2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
36 #include <sys/types.h>
38 #include <urcu/uatomic.h>
42 #include <common/common.h>
43 #include <common/compat/socket.h>
44 #include <common/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/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"
75 #define CONSUMERD_FILE "lttng-consumerd"
78 static const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
79 static int tracing_group_name_override
;
80 static char *opt_pidfile
;
81 static int opt_sig_parent
;
82 static int opt_verbose_consumer
;
83 static int opt_daemon
, opt_background
;
84 static int opt_no_kernel
;
85 static char *opt_load_session_path
;
86 static pid_t ppid
; /* Parent PID for --sig-parent option */
87 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
89 static int lockfile_fd
= -1;
91 /* Set to 1 when a SIGUSR1 signal is received. */
92 static int recv_child_signal
;
95 * Consumer daemon specific control data. Every value not initialized here is
96 * set to 0 by the static definition.
98 static struct consumer_data kconsumer_data
= {
99 .type
= LTTNG_CONSUMER_KERNEL
,
100 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
101 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
104 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
105 .lock
= PTHREAD_MUTEX_INITIALIZER
,
106 .cond
= PTHREAD_COND_INITIALIZER
,
107 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
109 static struct consumer_data ustconsumer64_data
= {
110 .type
= LTTNG_CONSUMER64_UST
,
111 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
112 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
115 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
116 .lock
= PTHREAD_MUTEX_INITIALIZER
,
117 .cond
= PTHREAD_COND_INITIALIZER
,
118 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
120 static struct consumer_data ustconsumer32_data
= {
121 .type
= LTTNG_CONSUMER32_UST
,
122 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
123 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
126 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
127 .lock
= PTHREAD_MUTEX_INITIALIZER
,
128 .cond
= PTHREAD_COND_INITIALIZER
,
129 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
132 /* Command line options */
133 static const struct option long_options
[] = {
134 { "client-sock", 1, 0, 'c' },
135 { "apps-sock", 1, 0, 'a' },
136 { "kconsumerd-cmd-sock", 1, 0, 'C' },
137 { "kconsumerd-err-sock", 1, 0, 'E' },
138 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
139 { "ustconsumerd32-err-sock", 1, 0, 'H' },
140 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
141 { "ustconsumerd64-err-sock", 1, 0, 'F' },
142 { "consumerd32-path", 1, 0, 'u' },
143 { "consumerd32-libdir", 1, 0, 'U' },
144 { "consumerd64-path", 1, 0, 't' },
145 { "consumerd64-libdir", 1, 0, 'T' },
146 { "daemonize", 0, 0, 'd' },
147 { "background", 0, 0, 'b' },
148 { "sig-parent", 0, 0, 'S' },
149 { "help", 0, 0, 'h' },
150 { "group", 1, 0, 'g' },
151 { "version", 0, 0, 'V' },
152 { "quiet", 0, 0, 'q' },
153 { "verbose", 0, 0, 'v' },
154 { "verbose-consumer", 0, 0, 'Z' },
155 { "no-kernel", 0, 0, 'N' },
156 { "pidfile", 1, 0, 'p' },
157 { "agent-tcp-port", 1, 0, 'J' },
158 { "config", 1, 0, 'f' },
159 { "load", 1, 0, 'l' },
160 { "kmod-probes", 1, 0, 'P' },
161 { "extra-kmod-probes", 1, 0, 'e' },
165 /* Command line options to ignore from configuration file */
166 static const char *config_ignore_options
[] = { "help", "version", "config" };
168 /* Shared between threads */
169 static int dispatch_thread_exit
;
171 /* Global application Unix socket path */
172 static char apps_unix_sock_path
[PATH_MAX
];
173 /* Global client Unix socket path */
174 static char client_unix_sock_path
[PATH_MAX
];
175 /* global wait shm path for UST */
176 static char wait_shm_path
[PATH_MAX
];
177 /* Global health check unix path */
178 static char health_unix_sock_path
[PATH_MAX
];
180 /* Sockets and FDs */
181 static int client_sock
= -1;
182 static int apps_sock
= -1;
183 int kernel_tracer_fd
= -1;
184 static int kernel_poll_pipe
[2] = { -1, -1 };
187 * Quit pipe for all threads. This permits a single cancellation point
188 * for all threads when receiving an event on the pipe.
190 static int thread_quit_pipe
[2] = { -1, -1 };
193 * This pipe is used to inform the thread managing application communication
194 * that a command is queued and ready to be processed.
196 static int apps_cmd_pipe
[2] = { -1, -1 };
198 int apps_cmd_notify_pipe
[2] = { -1, -1 };
200 /* Pthread, Mutexes and Semaphores */
201 static pthread_t apps_thread
;
202 static pthread_t apps_notify_thread
;
203 static pthread_t reg_apps_thread
;
204 static pthread_t client_thread
;
205 static pthread_t kernel_thread
;
206 static pthread_t dispatch_thread
;
207 static pthread_t health_thread
;
208 static pthread_t ht_cleanup_thread
;
209 static pthread_t agent_reg_thread
;
210 static pthread_t load_session_thread
;
213 * UST registration command queue. This queue is tied with a futex and uses a N
214 * wakers / 1 waiter implemented and detailed in futex.c/.h
216 * The thread_registration_apps and thread_dispatch_ust_registration uses this
217 * queue along with the wait/wake scheme. The thread_manage_apps receives down
218 * the line new application socket and monitors it for any I/O error or clean
219 * close that triggers an unregistration of the application.
221 static struct ust_cmd_queue ust_cmd_queue
;
224 * Pointer initialized before thread creation.
226 * This points to the tracing session list containing the session count and a
227 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
228 * MUST NOT be taken if you call a public function in session.c.
230 * The lock is nested inside the structure: session_list_ptr->lock. Please use
231 * session_lock_list and session_unlock_list for lock acquisition.
233 static struct ltt_session_list
*session_list_ptr
;
235 int ust_consumerd64_fd
= -1;
236 int ust_consumerd32_fd
= -1;
238 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
239 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
240 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
241 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
242 static int consumerd32_bin_override
;
243 static int consumerd64_bin_override
;
244 static int consumerd32_libdir_override
;
245 static int consumerd64_libdir_override
;
247 static const char *module_proc_lttng
= "/proc/lttng";
250 * Consumer daemon state which is changed when spawning it, killing it or in
251 * case of a fatal error.
253 enum consumerd_state
{
254 CONSUMER_STARTED
= 1,
255 CONSUMER_STOPPED
= 2,
260 * This consumer daemon state is used to validate if a client command will be
261 * able to reach the consumer. If not, the client is informed. For instance,
262 * doing a "lttng start" when the consumer state is set to ERROR will return an
263 * error to the client.
265 * The following example shows a possible race condition of this scheme:
267 * consumer thread error happens
269 * client cmd checks state -> still OK
270 * consumer thread exit, sets error
271 * client cmd try to talk to consumer
274 * However, since the consumer is a different daemon, we have no way of making
275 * sure the command will reach it safely even with this state flag. This is why
276 * we consider that up to the state validation during command processing, the
277 * command is safe. After that, we can not guarantee the correctness of the
278 * client request vis-a-vis the consumer.
280 static enum consumerd_state ust_consumerd_state
;
281 static enum consumerd_state kernel_consumerd_state
;
284 * Socket timeout for receiving and sending in seconds.
286 static int app_socket_timeout
;
288 /* Set in main() with the current page size. */
291 /* Application health monitoring */
292 struct health_app
*health_sessiond
;
294 /* Agent TCP port for registration. Used by the agent thread. */
295 unsigned int agent_tcp_port
= DEFAULT_AGENT_TCP_PORT
;
297 /* Am I root or not. */
298 int is_root
; /* Set to 1 if the daemon is running as root */
300 const char * const config_section_name
= "sessiond";
302 /* Load session thread information to operate. */
303 struct load_session_thread_data
*load_info
;
306 * Whether sessiond is ready for commands/health check requests.
307 * NR_LTTNG_SESSIOND_READY must match the number of calls to
308 * sessiond_notify_ready().
310 #define NR_LTTNG_SESSIOND_READY 3
311 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
313 /* Notify parents that we are ready for cmd and health check */
315 void sessiond_notify_ready(void)
317 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
319 * Notify parent pid that we are ready to accept command
320 * for client side. This ppid is the one from the
321 * external process that spawned us.
323 if (opt_sig_parent
) {
328 * Notify the parent of the fork() process that we are
331 if (opt_daemon
|| opt_background
) {
332 kill(child_ppid
, SIGUSR1
);
338 void setup_consumerd_path(void)
340 const char *bin
, *libdir
;
343 * Allow INSTALL_BIN_PATH to be used as a target path for the
344 * native architecture size consumer if CONFIG_CONSUMER*_PATH
345 * has not been defined.
347 #if (CAA_BITS_PER_LONG == 32)
348 if (!consumerd32_bin
[0]) {
349 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
351 if (!consumerd32_libdir
[0]) {
352 consumerd32_libdir
= INSTALL_LIB_PATH
;
354 #elif (CAA_BITS_PER_LONG == 64)
355 if (!consumerd64_bin
[0]) {
356 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
358 if (!consumerd64_libdir
[0]) {
359 consumerd64_libdir
= INSTALL_LIB_PATH
;
362 #error "Unknown bitness"
366 * runtime env. var. overrides the build default.
368 bin
= getenv("LTTNG_CONSUMERD32_BIN");
370 consumerd32_bin
= bin
;
372 bin
= getenv("LTTNG_CONSUMERD64_BIN");
374 consumerd64_bin
= bin
;
376 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
378 consumerd32_libdir
= libdir
;
380 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
382 consumerd64_libdir
= libdir
;
387 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
389 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
395 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
401 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
413 * Check if the thread quit pipe was triggered.
415 * Return 1 if it was triggered else 0;
417 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
419 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
427 * Init thread quit pipe.
429 * Return -1 on error or 0 if all pipes are created.
431 static int init_thread_quit_pipe(void)
435 ret
= pipe(thread_quit_pipe
);
437 PERROR("thread quit pipe");
441 for (i
= 0; i
< 2; i
++) {
442 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
454 * Stop all threads by closing the thread quit pipe.
456 static void stop_threads(void)
460 /* Stopping all threads */
461 DBG("Terminating all threads");
462 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
464 ERR("write error on thread quit pipe");
467 /* Dispatch thread */
468 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
469 futex_nto1_wake(&ust_cmd_queue
.futex
);
473 * Close every consumer sockets.
475 static void close_consumer_sockets(void)
479 if (kconsumer_data
.err_sock
>= 0) {
480 ret
= close(kconsumer_data
.err_sock
);
482 PERROR("kernel consumer err_sock close");
485 if (ustconsumer32_data
.err_sock
>= 0) {
486 ret
= close(ustconsumer32_data
.err_sock
);
488 PERROR("UST consumerd32 err_sock close");
491 if (ustconsumer64_data
.err_sock
>= 0) {
492 ret
= close(ustconsumer64_data
.err_sock
);
494 PERROR("UST consumerd64 err_sock close");
497 if (kconsumer_data
.cmd_sock
>= 0) {
498 ret
= close(kconsumer_data
.cmd_sock
);
500 PERROR("kernel consumer cmd_sock close");
503 if (ustconsumer32_data
.cmd_sock
>= 0) {
504 ret
= close(ustconsumer32_data
.cmd_sock
);
506 PERROR("UST consumerd32 cmd_sock close");
509 if (ustconsumer64_data
.cmd_sock
>= 0) {
510 ret
= close(ustconsumer64_data
.cmd_sock
);
512 PERROR("UST consumerd64 cmd_sock close");
518 * Generate the full lock file path using the rundir.
520 * Return the snprintf() return value thus a negative value is an error.
522 static int generate_lock_file_path(char *path
, size_t len
)
529 /* Build lockfile path from rundir. */
530 ret
= snprintf(path
, len
, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE
, rundir
);
532 PERROR("snprintf lockfile path");
541 static void cleanup(void)
544 struct ltt_session
*sess
, *stmp
;
550 * Close the thread quit pipe. It has already done its job,
551 * since we are now called.
553 utils_close_pipe(thread_quit_pipe
);
556 * If opt_pidfile is undefined, the default file will be wiped when
557 * removing the rundir.
