
;
3712 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3713 cmd_ctx
->lsm
->domain
.type
, &pids
);
3715 /* Return value is a negative lttng_error_code. */
3721 * Setup lttng message with payload size set to the event list size in
3722 * bytes and then copy list into the llm payload.
3724 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, pids
,
3725 sizeof(int32_t) * nr_pids
);
3735 case LTTNG_SET_CONSUMER_URI
:
3738 struct lttng_uri
*uris
;
3740 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3741 len
= nb_uri
* sizeof(struct lttng_uri
);
3744 ret
= LTTNG_ERR_INVALID
;
3748 uris
= zmalloc(len
);
3750 ret
= LTTNG_ERR_FATAL
;
3754 /* Receive variable len data */
3755 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3756 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3758 DBG("No URIs received from client... continuing");
3760 ret
= LTTNG_ERR_SESSION_FAIL
;
3765 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3767 if (ret
!= LTTNG_OK
) {
3774 case LTTNG_START_TRACE
:
3776 ret
= cmd_start_trace(cmd_ctx
->session
);
3779 case LTTNG_STOP_TRACE
:
3781 ret
= cmd_stop_trace(cmd_ctx
->session
);
3784 case LTTNG_CREATE_SESSION
:
3787 struct lttng_uri
*uris
= NULL
;
3789 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3790 len
= nb_uri
* sizeof(struct lttng_uri
);
3793 uris
= zmalloc(len
);
3795 ret
= LTTNG_ERR_FATAL
;
3799 /* Receive variable len data */
3800 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3801 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3803 DBG("No URIs received from client... continuing");
3805 ret
= LTTNG_ERR_SESSION_FAIL
;
3810 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3811 DBG("Creating session with ONE network URI is a bad call");
3812 ret
= LTTNG_ERR_SESSION_FAIL
;
3818 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3819 &cmd_ctx
->creds
, 0);
3825 case LTTNG_DESTROY_SESSION
:
3827 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3829 /* Set session to NULL so we do not unlock it after free. */
3830 cmd_ctx
->session
= NULL
;
3833 case LTTNG_LIST_DOMAINS
:
3836 struct lttng_domain
*domains
= NULL
;
3838 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3840 /* Return value is a negative lttng_error_code. */
3845 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, domains
,
3846 nb_dom
* sizeof(struct lttng_domain
));
3856 case LTTNG_LIST_CHANNELS
:
3858 ssize_t payload_size
;
3859 struct lttng_channel
*channels
= NULL
;
3861 payload_size
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3862 cmd_ctx
->session
, &channels
);
3863 if (payload_size
< 0) {
3864 /* Return value is a negative lttng_error_code. */
3865 ret
= -payload_size
;
3869 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, channels
,
3880 case LTTNG_LIST_EVENTS
:
3883 struct lttng_event
*events
= NULL
;
3884 struct lttcomm_event_command_header cmd_header
;
3887 memset(&cmd_header
, 0, sizeof(cmd_header
));
3888 /* Extended infos are included at the end of events */
3889 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
,
3890 cmd_ctx
->session
, cmd_ctx
->lsm
->u
.list
.channel_name
,
3891 &events
, &total_size
);
3894 /* Return value is a negative lttng_error_code. */
3899 cmd_header
.nb_events
= nb_event
;
3900 ret
= setup_lttng_msg(cmd_ctx
, events
, total_size
,
3901 &cmd_header
, sizeof(cmd_header
));
3911 case LTTNG_LIST_SESSIONS
:
3913 unsigned int nr_sessions
;
3914 void *sessions_payload
;
3917 session_lock_list();
3918 nr_sessions
= lttng_sessions_count(
3919 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3920 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3921 payload_len
= sizeof(struct lttng_session
) * nr_sessions
;
3922 sessions_payload
= zmalloc(payload_len
);
3924 if (!sessions_payload
) {
3925 session_unlock_list();
3930 cmd_list_lttng_sessions(sessions_payload
,
3931 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3932 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3933 session_unlock_list();
3935 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, sessions_payload
,
3937 free(sessions_payload
);
3946 case LTTNG_CALIBRATE
:
3948 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3949 &cmd_ctx
->lsm
->u
.calibrate
);
3952 case LTTNG_REGISTER_CONSUMER
:
3954 struct consumer_data
*cdata
;
3956 switch (cmd_ctx
->lsm
->domain
.type
) {
3957 case LTTNG_DOMAIN_KERNEL
:
3958 cdata
= &kconsumer_data
;
3961 ret
= LTTNG_ERR_UND
;
3965 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3966 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3969 case LTTNG_DATA_PENDING
:
3972 uint8_t pending_ret_byte
;
3974 pending_ret
= cmd_data_pending(cmd_ctx
->session
);
3979 * This function may returns 0 or 1 to indicate whether or not
3980 * there is data pending. In case of error, it should return an
3981 * LTTNG_ERR code. However, some code paths may still return
3982 * a nondescript error code, which we handle by returning an
3985 if (pending_ret
== 0 || pending_ret
== 1) {
3987 * ret will be set to LTTNG_OK at the end of
3990 } else if (pending_ret
< 0) {
3991 ret
= LTTNG_ERR_UNK
;
3998 pending_ret_byte
= (uint8_t) pending_ret
;
4000 /* 1 byte to return whether or not data is pending */
4001 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
,
4002 &pending_ret_byte
, 1);
4011 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
4013 struct lttcomm_lttng_output_id reply
;
4015 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
4016 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
4017 if (ret
!= LTTNG_OK
) {
4021 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &reply
,
4027 /* Copy output list into message payload */
4031 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
4033 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
4034 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
4037 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
4040 struct lttng_snapshot_output
*outputs
= NULL
;
4042 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
4043 if (nb_output
< 0) {
4048 assert((nb_output
> 0 && outputs
) || nb_output
== 0);
4049 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, outputs
,
4050 nb_output
* sizeof(struct lttng_snapshot_output
));
4060 case LTTNG_SNAPSHOT_RECORD
:
4062 ret
= cmd_snapshot_record(cmd_ctx
->session
,