560 ret
= remove(opt_pidfile
);
562 PERROR("remove pidfile %s", opt_pidfile
);
566 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
569 snprintf(path
, PATH_MAX
,
571 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
572 DBG("Removing %s", path
);
575 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
576 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
);
577 DBG("Removing %s", path
);
581 snprintf(path
, PATH_MAX
,
582 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
584 DBG("Removing %s", path
);
587 snprintf(path
, PATH_MAX
,
588 DEFAULT_KCONSUMERD_PATH
,
590 DBG("Removing directory %s", path
);
593 /* ust consumerd 32 */
594 snprintf(path
, PATH_MAX
,
595 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
597 DBG("Removing %s", path
);
600 snprintf(path
, PATH_MAX
,
601 DEFAULT_USTCONSUMERD32_PATH
,
603 DBG("Removing directory %s", path
);
606 /* ust consumerd 64 */
607 snprintf(path
, PATH_MAX
,
608 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
610 DBG("Removing %s", path
);
613 snprintf(path
, PATH_MAX
,
614 DEFAULT_USTCONSUMERD64_PATH
,
616 DBG("Removing directory %s", path
);
619 DBG("Cleaning up all sessions");
621 /* Destroy session list mutex */
622 if (session_list_ptr
!= NULL
) {
623 pthread_mutex_destroy(&session_list_ptr
->lock
);
625 /* Cleanup ALL session */
626 cds_list_for_each_entry_safe(sess
, stmp
,
627 &session_list_ptr
->head
, list
) {
628 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
632 DBG("Cleaning up all agent apps");
633 agent_app_ht_clean();
635 DBG("Closing all UST sockets");
636 ust_app_clean_list();
637 buffer_reg_destroy_registries();
639 if (is_root
&& !opt_no_kernel
) {
640 DBG2("Closing kernel fd");
641 if (kernel_tracer_fd
>= 0) {
642 ret
= close(kernel_tracer_fd
);
647 DBG("Unloading kernel modules");
648 modprobe_remove_lttng_all();
652 close_consumer_sockets();
655 * If the override option is set, the pointer points to a *non* const thus
656 * freeing it even though the variable type is set to const.
658 if (tracing_group_name_override
) {
659 free((void *) tracing_group_name
);
661 if (consumerd32_bin_override
) {
662 free((void *) consumerd32_bin
);
664 if (consumerd64_bin_override
) {
665 free((void *) consumerd64_bin
);
667 if (consumerd32_libdir_override
) {
668 free((void *) consumerd32_libdir
);
670 if (consumerd64_libdir_override
) {
671 free((void *) consumerd64_libdir
);
678 if (opt_load_session_path
) {
679 free(opt_load_session_path
);
683 load_session_destroy_data(load_info
);
688 * Cleanup lock file by deleting it and finaly closing it which will
689 * release the file system lock.
691 if (lockfile_fd
>= 0) {
692 char lockfile_path
[PATH_MAX
];
694 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
696 ret
= remove(lockfile_path
);
698 PERROR("remove lock file");
700 ret
= close(lockfile_fd
);
702 PERROR("close lock file");
708 * We do NOT rmdir rundir because there are other processes
709 * using it, for instance lttng-relayd, which can start in
710 * parallel with this teardown.
716 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
717 "Matthew, BEET driven development works!%c[%dm",
718 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
723 * Send data on a unix socket using the liblttsessiondcomm API.
725 * Return lttcomm error code.
727 static int send_unix_sock(int sock
, void *buf
, size_t len
)
729 /* Check valid length */
734 return lttcomm_send_unix_sock(sock
, buf
, len
);
738 * Free memory of a command context structure.
740 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
742 DBG("Clean command context structure");
744 if ((*cmd_ctx
)->llm
) {
745 free((*cmd_ctx
)->llm
);
747 if ((*cmd_ctx
)->lsm
) {
748 free((*cmd_ctx
)->lsm
);
756 * Notify UST applications using the shm mmap futex.
758 static int notify_ust_apps(int active
)
762 DBG("Notifying applications of session daemon state: %d", active
);
764 /* See shm.c for this call implying mmap, shm and futex calls */
765 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
766 if (wait_shm_mmap
== NULL
) {
770 /* Wake waiting process */
771 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
773 /* Apps notified successfully */
781 * Setup the outgoing data buffer for the response (llm) by allocating the
782 * right amount of memory and copying the original information from the lsm
785 * Return total size of the buffer pointed by buf.
787 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
793 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
794 if (cmd_ctx
->llm
== NULL
) {
800 /* Copy common data */
801 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
802 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
804 cmd_ctx
->llm
->data_size
= size
;
805 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
814 * Update the kernel poll set of all channel fd available over all tracing
815 * session. Add the wakeup pipe at the end of the set.
817 static int update_kernel_poll(struct lttng_poll_event
*events
)
820 struct ltt_session
*session
;
821 struct ltt_kernel_channel
*channel
;
823 DBG("Updating kernel poll set");
826 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
827 session_lock(session
);
828 if (session
->kernel_session
== NULL
) {
829 session_unlock(session
);
833 cds_list_for_each_entry(channel
,
834 &session
->kernel_session
->channel_list
.head
, list
) {
835 /* Add channel fd to the kernel poll set */
836 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
838 session_unlock(session
);
841 DBG("Channel fd %d added to kernel set", channel
->fd
);
843 session_unlock(session
);
845 session_unlock_list();
850 session_unlock_list();
855 * Find the channel fd from 'fd' over all tracing session. When found, check
856 * for new channel stream and send those stream fds to the kernel consumer.
858 * Useful for CPU hotplug feature.
860 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
863 struct ltt_session
*session
;
864 struct ltt_kernel_session
*ksess
;
865 struct ltt_kernel_channel
*channel
;
867 DBG("Updating kernel streams for channel fd %d", fd
);
870 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
871 session_lock(session
);
872 if (session
->kernel_session
== NULL
) {
873 session_unlock(session
);
876 ksess
= session
->kernel_session
;
878 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
879 if (channel
->fd
== fd
) {
880 DBG("Channel found, updating kernel streams");
881 ret
= kernel_open_channel_stream(channel
);
885 /* Update the stream global counter */
886 ksess
->stream_count_global
+= ret
;
889 * Have we already sent fds to the consumer? If yes, it means
890 * that tracing is started so it is safe to send our updated
893 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
894 struct lttng_ht_iter iter
;
895 struct consumer_socket
*socket
;
898 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
899 &iter
.iter
, socket
, node
.node
) {
900 pthread_mutex_lock(socket
->lock
);
901 ret
= kernel_consumer_send_channel_stream(socket
,
903 session
->output_traces
? 1 : 0);
904 pthread_mutex_unlock(socket
->lock
);
915 session_unlock(session
);
917 session_unlock_list();
921 session_unlock(session
);
922 session_unlock_list();
927 * For each tracing session, update newly registered apps. The session list
928 * lock MUST be acquired before calling this.
930 static void update_ust_app(int app_sock
)
932 struct ltt_session
*sess
, *stmp
;
934 /* Consumer is in an ERROR state. Stop any application update. */
935 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
936 /* Stop the update process since the consumer is dead. */
940 /* For all tracing session(s) */
941 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
943 if (sess
->ust_session
) {
944 ust_app_global_update(sess
->ust_session
, app_sock
);
946 session_unlock(sess
);
951 * This thread manage event coming from the kernel.
953 * Features supported in this thread:
956 static void *thread_manage_kernel(void *data
)
958 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
959 uint32_t revents
, nb_fd
;
961 struct lttng_poll_event events
;
963 DBG("[thread] Thread manage kernel started");
965 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
968 * This first step of the while is to clean this structure which could free
969 * non NULL pointers so initialize it before the loop.
971 lttng_poll_init(&events
);
973 if (testpoint(sessiond_thread_manage_kernel
)) {
974 goto error_testpoint
;
977 health_code_update();
979 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
980 goto error_testpoint
;
984 health_code_update();
986 if (update_poll_flag
== 1) {
987 /* Clean events object. We are about to populate it again. */
988 lttng_poll_clean(&events
);
990 ret
= sessiond_set_thread_pollset(&events
, 2);
992 goto error_poll_create
;
995 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1000 /* This will add the available kernel channel if any. */
1001 ret
= update_kernel_poll(&events
);
1005 update_poll_flag
= 0;
1008 DBG("Thread kernel polling");
1010 /* Poll infinite value of time */
1012 health_poll_entry();
1013 ret
= lttng_poll_wait(&events
, -1);
1014 DBG("Thread kernel return from poll on %d fds",
1015 LTTNG_POLL_GETNB(&events
));
1019 * Restart interrupted system call.
1021 if (errno
== EINTR
) {
1025 } else if (ret
== 0) {
1026 /* Should not happen since timeout is infinite */
1027 ERR("Return value of poll is 0 with an infinite timeout.\n"
1028 "This should not have happened! Continuing...");
1034 for (i
= 0; i
< nb_fd
; i
++) {
1035 /* Fetch once the poll data */
1036 revents
= LTTNG_POLL_GETEV(&events
, i
);
1037 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1039 health_code_update();
1042 /* No activity for this FD (poll implementation). */
1046 /* Thread quit pipe has been closed. Killing thread. */
1047 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1053 /* Check for data on kernel pipe */
1054 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
1055 (void) lttng_read(kernel_poll_pipe
[0],
1058 * Ret value is useless here, if this pipe gets any actions an
1059 * update is required anyway.
1061 update_poll_flag
= 1;
1065 * New CPU detected by the kernel. Adding kernel stream to
1066 * kernel session and updating the kernel consumer
1068 if (revents
& LPOLLIN
) {
1069 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1075 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1076 * and unregister kernel stream at this point.
1085 lttng_poll_clean(&events
);
1088 utils_close_pipe(kernel_poll_pipe
);
1089 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1092 ERR("Health error occurred in %s", __func__
);
1093 WARN("Kernel thread died unexpectedly. "
1094 "Kernel tracing can continue but CPU hotplug is disabled.");
1096 health_unregister(health_sessiond
);
1097 DBG("Kernel thread dying");
1102 * Signal pthread condition of the consumer data that the thread.
1104 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1106 pthread_mutex_lock(&data
->cond_mutex
);
1109 * The state is set before signaling. It can be any value, it's the waiter
1110 * job to correctly interpret this condition variable associated to the
1111 * consumer pthread_cond.
1113 * A value of 0 means that the corresponding thread of the consumer data
1114 * was not started. 1 indicates that the thread has started and is ready
1115 * for action. A negative value means that there was an error during the
1118 data
->consumer_thread_is_ready
= state
;
1119 (void) pthread_cond_signal(&data
->cond
);
1121 pthread_mutex_unlock(&data
->cond_mutex
);
1125 * This thread manage the consumer error sent back to the session daemon.
1127 static void *thread_manage_consumer(void *data
)
1129 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1130 uint32_t revents
, nb_fd
;
1131 enum lttcomm_return_code code
;
1132 struct lttng_poll_event events
;
1133 struct consumer_data
*consumer_data
= data
;
1135 DBG("[thread] Manage consumer started");
1137 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1139 health_code_update();
1142 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1143 * metadata_sock. Nothing more will be added to this poll set.
1145 ret
= sessiond_set_thread_pollset(&events
, 3);
1151 * The error socket here is already in a listening state which was done
1152 * just before spawning this thread to avoid a race between the consumer
1153 * daemon exec trying to connect and the listen() call.
1155 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1160 health_code_update();
1162 /* Infinite blocking call, waiting for transmission */
1164 health_poll_entry();
1166 if (testpoint(sessiond_thread_manage_consumer
)) {
1170 ret
= lttng_poll_wait(&events
, -1);
1174 * Restart interrupted system call.
1176 if (errno
== EINTR
) {
1184 for (i
= 0; i
< nb_fd
; i
++) {
1185 /* Fetch once the poll data */
1186 revents
= LTTNG_POLL_GETEV(&events
, i
);
1187 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1189 health_code_update();
1192 /* No activity for this FD (poll implementation). */
1196 /* Thread quit pipe has been closed. Killing thread. */
1197 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1203 /* Event on the registration socket */
1204 if (pollfd
== consumer_data
->err_sock
) {
1205 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1206 ERR("consumer err socket poll error");
1212 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1218 * Set the CLOEXEC flag. Return code is useless because either way, the
1221 (void) utils_set_fd_cloexec(sock
);
1223 health_code_update();
1225 DBG2("Receiving code from consumer err_sock");
1227 /* Getting status code from kconsumerd */
1228 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1229 sizeof(enum lttcomm_return_code
));
1234 health_code_update();
1235 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1236 /* Connect both socket, command and metadata. */
1237 consumer_data
->cmd_sock
=
1238 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1239 consumer_data
->metadata_fd
=
1240 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1241 if (consumer_data
->cmd_sock
< 0
1242 || consumer_data
->metadata_fd
< 0) {
1243 PERROR("consumer connect cmd socket");
1244 /* On error, signal condition and quit. */
1245 signal_consumer_condition(consumer_data
, -1);
1248 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1249 /* Create metadata socket lock. */
1250 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1251 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1252 PERROR("zmalloc pthread mutex");
1256 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1258 signal_consumer_condition(consumer_data
, 1);
1259 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1260 DBG("Consumer metadata socket ready (fd: %d)",
1261 consumer_data
->metadata_fd
);
1263 ERR("consumer error when waiting for SOCK_READY : %s",
1264 lttcomm_get_readable_code(-code
));
1268 /* Remove the consumerd error sock since we've established a connexion */
1269 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1274 /* Add new accepted error socket. */
1275 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1280 /* Add metadata socket that is successfully connected. */
1281 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1282 LPOLLIN
| LPOLLRDHUP
);
1287 health_code_update();
1289 /* Infinite blocking call, waiting for transmission */
1292 health_code_update();
1294 /* Exit the thread because the thread quit pipe has been triggered. */
1296 /* Not a health error. */
1301 health_poll_entry();
1302 ret
= lttng_poll_wait(&events
, -1);
1306 * Restart interrupted system call.