4063 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
4064 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
4067 case LTTNG_CREATE_SESSION_SNAPSHOT
:
4070 struct lttng_uri
*uris
= NULL
;
4072 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4073 len
= nb_uri
* sizeof(struct lttng_uri
);
4076 uris
= zmalloc(len
);
4078 ret
= LTTNG_ERR_FATAL
;
4082 /* Receive variable len data */
4083 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4084 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4086 DBG("No URIs received from client... continuing");
4088 ret
= LTTNG_ERR_SESSION_FAIL
;
4093 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4094 DBG("Creating session with ONE network URI is a bad call");
4095 ret
= LTTNG_ERR_SESSION_FAIL
;
4101 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
4102 nb_uri
, &cmd_ctx
->creds
);
4106 case LTTNG_CREATE_SESSION_LIVE
:
4109 struct lttng_uri
*uris
= NULL
;
4111 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4112 len
= nb_uri
* sizeof(struct lttng_uri
);
4115 uris
= zmalloc(len
);
4117 ret
= LTTNG_ERR_FATAL
;
4121 /* Receive variable len data */
4122 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4123 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4125 DBG("No URIs received from client... continuing");
4127 ret
= LTTNG_ERR_SESSION_FAIL
;
4132 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4133 DBG("Creating session with ONE network URI is a bad call");
4134 ret
= LTTNG_ERR_SESSION_FAIL
;
4140 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4141 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4145 case LTTNG_SAVE_SESSION
:
4147 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4151 case LTTNG_SET_SESSION_SHM_PATH
:
4153 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4154 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4157 case LTTNG_METADATA_REGENERATE
:
4159 ret
= cmd_metadata_regenerate(cmd_ctx
->session
);
4163 ret
= LTTNG_ERR_UND
;
4168 if (cmd_ctx
->llm
== NULL
) {
4169 DBG("Missing llm structure. Allocating one.");
4170 if (setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0) < 0) {
4174 /* Set return code */
4175 cmd_ctx
->llm
->ret_code
= ret
;
4177 if (cmd_ctx
->session
) {
4178 session_unlock(cmd_ctx
->session
);
4180 if (need_tracing_session
) {
4181 session_unlock_list();
4184 assert(!rcu_read_ongoing());
4189 * Thread managing health check socket.
4191 static void *thread_manage_health(void *data
)
4193 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4194 uint32_t revents
, nb_fd
;
4195 struct lttng_poll_event events
;
4196 struct health_comm_msg msg
;
4197 struct health_comm_reply reply
;
4199 DBG("[thread] Manage health check started");
4201 rcu_register_thread();
4203 /* We might hit an error path before this is created. */
4204 lttng_poll_init(&events
);
4206 /* Create unix socket */
4207 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
4209 ERR("Unable to create health check Unix socket");
4214 /* lttng health client socket path permissions */
4215 ret
= chown(health_unix_sock_path
, 0,
4216 utils_get_group_id(tracing_group_name
));
4218 ERR("Unable to set group on %s", health_unix_sock_path
);
4223 ret
= chmod(health_unix_sock_path
,
4224 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4226 ERR("Unable to set permissions on %s", health_unix_sock_path
);
4233 * Set the CLOEXEC flag. Return code is useless because either way, the
4236 (void) utils_set_fd_cloexec(sock
);
4238 ret
= lttcomm_listen_unix_sock(sock
);
4244 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4245 * more will be added to this poll set.
4247 ret
= sessiond_set_thread_pollset(&events
, 2);
4252 /* Add the application registration socket */
4253 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4258 sessiond_notify_ready();
4261 DBG("Health check ready");
4263 /* Inifinite blocking call, waiting for transmission */
4265 ret
= lttng_poll_wait(&events
, -1);
4268 * Restart interrupted system call.
4270 if (errno
== EINTR
) {
4278 for (i
= 0; i
< nb_fd
; i
++) {
4279 /* Fetch once the poll data */
4280 revents
= LTTNG_POLL_GETEV(&events
, i
);
4281 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4284 /* No activity for this FD (poll implementation). */
4288 /* Thread quit pipe has been closed. Killing thread. */
4289 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4295 /* Event on the registration socket */
4296 if (pollfd
== sock
) {
4297 if (revents
& LPOLLIN
) {
4299 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4300 ERR("Health socket poll error");
4303 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4309 new_sock
= lttcomm_accept_unix_sock(sock
);
4315 * Set the CLOEXEC flag. Return code is useless because either way, the
4318 (void) utils_set_fd_cloexec(new_sock
);
4320 DBG("Receiving data from client for health...");
4321 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4323 DBG("Nothing recv() from client... continuing");
4324 ret
= close(new_sock
);
4332 rcu_thread_online();
4334 memset(&reply
, 0, sizeof(reply
));
4335 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4337 * health_check_state returns 0 if health is
4340 if (!health_check_state(health_sessiond
, i
)) {
4341 reply
.ret_code
|= 1ULL << i
;
4345 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4347 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4349 ERR("Failed to send health data back to client");
4352 /* End of transmission */
4353 ret
= close(new_sock
);
4363 ERR("Health error occurred in %s", __func__
);
4365 DBG("Health check thread dying");
4366 unlink(health_unix_sock_path
);
4374 lttng_poll_clean(&events
);
4376 rcu_unregister_thread();
4381 * This thread manage all clients request using the unix client socket for
4384 static void *thread_manage_clients(void *data
)
4386 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4388 uint32_t revents
, nb_fd
;
4389 struct command_ctx
*cmd_ctx
= NULL
;
4390 struct lttng_poll_event events
;
4392 DBG("[thread] Manage client started");
4394 rcu_register_thread();
4396 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4398 health_code_update();
4400 ret
= lttcomm_listen_unix_sock(client_sock
);
4406 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4407 * more will be added to this poll set.