1308 if (errno
== EINTR
) {
1316 for (i
= 0; i
< nb_fd
; i
++) {
1317 /* Fetch once the poll data */
1318 revents
= LTTNG_POLL_GETEV(&events
, i
);
1319 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1321 health_code_update();
1324 /* No activity for this FD (poll implementation). */
1329 * Thread quit pipe has been triggered, flag that we should stop
1330 * but continue the current loop to handle potential data from
1333 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1335 if (pollfd
== sock
) {
1336 /* Event on the consumerd socket */
1337 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1338 ERR("consumer err socket second poll error");
1341 health_code_update();
1342 /* Wait for any kconsumerd error */
1343 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1344 sizeof(enum lttcomm_return_code
));
1346 ERR("consumer closed the command socket");
1350 ERR("consumer return code : %s",
1351 lttcomm_get_readable_code(-code
));
1354 } else if (pollfd
== consumer_data
->metadata_fd
) {
1355 /* UST metadata requests */
1356 ret
= ust_consumer_metadata_request(
1357 &consumer_data
->metadata_sock
);
1359 ERR("Handling metadata request");
1363 /* No need for an else branch all FDs are tested prior. */
1365 health_code_update();
1371 * We lock here because we are about to close the sockets and some other
1372 * thread might be using them so get exclusive access which will abort all
1373 * other consumer command by other threads.
1375 pthread_mutex_lock(&consumer_data
->lock
);
1377 /* Immediately set the consumerd state to stopped */
1378 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1379 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1380 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1381 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1382 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1384 /* Code flow error... */
1388 if (consumer_data
->err_sock
>= 0) {
1389 ret
= close(consumer_data
->err_sock
);
1393 consumer_data
->err_sock
= -1;
1395 if (consumer_data
->cmd_sock
>= 0) {
1396 ret
= close(consumer_data
->cmd_sock
);
1400 consumer_data
->cmd_sock
= -1;
1402 if (consumer_data
->metadata_sock
.fd_ptr
&&
1403 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1404 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1416 unlink(consumer_data
->err_unix_sock_path
);
1417 unlink(consumer_data
->cmd_unix_sock_path
);
1418 consumer_data
->pid
= 0;
1419 pthread_mutex_unlock(&consumer_data
->lock
);
1421 /* Cleanup metadata socket mutex. */
1422 if (consumer_data
->metadata_sock
.lock
) {
1423 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1424 free(consumer_data
->metadata_sock
.lock
);
1426 lttng_poll_clean(&events
);
1430 ERR("Health error occurred in %s", __func__
);
1432 health_unregister(health_sessiond
);
1433 DBG("consumer thread cleanup completed");
1439 * This thread manage application communication.
1441 static void *thread_manage_apps(void *data
)
1443 int i
, ret
, pollfd
, err
= -1;
1445 uint32_t revents
, nb_fd
;
1446 struct lttng_poll_event events
;
1448 DBG("[thread] Manage application started");
1450 rcu_register_thread();
1451 rcu_thread_online();
1453 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1455 if (testpoint(sessiond_thread_manage_apps
)) {
1456 goto error_testpoint
;
1459 health_code_update();
1461 ret
= sessiond_set_thread_pollset(&events
, 2);
1463 goto error_poll_create
;
1466 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1471 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1475 health_code_update();
1478 DBG("Apps thread polling");
1480 /* Inifinite blocking call, waiting for transmission */
1482 health_poll_entry();
1483 ret
= lttng_poll_wait(&events
, -1);
1484 DBG("Apps thread return from poll on %d fds",
1485 LTTNG_POLL_GETNB(&events
));
1489 * Restart interrupted system call.
1491 if (errno
== EINTR
) {
1499 for (i
= 0; i
< nb_fd
; i
++) {
1500 /* Fetch once the poll data */
1501 revents
= LTTNG_POLL_GETEV(&events
, i
);
1502 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1504 health_code_update();
1507 /* No activity for this FD (poll implementation). */
1511 /* Thread quit pipe has been closed. Killing thread. */
1512 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1518 /* Inspect the apps cmd pipe */
1519 if (pollfd
== apps_cmd_pipe
[0]) {
1520 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1521 ERR("Apps command pipe error");
1523 } else if (revents
& LPOLLIN
) {
1527 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1528 if (size_ret
< sizeof(sock
)) {
1529 PERROR("read apps cmd pipe");
1533 health_code_update();
1536 * We only monitor the error events of the socket. This
1537 * thread does not handle any incoming data from UST
1540 ret
= lttng_poll_add(&events
, sock
,
1541 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1546 DBG("Apps with sock %d added to poll set", sock
);
1550 * At this point, we know that a registered application made
1551 * the event at poll_wait.
1553 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1554 /* Removing from the poll set */
1555 ret
= lttng_poll_del(&events
, pollfd
);
1560 /* Socket closed on remote end. */
1561 ust_app_unregister(pollfd
);
1565 health_code_update();
1571 lttng_poll_clean(&events
);
1574 utils_close_pipe(apps_cmd_pipe
);
1575 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1578 * We don't clean the UST app hash table here since already registered
1579 * applications can still be controlled so let them be until the session
1580 * daemon dies or the applications stop.
1585 ERR("Health error occurred in %s", __func__
);
1587 health_unregister(health_sessiond
);
1588 DBG("Application communication apps thread cleanup complete");
1589 rcu_thread_offline();
1590 rcu_unregister_thread();
1595 * Send a socket to a thread This is called from the dispatch UST registration
1596 * thread once all sockets are set for the application.
1598 * The sock value can be invalid, we don't really care, the thread will handle
1599 * it and make the necessary cleanup if so.
1601 * On success, return 0 else a negative value being the errno message of the
1604 static int send_socket_to_thread(int fd
, int sock
)
1609 * It's possible that the FD is set as invalid with -1 concurrently just
1610 * before calling this function being a shutdown state of the thread.
1617 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1618 if (ret
< sizeof(sock
)) {
1619 PERROR("write apps pipe %d", fd
);
1626 /* All good. Don't send back the write positive ret value. */
1633 * Sanitize the wait queue of the dispatch registration thread meaning removing
1634 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1635 * notify socket is never received.
1637 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1639 int ret
, nb_fd
= 0, i
;
1640 unsigned int fd_added
= 0;
1641 struct lttng_poll_event events
;
1642 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1646 lttng_poll_init(&events
);
1648 /* Just skip everything for an empty queue. */
1649 if (!wait_queue
->count
) {
1653 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1658 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1659 &wait_queue
->head
, head
) {
1660 assert(wait_node
->app
);
1661 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1662 LPOLLHUP
| LPOLLERR
);
1675 * Poll but don't block so we can quickly identify the faulty events and
1676 * clean them afterwards from the wait queue.
1678 ret
= lttng_poll_wait(&events
, 0);
1684 for (i
= 0; i
< nb_fd
; i
++) {
1685 /* Get faulty FD. */
1686 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1687 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1690 /* No activity for this FD (poll implementation). */
1694 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1695 &wait_queue
->head
, head
) {
1696 if (pollfd
== wait_node
->app
->sock
&&
1697 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1698 cds_list_del(&wait_node
->head
);
1699 wait_queue
->count
--;
1700 ust_app_destroy(wait_node
->app
);
1708 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1712 lttng_poll_clean(&events
);
1716 lttng_poll_clean(&events
);
1718 ERR("Unable to sanitize wait queue");
1723 * Dispatch request from the registration threads to the application
1724 * communication thread.
1726 static void *thread_dispatch_ust_registration(void *data
)
1729 struct cds_wfcq_node
*node
;
1730 struct ust_command
*ust_cmd
= NULL
;
1731 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1732 struct ust_reg_wait_queue wait_queue
= {
1736 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1738 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1739 goto error_testpoint
;
1742 health_code_update();
1744 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1746 DBG("[thread] Dispatch UST command started");
1748 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1749 health_code_update();
1751 /* Atomically prepare the queue futex */
1752 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1755 struct ust_app
*app
= NULL
;
1759 * Make sure we don't have node(s) that have hung up before receiving
1760 * the notify socket. This is to clean the list in order to avoid
1761 * memory leaks from notify socket that are never seen.
1763 sanitize_wait_queue(&wait_queue
);
1765 health_code_update();
1766 /* Dequeue command for registration */
1767 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1769 DBG("Woken up but nothing in the UST command queue");
1770 /* Continue thread execution */
1774 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1776 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1777 " gid:%d sock:%d name:%s (version %d.%d)",
1778 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1779 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1780 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1781 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1783 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1784 wait_node
= zmalloc(sizeof(*wait_node
));
1786 PERROR("zmalloc wait_node dispatch");
1787 ret
= close(ust_cmd
->sock
);
1789 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1791 lttng_fd_put(LTTNG_FD_APPS
, 1);
1795 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1797 /* Create application object if socket is CMD. */
1798 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1800 if (!wait_node
->app
) {
1801 ret
= close(ust_cmd
->sock
);
1803 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1805 lttng_fd_put(LTTNG_FD_APPS
, 1);
1811 * Add application to the wait queue so we can set the notify
1812 * socket before putting this object in the global ht.
1814 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1819 * We have to continue here since we don't have the notify
1820 * socket and the application MUST be added to the hash table
1821 * only at that moment.
1826 * Look for the application in the local wait queue and set the
1827 * notify socket if found.
1829 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1830 &wait_queue
.head
, head
) {
1831 health_code_update();
1832 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1833 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1834 cds_list_del(&wait_node
->head
);
1836 app
= wait_node
->app
;
1838 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1844 * With no application at this stage the received socket is
1845 * basically useless so close it before we free the cmd data
1846 * structure for good.
1849 ret
= close(ust_cmd
->sock
);
1851 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1853 lttng_fd_put(LTTNG_FD_APPS
, 1);
1860 * @session_lock_list
1862 * Lock the global session list so from the register up to the
1863 * registration done message, no thread can see the application
1864 * and change its state.
1866 session_lock_list();
1870 * Add application to the global hash table. This needs to be
1871 * done before the update to the UST registry can locate the
1876 /* Set app version. This call will print an error if needed. */
1877 (void) ust_app_version(app
);
1879 /* Send notify socket through the notify pipe. */
1880 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1884 session_unlock_list();
1886 * No notify thread, stop the UST tracing. However, this is
1887 * not an internal error of the this thread thus setting
1888 * the health error code to a normal exit.
1895 * Update newly registered application with the tracing
1896 * registry info already enabled information.
1898 update_ust_app(app
->sock
);
1901 * Don't care about return value. Let the manage apps threads
1902 * handle app unregistration upon socket close.
1904 (void) ust_app_register_done(app
->sock
);
1907 * Even if the application socket has been closed, send the app
1908 * to the thread and unregistration will take place at that
1911 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1914 session_unlock_list();
1916 * No apps. thread, stop the UST tracing. However, this is
1917 * not an internal error of the this thread thus setting
1918 * the health error code to a normal exit.
1925 session_unlock_list();
1927 } while (node
!= NULL
);
1929 health_poll_entry();
1930 /* Futex wait on queue. Blocking call on futex() */
1931 futex_nto1_wait(&ust_cmd_queue
.futex
);
1934 /* Normal exit, no error */
1938 /* Clean up wait queue. */
1939 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1940 &wait_queue
.head
, head
) {
1941 cds_list_del(&wait_node
->head
);
1947 DBG("Dispatch thread dying");
1950 ERR("Health error occurred in %s", __func__
);
1952 health_unregister(health_sessiond
);
1957 * This thread manage application registration.
1959 static void *thread_registration_apps(void *data
)
1961 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1962 uint32_t revents
, nb_fd
;
1963 struct lttng_poll_event events
;
1965 * Get allocated in this thread, enqueued to a global queue, dequeued and
1966 * freed in the manage apps thread.
1968 struct ust_command
*ust_cmd
= NULL
;
1970 DBG("[thread] Manage application registration started");
1972 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
1974 if (testpoint(sessiond_thread_registration_apps
)) {
1975 goto error_testpoint
;
1978 ret
= lttcomm_listen_unix_sock(apps_sock
);
1984 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1985 * more will be added to this poll set.
1987 ret
= sessiond_set_thread_pollset(&events
, 2);
1989 goto error_create_poll
;
1992 /* Add the application registration socket */
1993 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1995 goto error_poll_add
;
1998 /* Notify all applications to register */
1999 ret
= notify_ust_apps(1);
2001 ERR("Failed to notify applications or create the wait shared memory.\n"
2002 "Execution continues but there might be problem for already\n"
2003 "running applications that wishes to register.");
2007 DBG("Accepting application registration");
2009 /* Inifinite blocking call, waiting for transmission */
2011 health_poll_entry();
2012 ret
= lttng_poll_wait(&events
, -1);
2016 * Restart interrupted system call.