4409 ret
= sessiond_set_thread_pollset(&events
, 2);
4411 goto error_create_poll
;
4414 /* Add the application registration socket */
4415 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4420 sessiond_notify_ready();
4421 ret
= sem_post(&load_info
->message_thread_ready
);
4423 PERROR("sem_post message_thread_ready");
4427 /* This testpoint is after we signal readiness to the parent. */
4428 if (testpoint(sessiond_thread_manage_clients
)) {
4432 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4436 health_code_update();
4439 DBG("Accepting client command ...");
4441 /* Inifinite blocking call, waiting for transmission */
4443 health_poll_entry();
4444 ret
= lttng_poll_wait(&events
, -1);
4448 * Restart interrupted system call.
4450 if (errno
== EINTR
) {
4458 for (i
= 0; i
< nb_fd
; i
++) {
4459 /* Fetch once the poll data */
4460 revents
= LTTNG_POLL_GETEV(&events
, i
);
4461 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4463 health_code_update();
4466 /* No activity for this FD (poll implementation). */
4470 /* Thread quit pipe has been closed. Killing thread. */
4471 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4477 /* Event on the registration socket */
4478 if (pollfd
== client_sock
) {
4479 if (revents
& LPOLLIN
) {
4481 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4482 ERR("Client socket poll error");
4485 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4491 DBG("Wait for client response");
4493 health_code_update();
4495 sock
= lttcomm_accept_unix_sock(client_sock
);
4501 * Set the CLOEXEC flag. Return code is useless because either way, the
4504 (void) utils_set_fd_cloexec(sock
);
4506 /* Set socket option for credentials retrieval */
4507 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4512 /* Allocate context command to process the client request */
4513 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4514 if (cmd_ctx
== NULL
) {
4515 PERROR("zmalloc cmd_ctx");
4519 /* Allocate data buffer for reception */
4520 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4521 if (cmd_ctx
->lsm
== NULL
) {
4522 PERROR("zmalloc cmd_ctx->lsm");
4526 cmd_ctx
->llm
= NULL
;
4527 cmd_ctx
->session
= NULL
;
4529 health_code_update();
4532 * Data is received from the lttng client. The struct
4533 * lttcomm_session_msg (lsm) contains the command and data request of
4536 DBG("Receiving data from client ...");
4537 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4538 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4540 DBG("Nothing recv() from client... continuing");
4546 clean_command_ctx(&cmd_ctx
);
4550 health_code_update();
4552 // TODO: Validate cmd_ctx including sanity check for
4553 // security purpose.
4555 rcu_thread_online();
4557 * This function dispatch the work to the kernel or userspace tracer
4558 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4559 * informations for the client. The command context struct contains
4560 * everything this function may needs.
4562 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4563 rcu_thread_offline();
4571 * TODO: Inform client somehow of the fatal error. At
4572 * this point, ret < 0 means that a zmalloc failed
4573 * (ENOMEM). Error detected but still accept
4574 * command, unless a socket error has been
4577 clean_command_ctx(&cmd_ctx
);
4581 health_code_update();
4583 DBG("Sending response (size: %d, retcode: %s (%d))",
4584 cmd_ctx
->lttng_msg_size
,
4585 lttng_strerror(-cmd_ctx
->llm
->ret_code
),
4586 cmd_ctx
->llm
->ret_code
);
4587 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4589 ERR("Failed to send data back to client");
4592 /* End of transmission */
4599 clean_command_ctx(&cmd_ctx
);
4601 health_code_update();
4613 lttng_poll_clean(&events
);
4614 clean_command_ctx(&cmd_ctx
);
4618 unlink(client_unix_sock_path
);
4619 if (client_sock
>= 0) {
4620 ret
= close(client_sock
);
4628 ERR("Health error occurred in %s", __func__
);
4631 health_unregister(health_sessiond
);
4633 DBG("Client thread dying");
4635 rcu_unregister_thread();
4638 * Since we are creating the consumer threads, we own them, so we need
4639 * to join them before our thread exits.
4641 ret
= join_consumer_thread(&kconsumer_data
);
4644 PERROR("join_consumer");
4647 ret
= join_consumer_thread(&ustconsumer32_data
);
4650 PERROR("join_consumer ust32");
4653 ret
= join_consumer_thread(&ustconsumer64_data
);
4656 PERROR("join_consumer ust64");
4661 static int string_match(const char *str1
, const char *str2
)
4663 return (str1
&& str2
) && !strcmp(str1
, str2
);
4667 * Take an option from the getopt output and set it in the right variable to be
4670 * Return 0 on success else a negative value.