2018 if (errno
== EINTR
) {
2026 for (i
= 0; i
< nb_fd
; i
++) {
2027 health_code_update();
2029 /* Fetch once the poll data */
2030 revents
= LTTNG_POLL_GETEV(&events
, i
);
2031 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2034 /* No activity for this FD (poll implementation). */
2038 /* Thread quit pipe has been closed. Killing thread. */
2039 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2045 /* Event on the registration socket */
2046 if (pollfd
== apps_sock
) {
2047 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2048 ERR("Register apps socket poll error");
2050 } else if (revents
& LPOLLIN
) {
2051 sock
= lttcomm_accept_unix_sock(apps_sock
);
2057 * Set socket timeout for both receiving and ending.
2058 * app_socket_timeout is in seconds, whereas
2059 * lttcomm_setsockopt_rcv_timeout and
2060 * lttcomm_setsockopt_snd_timeout expect msec as
2063 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2064 app_socket_timeout
* 1000);
2065 (void) lttcomm_setsockopt_snd_timeout(sock
,
2066 app_socket_timeout
* 1000);
2069 * Set the CLOEXEC flag. Return code is useless because
2070 * either way, the show must go on.
2072 (void) utils_set_fd_cloexec(sock
);
2074 /* Create UST registration command for enqueuing */
2075 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2076 if (ust_cmd
== NULL
) {
2077 PERROR("ust command zmalloc");
2082 * Using message-based transmissions to ensure we don't
2083 * have to deal with partially received messages.
2085 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2087 ERR("Exhausted file descriptors allowed for applications.");
2097 health_code_update();
2098 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2101 /* Close socket of the application. */
2106 lttng_fd_put(LTTNG_FD_APPS
, 1);
2110 health_code_update();
2112 ust_cmd
->sock
= sock
;
2115 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2116 " gid:%d sock:%d name:%s (version %d.%d)",
2117 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2118 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2119 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2120 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2123 * Lock free enqueue the registration request. The red pill
2124 * has been taken! This apps will be part of the *system*.
2126 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2129 * Wake the registration queue futex. Implicit memory
2130 * barrier with the exchange in cds_wfcq_enqueue.
2132 futex_nto1_wake(&ust_cmd_queue
.futex
);
2140 /* Notify that the registration thread is gone */
2143 if (apps_sock
>= 0) {
2144 ret
= close(apps_sock
);
2154 lttng_fd_put(LTTNG_FD_APPS
, 1);
2156 unlink(apps_unix_sock_path
);
2159 lttng_poll_clean(&events
);
2163 DBG("UST Registration thread cleanup complete");
2166 ERR("Health error occurred in %s", __func__
);
2168 health_unregister(health_sessiond
);
2174 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2175 * exec or it will fails.
2177 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2180 struct timespec timeout
;
2182 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2183 consumer_data
->consumer_thread_is_ready
= 0;
2185 /* Setup pthread condition */
2186 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2189 PERROR("pthread_condattr_init consumer data");
2194 * Set the monotonic clock in order to make sure we DO NOT jump in time
2195 * between the clock_gettime() call and the timedwait call. See bug #324
2196 * for a more details and how we noticed it.
2198 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2201 PERROR("pthread_condattr_setclock consumer data");
2205 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2208 PERROR("pthread_cond_init consumer data");
2212 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2215 PERROR("pthread_create consumer");
2220 /* We are about to wait on a pthread condition */
2221 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2223 /* Get time for sem_timedwait absolute timeout */
2224 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2226 * Set the timeout for the condition timed wait even if the clock gettime
2227 * call fails since we might loop on that call and we want to avoid to
2228 * increment the timeout too many times.
2230 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2233 * The following loop COULD be skipped in some conditions so this is why we
2234 * set ret to 0 in order to make sure at least one round of the loop is
2240 * Loop until the condition is reached or when a timeout is reached. Note
2241 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2242 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2243 * possible. This loop does not take any chances and works with both of
2246 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2247 if (clock_ret
< 0) {
2248 PERROR("clock_gettime spawn consumer");
2249 /* Infinite wait for the consumerd thread to be ready */
2250 ret
= pthread_cond_wait(&consumer_data
->cond
,
2251 &consumer_data
->cond_mutex
);
2253 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2254 &consumer_data
->cond_mutex
, &timeout
);
2258 /* Release the pthread condition */
2259 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2263 if (ret
== ETIMEDOUT
) {
2267 * Call has timed out so we kill the kconsumerd_thread and return
2270 ERR("Condition timed out. The consumer thread was never ready."
2272 pth_ret
= pthread_cancel(consumer_data
->thread
);
2274 PERROR("pthread_cancel consumer thread");
2277 PERROR("pthread_cond_wait failed consumer thread");
2279 /* Caller is expecting a negative value on failure. */
2284 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2285 if (consumer_data
->pid
== 0) {
2286 ERR("Consumerd did not start");
2287 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2290 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2299 * Join consumer thread
2301 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2305 /* Consumer pid must be a real one. */
2306 if (consumer_data
->pid
> 0) {
2308 ret
= kill(consumer_data
->pid
, SIGTERM
);
2310 ERR("Error killing consumer daemon");
2313 return pthread_join(consumer_data
->thread
, &status
);
2320 * Fork and exec a consumer daemon (consumerd).
2322 * Return pid if successful else -1.
2324 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2328 const char *consumer_to_use
;
2329 const char *verbosity
;
2332 DBG("Spawning consumerd");
2339 if (opt_verbose_consumer
) {
2340 verbosity
= "--verbose";
2341 } else if (lttng_opt_quiet
) {
2342 verbosity
= "--quiet";
2347 switch (consumer_data
->type
) {
2348 case LTTNG_CONSUMER_KERNEL
:
2350 * Find out which consumerd to execute. We will first try the
2351 * 64-bit path, then the sessiond's installation directory, and
2352 * fallback on the 32-bit one,
2354 DBG3("Looking for a kernel consumer at these locations:");
2355 DBG3(" 1) %s", consumerd64_bin
);
2356 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2357 DBG3(" 3) %s", consumerd32_bin
);
2358 if (stat(consumerd64_bin
, &st
) == 0) {
2359 DBG3("Found location #1");
2360 consumer_to_use
= consumerd64_bin
;
2361 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2362 DBG3("Found location #2");
2363 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2364 } else if (stat(consumerd32_bin
, &st
) == 0) {
2365 DBG3("Found location #3");
2366 consumer_to_use
= consumerd32_bin
;
2368 DBG("Could not find any valid consumerd executable");
2372 DBG("Using kernel consumer at: %s", consumer_to_use
);
2373 ret
= execl(consumer_to_use
,
2374 "lttng-consumerd", verbosity
, "-k",
2375 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2376 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2377 "--group", tracing_group_name
,
2380 case LTTNG_CONSUMER64_UST
:
2382 char *tmpnew
= NULL
;
2384 if (consumerd64_libdir
[0] != '\0') {
2388 tmp
= getenv("LD_LIBRARY_PATH");
2392 tmplen
= strlen("LD_LIBRARY_PATH=")
2393 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2394 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2399 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2400 strcat(tmpnew
, consumerd64_libdir
);
2401 if (tmp
[0] != '\0') {
2402 strcat(tmpnew
, ":");
2403 strcat(tmpnew
, tmp
);
2405 ret
= putenv(tmpnew
);
2412 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2413 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2414 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2415 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2416 "--group", tracing_group_name
,
2418 if (consumerd64_libdir
[0] != '\0') {
2423 case LTTNG_CONSUMER32_UST
:
2425 char *tmpnew
= NULL
;
2427 if (consumerd32_libdir
[0] != '\0') {
2431 tmp
= getenv("LD_LIBRARY_PATH");
2435 tmplen
= strlen("LD_LIBRARY_PATH=")
2436 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2437 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2442 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2443 strcat(tmpnew
, consumerd32_libdir
);
2444 if (tmp
[0] != '\0') {
2445 strcat(tmpnew
, ":");
2446 strcat(tmpnew
, tmp
);
2448 ret
= putenv(tmpnew
);
2455 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2456 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2457 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2458 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2459 "--group", tracing_group_name
,
2461 if (consumerd32_libdir
[0] != '\0') {
2467 PERROR("unknown consumer type");
2471 PERROR("Consumer execl()");
2473 /* Reaching this point, we got a failure on our execl(). */
2475 } else if (pid
> 0) {
2478 PERROR("start consumer fork");
2486 * Spawn the consumerd daemon and session daemon thread.
2488 static int start_consumerd(struct consumer_data
*consumer_data
)
2493 * Set the listen() state on the socket since there is a possible race
2494 * between the exec() of the consumer daemon and this call if place in the
2495 * consumer thread. See bug #366 for more details.
2497 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2502 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2503 if (consumer_data
->pid
!= 0) {
2504 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2508 ret
= spawn_consumerd(consumer_data
);
2510 ERR("Spawning consumerd failed");
2511 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2515 /* Setting up the consumer_data pid */
2516 consumer_data
->pid
= ret
;
2517 DBG2("Consumer pid %d", consumer_data
->pid
);
2518 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2520 DBG2("Spawning consumer control thread");
2521 ret
= spawn_consumer_thread(consumer_data
);
2523 ERR("Fatal error spawning consumer control thread");
2531 /* Cleanup already created sockets on error. */
2532 if (consumer_data
->err_sock
>= 0) {
2535 err
= close(consumer_data
->err_sock
);
2537 PERROR("close consumer data error socket");
2544 * Setup necessary data for kernel tracer action.
2546 static int init_kernel_tracer(void)
2550 /* Modprobe lttng kernel modules */
2551 ret
= modprobe_lttng_control();
2556 /* Open debugfs lttng */
2557 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2558 if (kernel_tracer_fd
< 0) {
2559 DBG("Failed to open %s", module_proc_lttng
);
2564 /* Validate kernel version */
2565 ret
= kernel_validate_version(kernel_tracer_fd
);
2570 ret
= modprobe_lttng_data();
2575 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2579 modprobe_remove_lttng_control();
2580 ret
= close(kernel_tracer_fd
);
2584 kernel_tracer_fd
= -1;
2585 return LTTNG_ERR_KERN_VERSION
;
2588 ret
= close(kernel_tracer_fd
);
2594 modprobe_remove_lttng_control();
2597 WARN("No kernel tracer available");
2598 kernel_tracer_fd
= -1;
2600 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2602 return LTTNG_ERR_KERN_NA
;
2608 * Copy consumer output from the tracing session to the domain session. The
2609 * function also applies the right modification on a per domain basis for the
2610 * trace files destination directory.
2612 * Should *NOT* be called with RCU read-side lock held.
2614 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2617 const char *dir_name
;
2618 struct consumer_output
*consumer
;
2621 assert(session
->consumer
);
2624 case LTTNG_DOMAIN_KERNEL
:
2625 DBG3("Copying tracing session consumer output in kernel session");
2627 * XXX: We should audit the session creation and what this function
2628 * does "extra" in order to avoid a destroy since this function is used
2629 * in the domain session creation (kernel and ust) only. Same for UST
2632 if (session
->kernel_session
->consumer
) {
2633 consumer_destroy_output(session
->kernel_session
->consumer
);
2635 session
->kernel_session
->consumer
=
2636 consumer_copy_output(session
->consumer
);
2637 /* Ease our life a bit for the next part */
2638 consumer
= session
->kernel_session
->consumer
;
2639 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2641 case LTTNG_DOMAIN_JUL
:
2642 case LTTNG_DOMAIN_LOG4J
:
2643 case LTTNG_DOMAIN_UST
:
2644 DBG3("Copying tracing session consumer output in UST session");
2645 if (session
->ust_session
->consumer
) {
2646 consumer_destroy_output(session
->ust_session
->consumer
);
2648 session
->ust_session
->consumer
=
2649 consumer_copy_output(session
->consumer
);
2650 /* Ease our life a bit for the next part */
2651 consumer
= session
->ust_session
->consumer
;
2652 dir_name
= DEFAULT_UST_TRACE_DIR
;
2655 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2659 /* Append correct directory to subdir */
2660 strncat(consumer
->subdir
, dir_name
,
2661 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2662 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2671 * Create an UST session and add it to the session ust list.
2673 * Should *NOT* be called with RCU read-side lock held.
2675 static int create_ust_session(struct ltt_session
*session
,
2676 struct lttng_domain
*domain
)
2679 struct ltt_ust_session
*lus
= NULL
;
2683 assert(session
->consumer
);
2685 switch (domain
->type
) {
2686 case LTTNG_DOMAIN_JUL
:
2687 case LTTNG_DOMAIN_LOG4J
:
2688 case LTTNG_DOMAIN_UST
:
2691 ERR("Unknown UST domain on create session %d", domain
->type
);
2692 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2696 DBG("Creating UST session");
2698 lus
= trace_ust_create_session(session
->id
);
2700 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2704 lus
->uid
= session
->uid
;
2705 lus
->gid
= session
->gid
;
2706 lus
->output_traces
= session
->output_traces
;
2707 lus
->snapshot_mode
= session
->snapshot_mode
;
2708 lus
->live_timer_interval
= session
->live_timer
;
2709 session
->ust_session
= lus
;
2711 /* Copy session output to the newly created UST session */
2712 ret
= copy_session_consumer(domain
->type
, session
);
2713 if (ret
!= LTTNG_OK
) {
2721 session
->ust_session
= NULL
;
2726 * Create a kernel tracer session then create the default channel.
2728 static int create_kernel_session(struct ltt_session
*session
)
2732 DBG("Creating kernel session");
2734 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2736 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2740 /* Code flow safety */
2741 assert(session
->kernel_session
);
2743 /* Copy session output to the newly created Kernel session */
2744 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2745 if (ret
!= LTTNG_OK
) {
2749 /* Create directory(ies) on local filesystem. */
2750 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2751 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2752 ret
= run_as_mkdir_recursive(
2753 session
->kernel_session
->consumer
->dst
.trace_path
,
2754 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2756 if (ret
!= -EEXIST
) {
2757 ERR("Trace directory creation error");
2763 session
->kernel_session
->uid
= session
->uid
;
2764 session
->kernel_session
->gid
= session
->gid
;
2765 session
->kernel_session
->output_traces
= session
->output_traces
;
2766 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2771 trace_kernel_destroy_session(session
->kernel_session
);
2772 session
->kernel_session
= NULL
;
2777 * Count number of session permitted by uid/gid.
2779 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2782 struct ltt_session
*session
;
2784 DBG("Counting number of available session for UID %d GID %d",
2786 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2788 * Only list the sessions the user can control.
2790 if (!session_access_ok(session
, uid
, gid
)) {
2799 * Process the command requested by the lttng client within the command
2800 * context structure. This function make sure that the return structure (llm)
2801 * is set and ready for transmission before returning.
2803 * Return any error encountered or 0 for success.
2805 * "sock" is only used for special-case var. len data.
2807 * Should *NOT* be called with RCU read-side lock held.
2809 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2813 int need_tracing_session
= 1;
2816 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2820 switch (cmd_ctx
->lsm
->cmd_type
) {
2821 case LTTNG_CREATE_SESSION
:
2822 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2823 case LTTNG_CREATE_SESSION_LIVE
:
2824 case LTTNG_DESTROY_SESSION
:
2825 case LTTNG_LIST_SESSIONS
:
2826 case LTTNG_LIST_DOMAINS
:
2827 case LTTNG_START_TRACE
:
2828 case LTTNG_STOP_TRACE
:
2829 case LTTNG_DATA_PENDING
:
2830 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2831 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2832 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2833 case LTTNG_SNAPSHOT_RECORD
:
2834 case LTTNG_SAVE_SESSION
:
2841 if (opt_no_kernel
&& need_domain
2842 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2844 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2846 ret
= LTTNG_ERR_KERN_NA
;
2851 /* Deny register consumer if we already have a spawned consumer. */
2852 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2853 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2854 if (kconsumer_data
.pid
> 0) {
2855 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2856 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2859 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2863 * Check for command that don't needs to allocate a returned payload. We do
2864 * this here so we don't have to make the call for no payload at each
2867 switch(cmd_ctx
->lsm
->cmd_type
) {
2868 case LTTNG_LIST_SESSIONS
:
2869 case LTTNG_LIST_TRACEPOINTS
:
2870 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2871 case LTTNG_LIST_DOMAINS
:
2872 case LTTNG_LIST_CHANNELS
:
2873 case LTTNG_LIST_EVENTS
:
2874 case LTTNG_LIST_SYSCALLS
:
2877 /* Setup lttng message with no payload */
2878 ret
= setup_lttng_msg(cmd_ctx
, 0);
2880 /* This label does not try to unlock the session */
2881 goto init_setup_error
;
2885 /* Commands that DO NOT need a session. */
2886 switch (cmd_ctx
->lsm
->cmd_type
) {
2887 case LTTNG_CREATE_SESSION
:
2888 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2889 case LTTNG_CREATE_SESSION_LIVE
:
2890 case LTTNG_CALIBRATE
:
2891 case LTTNG_LIST_SESSIONS
:
2892 case LTTNG_LIST_TRACEPOINTS
:
2893 case LTTNG_LIST_SYSCALLS
:
2894 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2895 case LTTNG_SAVE_SESSION
:
2896 need_tracing_session
= 0;
2899 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2901 * We keep the session list lock across _all_ commands
2902 * for now, because the per-session lock does not
2903 * handle teardown properly.
2905 session_lock_list();
2906 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2907 if (cmd_ctx
->session
== NULL
) {
2908 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2911 /* Acquire lock for the session */
2912 session_lock(cmd_ctx
->session
);
2918 * Commands that need a valid session but should NOT create one if none
2919 * exists. Instead of creating one and destroying it when the command is
2920 * handled, process that right before so we save some round trip in useless
2923 switch (cmd_ctx
->lsm
->cmd_type
) {
2924 case LTTNG_DISABLE_CHANNEL
:
2925 case LTTNG_DISABLE_EVENT
:
2926 switch (cmd_ctx
->lsm
->domain
.type
) {
2927 case LTTNG_DOMAIN_KERNEL
:
2928 if (!cmd_ctx
->session
->kernel_session
) {
2929 ret
= LTTNG_ERR_NO_CHANNEL
;
2933 case LTTNG_DOMAIN_JUL
:
2934 case LTTNG_DOMAIN_LOG4J
:
2935 case LTTNG_DOMAIN_UST
:
2936 if (!cmd_ctx
->session
->ust_session
) {
2937 ret
= LTTNG_ERR_NO_CHANNEL
;
2942 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2954 * Check domain type for specific "pre-action".
2956 switch (cmd_ctx
->lsm
->domain
.type
) {
2957 case LTTNG_DOMAIN_KERNEL
:
2959 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2963 /* Kernel tracer check */
2964 if (kernel_tracer_fd
== -1) {
2965 /* Basically, load kernel tracer modules */
2966 ret
= init_kernel_tracer();
2972 /* Consumer is in an ERROR state. Report back to client */
2973 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2974 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2978 /* Need a session for kernel command */
2979 if (need_tracing_session
) {
2980 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2981 ret
= create_kernel_session(cmd_ctx
->session
);
2983 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2988 /* Start the kernel consumer daemon */
2989 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2990 if (kconsumer_data
.pid
== 0 &&
2991 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2992 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2993 ret
= start_consumerd(&kconsumer_data
);
2995 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2998 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3000 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3004 * The consumer was just spawned so we need to add the socket to
3005 * the consumer output of the session if exist.
3007 ret
= consumer_create_socket(&kconsumer_data
,
3008 cmd_ctx
->session
->kernel_session
->consumer
);
3015 case LTTNG_DOMAIN_JUL
:
3016 case LTTNG_DOMAIN_LOG4J
:
3017 case LTTNG_DOMAIN_UST
:
3019 if (!ust_app_supported()) {
3020 ret
= LTTNG_ERR_NO_UST
;
3023 /* Consumer is in an ERROR state. Report back to client */
3024 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3025 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3029 if (need_tracing_session
) {
3030 /* Create UST session if none exist. */
3031 if (cmd_ctx
->session
->ust_session
== NULL
) {
3032 ret
= create_ust_session(cmd_ctx
->session
,
3033 &cmd_ctx
->lsm
->domain
);
3034 if (ret
!= LTTNG_OK
) {
3039 /* Start the UST consumer daemons */
3041 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3042 if (consumerd64_bin
[0] != '\0' &&
3043 ustconsumer64_data
.pid
== 0 &&
3044 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3045 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3046 ret
= start_consumerd(&ustconsumer64_data
);
3048 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3049 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3053 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3054 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3056 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3060 * Setup socket for consumer 64 bit. No need for atomic access
3061 * since it was set above and can ONLY be set in this thread.
3063 ret
= consumer_create_socket(&ustconsumer64_data
,
3064 cmd_ctx
->session
->ust_session
->consumer
);
3070 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3071 if (consumerd32_bin
[0] != '\0' &&
3072 ustconsumer32_data
.pid
== 0 &&
3073 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3074 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3075 ret
= start_consumerd(&ustconsumer32_data
);
3077 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3078 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3082 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3083 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3085 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3089 * Setup socket for consumer 64 bit. No need for atomic access
3090 * since it was set above and can ONLY be set in this thread.
3092 ret
= consumer_create_socket(&ustconsumer32_data
,
3093 cmd_ctx
->session
->ust_session
->consumer
);
3105 /* Validate consumer daemon state when start/stop trace command */
3106 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3107 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3108 switch (cmd_ctx
->lsm
->domain
.type
) {
3109 case LTTNG_DOMAIN_JUL
:
3110 case LTTNG_DOMAIN_LOG4J
:
3111 case LTTNG_DOMAIN_UST
:
3112 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3113 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3117 case LTTNG_DOMAIN_KERNEL
:
3118 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3119 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3127 * Check that the UID or GID match that of the tracing session.
3128 * The root user can interact with all sessions.
3130 if (need_tracing_session
) {
3131 if (!session_access_ok(cmd_ctx
->session
,
3132 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3133 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3134 ret
= LTTNG_ERR_EPERM
;
3140 * Send relayd information to consumer as soon as we have a domain and a
3143 if (cmd_ctx
->session
&& need_domain
) {
3145 * Setup relayd if not done yet. If the relayd information was already
3146 * sent to the consumer, this call will gracefully return.
3148 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3149 if (ret
!= LTTNG_OK
) {
3154 /* Process by command type */
3155 switch (cmd_ctx
->lsm
->cmd_type
) {
3156 case LTTNG_ADD_CONTEXT
:
3158 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3159 cmd_ctx
->lsm
->u
.context
.channel_name
,
3160 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
3163 case LTTNG_DISABLE_CHANNEL
:
3165 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3166 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3169 case LTTNG_DISABLE_EVENT
:
3171 /* FIXME: passing packed structure to non-packed pointer */
3172 /* TODO: handle filter */
3173 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3174 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3175 &cmd_ctx
->lsm
->u
.disable
.event
);
3178 case LTTNG_ENABLE_CHANNEL
:
3180 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3181 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3184 case LTTNG_ENABLE_EVENT
:
3186 struct lttng_event_exclusion
*exclusion
= NULL
;
3187 struct lttng_filter_bytecode
*bytecode
= NULL
;
3188 char *filter_expression
= NULL
;
3190 /* Handle exclusion events and receive it from the client. */
3191 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3192 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3194 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3195 (count
* LTTNG_SYMBOL_NAME_LEN
));
3197 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3201 DBG("Receiving var len exclusion event list from client ...");
3202 exclusion
->count
= count
;
3203 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3204 count
* LTTNG_SYMBOL_NAME_LEN
);
3206 DBG("Nothing recv() from client var len data... continuing");
3209 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3214 /* Get filter expression from client. */
3215 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3216 size_t expression_len
=
3217 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3219 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3220 ret
= LTTNG_ERR_FILTER_INVAL
;
3225 filter_expression
= zmalloc(expression_len
);
3226 if (!filter_expression
) {
3228 ret
= LTTNG_ERR_FILTER_NOMEM
;
3232 /* Receive var. len. data */
3233 DBG("Receiving var len filter's expression from client ...");
3234 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3237 DBG("Nothing recv() from client car len data... continuing");
3239 free(filter_expression
);
3241 ret
= LTTNG_ERR_FILTER_INVAL
;
3246 /* Handle filter and get bytecode from client. */
3247 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3248 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3250 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3251 ret
= LTTNG_ERR_FILTER_INVAL
;
3252 free(filter_expression
);
3257 bytecode
= zmalloc(bytecode_len
);
3259 free(filter_expression
);
3261 ret
= LTTNG_ERR_FILTER_NOMEM
;
3265 /* Receive var. len. data */
3266 DBG("Receiving var len filter's bytecode from client ...");
3267 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3269 DBG("Nothing recv() from client car len data... continuing");
3271 free(filter_expression
);
3274 ret
= LTTNG_ERR_FILTER_INVAL
;
3278 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3279 free(filter_expression
);
3282 ret
= LTTNG_ERR_FILTER_INVAL
;
3287 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3288 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3289 &cmd_ctx
->lsm
->u
.enable
.event
,
3290 filter_expression
, bytecode
, exclusion
,
3291 kernel_poll_pipe
[1]);
3294 case LTTNG_LIST_TRACEPOINTS
:
3296 struct lttng_event
*events
;
3299 session_lock_list();
3300 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3301 session_unlock_list();
3302 if (nb_events
< 0) {
3303 /* Return value is a negative lttng_error_code. */
3309 * Setup lttng message with payload size set to the event list size in
3310 * bytes and then copy list into the llm payload.
3312 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3318 /* Copy event list into message payload */
3319 memcpy(cmd_ctx
->llm
->payload
, events
,
3320 sizeof(struct lttng_event
) * nb_events
);
3327 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3329 struct lttng_event_field
*fields
;
3332 session_lock_list();
3333 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3335 session_unlock_list();
3336 if (nb_fields
< 0) {
3337 /* Return value is a negative lttng_error_code. */
3343 * Setup lttng message with payload size set to the event list size in
3344 * bytes and then copy list into the llm payload.