4672 static int set_option(int opt
, const char *arg
, const char *optname
)
4676 if (string_match(optname
, "client-sock") || opt
== 'c') {
4677 if (!arg
|| *arg
== '\0') {
4681 if (lttng_is_setuid_setgid()) {
4682 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4683 "-c, --client-sock");
4685 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4687 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4688 if (!arg
|| *arg
== '\0') {
4692 if (lttng_is_setuid_setgid()) {
4693 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4696 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4698 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4700 } else if (string_match(optname
, "background") || opt
== 'b') {
4702 } else if (string_match(optname
, "group") || opt
== 'g') {
4703 if (!arg
|| *arg
== '\0') {
4707 if (lttng_is_setuid_setgid()) {
4708 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4712 * If the override option is set, the pointer points to a
4713 * *non* const thus freeing it even though the variable type is
4716 if (tracing_group_name_override
) {
4717 free((void *) tracing_group_name
);
4719 tracing_group_name
= strdup(arg
);
4720 if (!tracing_group_name
) {
4724 tracing_group_name_override
= 1;
4726 } else if (string_match(optname
, "help") || opt
== 'h') {
4727 ret
= utils_show_man_page(8, "lttng-sessiond");
4729 ERR("Cannot view man page lttng-sessiond(8)");
4732 exit(ret
? EXIT_FAILURE
: EXIT_SUCCESS
);
4733 } else if (string_match(optname
, "version") || opt
== 'V') {
4734 fprintf(stdout
, "%s\n", VERSION
);
4736 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4738 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4739 if (!arg
|| *arg
== '\0') {
4743 if (lttng_is_setuid_setgid()) {
4744 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4745 "--kconsumerd-err-sock");
4747 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4749 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4750 if (!arg
|| *arg
== '\0') {
4754 if (lttng_is_setuid_setgid()) {
4755 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4756 "--kconsumerd-cmd-sock");
4758 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4760 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4761 if (!arg
|| *arg
== '\0') {
4765 if (lttng_is_setuid_setgid()) {
4766 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4767 "--ustconsumerd64-err-sock");
4769 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4771 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4772 if (!arg
|| *arg
== '\0') {
4776 if (lttng_is_setuid_setgid()) {
4777 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4778 "--ustconsumerd64-cmd-sock");
4780 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4782 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4783 if (!arg
|| *arg
== '\0') {
4787 if (lttng_is_setuid_setgid()) {
4788 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4789 "--ustconsumerd32-err-sock");
4791 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4793 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4794 if (!arg
|| *arg
== '\0') {
4798 if (lttng_is_setuid_setgid()) {
4799 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4800 "--ustconsumerd32-cmd-sock");
4802 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4804 } else if (string_match(optname
, "no-kernel")) {
4806 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4807 lttng_opt_quiet
= 1;
4808 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4809 /* Verbose level can increase using multiple -v */
4811 /* Value obtained from config file */
4812 lttng_opt_verbose
= config_parse_value(arg
);
4814 /* -v used on command line */
4815 lttng_opt_verbose
++;
4817 /* Clamp value to [0, 3] */
4818 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4819 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4820 } else if (string_match(optname
, "verbose-consumer")) {
4822 opt_verbose_consumer
= config_parse_value(arg
);
4824 opt_verbose_consumer
++;
4826 } else if (string_match(optname
, "consumerd32-path")) {
4827 if (!arg
|| *arg
== '\0') {
4831 if (lttng_is_setuid_setgid()) {
4832 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4833 "--consumerd32-path");
4835 if (consumerd32_bin_override
) {
4836 free((void *) consumerd32_bin
);
4838 consumerd32_bin
= strdup(arg
);
4839 if (!consumerd32_bin
) {
4843 consumerd32_bin_override
= 1;
4845 } else if (string_match(optname
, "consumerd32-libdir")) {
4846 if (!arg
|| *arg
== '\0') {
4850 if (lttng_is_setuid_setgid()) {
4851 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4852 "--consumerd32-libdir");
4854 if (consumerd32_libdir_override
) {
4855 free((void *) consumerd32_libdir
);
4857 consumerd32_libdir
= strdup(arg
);
4858 if (!consumerd32_libdir
) {
4862 consumerd32_libdir_override
= 1;
4864 } else if (string_match(optname
, "consumerd64-path")) {
4865 if (!arg
|| *arg
== '\0') {
4869 if (lttng_is_setuid_setgid()) {
4870 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4871 "--consumerd64-path");
4873 if (consumerd64_bin_override
) {
4874 free((void *) consumerd64_bin
);
4876 consumerd64_bin
= strdup(arg
);
4877 if (!consumerd64_bin
) {
4881 consumerd64_bin_override
= 1;
4883 } else if (string_match(optname
, "consumerd64-libdir")) {
4884 if (!arg
|| *arg
== '\0') {
4888 if (lttng_is_setuid_setgid()) {
4889 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4890 "--consumerd64-libdir");
4892 if (consumerd64_libdir_override
) {
4893 free((void *) consumerd64_libdir
);
4895 consumerd64_libdir
= strdup(arg
);
4896 if (!consumerd64_libdir
) {
4900 consumerd64_libdir_override
= 1;
4902 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4903 if (!arg
|| *arg
== '\0') {
4907 if (lttng_is_setuid_setgid()) {
4908 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4912 opt_pidfile
= strdup(arg
);
4918 } else if (string_match(optname
, "agent-tcp-port")) {
4919 if (!arg
|| *arg
== '\0') {
4923 if (lttng_is_setuid_setgid()) {
4924 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4925 "--agent-tcp-port");
4930 v
= strtoul(arg
, NULL
, 0);
4931 if (errno
!= 0 || !isdigit(arg
[0])) {
4932 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4935 if (v
== 0 || v
>= 65535) {
4936 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4939 agent_tcp_port
= (uint32_t) v
;
4940 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4942 } else if (string_match(optname
, "load") || opt
== 'l') {
4943 if (!arg
|| *arg
== '\0') {
4947 if (lttng_is_setuid_setgid()) {
4948 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4951 free(opt_load_session_path
);
4952 opt_load_session_path
= strdup(arg
);
4953 if (!opt_load_session_path
) {
4958 } else if (string_match(optname
, "kmod-probes")) {
4959 if (!arg
|| *arg
== '\0') {
4963 if (lttng_is_setuid_setgid()) {
4964 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4967 free(kmod_probes_list
);
4968 kmod_probes_list
= strdup(arg
);
4969 if (!kmod_probes_list
) {
4974 } else if (string_match(optname
, "extra-kmod-probes")) {
4975 if (!arg
|| *arg
== '\0') {
4979 if (lttng_is_setuid_setgid()) {
4980 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4981 "--extra-kmod-probes");
4983 free(kmod_extra_probes_list
);
4984 kmod_extra_probes_list
= strdup(arg
);
4985 if (!kmod_extra_probes_list
) {
4990 } else if (string_match(optname
, "config") || opt
== 'f') {
4991 /* This is handled in set_options() thus silent skip. */
4994 /* Unknown option or other error.
4995 * Error is printed by getopt, just return */
5000 if (ret
== -EINVAL
) {
5001 const char *opt_name
= "unknown";
5004 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
5006 if (opt
== long_options
[i
].val
) {
5007 opt_name
= long_options
[i
].name
;
5012 WARN("Invalid argument provided for option \"%s\", using default value.",
5020 * config_entry_handler_cb used to handle options read from a config file.
5021 * See config_entry_handler_cb comment in common/config/session-config.h for the
5022 * return value conventions.
5024 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
5028 if (!entry
|| !entry
->name
|| !entry
->value
) {
5033 /* Check if the option is to be ignored */
5034 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
5035 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
5040 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
5043 /* Ignore if not fully matched. */
5044 if (strcmp(entry
->name
, long_options
[i
].name
)) {
5049 * If the option takes no argument on the command line, we have to
5050 * check if the value is "true". We support non-zero numeric values,
5053 if (!long_options
[i
].has_arg
) {
5054 ret
= config_parse_value(entry
->value
);
5057 WARN("Invalid configuration value \"%s\" for option %s",
5058 entry
->value
, entry
->name
);
5060 /* False, skip boolean config option. */
5065 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
5069 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
5076 * daemon configuration loading and argument parsing
5078 static int set_options(int argc
, char **argv
)
5080 int ret
= 0, c
= 0, option_index
= 0;
5081 int orig_optopt
= optopt
, orig_optind
= optind
;
5083 const char *config_path
= NULL
;
5085 optstring
= utils_generate_optstring(long_options
,
5086 sizeof(long_options
) / sizeof(struct option
));
5092 /* Check for the --config option */
5093 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
5094 &option_index
)) != -1) {
5098 } else if (c
!= 'f') {
5099 /* if not equal to --config option. */
5103 if (lttng_is_setuid_setgid()) {
5104 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5107 config_path
= utils_expand_path(optarg
);
5109 ERR("Failed to resolve path: %s", optarg
);
5114 ret
= config_get_section_entries(config_path
, config_section_name
,
5115 config_entry_handler
, NULL
);
5118 ERR("Invalid configuration option at line %i", ret
);
5124 /* Reset getopt's global state */
5125 optopt
= orig_optopt
;
5126 optind
= orig_optind
;
5130 * getopt_long() will not set option_index if it encounters a
5133 c
= getopt_long(argc
, argv
, optstring
, long_options
,
5140 * Pass NULL as the long option name if popt left the index
5143 ret
= set_option(c
, optarg
,
5144 option_index
< 0 ? NULL
:
5145 long_options
[option_index
].name
);
5157 * Creates the two needed socket by the daemon.
5158 * apps_sock - The communication socket for all UST apps.
5159 * client_sock - The communication of the cli tool (lttng).
5161 static int init_daemon_socket(void)
5166 old_umask
= umask(0);
5168 /* Create client tool unix socket */
5169 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
5170 if (client_sock
< 0) {
5171 ERR("Create unix sock failed: %s", client_unix_sock_path
);
5176 /* Set the cloexec flag */
5177 ret
= utils_set_fd_cloexec(client_sock
);
5179 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5180 "Continuing but note that the consumer daemon will have a "
5181 "reference to this socket on exec()", client_sock
);
5184 /* File permission MUST be 660 */
5185 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5187 ERR("Set file permissions failed: %s", client_unix_sock_path
);
5192 /* Create the application unix socket */
5193 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
5194 if (apps_sock
< 0) {
5195 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
5200 /* Set the cloexec flag */
5201 ret
= utils_set_fd_cloexec(apps_sock
);
5203 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5204 "Continuing but note that the consumer daemon will have a "
5205 "reference to this socket on exec()", apps_sock
);
5208 /* File permission MUST be 666 */
5209 ret
= chmod(apps_unix_sock_path
,
5210 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5212 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
5217 DBG3("Session daemon client socket %d and application socket %d created",
5218 client_sock
, apps_sock
);
5226 * Check if the global socket is available, and if a daemon is answering at the
5227 * other side. If yes, error is returned.
5229 static int check_existing_daemon(void)
5231 /* Is there anybody out there ? */
5232 if (lttng_session_daemon_alive()) {
5240 * Set the tracing group gid onto the client socket.
5242 * Race window between mkdir and chown is OK because we are going from more
5243 * permissive (root.root) to less permissive (root.tracing).
5245 static int set_permissions(char *rundir
)
5250 gid
= utils_get_group_id(tracing_group_name
);
5252 /* Set lttng run dir */
5253 ret
= chown(rundir
, 0, gid
);
5255 ERR("Unable to set group on %s", rundir
);
5260 * Ensure all applications and tracing group can search the run
5261 * dir. Allow everyone to read the directory, since it does not
5262 * buy us anything to hide its content.
5264 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5266 ERR("Unable to set permissions on %s", rundir
);
5270 /* lttng client socket path */
5271 ret
= chown(client_unix_sock_path
, 0, gid
);
5273 ERR("Unable to set group on %s", client_unix_sock_path
);
5277 /* kconsumer error socket path */
5278 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5280 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5284 /* 64-bit ustconsumer error socket path */
5285 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5287 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5291 /* 32-bit ustconsumer compat32 error socket path */
5292 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5294 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5298 DBG("All permissions are set");
5304 * Create the lttng run directory needed for all global sockets and pipe.