3346 ret
= setup_lttng_msg(cmd_ctx
,
3347 sizeof(struct lttng_event_field
) * nb_fields
);
3353 /* Copy event list into message payload */
3354 memcpy(cmd_ctx
->llm
->payload
, fields
,
3355 sizeof(struct lttng_event_field
) * nb_fields
);
3362 case LTTNG_LIST_SYSCALLS
:
3364 struct lttng_event
*events
;
3367 nb_events
= cmd_list_syscalls(&events
);
3368 if (nb_events
< 0) {
3369 /* Return value is a negative lttng_error_code. */
3375 * Setup lttng message with payload size set to the event list size in
3376 * bytes and then copy list into the llm payload.
3378 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3384 /* Copy event list into message payload */
3385 memcpy(cmd_ctx
->llm
->payload
, events
,
3386 sizeof(struct lttng_event
) * nb_events
);
3393 case LTTNG_SET_CONSUMER_URI
:
3396 struct lttng_uri
*uris
;
3398 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3399 len
= nb_uri
* sizeof(struct lttng_uri
);
3402 ret
= LTTNG_ERR_INVALID
;
3406 uris
= zmalloc(len
);
3408 ret
= LTTNG_ERR_FATAL
;
3412 /* Receive variable len data */
3413 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3414 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3416 DBG("No URIs received from client... continuing");
3418 ret
= LTTNG_ERR_SESSION_FAIL
;
3423 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3425 if (ret
!= LTTNG_OK
) {
3432 case LTTNG_START_TRACE
:
3434 ret
= cmd_start_trace(cmd_ctx
->session
);
3437 case LTTNG_STOP_TRACE
:
3439 ret
= cmd_stop_trace(cmd_ctx
->session
);
3442 case LTTNG_CREATE_SESSION
:
3445 struct lttng_uri
*uris
= NULL
;
3447 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3448 len
= nb_uri
* sizeof(struct lttng_uri
);
3451 uris
= zmalloc(len
);
3453 ret
= LTTNG_ERR_FATAL
;
3457 /* Receive variable len data */
3458 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3459 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3461 DBG("No URIs received from client... continuing");
3463 ret
= LTTNG_ERR_SESSION_FAIL
;
3468 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3469 DBG("Creating session with ONE network URI is a bad call");
3470 ret
= LTTNG_ERR_SESSION_FAIL
;
3476 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3477 &cmd_ctx
->creds
, 0);
3483 case LTTNG_DESTROY_SESSION
:
3485 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3487 /* Set session to NULL so we do not unlock it after free. */
3488 cmd_ctx
->session
= NULL
;
3491 case LTTNG_LIST_DOMAINS
:
3494 struct lttng_domain
*domains
;
3496 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3498 /* Return value is a negative lttng_error_code. */
3503 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3509 /* Copy event list into message payload */
3510 memcpy(cmd_ctx
->llm
->payload
, domains
,
3511 nb_dom
* sizeof(struct lttng_domain
));
3518 case LTTNG_LIST_CHANNELS
:
3521 struct lttng_channel
*channels
= NULL
;
3523 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3524 cmd_ctx
->session
, &channels
);
3526 /* Return value is a negative lttng_error_code. */
3531 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3537 /* Copy event list into message payload */
3538 memcpy(cmd_ctx
->llm
->payload
, channels
,
3539 nb_chan
* sizeof(struct lttng_channel
));
3546 case LTTNG_LIST_EVENTS
:
3549 struct lttng_event
*events
= NULL
;
3551 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3552 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3554 /* Return value is a negative lttng_error_code. */
3559 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3565 /* Copy event list into message payload */
3566 memcpy(cmd_ctx
->llm
->payload
, events
,
3567 nb_event
* sizeof(struct lttng_event
));
3574 case LTTNG_LIST_SESSIONS
:
3576 unsigned int nr_sessions
;
3578 session_lock_list();
3579 nr_sessions
= lttng_sessions_count(
3580 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3581 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3583 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3585 session_unlock_list();
3589 /* Filled the session array */
3590 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3591 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3592 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3594 session_unlock_list();
3599 case LTTNG_CALIBRATE
:
3601 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3602 &cmd_ctx
->lsm
->u
.calibrate
);
3605 case LTTNG_REGISTER_CONSUMER
:
3607 struct consumer_data
*cdata
;
3609 switch (cmd_ctx
->lsm
->domain
.type
) {
3610 case LTTNG_DOMAIN_KERNEL
:
3611 cdata
= &kconsumer_data
;
3614 ret
= LTTNG_ERR_UND
;
3618 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3619 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3622 case LTTNG_DATA_PENDING
:
3624 ret
= cmd_data_pending(cmd_ctx
->session
);
3627 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3629 struct lttcomm_lttng_output_id reply
;
3631 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3632 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3633 if (ret
!= LTTNG_OK
) {
3637 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3642 /* Copy output list into message payload */
3643 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3647 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3649 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3650 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3653 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3656 struct lttng_snapshot_output
*outputs
= NULL
;
3658 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3659 if (nb_output
< 0) {
3664 ret
= setup_lttng_msg(cmd_ctx
,
3665 nb_output
* sizeof(struct lttng_snapshot_output
));
3672 /* Copy output list into message payload */
3673 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3674 nb_output
* sizeof(struct lttng_snapshot_output
));
3681 case LTTNG_SNAPSHOT_RECORD
:
3683 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3684 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3685 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3688 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3691 struct lttng_uri
*uris
= NULL
;
3693 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3694 len
= nb_uri
* sizeof(struct lttng_uri
);
3697 uris
= zmalloc(len
);
3699 ret
= LTTNG_ERR_FATAL
;
3703 /* Receive variable len data */
3704 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3705 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3707 DBG("No URIs received from client... continuing");
3709 ret
= LTTNG_ERR_SESSION_FAIL
;
3714 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3715 DBG("Creating session with ONE network URI is a bad call");
3716 ret
= LTTNG_ERR_SESSION_FAIL
;
3722 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3723 nb_uri
, &cmd_ctx
->creds
);
3727 case LTTNG_CREATE_SESSION_LIVE
:
3730 struct lttng_uri
*uris
= NULL
;
3732 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3733 len
= nb_uri
* sizeof(struct lttng_uri
);
3736 uris
= zmalloc(len
);
3738 ret
= LTTNG_ERR_FATAL
;
3742 /* Receive variable len data */
3743 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3744 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3746 DBG("No URIs received from client... continuing");
3748 ret
= LTTNG_ERR_SESSION_FAIL
;
3753 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3754 DBG("Creating session with ONE network URI is a bad call");
3755 ret
= LTTNG_ERR_SESSION_FAIL
;
3761 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
3762 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
3766 case LTTNG_SAVE_SESSION
:
3768 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
3773 ret
= LTTNG_ERR_UND
;
3778 if (cmd_ctx
->llm
== NULL
) {
3779 DBG("Missing llm structure. Allocating one.");
3780 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3784 /* Set return code */
3785 cmd_ctx
->llm
->ret_code
= ret
;
3787 if (cmd_ctx
->session
) {
3788 session_unlock(cmd_ctx
->session
);
3790 if (need_tracing_session
) {
3791 session_unlock_list();
3798 * Thread managing health check socket.
3800 static void *thread_manage_health(void *data
)
3802 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3803 uint32_t revents
, nb_fd
;
3804 struct lttng_poll_event events
;
3805 struct health_comm_msg msg
;
3806 struct health_comm_reply reply
;
3808 DBG("[thread] Manage health check started");
3810 rcu_register_thread();
3812 /* We might hit an error path before this is created. */
3813 lttng_poll_init(&events
);
3815 /* Create unix socket */
3816 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3818 ERR("Unable to create health check Unix socket");
3824 /* lttng health client socket path permissions */
3825 ret
= chown(health_unix_sock_path
, 0,
3826 utils_get_group_id(tracing_group_name
));
3828 ERR("Unable to set group on %s", health_unix_sock_path
);
3834 ret
= chmod(health_unix_sock_path
,
3835 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3837 ERR("Unable to set permissions on %s", health_unix_sock_path
);
3845 * Set the CLOEXEC flag. Return code is useless because either way, the
3848 (void) utils_set_fd_cloexec(sock
);
3850 ret
= lttcomm_listen_unix_sock(sock
);
3856 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3857 * more will be added to this poll set.
3859 ret
= sessiond_set_thread_pollset(&events
, 2);
3864 /* Add the application registration socket */
3865 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3870 sessiond_notify_ready();
3873 DBG("Health check ready");
3875 /* Inifinite blocking call, waiting for transmission */
3877 ret
= lttng_poll_wait(&events
, -1);
3880 * Restart interrupted system call.
3882 if (errno
== EINTR
) {
3890 for (i
= 0; i
< nb_fd
; i
++) {
3891 /* Fetch once the poll data */
3892 revents
= LTTNG_POLL_GETEV(&events
, i
);
3893 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3896 /* No activity for this FD (poll implementation). */
3900 /* Thread quit pipe has been closed. Killing thread. */
3901 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3907 /* Event on the registration socket */
3908 if (pollfd
== sock
) {
3909 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3910 ERR("Health socket poll error");
3916 new_sock
= lttcomm_accept_unix_sock(sock
);
3922 * Set the CLOEXEC flag. Return code is useless because either way, the
3925 (void) utils_set_fd_cloexec(new_sock
);
3927 DBG("Receiving data from client for health...");
3928 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3930 DBG("Nothing recv() from client... continuing");
3931 ret
= close(new_sock
);
3939 rcu_thread_online();
3941 memset(&reply
, 0, sizeof(reply
));
3942 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
3944 * health_check_state returns 0 if health is
3947 if (!health_check_state(health_sessiond
, i
)) {
3948 reply
.ret_code
|= 1ULL << i
;
3952 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
3954 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3956 ERR("Failed to send health data back to client");
3959 /* End of transmission */
3960 ret
= close(new_sock
);
3970 ERR("Health error occurred in %s", __func__
);
3972 DBG("Health check thread dying");
3973 unlink(health_unix_sock_path
);
3981 lttng_poll_clean(&events
);
3983 rcu_unregister_thread();
3988 * This thread manage all clients request using the unix client socket for
3991 static void *thread_manage_clients(void *data
)
3993 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3995 uint32_t revents
, nb_fd
;
3996 struct command_ctx
*cmd_ctx
= NULL
;
3997 struct lttng_poll_event events
;
3999 DBG("[thread] Manage client started");
4001 rcu_register_thread();
4003 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4005 health_code_update();
4007 ret
= lttcomm_listen_unix_sock(client_sock
);
4013 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4014 * more will be added to this poll set.
4016 ret
= sessiond_set_thread_pollset(&events
, 2);
4018 goto error_create_poll
;
4021 /* Add the application registration socket */
4022 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4027 sessiond_notify_ready();
4028 ret
= sem_post(&load_info
->message_thread_ready
);
4030 PERROR("sem_post message_thread_ready");
4034 /* This testpoint is after we signal readiness to the parent. */
4035 if (testpoint(sessiond_thread_manage_clients
)) {
4039 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4043 health_code_update();
4046 DBG("Accepting client command ...");
4048 /* Inifinite blocking call, waiting for transmission */
4050 health_poll_entry();
4051 ret
= lttng_poll_wait(&events
, -1);
4055 * Restart interrupted system call.
4057 if (errno
== EINTR
) {
4065 for (i
= 0; i
< nb_fd
; i
++) {
4066 /* Fetch once the poll data */
4067 revents
= LTTNG_POLL_GETEV(&events
, i
);
4068 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4070 health_code_update();
4073 /* No activity for this FD (poll implementation). */
4077 /* Thread quit pipe has been closed. Killing thread. */
4078 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4084 /* Event on the registration socket */
4085 if (pollfd
== client_sock
) {
4086 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4087 ERR("Client socket poll error");
4093 DBG("Wait for client response");
4095 health_code_update();
4097 sock
= lttcomm_accept_unix_sock(client_sock
);
4103 * Set the CLOEXEC flag. Return code is useless because either way, the
4106 (void) utils_set_fd_cloexec(sock
);
4108 /* Set socket option for credentials retrieval */
4109 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4114 /* Allocate context command to process the client request */
4115 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4116 if (cmd_ctx
== NULL
) {
4117 PERROR("zmalloc cmd_ctx");
4121 /* Allocate data buffer for reception */
4122 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4123 if (cmd_ctx
->lsm
== NULL
) {
4124 PERROR("zmalloc cmd_ctx->lsm");
4128 cmd_ctx
->llm
= NULL
;
4129 cmd_ctx
->session
= NULL
;
4131 health_code_update();
4134 * Data is received from the lttng client. The struct
4135 * lttcomm_session_msg (lsm) contains the command and data request of
4138 DBG("Receiving data from client ...");
4139 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4140 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4142 DBG("Nothing recv() from client... continuing");
4148 clean_command_ctx(&cmd_ctx
);
4152 health_code_update();
4154 // TODO: Validate cmd_ctx including sanity check for
4155 // security purpose.
4157 rcu_thread_online();
4159 * This function dispatch the work to the kernel or userspace tracer
4160 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4161 * informations for the client. The command context struct contains
4162 * everything this function may needs.
4164 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4165 rcu_thread_offline();
4173 * TODO: Inform client somehow of the fatal error. At
4174 * this point, ret < 0 means that a zmalloc failed
4175 * (ENOMEM). Error detected but still accept
4176 * command, unless a socket error has been
4179 clean_command_ctx(&cmd_ctx
);
4183 health_code_update();
4185 DBG("Sending response (size: %d, retcode: %s)",
4186 cmd_ctx
->lttng_msg_size
,
4187 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
4188 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4190 ERR("Failed to send data back to client");
4193 /* End of transmission */
4200 clean_command_ctx(&cmd_ctx
);
4202 health_code_update();
4214 lttng_poll_clean(&events
);
4215 clean_command_ctx(&cmd_ctx
);
4219 unlink(client_unix_sock_path
);
4220 if (client_sock
>= 0) {
4221 ret
= close(client_sock
);
4229 ERR("Health error occurred in %s", __func__
);
4232 health_unregister(health_sessiond
);
4234 DBG("Client thread dying");
4236 rcu_unregister_thread();
4242 * usage function on stderr
4244 static void usage(void)
4246 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4247 fprintf(stderr
, " -h, --help Display this usage.\n");
4248 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4249 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4250 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4251 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4252 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4253 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4254 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4255 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4256 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4257 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4258 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4259 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4260 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4261 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
4262 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4263 fprintf(stderr
, " -V, --version Show version number.\n");
4264 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4265 fprintf(stderr
, " -q, --quiet No output at all.\n");
4266 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4267 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4268 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4269 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4270 fprintf(stderr
, " --agent-tcp-port Agent registration TCP port\n");
4271 fprintf(stderr
, " -f --config PATH Load daemon configuration file\n");
4272 fprintf(stderr
, " -l --load PATH Load session configuration\n");
4273 fprintf(stderr
, " --kmod-probes Specify kernel module probes to load\n");
4274 fprintf(stderr
, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4278 * Take an option from the getopt output and set it in the right variable to be
4281 * Return 0 on success else a negative value.
4283 static int set_option(int opt
, const char *arg
, const char *optname
)
4287 if (arg
&& arg
[0] == '\0') {
4289 * This only happens if the value is read from daemon config
4290 * file. This means the option requires an argument and the
4291 * configuration file contains a line such as:
4300 fprintf(stderr
, "option %s", optname
);
4302 fprintf(stderr
, " with arg %s\n", arg
);
4306 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4309 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4319 * If the override option is set, the pointer points to a
4320 * *non* const thus freeing it even though the variable type is
4323 if (tracing_group_name_override
) {
4324 free((void *) tracing_group_name
);
4326 tracing_group_name
= strdup(arg
);
4327 if (!tracing_group_name
) {
4331 tracing_group_name_override
= 1;
4337 fprintf(stdout
, "%s\n", VERSION
);
4343 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4346 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4349 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4352 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4355 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4358 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4364 lttng_opt_quiet
= 1;
4367 /* Verbose level can increase using multiple -v */
4369 /* Value obtained from config file */
4370 lttng_opt_verbose
= config_parse_value(arg
);
4372 /* -v used on command line */
4373 lttng_opt_verbose
++;
4375 /* Clamp value to [0, 3] */
4376 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4377 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4381 opt_verbose_consumer
= config_parse_value(arg
);
4383 opt_verbose_consumer
+= 1;
4387 if (consumerd32_bin_override
) {
4388 free((void *) consumerd32_bin
);
4390 consumerd32_bin
= strdup(arg
);
4391 if (!consumerd32_bin
) {
4395 consumerd32_bin_override
= 1;
4398 if (consumerd32_libdir_override
) {
4399 free((void *) consumerd32_libdir
);
4401 consumerd32_libdir
= strdup(arg
);
4402 if (!consumerd32_libdir
) {
4406 consumerd32_libdir_override
= 1;
4409 if (consumerd64_bin_override
) {
4410 free((void *) consumerd64_bin
);
4412 consumerd64_bin
= strdup(arg
);
4413 if (!consumerd64_bin
) {
4417 consumerd64_bin_override
= 1;
4420 if (consumerd64_libdir_override
) {
4421 free((void *) consumerd64_libdir
);
4423 consumerd64_libdir
= strdup(arg
);
4424 if (!consumerd64_libdir
) {
4428 consumerd64_libdir_override
= 1;
4432 opt_pidfile
= strdup(arg
);
4438 case 'J': /* Agent TCP port. */
4443 v
= strtoul(arg
, NULL
, 0);
4444 if (errno
!= 0 || !isdigit(arg
[0])) {
4445 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4448 if (v
== 0 || v
>= 65535) {
4449 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4452 agent_tcp_port
= (uint32_t) v
;
4453 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4457 free(opt_load_session_path
);
4458 opt_load_session_path
= strdup(arg
);
4459 if (!opt_load_session_path
) {
4464 case 'P': /* probe modules list */
4465 free(kmod_probes_list
);
4466 kmod_probes_list
= strdup(arg
);
4467 if (!kmod_probes_list
) {
4473 free(kmod_extra_probes_list
);
4474 kmod_extra_probes_list
= strdup(arg
);
4475 if (!kmod_extra_probes_list
) {
4481 /* This is handled in set_options() thus silent break. */
4484 /* Unknown option or other error.
4485 * Error is printed by getopt, just return */
4490 if (ret
== -EINVAL
) {
4491 const char *opt_name
= "unknown";
4494 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
4496 if (opt
== long_options
[i
].val
) {
4497 opt_name
= long_options
[i
].name
;
4502 WARN("Invalid argument provided for option \"%s\", using default value.",
4510 * config_entry_handler_cb used to handle options read from a config file.
4511 * See config_entry_handler_cb comment in common/config/config.h for the
4512 * return value conventions.
4514 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4518 if (!entry
|| !entry
->name
|| !entry
->value
) {
4523 /* Check if the option is to be ignored */
4524 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4525 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4530 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4533 /* Ignore if not fully matched. */
4534 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4539 * If the option takes no argument on the command line, we have to
4540 * check if the value is "true". We support non-zero numeric values,
4543 if (!long_options
[i
].has_arg
) {
4544 ret
= config_parse_value(entry
->value
);
4547 WARN("Invalid configuration value \"%s\" for option %s",
4548 entry
->value
, entry
->name
);
4550 /* False, skip boolean config option. */
4555 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4559 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4566 * daemon configuration loading and argument parsing
4568 static int set_options(int argc
, char **argv
)
4570 int ret
= 0, c
= 0, option_index
= 0;
4571 int orig_optopt
= optopt
, orig_optind
= optind
;
4573 const char *config_path
= NULL
;
4575 optstring
= utils_generate_optstring(long_options
,
4576 sizeof(long_options
) / sizeof(struct option
));
4582 /* Check for the --config option */
4583 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
4584 &option_index
)) != -1) {
4588 } else if (c
!= 'f') {
4589 /* if not equal to --config option. */
4593 config_path
= utils_expand_path(optarg
);
4595 ERR("Failed to resolve path: %s", optarg
);
4599 ret
= config_get_section_entries(config_path
, config_section_name
,
4600 config_entry_handler
, NULL
);
4603 ERR("Invalid configuration option at line %i", ret
);
4609 /* Reset getopt's global state */
4610 optopt
= orig_optopt
;
4611 optind
= orig_optind
;
4613 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
4618 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
4630 * Creates the two needed socket by the daemon.
4631 * apps_sock - The communication socket for all UST apps.
4632 * client_sock - The communication of the cli tool (lttng).
4634 static int init_daemon_socket(void)
4639 old_umask
= umask(0);
4641 /* Create client tool unix socket */
4642 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4643 if (client_sock
< 0) {
4644 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4649 /* Set the cloexec flag */
4650 ret
= utils_set_fd_cloexec(client_sock
);
4652 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4653 "Continuing but note that the consumer daemon will have a "
4654 "reference to this socket on exec()", client_sock
);
4657 /* File permission MUST be 660 */
4658 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4660 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4665 /* Create the application unix socket */
4666 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4667 if (apps_sock
< 0) {
4668 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4673 /* Set the cloexec flag */
4674 ret
= utils_set_fd_cloexec(apps_sock
);
4676 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4677 "Continuing but note that the consumer daemon will have a "
4678 "reference to this socket on exec()", apps_sock
);
4681 /* File permission MUST be 666 */
4682 ret
= chmod(apps_unix_sock_path
,
4683 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4685 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4690 DBG3("Session daemon client socket %d and application socket %d created",
4691 client_sock
, apps_sock
);
4699 * Check if the global socket is available, and if a daemon is answering at the
4700 * other side. If yes, error is returned.
4702 static int check_existing_daemon(void)
4704 /* Is there anybody out there ? */
4705 if (lttng_session_daemon_alive()) {
4713 * Set the tracing group gid onto the client socket.
4715 * Race window between mkdir and chown is OK because we are going from more
4716 * permissive (root.root) to less permissive (root.tracing).
4718 static int set_permissions(char *rundir
)
4723 gid
= utils_get_group_id(tracing_group_name
);
4725 /* Set lttng run dir */
4726 ret
= chown(rundir
, 0, gid
);
4728 ERR("Unable to set group on %s", rundir
);
4733 * Ensure all applications and tracing group can search the run
4734 * dir. Allow everyone to read the directory, since it does not
4735 * buy us anything to hide its content.
4737 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
4739 ERR("Unable to set permissions on %s", rundir
);
4743 /* lttng client socket path */
4744 ret
= chown(client_unix_sock_path
, 0, gid
);
4746 ERR("Unable to set group on %s", client_unix_sock_path
);
4750 /* kconsumer error socket path */
4751 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
4753 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4757 /* 64-bit ustconsumer error socket path */
4758 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
4760 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4764 /* 32-bit ustconsumer compat32 error socket path */
4765 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
4767 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4771 DBG("All permissions are set");
4777 * Create the lttng run directory needed for all global sockets and pipe.
4779 static int create_lttng_rundir(const char *rundir
)
4783 DBG3("Creating LTTng run directory: %s", rundir
);
4785 ret
= mkdir(rundir
, S_IRWXU
);
4787 if (errno
!= EEXIST
) {
4788 ERR("Unable to create %s", rundir
);
4800 * Setup sockets and directory needed by the kconsumerd communication with the
4803 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4807 char path
[PATH_MAX
];
4809 switch (consumer_data
->type
) {
4810 case LTTNG_CONSUMER_KERNEL
:
4811 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4813 case LTTNG_CONSUMER64_UST
:
4814 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4816 case LTTNG_CONSUMER32_UST
:
4817 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4820 ERR("Consumer type unknown");
4825 DBG2("Creating consumer directory: %s", path
);
4827 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
4829 if (errno
!= EEXIST
) {
4831 ERR("Failed to create %s", path
);
4837 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
4839 ERR("Unable to set group on %s", path
);
4845 /* Create the kconsumerd error unix socket */
4846 consumer_data
->err_sock
=
4847 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4848 if (consumer_data
->err_sock
< 0) {
4849 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4855 * Set the CLOEXEC flag. Return code is useless because either way, the
4858 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4860 PERROR("utils_set_fd_cloexec");
4861 /* continue anyway */
4864 /* File permission MUST be 660 */
4865 ret
= chmod(consumer_data
->err_unix_sock_path
,
4866 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4868 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4878 * Signal handler for the daemon
4880 * Simply stop all worker threads, leaving main() return gracefully after
4881 * joining all threads and calling cleanup().
4883 static void sighandler(int sig
)
4887 DBG("SIGPIPE caught");
4890 DBG("SIGINT caught");
4894 DBG("SIGTERM caught");
4898 CMM_STORE_SHARED(recv_child_signal
, 1);
4906 * Setup signal handler for :
4907 * SIGINT, SIGTERM, SIGPIPE
4909 static int set_signal_handler(void)
4912 struct sigaction sa
;
4915 if ((ret
= sigemptyset(&sigset
)) < 0) {
4916 PERROR("sigemptyset");
4920 sa
.sa_handler
= sighandler
;
4921 sa
.sa_mask
= sigset
;
4923 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4924 PERROR("sigaction");
4928 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4929 PERROR("sigaction");
4933 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4934 PERROR("sigaction");
4938 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
4939 PERROR("sigaction");
4943 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4949 * Set open files limit to unlimited. This daemon can open a large number of
4950 * file descriptors in order to consumer multiple kernel traces.
4952 static void set_ulimit(void)
4957 /* The kernel does not allowed an infinite limit for open files */
4958 lim
.rlim_cur
= 65535;
4959 lim
.rlim_max
= 65535;
4961 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4963 PERROR("failed to set open files limit");
4968 * Write pidfile using the rundir and opt_pidfile.
4970 static void write_pidfile(void)
4973 char pidfile_path
[PATH_MAX
];
4978 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4980 /* Build pidfile path from rundir and opt_pidfile. */
4981 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4982 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4984 PERROR("snprintf pidfile path");
4990 * Create pid file in rundir. Return value is of no importance. The
4991 * execution will continue even though we are not able to write the file.
4993 (void) utils_create_pid_file(getpid(), pidfile_path
);
5000 * Create lockfile using the rundir and return its fd.
5002 static int create_lockfile(void)
5005 char lockfile_path
[PATH_MAX
];
5007 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5012 ret
= utils_create_lock_file(lockfile_path
);
5018 * Write agent TCP port using the rundir.
5020 static void write_agent_port(void)
5023 char path
[PATH_MAX
];
5027 ret
= snprintf(path
, sizeof(path
), "%s/"
5028 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5030 PERROR("snprintf agent port path");
5035 * Create TCP agent port file in rundir. Return value is of no importance.
5036 * The execution will continue even though we are not able to write the
5039 (void) utils_create_pid_file(agent_tcp_port
, path
);
5046 * Start the load session thread and dettach from it so the main thread can
5047 * continue. This does not return a value since whatever the outcome, the main
5048 * thread will continue.
5050 static void start_load_session_thread(void)
5054 /* Create session loading thread. */
5055 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
5058 PERROR("pthread_create load_session_thread");
5062 ret
= pthread_detach(load_session_thread
);
5064 PERROR("pthread_detach load_session_thread");
5067 /* Everything went well so don't cleanup anything. */
5070 /* The cleanup() function will destroy the load_info data. */
5077 int main(int argc
, char **argv
)
5081 const char *home_path
, *env_app_timeout
;
5083 /* Initialize agent apps ht global variable */
5084 agent_apps_ht_by_sock
= NULL
;
5086 init_kernel_workarounds();
5088 rcu_register_thread();
5090 if ((ret
= set_signal_handler()) < 0) {
5094 setup_consumerd_path();
5096 page_size
= sysconf(_SC_PAGESIZE
);
5097 if (page_size
< 0) {
5098 PERROR("sysconf _SC_PAGESIZE");
5099 page_size
= LONG_MAX
;
5100 WARN("Fallback page size to %ld", page_size
);
5103 /* Parse arguments and load the daemon configuration file */
5105 if ((ret
= set_options(argc
, argv
)) < 0) {
5110 if (opt_daemon
|| opt_background
) {
5113 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5120 * We are in the child. Make sure all other file descriptors are
5121 * closed, in case we are called with more opened file descriptors than
5122 * the standard ones.
5124 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5129 /* Create thread quit pipe */
5130 if ((ret
= init_thread_quit_pipe()) < 0) {
5134 /* Check if daemon is UID = 0 */
5135 is_root
= !getuid();
5138 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5144 /* Create global run dir with root access */
5145 ret
= create_lttng_rundir(rundir
);
5150 if (strlen(apps_unix_sock_path
) == 0) {
5151 snprintf(apps_unix_sock_path
, PATH_MAX
,
5152 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5155 if (strlen(client_unix_sock_path
) == 0) {
5156 snprintf(client_unix_sock_path
, PATH_MAX
,
5157 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5160 /* Set global SHM for ust */
5161 if (strlen(wait_shm_path
) == 0) {
5162 snprintf(wait_shm_path
, PATH_MAX
,
5163 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5166 if (strlen(health_unix_sock_path
) == 0) {
5167 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
5168 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5171 /* Setup kernel consumerd path */
5172 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5173 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5174 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5175 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5177 DBG2("Kernel consumer err path: %s",
5178 kconsumer_data
.err_unix_sock_path
);
5179 DBG2("Kernel consumer cmd path: %s",
5180 kconsumer_data
.cmd_unix_sock_path
);
5182 home_path
= utils_get_home_dir();
5183 if (home_path
== NULL
) {
5184 /* TODO: Add --socket PATH option */
5185 ERR("Can't get HOME directory for sockets creation.");
5191 * Create rundir from home path. This will create something like
5194 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5200 ret
= create_lttng_rundir(rundir
);
5205 if (strlen(apps_unix_sock_path
) == 0) {
5206 snprintf(apps_unix_sock_path
, PATH_MAX
,
5207 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
5210 /* Set the cli tool unix socket path */
5211 if (strlen(client_unix_sock_path
) == 0) {
5212 snprintf(client_unix_sock_path
, PATH_MAX
,
5213 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
5216 /* Set global SHM for ust */
5217 if (strlen(wait_shm_path
) == 0) {
5218 snprintf(wait_shm_path
, PATH_MAX
,
5219 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
5222 /* Set health check Unix path */
5223 if (strlen(health_unix_sock_path
) == 0) {
5224 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
5225 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
5229 lockfile_fd
= create_lockfile();
5230 if (lockfile_fd
< 0) {
5234 /* Set consumer initial state */
5235 kernel_consumerd_state
= CONSUMER_STOPPED
;
5236 ust_consumerd_state
= CONSUMER_STOPPED
;
5238 DBG("Client socket path %s", client_unix_sock_path
);
5239 DBG("Application socket path %s", apps_unix_sock_path
);
5240 DBG("Application wait path %s", wait_shm_path
);
5241 DBG("LTTng run directory path: %s", rundir
);
5243 /* 32 bits consumerd path setup */
5244 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5245 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5246 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5247 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5249 DBG2("UST consumer 32 bits err path: %s",
5250 ustconsumer32_data
.err_unix_sock_path
);
5251 DBG2("UST consumer 32 bits cmd path: %s",
5252 ustconsumer32_data
.cmd_unix_sock_path
);
5254 /* 64 bits consumerd path setup */
5255 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5256 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5257 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5258 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5260 DBG2("UST consumer 64 bits err path: %s",
5261 ustconsumer64_data
.err_unix_sock_path
);
5262 DBG2("UST consumer 64 bits cmd path: %s",
5263 ustconsumer64_data
.cmd_unix_sock_path
);
5266 * See if daemon already exist.
5268 if ((ret
= check_existing_daemon()) < 0) {
5269 ERR("Already running daemon.\n");
5271 * We do not goto exit because we must not cleanup()
5272 * because a daemon is already running.
5277 /* After this point, we can safely call cleanup() with "goto exit" */
5280 * Init UST app hash table. Alloc hash table before this point since
5281 * cleanup() can get called after that point.
5286 * Initialize agent app hash table. We allocate the hash table here
5287 * since cleanup() can get called after this point.
5289 if (agent_app_ht_alloc()) {
5290 ERR("Failed to allocate Agent app hash table");
5296 * These actions must be executed as root. We do that *after* setting up
5297 * the sockets path because we MUST make the check for another daemon using
5298 * those paths *before* trying to set the kernel consumer sockets and init
5302 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
5307 /* Setup kernel tracer */
5308 if (!opt_no_kernel
) {
5309 init_kernel_tracer();
5310 if (kernel_tracer_fd
>= 0) {
5311 ret
= syscall_init_table();
5313 ERR("Unable to populate syscall table. Syscall tracing"
5314 " won't work for this session daemon.");
5319 /* Set ulimit for open files */
5322 /* init lttng_fd tracking must be done after set_ulimit. */
5325 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
5330 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
5335 /* Setup the needed unix socket */
5336 if ((ret
= init_daemon_socket()) < 0) {
5340 /* Set credentials to socket */
5341 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
5345 /* Get parent pid if -S, --sig-parent is specified. */
5346 if (opt_sig_parent
) {
5350 /* Setup the kernel pipe for waking up the kernel thread */
5351 if (is_root
&& !opt_no_kernel
) {
5352 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
5357 /* Setup the thread ht_cleanup communication pipe. */
5358 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
5362 /* Setup the thread apps communication pipe. */
5363 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
5367 /* Setup the thread apps notify communication pipe. */
5368 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
5372 /* Initialize global buffer per UID and PID registry. */
5373 buffer_reg_init_uid_registry();
5374 buffer_reg_init_pid_registry();
5376 /* Init UST command queue. */
5377 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
5380 * Get session list pointer. This pointer MUST NOT be free(). This list is
5381 * statically declared in session.c
5383 session_list_ptr
= session_get_list();
5385 /* Set up max poll set size */
5386 lttng_poll_set_max_size();
5390 /* Check for the application socket timeout env variable. */
5391 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5392 if (env_app_timeout
) {
5393 app_socket_timeout
= atoi(env_app_timeout
);
5395 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5401 /* Initialize communication library */
5403 /* This is to get the TCP timeout value. */
5404 lttcomm_inet_init();
5406 if (load_session_init_data(&load_info
) < 0) {
5409 load_info
->path
= opt_load_session_path
;
5412 * Initialize the health check subsystem. This call should set the
5413 * appropriate time values.
5415 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5416 if (!health_sessiond
) {
5417 PERROR("health_app_create error");
5418 goto exit_health_sessiond_cleanup
;
5421 /* Create thread to clean up RCU hash tables */
5422 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5423 thread_ht_cleanup
, (void *) NULL
);
5425 PERROR("pthread_create ht_cleanup");
5426 goto exit_ht_cleanup
;
5429 /* Create health-check thread */
5430 ret
= pthread_create(&health_thread
, NULL
,
5431 thread_manage_health
, (void *) NULL
);
5433 PERROR("pthread_create health");
5437 /* Create thread to manage the client socket */
5438 ret
= pthread_create(&client_thread
, NULL
,
5439 thread_manage_clients
, (void *) NULL
);
5441 PERROR("pthread_create clients");
5445 /* Create thread to dispatch registration */
5446 ret
= pthread_create(&dispatch_thread
, NULL
,
5447 thread_dispatch_ust_registration
, (void *) NULL
);
5449 PERROR("pthread_create dispatch");
5453 /* Create thread to manage application registration. */
5454 ret
= pthread_create(®_apps_thread
, NULL
,
5455 thread_registration_apps
, (void *) NULL
);
5457 PERROR("pthread_create registration");
5461 /* Create thread to manage application socket */
5462 ret
= pthread_create(&apps_thread
, NULL
,
5463 thread_manage_apps
, (void *) NULL
);
5465 PERROR("pthread_create apps");
5469 /* Create thread to manage application notify socket */
5470 ret
= pthread_create(&apps_notify_thread
, NULL
,
5471 ust_thread_manage_notify
, (void *) NULL
);
5473 PERROR("pthread_create notify");
5474 goto exit_apps_notify
;
5477 /* Create agent registration thread. */
5478 ret
= pthread_create(&agent_reg_thread
, NULL
,
5479 agent_thread_manage_registration
, (void *) NULL
);
5481 PERROR("pthread_create agent");
5482 goto exit_agent_reg
;
5485 /* Don't start this thread if kernel tracing is not requested nor root */
5486 if (is_root
&& !opt_no_kernel
) {
5487 /* Create kernel thread to manage kernel event */
5488 ret
= pthread_create(&kernel_thread
, NULL
,
5489 thread_manage_kernel
, (void *) NULL
);
5491 PERROR("pthread_create kernel");
5496 /* Load possible session(s). */
5497 start_load_session_thread();
5499 if (is_root
&& !opt_no_kernel
) {
5500 ret
= pthread_join(kernel_thread
, &status
);
5502 PERROR("pthread_join");
5503 goto error
; /* join error, exit without cleanup */
5508 ret
= pthread_join(agent_reg_thread
, &status
);
5510 PERROR("pthread_join agent");
5511 goto error
; /* join error, exit without cleanup */
5515 ret
= pthread_join(apps_notify_thread
, &status
);
5517 PERROR("pthread_join apps notify");
5518 goto error
; /* join error, exit without cleanup */
5522 ret
= pthread_join(apps_thread
, &status
);
5524 PERROR("pthread_join apps");
5525 goto error
; /* join error, exit without cleanup */
5530 ret
= pthread_join(reg_apps_thread
, &status
);
5532 PERROR("pthread_join");
5533 goto error
; /* join error, exit without cleanup */
5537 ret
= pthread_join(dispatch_thread
, &status
);
5539 PERROR("pthread_join");
5540 goto error
; /* join error, exit without cleanup */
5544 ret
= pthread_join(client_thread
, &status
);
5546 PERROR("pthread_join");
5547 goto error
; /* join error, exit without cleanup */
5550 ret
= join_consumer_thread(&kconsumer_data
);
5552 PERROR("join_consumer");
5553 goto error
; /* join error, exit without cleanup */
5556 ret
= join_consumer_thread(&ustconsumer32_data
);
5558 PERROR("join_consumer ust32");
5559 goto error
; /* join error, exit without cleanup */
5562 ret
= join_consumer_thread(&ustconsumer64_data
);
5564 PERROR("join_consumer ust64");
5565 goto error
; /* join error, exit without cleanup */
5569 ret
= pthread_join(health_thread
, &status
);
5571 PERROR("pthread_join health thread");
5572 goto error
; /* join error, exit without cleanup */
5576 ret
= pthread_join(ht_cleanup_thread
, &status
);
5578 PERROR("pthread_join ht cleanup thread");
5579 goto error
; /* join error, exit without cleanup */
5582 health_app_destroy(health_sessiond
);
5583 exit_health_sessiond_cleanup
:
5586 * cleanup() is called when no other thread is running.
5588 rcu_thread_online();
5590 rcu_thread_offline();
5591 rcu_unregister_thread();