5306 static int create_lttng_rundir(const char *rundir
)
5310 DBG3("Creating LTTng run directory: %s", rundir
);
5312 ret
= mkdir(rundir
, S_IRWXU
);
5314 if (errno
!= EEXIST
) {
5315 ERR("Unable to create %s", rundir
);
5327 * Setup sockets and directory needed by the kconsumerd communication with the
5330 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5334 char path
[PATH_MAX
];
5336 switch (consumer_data
->type
) {
5337 case LTTNG_CONSUMER_KERNEL
:
5338 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5340 case LTTNG_CONSUMER64_UST
:
5341 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5343 case LTTNG_CONSUMER32_UST
:
5344 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5347 ERR("Consumer type unknown");
5352 DBG2("Creating consumer directory: %s", path
);
5354 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5356 if (errno
!= EEXIST
) {
5358 ERR("Failed to create %s", path
);
5364 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5366 ERR("Unable to set group on %s", path
);
5372 /* Create the kconsumerd error unix socket */
5373 consumer_data
->err_sock
=
5374 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5375 if (consumer_data
->err_sock
< 0) {
5376 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5382 * Set the CLOEXEC flag. Return code is useless because either way, the
5385 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5387 PERROR("utils_set_fd_cloexec");
5388 /* continue anyway */
5391 /* File permission MUST be 660 */
5392 ret
= chmod(consumer_data
->err_unix_sock_path
,
5393 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5395 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5405 * Signal handler for the daemon
5407 * Simply stop all worker threads, leaving main() return gracefully after
5408 * joining all threads and calling cleanup().
5410 static void sighandler(int sig
)
5414 DBG("SIGPIPE caught");
5417 DBG("SIGINT caught");
5421 DBG("SIGTERM caught");
5425 CMM_STORE_SHARED(recv_child_signal
, 1);
5433 * Setup signal handler for :
5434 * SIGINT, SIGTERM, SIGPIPE
5436 static int set_signal_handler(void)
5439 struct sigaction sa
;
5442 if ((ret
= sigemptyset(&sigset
)) < 0) {
5443 PERROR("sigemptyset");
5447 sa
.sa_handler
= sighandler
;
5448 sa
.sa_mask
= sigset
;
5450 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5451 PERROR("sigaction");
5455 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5456 PERROR("sigaction");
5460 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5461 PERROR("sigaction");
5465 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5466 PERROR("sigaction");
5470 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5476 * Set open files limit to unlimited. This daemon can open a large number of
5477 * file descriptors in order to consume multiple kernel traces.
5479 static void set_ulimit(void)
5484 /* The kernel does not allow an infinite limit for open files */
5485 lim
.rlim_cur
= 65535;
5486 lim
.rlim_max
= 65535;
5488 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5490 PERROR("failed to set open files limit");
5495 * Write pidfile using the rundir and opt_pidfile.
5497 static int write_pidfile(void)
5500 char pidfile_path
[PATH_MAX
];
5505 if (lttng_strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
))) {
5510 /* Build pidfile path from rundir and opt_pidfile. */
5511 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5512 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5514 PERROR("snprintf pidfile path");
5520 * Create pid file in rundir.
5522 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5528 * Create lockfile using the rundir and return its fd.
5530 static int create_lockfile(void)
5533 char lockfile_path
[PATH_MAX
];
5535 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5540 ret
= utils_create_lock_file(lockfile_path
);
5546 * Write agent TCP port using the rundir.
5548 static int write_agent_port(void)
5551 char path
[PATH_MAX
];
5555 ret
= snprintf(path
, sizeof(path
), "%s/"
5556 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5558 PERROR("snprintf agent port path");
5563 * Create TCP agent port file in rundir.
5565 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5574 int main(int argc
, char **argv
)
5576 int ret
= 0, retval
= 0;
5578 const char *home_path
, *env_app_timeout
;
5580 init_kernel_workarounds();
5582 rcu_register_thread();
5584 if (set_signal_handler()) {
5586 goto exit_set_signal_handler
;
5589 setup_consumerd_path();
5591 page_size
= sysconf(_SC_PAGESIZE
);
5592 if (page_size
< 0) {
5593 PERROR("sysconf _SC_PAGESIZE");
5594 page_size
= LONG_MAX
;
5595 WARN("Fallback page size to %ld", page_size
);
5599 * Parse arguments and load the daemon configuration file.
5601 * We have an exit_options exit path to free memory reserved by
5602 * set_options. This is needed because the rest of sessiond_cleanup()
5603 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5604 * depends on set_options.
5607 if (set_options(argc
, argv
)) {
5613 if (opt_daemon
|| opt_background
) {
5616 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5624 * We are in the child. Make sure all other file descriptors are
5625 * closed, in case we are called with more opened file
5626 * descriptors than the standard ones.
5628 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5633 if (run_as_create_worker(argv
[0]) < 0) {
5634 goto exit_create_run_as_worker_cleanup
;
5638 * Starting from here, we can create threads. This needs to be after
5639 * lttng_daemonize due to RCU.
5643 * Initialize the health check subsystem. This call should set the
5644 * appropriate time values.
5646 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5647 if (!health_sessiond
) {
5648 PERROR("health_app_create error");
5650 goto exit_health_sessiond_cleanup
;
5653 if (init_ht_cleanup_quit_pipe()) {
5655 goto exit_ht_cleanup_quit_pipe
;
5658 /* Setup the thread ht_cleanup communication pipe. */
5659 if (utils_create_pipe_cloexec(ht_cleanup_pipe
)) {
5661 goto exit_ht_cleanup_pipe
;
5664 /* Set up max poll set size */
5665 if (lttng_poll_set_max_size()) {
5667 goto exit_set_max_size
;
5670 /* Create thread to clean up RCU hash tables */
5671 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5672 thread_ht_cleanup
, (void *) NULL
);
5675 PERROR("pthread_create ht_cleanup");
5677 goto exit_ht_cleanup
;
5680 /* Create thread quit pipe */
5681 if (init_thread_quit_pipe()) {
5683 goto exit_init_data
;
5686 /* Check if daemon is UID = 0 */
5687 is_root
= !getuid();
5690 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5693 goto exit_init_data
;
5696 /* Create global run dir with root access */
5697 if (create_lttng_rundir(rundir
)) {
5699 goto exit_init_data
;
5702 if (strlen(apps_unix_sock_path
) == 0) {
5703 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5704 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5707 goto exit_init_data
;
5711 if (strlen(client_unix_sock_path
) == 0) {
5712 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5713 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5716 goto exit_init_data
;
5720 /* Set global SHM for ust */
5721 if (strlen(wait_shm_path
) == 0) {
5722 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5723 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5726 goto exit_init_data
;
5730 if (strlen(health_unix_sock_path
) == 0) {
5731 ret
= snprintf(health_unix_sock_path
,
5732 sizeof(health_unix_sock_path
),
5733 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5736 goto exit_init_data
;
5740 /* Setup kernel consumerd path */
5741 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5742 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5745 goto exit_init_data
;
5747 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5748 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5751 goto exit_init_data
;
5754 DBG2("Kernel consumer err path: %s",
5755 kconsumer_data
.err_unix_sock_path
);
5756 DBG2("Kernel consumer cmd path: %s",
5757 kconsumer_data
.cmd_unix_sock_path
);
5759 home_path
= utils_get_home_dir();
5760 if (home_path
== NULL
) {
5761 /* TODO: Add --socket PATH option */
5762 ERR("Can't get HOME directory for sockets creation.");
5764 goto exit_init_data
;
5768 * Create rundir from home path. This will create something like
5771 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5774 goto exit_init_data
;
5777 if (create_lttng_rundir(rundir
)) {
5779 goto exit_init_data
;
5782 if (strlen(apps_unix_sock_path
) == 0) {
5783 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5784 DEFAULT_HOME_APPS_UNIX_SOCK
,
5788 goto exit_init_data
;
5792 /* Set the cli tool unix socket path */
5793 if (strlen(client_unix_sock_path
) == 0) {
5794 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5795 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5799 goto exit_init_data
;
5803 /* Set global SHM for ust */
5804 if (strlen(wait_shm_path
) == 0) {
5805 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5806 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5810 goto exit_init_data
;
5814 /* Set health check Unix path */
5815 if (strlen(health_unix_sock_path
) == 0) {
5816 ret
= snprintf(health_unix_sock_path
,
5817 sizeof(health_unix_sock_path
),
5818 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5822 goto exit_init_data
;
5827 lockfile_fd
= create_lockfile();
5828 if (lockfile_fd
< 0) {
5830 goto exit_init_data
;
5833 /* Set consumer initial state */
5834 kernel_consumerd_state
= CONSUMER_STOPPED
;
5835 ust_consumerd_state
= CONSUMER_STOPPED
;
5837 DBG("Client socket path %s", client_unix_sock_path
);
5838 DBG("Application socket path %s", apps_unix_sock_path
);
5839 DBG("Application wait path %s", wait_shm_path
);
5840 DBG("LTTng run directory path: %s", rundir
);
5842 /* 32 bits consumerd path setup */
5843 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5844 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5846 PERROR("snprintf 32-bit consumer error socket path");
5848 goto exit_init_data
;
5850 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5851 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5853 PERROR("snprintf 32-bit consumer command socket path");
5855 goto exit_init_data
;
5858 DBG2("UST consumer 32 bits err path: %s",
5859 ustconsumer32_data
.err_unix_sock_path
);
5860 DBG2("UST consumer 32 bits cmd path: %s",
5861 ustconsumer32_data
.cmd_unix_sock_path
);
5863 /* 64 bits consumerd path setup */
5864 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5865 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5867 PERROR("snprintf 64-bit consumer error socket path");
5869 goto exit_init_data
;
5871 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5872 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5874 PERROR("snprintf 64-bit consumer command socket path");
5876 goto exit_init_data
;
5879 DBG2("UST consumer 64 bits err path: %s",
5880 ustconsumer64_data
.err_unix_sock_path
);
5881 DBG2("UST consumer 64 bits cmd path: %s",
5882 ustconsumer64_data
.cmd_unix_sock_path
);
5885 * See if daemon already exist.
5887 if (check_existing_daemon()) {
5888 ERR("Already running daemon.\n");
5890 * We do not goto exit because we must not cleanup()
5891 * because a daemon is already running.
5894 goto exit_init_data
;
5898 * Init UST app hash table. Alloc hash table before this point since
5899 * cleanup() can get called after that point.
5901 if (ust_app_ht_alloc()) {
5902 ERR("Failed to allocate UST app hash table");
5904 goto exit_init_data
;
5908 * Initialize agent app hash table. We allocate the hash table here
5909 * since cleanup() can get called after this point.
5911 if (agent_app_ht_alloc()) {
5912 ERR("Failed to allocate Agent app hash table");
5914 goto exit_init_data
;
5918 * These actions must be executed as root. We do that *after* setting up
5919 * the sockets path because we MUST make the check for another daemon using
5920 * those paths *before* trying to set the kernel consumer sockets and init
5924 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
5926 goto exit_init_data
;
5929 /* Setup kernel tracer */
5930 if (!opt_no_kernel
) {
5931 init_kernel_tracer();
5932 if (kernel_tracer_fd
>= 0) {
5933 ret
= syscall_init_table();
5935 ERR("Unable to populate syscall table. "
5936 "Syscall tracing won't work "
5937 "for this session daemon.");
5942 /* Set ulimit for open files */
5945 /* init lttng_fd tracking must be done after set_ulimit. */
5948 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
5950 goto exit_init_data
;
5953 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
5955 goto exit_init_data
;
5958 /* Setup the needed unix socket */
5959 if (init_daemon_socket()) {
5961 goto exit_init_data
;
5964 /* Set credentials to socket */
5965 if (is_root
&& set_permissions(rundir
)) {
5967 goto exit_init_data
;
5970 /* Get parent pid if -S, --sig-parent is specified. */
5971 if (opt_sig_parent
) {
5975 /* Setup the kernel pipe for waking up the kernel thread */
5976 if (is_root
&& !opt_no_kernel
) {
5977 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5979 goto exit_init_data
;
5983 /* Setup the thread apps communication pipe. */
5984 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
5986 goto exit_init_data
;
5989 /* Setup the thread apps notify communication pipe. */
5990 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
5992 goto exit_init_data
;
5995 /* Initialize global buffer per UID and PID registry. */
5996 buffer_reg_init_uid_registry();
5997 buffer_reg_init_pid_registry();
5999 /* Init UST command queue. */
6000 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
6003 * Get session list pointer. This pointer MUST NOT be free'd. This list
6004 * is statically declared in session.c
6006 session_list_ptr
= session_get_list();
6010 /* Check for the application socket timeout env variable. */
6011 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
6012 if (env_app_timeout
) {
6013 app_socket_timeout
= atoi(env_app_timeout
);
6015 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
6018 ret
= write_pidfile();
6020 ERR("Error in write_pidfile");
6022 goto exit_init_data
;
6024 ret
= write_agent_port();
6026 ERR("Error in write_agent_port");
6028 goto exit_init_data
;
6031 /* Initialize communication library */
6033 /* Initialize TCP timeout values */
6034 lttcomm_inet_init();
6036 if (load_session_init_data(&load_info
) < 0) {
6038 goto exit_init_data
;
6040 load_info
->path
= opt_load_session_path
;
6042 /* Create health-check thread */
6043 ret
= pthread_create(&health_thread
, NULL
,
6044 thread_manage_health
, (void *) NULL
);
6047 PERROR("pthread_create health");
6052 /* Create thread to manage the client socket */
6053 ret
= pthread_create(&client_thread
, NULL
,
6054 thread_manage_clients
, (void *) NULL
);
6057 PERROR("pthread_create clients");
6062 /* Create thread to dispatch registration */
6063 ret
= pthread_create(&dispatch_thread
, NULL
,
6064 thread_dispatch_ust_registration
, (void *) NULL
);
6067 PERROR("pthread_create dispatch");
6072 /* Create thread to manage application registration. */
6073 ret
= pthread_create(®_apps_thread
, NULL
,
6074 thread_registration_apps
, (void *) NULL
);
6077 PERROR("pthread_create registration");
6082 /* Create thread to manage application socket */
6083 ret
= pthread_create(&apps_thread
, NULL
,
6084 thread_manage_apps
, (void *) NULL
);
6087 PERROR("pthread_create apps");
6092 /* Create thread to manage application notify socket */
6093 ret
= pthread_create(&apps_notify_thread
, NULL
,
6094 ust_thread_manage_notify
, (void *) NULL
);
6097 PERROR("pthread_create notify");
6099 goto exit_apps_notify
;
6102 /* Create agent registration thread. */
6103 ret
= pthread_create(&agent_reg_thread
, NULL
,
6104 agent_thread_manage_registration
, (void *) NULL
);
6107 PERROR("pthread_create agent");
6109 goto exit_agent_reg
;
6112 /* Don't start this thread if kernel tracing is not requested nor root */
6113 if (is_root
&& !opt_no_kernel
) {
6114 /* Create kernel thread to manage kernel event */
6115 ret
= pthread_create(&kernel_thread
, NULL
,
6116 thread_manage_kernel
, (void *) NULL
);
6119 PERROR("pthread_create kernel");
6125 /* Create session loading thread. */
6126 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
6130 PERROR("pthread_create load_session_thread");
6132 goto exit_load_session
;
6136 * This is where we start awaiting program completion (e.g. through
6137 * signal that asks threads to teardown).
6140 ret
= pthread_join(load_session_thread
, &status
);
6143 PERROR("pthread_join load_session_thread");
6148 if (is_root
&& !opt_no_kernel
) {
6149 ret
= pthread_join(kernel_thread
, &status
);
6152 PERROR("pthread_join");
6158 ret
= pthread_join(agent_reg_thread
, &status
);
6161 PERROR("pthread_join agent");
6166 ret
= pthread_join(apps_notify_thread
, &status
);
6169 PERROR("pthread_join apps notify");
6174 ret
= pthread_join(apps_thread
, &status
);
6177 PERROR("pthread_join apps");
6182 ret
= pthread_join(reg_apps_thread
, &status
);
6185 PERROR("pthread_join");
6191 * Join dispatch thread after joining reg_apps_thread to ensure
6192 * we don't leak applications in the queue.
6194 ret
= pthread_join(dispatch_thread
, &status
);
6197 PERROR("pthread_join");
6202 ret
= pthread_join(client_thread
, &status
);
6205 PERROR("pthread_join");
6210 ret
= pthread_join(health_thread
, &status
);
6213 PERROR("pthread_join health thread");
6220 * sessiond_cleanup() is called when no other thread is running, except
6221 * the ht_cleanup thread, which is needed to destroy the hash tables.
6223 rcu_thread_online();
6225 rcu_thread_offline();
6226 rcu_unregister_thread();
6228 ret
= notify_thread_pipe(ht_cleanup_quit_pipe
[1]);
6230 ERR("write error on ht_cleanup quit pipe");
6234 ret
= pthread_join(ht_cleanup_thread
, &status
);
6237 PERROR("pthread_join ht cleanup thread");
6243 utils_close_pipe(ht_cleanup_pipe
);
6244 exit_ht_cleanup_pipe
:
6247 * Close the ht_cleanup quit pipe.
6249 utils_close_pipe(ht_cleanup_quit_pipe
);
6250 exit_ht_cleanup_quit_pipe
:
6252 health_app_destroy(health_sessiond
);
6253 exit_health_sessiond_cleanup
:
6254 exit_create_run_as_worker_cleanup
:
6257 /* Ensure all prior call_rcu are done. */
6260 sessiond_cleanup_options();
6262 exit_set_signal_handler
: