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("Closing all UST sockets");
633 ust_app_clean_list();
634 buffer_reg_destroy_registries();
636 if (is_root
&& !opt_no_kernel
) {
637 DBG2("Closing kernel fd");
638 if (kernel_tracer_fd
>= 0) {
639 ret
= close(kernel_tracer_fd
);
644 DBG("Unloading kernel modules");
645 modprobe_remove_lttng_all();
649 close_consumer_sockets();
652 * If the override option is set, the pointer points to a *non* const thus
653 * freeing it even though the variable type is set to const.
655 if (tracing_group_name_override
) {
656 free((void *) tracing_group_name
);
658 if (consumerd32_bin_override
) {
659 free((void *) consumerd32_bin
);
661 if (consumerd64_bin_override
) {
662 free((void *) consumerd64_bin
);
664 if (consumerd32_libdir_override
) {
665 free((void *) consumerd32_libdir
);
667 if (consumerd64_libdir_override
) {
668 free((void *) consumerd64_libdir
);
675 if (opt_load_session_path
) {
676 free(opt_load_session_path
);
680 load_session_destroy_data(load_info
);
685 * Cleanup lock file by deleting it and finaly closing it which will
686 * release the file system lock.
688 if (lockfile_fd
>= 0) {
689 char lockfile_path
[PATH_MAX
];
691 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
693 ret
= remove(lockfile_path
);
695 PERROR("remove lock file");
697 ret
= close(lockfile_fd
);
699 PERROR("close lock file");
705 * We do NOT rmdir rundir because there are other processes
706 * using it, for instance lttng-relayd, which can start in
707 * parallel with this teardown.
713 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
714 "Matthew, BEET driven development works!%c[%dm",
715 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
720 * Send data on a unix socket using the liblttsessiondcomm API.
722 * Return lttcomm error code.
724 static int send_unix_sock(int sock
, void *buf
, size_t len
)
726 /* Check valid length */
731 return lttcomm_send_unix_sock(sock
, buf
, len
);
735 * Free memory of a command context structure.
737 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
739 DBG("Clean command context structure");
741 if ((*cmd_ctx
)->llm
) {
742 free((*cmd_ctx
)->llm
);
744 if ((*cmd_ctx
)->lsm
) {
745 free((*cmd_ctx
)->lsm
);
753 * Notify UST applications using the shm mmap futex.
755 static int notify_ust_apps(int active
)
759 DBG("Notifying applications of session daemon state: %d", active
);
761 /* See shm.c for this call implying mmap, shm and futex calls */
762 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
763 if (wait_shm_mmap
== NULL
) {
767 /* Wake waiting process */
768 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
770 /* Apps notified successfully */
778 * Setup the outgoing data buffer for the response (llm) by allocating the
779 * right amount of memory and copying the original information from the lsm
782 * Return total size of the buffer pointed by buf.
784 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
790 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
791 if (cmd_ctx
->llm
== NULL
) {
797 /* Copy common data */
798 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
799 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
801 cmd_ctx
->llm
->data_size
= size
;
802 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
811 * Update the kernel poll set of all channel fd available over all tracing
812 * session. Add the wakeup pipe at the end of the set.
814 static int update_kernel_poll(struct lttng_poll_event
*events
)
817 struct ltt_session
*session
;
818 struct ltt_kernel_channel
*channel
;
820 DBG("Updating kernel poll set");
823 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
824 session_lock(session
);
825 if (session
->kernel_session
== NULL
) {
826 session_unlock(session
);
830 cds_list_for_each_entry(channel
,
831 &session
->kernel_session
->channel_list
.head
, list
) {
832 /* Add channel fd to the kernel poll set */
833 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
835 session_unlock(session
);
838 DBG("Channel fd %d added to kernel set", channel
->fd
);
840 session_unlock(session
);
842 session_unlock_list();
847 session_unlock_list();
852 * Find the channel fd from 'fd' over all tracing session. When found, check
853 * for new channel stream and send those stream fds to the kernel consumer.
855 * Useful for CPU hotplug feature.
857 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
860 struct ltt_session
*session
;
861 struct ltt_kernel_session
*ksess
;
862 struct ltt_kernel_channel
*channel
;
864 DBG("Updating kernel streams for channel fd %d", fd
);
867 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
868 session_lock(session
);
869 if (session
->kernel_session
== NULL
) {
870 session_unlock(session
);
873 ksess
= session
->kernel_session
;
875 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
876 if (channel
->fd
== fd
) {
877 DBG("Channel found, updating kernel streams");
878 ret
= kernel_open_channel_stream(channel
);
882 /* Update the stream global counter */
883 ksess
->stream_count_global
+= ret
;
886 * Have we already sent fds to the consumer? If yes, it means
887 * that tracing is started so it is safe to send our updated
890 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
891 struct lttng_ht_iter iter
;
892 struct consumer_socket
*socket
;
895 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
896 &iter
.iter
, socket
, node
.node
) {
897 pthread_mutex_lock(socket
->lock
);
898 ret
= kernel_consumer_send_channel_stream(socket
,
900 session
->output_traces
? 1 : 0);
901 pthread_mutex_unlock(socket
->lock
);
912 session_unlock(session
);
914 session_unlock_list();
918 session_unlock(session
);
919 session_unlock_list();
924 * For each tracing session, update newly registered apps. The session list
925 * lock MUST be acquired before calling this.
927 static void update_ust_app(int app_sock
)
929 struct ltt_session
*sess
, *stmp
;
931 /* Consumer is in an ERROR state. Stop any application update. */
932 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
933 /* Stop the update process since the consumer is dead. */
937 /* For all tracing session(s) */
938 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
940 if (sess
->ust_session
) {
941 ust_app_global_update(sess
->ust_session
, app_sock
);
943 session_unlock(sess
);
948 * This thread manage event coming from the kernel.
950 * Features supported in this thread:
953 static void *thread_manage_kernel(void *data
)
955 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
956 uint32_t revents
, nb_fd
;
958 struct lttng_poll_event events
;
960 DBG("[thread] Thread manage kernel started");
962 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
965 * This first step of the while is to clean this structure which could free
966 * non NULL pointers so initialize it before the loop.
968 lttng_poll_init(&events
);
970 if (testpoint(sessiond_thread_manage_kernel
)) {
971 goto error_testpoint
;
974 health_code_update();
976 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
977 goto error_testpoint
;
981 health_code_update();
983 if (update_poll_flag
== 1) {
984 /* Clean events object. We are about to populate it again. */
985 lttng_poll_clean(&events
);
987 ret
= sessiond_set_thread_pollset(&events
, 2);
989 goto error_poll_create
;
992 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
997 /* This will add the available kernel channel if any. */
998 ret
= update_kernel_poll(&events
);
1002 update_poll_flag
= 0;
1005 DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events
));
1007 /* Poll infinite value of time */
1009 health_poll_entry();
1010 ret
= lttng_poll_wait(&events
, -1);
1014 * Restart interrupted system call.
1016 if (errno
== EINTR
) {
1020 } else if (ret
== 0) {
1021 /* Should not happen since timeout is infinite */
1022 ERR("Return value of poll is 0 with an infinite timeout.\n"
1023 "This should not have happened! Continuing...");
1029 for (i
= 0; i
< nb_fd
; i
++) {
1030 /* Fetch once the poll data */
1031 revents
= LTTNG_POLL_GETEV(&events
, i
);
1032 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1034 health_code_update();
1036 /* Thread quit pipe has been closed. Killing thread. */
1037 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1043 /* Check for data on kernel pipe */
1044 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
1045 (void) lttng_read(kernel_poll_pipe
[0],
1048 * Ret value is useless here, if this pipe gets any actions an
1049 * update is required anyway.
1051 update_poll_flag
= 1;
1055 * New CPU detected by the kernel. Adding kernel stream to
1056 * kernel session and updating the kernel consumer
1058 if (revents
& LPOLLIN
) {
1059 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1065 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1066 * and unregister kernel stream at this point.
1075 lttng_poll_clean(&events
);
1078 utils_close_pipe(kernel_poll_pipe
);
1079 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1082 ERR("Health error occurred in %s", __func__
);
1083 WARN("Kernel thread died unexpectedly. "
1084 "Kernel tracing can continue but CPU hotplug is disabled.");
1086 health_unregister(health_sessiond
);
1087 DBG("Kernel thread dying");
1092 * Signal pthread condition of the consumer data that the thread.
1094 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1096 pthread_mutex_lock(&data
->cond_mutex
);
1099 * The state is set before signaling. It can be any value, it's the waiter
1100 * job to correctly interpret this condition variable associated to the
1101 * consumer pthread_cond.
1103 * A value of 0 means that the corresponding thread of the consumer data
1104 * was not started. 1 indicates that the thread has started and is ready
1105 * for action. A negative value means that there was an error during the
1108 data
->consumer_thread_is_ready
= state
;
1109 (void) pthread_cond_signal(&data
->cond
);
1111 pthread_mutex_unlock(&data
->cond_mutex
);
1115 * This thread manage the consumer error sent back to the session daemon.
1117 static void *thread_manage_consumer(void *data
)
1119 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1120 uint32_t revents
, nb_fd
;
1121 enum lttcomm_return_code code
;
1122 struct lttng_poll_event events
;
1123 struct consumer_data
*consumer_data
= data
;
1125 DBG("[thread] Manage consumer started");
1127 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1129 health_code_update();
1132 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1133 * metadata_sock. Nothing more will be added to this poll set.
1135 ret
= sessiond_set_thread_pollset(&events
, 3);
1141 * The error socket here is already in a listening state which was done
1142 * just before spawning this thread to avoid a race between the consumer
1143 * daemon exec trying to connect and the listen() call.
1145 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1150 health_code_update();
1152 /* Infinite blocking call, waiting for transmission */
1154 health_poll_entry();
1156 if (testpoint(sessiond_thread_manage_consumer
)) {
1160 ret
= lttng_poll_wait(&events
, -1);
1164 * Restart interrupted system call.
1166 if (errno
== EINTR
) {
1174 for (i
= 0; i
< nb_fd
; i
++) {
1175 /* Fetch once the poll data */
1176 revents
= LTTNG_POLL_GETEV(&events
, i
);
1177 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1179 health_code_update();
1181 /* Thread quit pipe has been closed. Killing thread. */
1182 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1188 /* Event on the registration socket */
1189 if (pollfd
== consumer_data
->err_sock
) {
1190 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1191 ERR("consumer err socket poll error");
1197 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1203 * Set the CLOEXEC flag. Return code is useless because either way, the
1206 (void) utils_set_fd_cloexec(sock
);
1208 health_code_update();
1210 DBG2("Receiving code from consumer err_sock");
1212 /* Getting status code from kconsumerd */
1213 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1214 sizeof(enum lttcomm_return_code
));
1219 health_code_update();
1220 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1221 /* Connect both socket, command and metadata. */
1222 consumer_data
->cmd_sock
=
1223 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1224 consumer_data
->metadata_fd
=
1225 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1226 if (consumer_data
->cmd_sock
< 0
1227 || consumer_data
->metadata_fd
< 0) {
1228 PERROR("consumer connect cmd socket");
1229 /* On error, signal condition and quit. */
1230 signal_consumer_condition(consumer_data
, -1);
1233 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1234 /* Create metadata socket lock. */
1235 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1236 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1237 PERROR("zmalloc pthread mutex");
1241 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1243 signal_consumer_condition(consumer_data
, 1);
1244 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1245 DBG("Consumer metadata socket ready (fd: %d)",
1246 consumer_data
->metadata_fd
);
1248 ERR("consumer error when waiting for SOCK_READY : %s",
1249 lttcomm_get_readable_code(-code
));
1253 /* Remove the consumerd error sock since we've established a connexion */
1254 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1259 /* Add new accepted error socket. */
1260 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1265 /* Add metadata socket that is successfully connected. */
1266 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1267 LPOLLIN
| LPOLLRDHUP
);
1272 health_code_update();
1274 /* Infinite blocking call, waiting for transmission */
1277 health_code_update();
1279 /* Exit the thread because the thread quit pipe has been triggered. */
1281 /* Not a health error. */
1286 health_poll_entry();
1287 ret
= lttng_poll_wait(&events
, -1);
1291 * Restart interrupted system call.
1293 if (errno
== EINTR
) {
1301 for (i
= 0; i
< nb_fd
; i
++) {
1302 /* Fetch once the poll data */
1303 revents
= LTTNG_POLL_GETEV(&events
, i
);
1304 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1306 health_code_update();
1309 * Thread quit pipe has been triggered, flag that we should stop
1310 * but continue the current loop to handle potential data from
1313 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1315 if (pollfd
== sock
) {
1316 /* Event on the consumerd socket */
1317 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1318 ERR("consumer err socket second poll error");
1321 health_code_update();
1322 /* Wait for any kconsumerd error */
1323 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1324 sizeof(enum lttcomm_return_code
));
1326 ERR("consumer closed the command socket");
1330 ERR("consumer return code : %s",
1331 lttcomm_get_readable_code(-code
));
1334 } else if (pollfd
== consumer_data
->metadata_fd
) {
1335 /* UST metadata requests */
1336 ret
= ust_consumer_metadata_request(
1337 &consumer_data
->metadata_sock
);
1339 ERR("Handling metadata request");
1343 /* No need for an else branch all FDs are tested prior. */
1345 health_code_update();
1351 * We lock here because we are about to close the sockets and some other
1352 * thread might be using them so get exclusive access which will abort all
1353 * other consumer command by other threads.
1355 pthread_mutex_lock(&consumer_data
->lock
);
1357 /* Immediately set the consumerd state to stopped */
1358 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1359 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1360 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1361 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1362 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1364 /* Code flow error... */
1368 if (consumer_data
->err_sock
>= 0) {
1369 ret
= close(consumer_data
->err_sock
);
1373 consumer_data
->err_sock
= -1;
1375 if (consumer_data
->cmd_sock
>= 0) {
1376 ret
= close(consumer_data
->cmd_sock
);
1380 consumer_data
->cmd_sock
= -1;
1382 if (consumer_data
->metadata_sock
.fd_ptr
&&
1383 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1384 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1396 unlink(consumer_data
->err_unix_sock_path
);
1397 unlink(consumer_data
->cmd_unix_sock_path
);
1398 consumer_data
->pid
= 0;
1399 pthread_mutex_unlock(&consumer_data
->lock
);
1401 /* Cleanup metadata socket mutex. */
1402 if (consumer_data
->metadata_sock
.lock
) {
1403 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1404 free(consumer_data
->metadata_sock
.lock
);
1406 lttng_poll_clean(&events
);
1410 ERR("Health error occurred in %s", __func__
);
1412 health_unregister(health_sessiond
);
1413 DBG("consumer thread cleanup completed");
1419 * This thread manage application communication.
1421 static void *thread_manage_apps(void *data
)
1423 int i
, ret
, pollfd
, err
= -1;
1425 uint32_t revents
, nb_fd
;
1426 struct lttng_poll_event events
;
1428 DBG("[thread] Manage application started");
1430 rcu_register_thread();
1431 rcu_thread_online();
1433 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1435 if (testpoint(sessiond_thread_manage_apps
)) {
1436 goto error_testpoint
;
1439 health_code_update();
1441 ret
= sessiond_set_thread_pollset(&events
, 2);
1443 goto error_poll_create
;
1446 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1451 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1455 health_code_update();
1458 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1460 /* Inifinite blocking call, waiting for transmission */
1462 health_poll_entry();
1463 ret
= lttng_poll_wait(&events
, -1);
1467 * Restart interrupted system call.
1469 if (errno
== EINTR
) {
1477 for (i
= 0; i
< nb_fd
; i
++) {
1478 /* Fetch once the poll data */
1479 revents
= LTTNG_POLL_GETEV(&events
, i
);
1480 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1482 health_code_update();
1484 /* Thread quit pipe has been closed. Killing thread. */
1485 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1491 /* Inspect the apps cmd pipe */
1492 if (pollfd
== apps_cmd_pipe
[0]) {
1493 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1494 ERR("Apps command pipe error");
1496 } else if (revents
& LPOLLIN
) {
1500 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1501 if (size_ret
< sizeof(sock
)) {
1502 PERROR("read apps cmd pipe");
1506 health_code_update();
1509 * We only monitor the error events of the socket. This
1510 * thread does not handle any incoming data from UST
1513 ret
= lttng_poll_add(&events
, sock
,
1514 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1519 DBG("Apps with sock %d added to poll set", sock
);
1523 * At this point, we know that a registered application made
1524 * the event at poll_wait.
1526 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1527 /* Removing from the poll set */
1528 ret
= lttng_poll_del(&events
, pollfd
);
1533 /* Socket closed on remote end. */
1534 ust_app_unregister(pollfd
);
1538 health_code_update();
1544 lttng_poll_clean(&events
);
1547 utils_close_pipe(apps_cmd_pipe
);
1548 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1551 * We don't clean the UST app hash table here since already registered
1552 * applications can still be controlled so let them be until the session
1553 * daemon dies or the applications stop.
1558 ERR("Health error occurred in %s", __func__
);
1560 health_unregister(health_sessiond
);
1561 DBG("Application communication apps thread cleanup complete");
1562 rcu_thread_offline();
1563 rcu_unregister_thread();
1568 * Send a socket to a thread This is called from the dispatch UST registration
1569 * thread once all sockets are set for the application.
1571 * The sock value can be invalid, we don't really care, the thread will handle
1572 * it and make the necessary cleanup if so.
1574 * On success, return 0 else a negative value being the errno message of the
1577 static int send_socket_to_thread(int fd
, int sock
)
1582 * It's possible that the FD is set as invalid with -1 concurrently just
1583 * before calling this function being a shutdown state of the thread.
1590 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1591 if (ret
< sizeof(sock
)) {
1592 PERROR("write apps pipe %d", fd
);
1599 /* All good. Don't send back the write positive ret value. */
1606 * Sanitize the wait queue of the dispatch registration thread meaning removing
1607 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1608 * notify socket is never received.
1610 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1612 int ret
, nb_fd
= 0, i
;
1613 unsigned int fd_added
= 0;
1614 struct lttng_poll_event events
;
1615 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1619 lttng_poll_init(&events
);
1621 /* Just skip everything for an empty queue. */
1622 if (!wait_queue
->count
) {
1626 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1631 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1632 &wait_queue
->head
, head
) {
1633 assert(wait_node
->app
);
1634 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1635 LPOLLHUP
| LPOLLERR
);
1648 * Poll but don't block so we can quickly identify the faulty events and
1649 * clean them afterwards from the wait queue.
1651 ret
= lttng_poll_wait(&events
, 0);
1657 for (i
= 0; i
< nb_fd
; i
++) {
1658 /* Get faulty FD. */
1659 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1660 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1662 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1663 &wait_queue
->head
, head
) {
1664 if (pollfd
== wait_node
->app
->sock
&&
1665 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1666 cds_list_del(&wait_node
->head
);
1667 wait_queue
->count
--;
1668 ust_app_destroy(wait_node
->app
);
1676 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1680 lttng_poll_clean(&events
);
1684 lttng_poll_clean(&events
);
1686 ERR("Unable to sanitize wait queue");
1691 * Dispatch request from the registration threads to the application
1692 * communication thread.
1694 static void *thread_dispatch_ust_registration(void *data
)
1697 struct cds_wfcq_node
*node
;
1698 struct ust_command
*ust_cmd
= NULL
;
1699 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1700 struct ust_reg_wait_queue wait_queue
= {
1704 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1706 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1707 goto error_testpoint
;
1710 health_code_update();
1712 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1714 DBG("[thread] Dispatch UST command started");
1716 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1717 health_code_update();
1719 /* Atomically prepare the queue futex */
1720 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1723 struct ust_app
*app
= NULL
;
1727 * Make sure we don't have node(s) that have hung up before receiving
1728 * the notify socket. This is to clean the list in order to avoid
1729 * memory leaks from notify socket that are never seen.
1731 sanitize_wait_queue(&wait_queue
);
1733 health_code_update();
1734 /* Dequeue command for registration */
1735 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1737 DBG("Woken up but nothing in the UST command queue");
1738 /* Continue thread execution */
1742 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1744 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1745 " gid:%d sock:%d name:%s (version %d.%d)",
1746 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1747 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1748 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1749 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1751 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1752 wait_node
= zmalloc(sizeof(*wait_node
));
1754 PERROR("zmalloc wait_node dispatch");
1755 ret
= close(ust_cmd
->sock
);
1757 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1759 lttng_fd_put(LTTNG_FD_APPS
, 1);
1763 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1765 /* Create application object if socket is CMD. */
1766 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1768 if (!wait_node
->app
) {
1769 ret
= close(ust_cmd
->sock
);
1771 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1773 lttng_fd_put(LTTNG_FD_APPS
, 1);
1779 * Add application to the wait queue so we can set the notify
1780 * socket before putting this object in the global ht.
1782 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1787 * We have to continue here since we don't have the notify
1788 * socket and the application MUST be added to the hash table
1789 * only at that moment.
1794 * Look for the application in the local wait queue and set the
1795 * notify socket if found.
1797 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1798 &wait_queue
.head
, head
) {
1799 health_code_update();
1800 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1801 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1802 cds_list_del(&wait_node
->head
);
1804 app
= wait_node
->app
;
1806 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1812 * With no application at this stage the received socket is
1813 * basically useless so close it before we free the cmd data
1814 * structure for good.
1817 ret
= close(ust_cmd
->sock
);
1819 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1821 lttng_fd_put(LTTNG_FD_APPS
, 1);
1828 * @session_lock_list
1830 * Lock the global session list so from the register up to the
1831 * registration done message, no thread can see the application
1832 * and change its state.
1834 session_lock_list();
1838 * Add application to the global hash table. This needs to be
1839 * done before the update to the UST registry can locate the
1844 /* Set app version. This call will print an error if needed. */
1845 (void) ust_app_version(app
);
1847 /* Send notify socket through the notify pipe. */
1848 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1852 session_unlock_list();
1854 * No notify thread, stop the UST tracing. However, this is
1855 * not an internal error of the this thread thus setting
1856 * the health error code to a normal exit.
1863 * Update newly registered application with the tracing
1864 * registry info already enabled information.
1866 update_ust_app(app
->sock
);
1869 * Don't care about return value. Let the manage apps threads
1870 * handle app unregistration upon socket close.
1872 (void) ust_app_register_done(app
->sock
);
1875 * Even if the application socket has been closed, send the app
1876 * to the thread and unregistration will take place at that
1879 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1882 session_unlock_list();
1884 * No apps. thread, stop the UST tracing. However, this is
1885 * not an internal error of the this thread thus setting
1886 * the health error code to a normal exit.
1893 session_unlock_list();
1895 } while (node
!= NULL
);
1897 health_poll_entry();
1898 /* Futex wait on queue. Blocking call on futex() */
1899 futex_nto1_wait(&ust_cmd_queue
.futex
);
1902 /* Normal exit, no error */
1906 /* Clean up wait queue. */
1907 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1908 &wait_queue
.head
, head
) {
1909 cds_list_del(&wait_node
->head
);
1915 DBG("Dispatch thread dying");
1918 ERR("Health error occurred in %s", __func__
);
1920 health_unregister(health_sessiond
);
1925 * This thread manage application registration.
1927 static void *thread_registration_apps(void *data
)
1929 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1930 uint32_t revents
, nb_fd
;
1931 struct lttng_poll_event events
;
1933 * Get allocated in this thread, enqueued to a global queue, dequeued and
1934 * freed in the manage apps thread.
1936 struct ust_command
*ust_cmd
= NULL
;
1938 DBG("[thread] Manage application registration started");
1940 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
1942 if (testpoint(sessiond_thread_registration_apps
)) {
1943 goto error_testpoint
;
1946 ret
= lttcomm_listen_unix_sock(apps_sock
);
1952 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1953 * more will be added to this poll set.
1955 ret
= sessiond_set_thread_pollset(&events
, 2);
1957 goto error_create_poll
;
1960 /* Add the application registration socket */
1961 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1963 goto error_poll_add
;
1966 /* Notify all applications to register */
1967 ret
= notify_ust_apps(1);
1969 ERR("Failed to notify applications or create the wait shared memory.\n"
1970 "Execution continues but there might be problem for already\n"
1971 "running applications that wishes to register.");
1975 DBG("Accepting application registration");
1977 /* Inifinite blocking call, waiting for transmission */
1979 health_poll_entry();
1980 ret
= lttng_poll_wait(&events
, -1);
1984 * Restart interrupted system call.
1986 if (errno
== EINTR
) {
1994 for (i
= 0; i
< nb_fd
; i
++) {
1995 health_code_update();
1997 /* Fetch once the poll data */
1998 revents
= LTTNG_POLL_GETEV(&events
, i
);
1999 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2001 /* Thread quit pipe has been closed. Killing thread. */
2002 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2008 /* Event on the registration socket */
2009 if (pollfd
== apps_sock
) {
2010 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2011 ERR("Register apps socket poll error");
2013 } else if (revents
& LPOLLIN
) {
2014 sock
= lttcomm_accept_unix_sock(apps_sock
);
2020 * Set socket timeout for both receiving and ending.
2021 * app_socket_timeout is in seconds, whereas
2022 * lttcomm_setsockopt_rcv_timeout and
2023 * lttcomm_setsockopt_snd_timeout expect msec as
2026 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2027 app_socket_timeout
* 1000);
2028 (void) lttcomm_setsockopt_snd_timeout(sock
,
2029 app_socket_timeout
* 1000);
2032 * Set the CLOEXEC flag. Return code is useless because
2033 * either way, the show must go on.
2035 (void) utils_set_fd_cloexec(sock
);
2037 /* Create UST registration command for enqueuing */
2038 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2039 if (ust_cmd
== NULL
) {
2040 PERROR("ust command zmalloc");
2045 * Using message-based transmissions to ensure we don't
2046 * have to deal with partially received messages.
2048 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2050 ERR("Exhausted file descriptors allowed for applications.");
2060 health_code_update();
2061 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2064 /* Close socket of the application. */
2069 lttng_fd_put(LTTNG_FD_APPS
, 1);
2073 health_code_update();
2075 ust_cmd
->sock
= sock
;
2078 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2079 " gid:%d sock:%d name:%s (version %d.%d)",
2080 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2081 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2082 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2083 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2086 * Lock free enqueue the registration request. The red pill
2087 * has been taken! This apps will be part of the *system*.
2089 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2092 * Wake the registration queue futex. Implicit memory
2093 * barrier with the exchange in cds_wfcq_enqueue.
2095 futex_nto1_wake(&ust_cmd_queue
.futex
);
2103 /* Notify that the registration thread is gone */
2106 if (apps_sock
>= 0) {
2107 ret
= close(apps_sock
);
2117 lttng_fd_put(LTTNG_FD_APPS
, 1);
2119 unlink(apps_unix_sock_path
);
2122 lttng_poll_clean(&events
);
2126 DBG("UST Registration thread cleanup complete");
2129 ERR("Health error occurred in %s", __func__
);
2131 health_unregister(health_sessiond
);
2137 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2138 * exec or it will fails.
2140 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2143 struct timespec timeout
;
2145 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2146 consumer_data
->consumer_thread_is_ready
= 0;
2148 /* Setup pthread condition */
2149 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2152 PERROR("pthread_condattr_init consumer data");
2157 * Set the monotonic clock in order to make sure we DO NOT jump in time
2158 * between the clock_gettime() call and the timedwait call. See bug #324
2159 * for a more details and how we noticed it.
2161 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2164 PERROR("pthread_condattr_setclock consumer data");
2168 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2171 PERROR("pthread_cond_init consumer data");
2175 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2178 PERROR("pthread_create consumer");
2183 /* We are about to wait on a pthread condition */
2184 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2186 /* Get time for sem_timedwait absolute timeout */
2187 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2189 * Set the timeout for the condition timed wait even if the clock gettime
2190 * call fails since we might loop on that call and we want to avoid to
2191 * increment the timeout too many times.
2193 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2196 * The following loop COULD be skipped in some conditions so this is why we
2197 * set ret to 0 in order to make sure at least one round of the loop is
2203 * Loop until the condition is reached or when a timeout is reached. Note
2204 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2205 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2206 * possible. This loop does not take any chances and works with both of
2209 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2210 if (clock_ret
< 0) {
2211 PERROR("clock_gettime spawn consumer");
2212 /* Infinite wait for the consumerd thread to be ready */
2213 ret
= pthread_cond_wait(&consumer_data
->cond
,
2214 &consumer_data
->cond_mutex
);
2216 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2217 &consumer_data
->cond_mutex
, &timeout
);
2221 /* Release the pthread condition */
2222 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2226 if (ret
== ETIMEDOUT
) {
2230 * Call has timed out so we kill the kconsumerd_thread and return
2233 ERR("Condition timed out. The consumer thread was never ready."
2235 pth_ret
= pthread_cancel(consumer_data
->thread
);
2237 PERROR("pthread_cancel consumer thread");
2240 PERROR("pthread_cond_wait failed consumer thread");
2242 /* Caller is expecting a negative value on failure. */
2247 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2248 if (consumer_data
->pid
== 0) {
2249 ERR("Consumerd did not start");
2250 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2253 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2262 * Join consumer thread
2264 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2268 /* Consumer pid must be a real one. */
2269 if (consumer_data
->pid
> 0) {
2271 ret
= kill(consumer_data
->pid
, SIGTERM
);
2273 ERR("Error killing consumer daemon");
2276 return pthread_join(consumer_data
->thread
, &status
);
2283 * Fork and exec a consumer daemon (consumerd).
2285 * Return pid if successful else -1.
2287 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2291 const char *consumer_to_use
;
2292 const char *verbosity
;
2295 DBG("Spawning consumerd");
2302 if (opt_verbose_consumer
) {
2303 verbosity
= "--verbose";
2304 } else if (lttng_opt_quiet
) {
2305 verbosity
= "--quiet";
2310 switch (consumer_data
->type
) {
2311 case LTTNG_CONSUMER_KERNEL
:
2313 * Find out which consumerd to execute. We will first try the
2314 * 64-bit path, then the sessiond's installation directory, and
2315 * fallback on the 32-bit one,
2317 DBG3("Looking for a kernel consumer at these locations:");
2318 DBG3(" 1) %s", consumerd64_bin
);
2319 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2320 DBG3(" 3) %s", consumerd32_bin
);
2321 if (stat(consumerd64_bin
, &st
) == 0) {
2322 DBG3("Found location #1");
2323 consumer_to_use
= consumerd64_bin
;
2324 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2325 DBG3("Found location #2");
2326 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2327 } else if (stat(consumerd32_bin
, &st
) == 0) {
2328 DBG3("Found location #3");
2329 consumer_to_use
= consumerd32_bin
;
2331 DBG("Could not find any valid consumerd executable");
2335 DBG("Using kernel consumer at: %s", consumer_to_use
);
2336 ret
= execl(consumer_to_use
,
2337 "lttng-consumerd", verbosity
, "-k",
2338 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2339 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2340 "--group", tracing_group_name
,
2343 case LTTNG_CONSUMER64_UST
:
2345 char *tmpnew
= NULL
;
2347 if (consumerd64_libdir
[0] != '\0') {
2351 tmp
= getenv("LD_LIBRARY_PATH");
2355 tmplen
= strlen("LD_LIBRARY_PATH=")
2356 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2357 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2362 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2363 strcat(tmpnew
, consumerd64_libdir
);
2364 if (tmp
[0] != '\0') {
2365 strcat(tmpnew
, ":");
2366 strcat(tmpnew
, tmp
);
2368 ret
= putenv(tmpnew
);
2375 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2376 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2377 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2378 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2379 "--group", tracing_group_name
,
2381 if (consumerd64_libdir
[0] != '\0') {
2386 case LTTNG_CONSUMER32_UST
:
2388 char *tmpnew
= NULL
;
2390 if (consumerd32_libdir
[0] != '\0') {
2394 tmp
= getenv("LD_LIBRARY_PATH");
2398 tmplen
= strlen("LD_LIBRARY_PATH=")
2399 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2400 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2405 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2406 strcat(tmpnew
, consumerd32_libdir
);
2407 if (tmp
[0] != '\0') {
2408 strcat(tmpnew
, ":");
2409 strcat(tmpnew
, tmp
);
2411 ret
= putenv(tmpnew
);
2418 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2419 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2420 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2421 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2422 "--group", tracing_group_name
,
2424 if (consumerd32_libdir
[0] != '\0') {
2430 PERROR("unknown consumer type");
2434 PERROR("Consumer execl()");
2436 /* Reaching this point, we got a failure on our execl(). */
2438 } else if (pid
> 0) {
2441 PERROR("start consumer fork");
2449 * Spawn the consumerd daemon and session daemon thread.
2451 static int start_consumerd(struct consumer_data
*consumer_data
)
2456 * Set the listen() state on the socket since there is a possible race
2457 * between the exec() of the consumer daemon and this call if place in the
2458 * consumer thread. See bug #366 for more details.
2460 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2465 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2466 if (consumer_data
->pid
!= 0) {
2467 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2471 ret
= spawn_consumerd(consumer_data
);
2473 ERR("Spawning consumerd failed");
2474 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2478 /* Setting up the consumer_data pid */
2479 consumer_data
->pid
= ret
;
2480 DBG2("Consumer pid %d", consumer_data
->pid
);
2481 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2483 DBG2("Spawning consumer control thread");
2484 ret
= spawn_consumer_thread(consumer_data
);
2486 ERR("Fatal error spawning consumer control thread");
2494 /* Cleanup already created sockets on error. */
2495 if (consumer_data
->err_sock
>= 0) {
2498 err
= close(consumer_data
->err_sock
);
2500 PERROR("close consumer data error socket");
2507 * Setup necessary data for kernel tracer action.
2509 static int init_kernel_tracer(void)
2513 /* Modprobe lttng kernel modules */
2514 ret
= modprobe_lttng_control();
2519 /* Open debugfs lttng */
2520 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2521 if (kernel_tracer_fd
< 0) {
2522 DBG("Failed to open %s", module_proc_lttng
);
2527 /* Validate kernel version */
2528 ret
= kernel_validate_version(kernel_tracer_fd
);
2533 ret
= modprobe_lttng_data();
2538 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2542 modprobe_remove_lttng_control();
2543 ret
= close(kernel_tracer_fd
);
2547 kernel_tracer_fd
= -1;
2548 return LTTNG_ERR_KERN_VERSION
;
2551 ret
= close(kernel_tracer_fd
);
2557 modprobe_remove_lttng_control();
2560 WARN("No kernel tracer available");
2561 kernel_tracer_fd
= -1;
2563 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2565 return LTTNG_ERR_KERN_NA
;
2571 * Copy consumer output from the tracing session to the domain session. The
2572 * function also applies the right modification on a per domain basis for the
2573 * trace files destination directory.
2575 * Should *NOT* be called with RCU read-side lock held.
2577 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2580 const char *dir_name
;
2581 struct consumer_output
*consumer
;
2584 assert(session
->consumer
);
2587 case LTTNG_DOMAIN_KERNEL
:
2588 DBG3("Copying tracing session consumer output in kernel session");
2590 * XXX: We should audit the session creation and what this function
2591 * does "extra" in order to avoid a destroy since this function is used
2592 * in the domain session creation (kernel and ust) only. Same for UST
2595 if (session
->kernel_session
->consumer
) {
2596 consumer_destroy_output(session
->kernel_session
->consumer
);
2598 session
->kernel_session
->consumer
=
2599 consumer_copy_output(session
->consumer
);
2600 /* Ease our life a bit for the next part */
2601 consumer
= session
->kernel_session
->consumer
;
2602 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2604 case LTTNG_DOMAIN_JUL
:
2605 case LTTNG_DOMAIN_LOG4J
:
2606 case LTTNG_DOMAIN_UST
:
2607 DBG3("Copying tracing session consumer output in UST session");
2608 if (session
->ust_session
->consumer
) {
2609 consumer_destroy_output(session
->ust_session
->consumer
);
2611 session
->ust_session
->consumer
=
2612 consumer_copy_output(session
->consumer
);
2613 /* Ease our life a bit for the next part */
2614 consumer
= session
->ust_session
->consumer
;
2615 dir_name
= DEFAULT_UST_TRACE_DIR
;
2618 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2622 /* Append correct directory to subdir */
2623 strncat(consumer
->subdir
, dir_name
,
2624 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2625 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2634 * Create an UST session and add it to the session ust list.
2636 * Should *NOT* be called with RCU read-side lock held.
2638 static int create_ust_session(struct ltt_session
*session
,
2639 struct lttng_domain
*domain
)
2642 struct ltt_ust_session
*lus
= NULL
;
2646 assert(session
->consumer
);
2648 switch (domain
->type
) {
2649 case LTTNG_DOMAIN_JUL
:
2650 case LTTNG_DOMAIN_LOG4J
:
2651 case LTTNG_DOMAIN_UST
:
2654 ERR("Unknown UST domain on create session %d", domain
->type
);
2655 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2659 DBG("Creating UST session");
2661 lus
= trace_ust_create_session(session
->id
);
2663 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2667 lus
->uid
= session
->uid
;
2668 lus
->gid
= session
->gid
;
2669 lus
->output_traces
= session
->output_traces
;
2670 lus
->snapshot_mode
= session
->snapshot_mode
;
2671 lus
->live_timer_interval
= session
->live_timer
;
2672 session
->ust_session
= lus
;
2674 /* Copy session output to the newly created UST session */
2675 ret
= copy_session_consumer(domain
->type
, session
);
2676 if (ret
!= LTTNG_OK
) {
2684 session
->ust_session
= NULL
;
2689 * Create a kernel tracer session then create the default channel.
2691 static int create_kernel_session(struct ltt_session
*session
)
2695 DBG("Creating kernel session");
2697 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2699 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2703 /* Code flow safety */
2704 assert(session
->kernel_session
);
2706 /* Copy session output to the newly created Kernel session */
2707 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2708 if (ret
!= LTTNG_OK
) {
2712 /* Create directory(ies) on local filesystem. */
2713 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2714 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2715 ret
= run_as_mkdir_recursive(
2716 session
->kernel_session
->consumer
->dst
.trace_path
,
2717 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2719 if (ret
!= -EEXIST
) {
2720 ERR("Trace directory creation error");
2726 session
->kernel_session
->uid
= session
->uid
;
2727 session
->kernel_session
->gid
= session
->gid
;
2728 session
->kernel_session
->output_traces
= session
->output_traces
;
2729 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2734 trace_kernel_destroy_session(session
->kernel_session
);
2735 session
->kernel_session
= NULL
;
2740 * Count number of session permitted by uid/gid.
2742 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2745 struct ltt_session
*session
;
2747 DBG("Counting number of available session for UID %d GID %d",
2749 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2751 * Only list the sessions the user can control.
2753 if (!session_access_ok(session
, uid
, gid
)) {
2762 * Process the command requested by the lttng client within the command
2763 * context structure. This function make sure that the return structure (llm)
2764 * is set and ready for transmission before returning.
2766 * Return any error encountered or 0 for success.
2768 * "sock" is only used for special-case var. len data.
2770 * Should *NOT* be called with RCU read-side lock held.
2772 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2776 int need_tracing_session
= 1;
2779 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2783 switch (cmd_ctx
->lsm
->cmd_type
) {
2784 case LTTNG_CREATE_SESSION
:
2785 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2786 case LTTNG_CREATE_SESSION_LIVE
:
2787 case LTTNG_DESTROY_SESSION
:
2788 case LTTNG_LIST_SESSIONS
:
2789 case LTTNG_LIST_DOMAINS
:
2790 case LTTNG_START_TRACE
:
2791 case LTTNG_STOP_TRACE
:
2792 case LTTNG_DATA_PENDING
:
2793 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2794 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2795 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2796 case LTTNG_SNAPSHOT_RECORD
:
2797 case LTTNG_SAVE_SESSION
:
2804 if (opt_no_kernel
&& need_domain
2805 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2807 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2809 ret
= LTTNG_ERR_KERN_NA
;
2814 /* Deny register consumer if we already have a spawned consumer. */
2815 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2816 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2817 if (kconsumer_data
.pid
> 0) {
2818 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2819 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2822 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2826 * Check for command that don't needs to allocate a returned payload. We do
2827 * this here so we don't have to make the call for no payload at each
2830 switch(cmd_ctx
->lsm
->cmd_type
) {
2831 case LTTNG_LIST_SESSIONS
:
2832 case LTTNG_LIST_TRACEPOINTS
:
2833 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2834 case LTTNG_LIST_DOMAINS
:
2835 case LTTNG_LIST_CHANNELS
:
2836 case LTTNG_LIST_EVENTS
:
2837 case LTTNG_LIST_SYSCALLS
:
2840 /* Setup lttng message with no payload */
2841 ret
= setup_lttng_msg(cmd_ctx
, 0);
2843 /* This label does not try to unlock the session */
2844 goto init_setup_error
;
2848 /* Commands that DO NOT need a session. */
2849 switch (cmd_ctx
->lsm
->cmd_type
) {
2850 case LTTNG_CREATE_SESSION
:
2851 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2852 case LTTNG_CREATE_SESSION_LIVE
:
2853 case LTTNG_CALIBRATE
:
2854 case LTTNG_LIST_SESSIONS
:
2855 case LTTNG_LIST_TRACEPOINTS
:
2856 case LTTNG_LIST_SYSCALLS
:
2857 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2858 case LTTNG_SAVE_SESSION
:
2859 need_tracing_session
= 0;
2862 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2864 * We keep the session list lock across _all_ commands
2865 * for now, because the per-session lock does not
2866 * handle teardown properly.
2868 session_lock_list();
2869 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2870 if (cmd_ctx
->session
== NULL
) {
2871 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2874 /* Acquire lock for the session */
2875 session_lock(cmd_ctx
->session
);
2881 * Commands that need a valid session but should NOT create one if none
2882 * exists. Instead of creating one and destroying it when the command is
2883 * handled, process that right before so we save some round trip in useless
2886 switch (cmd_ctx
->lsm
->cmd_type
) {
2887 case LTTNG_DISABLE_CHANNEL
:
2888 case LTTNG_DISABLE_EVENT
:
2889 switch (cmd_ctx
->lsm
->domain
.type
) {
2890 case LTTNG_DOMAIN_KERNEL
:
2891 if (!cmd_ctx
->session
->kernel_session
) {
2892 ret
= LTTNG_ERR_NO_CHANNEL
;
2896 case LTTNG_DOMAIN_JUL
:
2897 case LTTNG_DOMAIN_LOG4J
:
2898 case LTTNG_DOMAIN_UST
:
2899 if (!cmd_ctx
->session
->ust_session
) {
2900 ret
= LTTNG_ERR_NO_CHANNEL
;
2905 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2917 * Check domain type for specific "pre-action".
2919 switch (cmd_ctx
->lsm
->domain
.type
) {
2920 case LTTNG_DOMAIN_KERNEL
:
2922 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2926 /* Kernel tracer check */
2927 if (kernel_tracer_fd
== -1) {
2928 /* Basically, load kernel tracer modules */
2929 ret
= init_kernel_tracer();
2935 /* Consumer is in an ERROR state. Report back to client */
2936 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2937 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2941 /* Need a session for kernel command */
2942 if (need_tracing_session
) {
2943 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2944 ret
= create_kernel_session(cmd_ctx
->session
);
2946 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2951 /* Start the kernel consumer daemon */
2952 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2953 if (kconsumer_data
.pid
== 0 &&
2954 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2955 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2956 ret
= start_consumerd(&kconsumer_data
);
2958 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2961 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2963 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2967 * The consumer was just spawned so we need to add the socket to
2968 * the consumer output of the session if exist.
2970 ret
= consumer_create_socket(&kconsumer_data
,
2971 cmd_ctx
->session
->kernel_session
->consumer
);
2978 case LTTNG_DOMAIN_JUL
:
2979 case LTTNG_DOMAIN_LOG4J
:
2980 case LTTNG_DOMAIN_UST
:
2982 if (!ust_app_supported()) {
2983 ret
= LTTNG_ERR_NO_UST
;
2986 /* Consumer is in an ERROR state. Report back to client */
2987 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2988 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2992 if (need_tracing_session
) {
2993 /* Create UST session if none exist. */
2994 if (cmd_ctx
->session
->ust_session
== NULL
) {
2995 ret
= create_ust_session(cmd_ctx
->session
,
2996 &cmd_ctx
->lsm
->domain
);
2997 if (ret
!= LTTNG_OK
) {
3002 /* Start the UST consumer daemons */
3004 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3005 if (consumerd64_bin
[0] != '\0' &&
3006 ustconsumer64_data
.pid
== 0 &&
3007 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3008 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3009 ret
= start_consumerd(&ustconsumer64_data
);
3011 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3012 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3016 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3017 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3019 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3023 * Setup socket for consumer 64 bit. No need for atomic access
3024 * since it was set above and can ONLY be set in this thread.
3026 ret
= consumer_create_socket(&ustconsumer64_data
,
3027 cmd_ctx
->session
->ust_session
->consumer
);
3033 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3034 if (consumerd32_bin
[0] != '\0' &&
3035 ustconsumer32_data
.pid
== 0 &&
3036 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3037 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3038 ret
= start_consumerd(&ustconsumer32_data
);
3040 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3041 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3045 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3046 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3048 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3052 * Setup socket for consumer 64 bit. No need for atomic access
3053 * since it was set above and can ONLY be set in this thread.
3055 ret
= consumer_create_socket(&ustconsumer32_data
,
3056 cmd_ctx
->session
->ust_session
->consumer
);
3068 /* Validate consumer daemon state when start/stop trace command */
3069 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3070 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3071 switch (cmd_ctx
->lsm
->domain
.type
) {
3072 case LTTNG_DOMAIN_JUL
:
3073 case LTTNG_DOMAIN_LOG4J
:
3074 case LTTNG_DOMAIN_UST
:
3075 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3076 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3080 case LTTNG_DOMAIN_KERNEL
:
3081 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3082 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3090 * Check that the UID or GID match that of the tracing session.
3091 * The root user can interact with all sessions.
3093 if (need_tracing_session
) {
3094 if (!session_access_ok(cmd_ctx
->session
,
3095 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3096 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3097 ret
= LTTNG_ERR_EPERM
;
3103 * Send relayd information to consumer as soon as we have a domain and a
3106 if (cmd_ctx
->session
&& need_domain
) {
3108 * Setup relayd if not done yet. If the relayd information was already
3109 * sent to the consumer, this call will gracefully return.
3111 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3112 if (ret
!= LTTNG_OK
) {
3117 /* Process by command type */
3118 switch (cmd_ctx
->lsm
->cmd_type
) {
3119 case LTTNG_ADD_CONTEXT
:
3121 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3122 cmd_ctx
->lsm
->u
.context
.channel_name
,
3123 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
3126 case LTTNG_DISABLE_CHANNEL
:
3128 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3129 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3132 case LTTNG_DISABLE_EVENT
:
3134 /* FIXME: passing packed structure to non-packed pointer */
3135 /* TODO: handle filter */
3136 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3137 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3138 &cmd_ctx
->lsm
->u
.disable
.event
);
3141 case LTTNG_ENABLE_CHANNEL
:
3143 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3144 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3147 case LTTNG_ENABLE_EVENT
:
3149 struct lttng_event_exclusion
*exclusion
= NULL
;
3150 struct lttng_filter_bytecode
*bytecode
= NULL
;
3151 char *filter_expression
= NULL
;
3153 /* Handle exclusion events and receive it from the client. */
3154 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3155 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3157 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3158 (count
* LTTNG_SYMBOL_NAME_LEN
));
3160 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3164 DBG("Receiving var len exclusion event list from client ...");
3165 exclusion
->count
= count
;
3166 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3167 count
* LTTNG_SYMBOL_NAME_LEN
);
3169 DBG("Nothing recv() from client var len data... continuing");
3172 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3177 /* Get filter expression from client. */
3178 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3179 size_t expression_len
=
3180 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3182 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3183 ret
= LTTNG_ERR_FILTER_INVAL
;
3188 filter_expression
= zmalloc(expression_len
);
3189 if (!filter_expression
) {
3191 ret
= LTTNG_ERR_FILTER_NOMEM
;
3195 /* Receive var. len. data */
3196 DBG("Receiving var len filter's expression from client ...");
3197 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3200 DBG("Nothing recv() from client car len data... continuing");
3202 free(filter_expression
);
3204 ret
= LTTNG_ERR_FILTER_INVAL
;
3209 /* Handle filter and get bytecode from client. */
3210 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3211 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3213 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3214 ret
= LTTNG_ERR_FILTER_INVAL
;
3215 free(filter_expression
);
3220 bytecode
= zmalloc(bytecode_len
);
3222 free(filter_expression
);
3224 ret
= LTTNG_ERR_FILTER_NOMEM
;
3228 /* Receive var. len. data */
3229 DBG("Receiving var len filter's bytecode from client ...");
3230 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3232 DBG("Nothing recv() from client car len data... continuing");
3234 free(filter_expression
);
3237 ret
= LTTNG_ERR_FILTER_INVAL
;
3241 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3242 free(filter_expression
);
3245 ret
= LTTNG_ERR_FILTER_INVAL
;
3250 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3251 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3252 &cmd_ctx
->lsm
->u
.enable
.event
,
3253 filter_expression
, bytecode
, exclusion
,
3254 kernel_poll_pipe
[1]);
3257 case LTTNG_LIST_TRACEPOINTS
:
3259 struct lttng_event
*events
;
3262 session_lock_list();
3263 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3264 session_unlock_list();
3265 if (nb_events
< 0) {
3266 /* Return value is a negative lttng_error_code. */
3272 * Setup lttng message with payload size set to the event list size in
3273 * bytes and then copy list into the llm payload.
3275 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3281 /* Copy event list into message payload */
3282 memcpy(cmd_ctx
->llm
->payload
, events
,
3283 sizeof(struct lttng_event
) * nb_events
);
3290 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3292 struct lttng_event_field
*fields
;
3295 session_lock_list();
3296 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3298 session_unlock_list();
3299 if (nb_fields
< 0) {
3300 /* Return value is a negative lttng_error_code. */
3306 * Setup lttng message with payload size set to the event list size in
3307 * bytes and then copy list into the llm payload.
3309 ret
= setup_lttng_msg(cmd_ctx
,
3310 sizeof(struct lttng_event_field
) * nb_fields
);
3316 /* Copy event list into message payload */
3317 memcpy(cmd_ctx
->llm
->payload
, fields
,
3318 sizeof(struct lttng_event_field
) * nb_fields
);
3325 case LTTNG_LIST_SYSCALLS
:
3327 struct lttng_event
*events
;
3330 nb_events
= cmd_list_syscalls(&events
);
3331 if (nb_events
< 0) {
3332 /* Return value is a negative lttng_error_code. */
3338 * Setup lttng message with payload size set to the event list size in
3339 * bytes and then copy list into the llm payload.
3341 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3347 /* Copy event list into message payload */
3348 memcpy(cmd_ctx
->llm
->payload
, events
,
3349 sizeof(struct lttng_event
) * nb_events
);
3356 case LTTNG_SET_CONSUMER_URI
:
3359 struct lttng_uri
*uris
;
3361 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3362 len
= nb_uri
* sizeof(struct lttng_uri
);
3365 ret
= LTTNG_ERR_INVALID
;
3369 uris
= zmalloc(len
);
3371 ret
= LTTNG_ERR_FATAL
;
3375 /* Receive variable len data */
3376 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3377 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3379 DBG("No URIs received from client... continuing");
3381 ret
= LTTNG_ERR_SESSION_FAIL
;
3386 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3388 if (ret
!= LTTNG_OK
) {
3395 case LTTNG_START_TRACE
:
3397 ret
= cmd_start_trace(cmd_ctx
->session
);
3400 case LTTNG_STOP_TRACE
:
3402 ret
= cmd_stop_trace(cmd_ctx
->session
);
3405 case LTTNG_CREATE_SESSION
:
3408 struct lttng_uri
*uris
= NULL
;
3410 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3411 len
= nb_uri
* sizeof(struct lttng_uri
);
3414 uris
= zmalloc(len
);
3416 ret
= LTTNG_ERR_FATAL
;
3420 /* Receive variable len data */
3421 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3422 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3424 DBG("No URIs received from client... continuing");
3426 ret
= LTTNG_ERR_SESSION_FAIL
;
3431 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3432 DBG("Creating session with ONE network URI is a bad call");
3433 ret
= LTTNG_ERR_SESSION_FAIL
;
3439 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3440 &cmd_ctx
->creds
, 0);
3446 case LTTNG_DESTROY_SESSION
:
3448 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3450 /* Set session to NULL so we do not unlock it after free. */
3451 cmd_ctx
->session
= NULL
;
3454 case LTTNG_LIST_DOMAINS
:
3457 struct lttng_domain
*domains
;
3459 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3461 /* Return value is a negative lttng_error_code. */
3466 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3472 /* Copy event list into message payload */
3473 memcpy(cmd_ctx
->llm
->payload
, domains
,
3474 nb_dom
* sizeof(struct lttng_domain
));
3481 case LTTNG_LIST_CHANNELS
:
3484 struct lttng_channel
*channels
= NULL
;
3486 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3487 cmd_ctx
->session
, &channels
);
3489 /* Return value is a negative lttng_error_code. */
3494 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3500 /* Copy event list into message payload */
3501 memcpy(cmd_ctx
->llm
->payload
, channels
,
3502 nb_chan
* sizeof(struct lttng_channel
));
3509 case LTTNG_LIST_EVENTS
:
3512 struct lttng_event
*events
= NULL
;
3514 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3515 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3517 /* Return value is a negative lttng_error_code. */
3522 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3528 /* Copy event list into message payload */
3529 memcpy(cmd_ctx
->llm
->payload
, events
,
3530 nb_event
* sizeof(struct lttng_event
));
3537 case LTTNG_LIST_SESSIONS
:
3539 unsigned int nr_sessions
;
3541 session_lock_list();
3542 nr_sessions
= lttng_sessions_count(
3543 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3544 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3546 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3548 session_unlock_list();
3552 /* Filled the session array */
3553 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3554 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3555 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3557 session_unlock_list();
3562 case LTTNG_CALIBRATE
:
3564 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3565 &cmd_ctx
->lsm
->u
.calibrate
);
3568 case LTTNG_REGISTER_CONSUMER
:
3570 struct consumer_data
*cdata
;
3572 switch (cmd_ctx
->lsm
->domain
.type
) {
3573 case LTTNG_DOMAIN_KERNEL
:
3574 cdata
= &kconsumer_data
;
3577 ret
= LTTNG_ERR_UND
;
3581 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3582 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3585 case LTTNG_DATA_PENDING
:
3587 ret
= cmd_data_pending(cmd_ctx
->session
);
3590 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3592 struct lttcomm_lttng_output_id reply
;
3594 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3595 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3596 if (ret
!= LTTNG_OK
) {
3600 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3605 /* Copy output list into message payload */
3606 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3610 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3612 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3613 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3616 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3619 struct lttng_snapshot_output
*outputs
= NULL
;
3621 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3622 if (nb_output
< 0) {
3627 ret
= setup_lttng_msg(cmd_ctx
,
3628 nb_output
* sizeof(struct lttng_snapshot_output
));
3635 /* Copy output list into message payload */
3636 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3637 nb_output
* sizeof(struct lttng_snapshot_output
));
3644 case LTTNG_SNAPSHOT_RECORD
:
3646 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3647 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3648 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3651 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3654 struct lttng_uri
*uris
= NULL
;
3656 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3657 len
= nb_uri
* sizeof(struct lttng_uri
);
3660 uris
= zmalloc(len
);
3662 ret
= LTTNG_ERR_FATAL
;
3666 /* Receive variable len data */
3667 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3668 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3670 DBG("No URIs received from client... continuing");
3672 ret
= LTTNG_ERR_SESSION_FAIL
;
3677 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3678 DBG("Creating session with ONE network URI is a bad call");
3679 ret
= LTTNG_ERR_SESSION_FAIL
;
3685 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3686 nb_uri
, &cmd_ctx
->creds
);
3690 case LTTNG_CREATE_SESSION_LIVE
:
3693 struct lttng_uri
*uris
= NULL
;
3695 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3696 len
= nb_uri
* sizeof(struct lttng_uri
);
3699 uris
= zmalloc(len
);
3701 ret
= LTTNG_ERR_FATAL
;
3705 /* Receive variable len data */
3706 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3707 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3709 DBG("No URIs received from client... continuing");
3711 ret
= LTTNG_ERR_SESSION_FAIL
;
3716 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3717 DBG("Creating session with ONE network URI is a bad call");
3718 ret
= LTTNG_ERR_SESSION_FAIL
;
3724 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
3725 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
3729 case LTTNG_SAVE_SESSION
:
3731 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
3736 ret
= LTTNG_ERR_UND
;
3741 if (cmd_ctx
->llm
== NULL
) {
3742 DBG("Missing llm structure. Allocating one.");
3743 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3747 /* Set return code */
3748 cmd_ctx
->llm
->ret_code
= ret
;
3750 if (cmd_ctx
->session
) {
3751 session_unlock(cmd_ctx
->session
);
3753 if (need_tracing_session
) {
3754 session_unlock_list();
3761 * Thread managing health check socket.
3763 static void *thread_manage_health(void *data
)
3765 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3766 uint32_t revents
, nb_fd
;
3767 struct lttng_poll_event events
;
3768 struct health_comm_msg msg
;
3769 struct health_comm_reply reply
;
3771 DBG("[thread] Manage health check started");
3773 rcu_register_thread();
3775 /* We might hit an error path before this is created. */
3776 lttng_poll_init(&events
);
3778 /* Create unix socket */
3779 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3781 ERR("Unable to create health check Unix socket");
3787 /* lttng health client socket path permissions */
3788 ret
= chown(health_unix_sock_path
, 0,
3789 utils_get_group_id(tracing_group_name
));
3791 ERR("Unable to set group on %s", health_unix_sock_path
);
3797 ret
= chmod(health_unix_sock_path
,
3798 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3800 ERR("Unable to set permissions on %s", health_unix_sock_path
);
3808 * Set the CLOEXEC flag. Return code is useless because either way, the
3811 (void) utils_set_fd_cloexec(sock
);
3813 ret
= lttcomm_listen_unix_sock(sock
);
3819 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3820 * more will be added to this poll set.
3822 ret
= sessiond_set_thread_pollset(&events
, 2);
3827 /* Add the application registration socket */
3828 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3833 sessiond_notify_ready();
3836 DBG("Health check ready");
3838 /* Inifinite blocking call, waiting for transmission */
3840 ret
= lttng_poll_wait(&events
, -1);
3843 * Restart interrupted system call.
3845 if (errno
== EINTR
) {
3853 for (i
= 0; i
< nb_fd
; i
++) {
3854 /* Fetch once the poll data */
3855 revents
= LTTNG_POLL_GETEV(&events
, i
);
3856 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3858 /* Thread quit pipe has been closed. Killing thread. */
3859 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3865 /* Event on the registration socket */
3866 if (pollfd
== sock
) {
3867 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3868 ERR("Health socket poll error");
3874 new_sock
= lttcomm_accept_unix_sock(sock
);
3880 * Set the CLOEXEC flag. Return code is useless because either way, the
3883 (void) utils_set_fd_cloexec(new_sock
);
3885 DBG("Receiving data from client for health...");
3886 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3888 DBG("Nothing recv() from client... continuing");
3889 ret
= close(new_sock
);
3897 rcu_thread_online();
3899 memset(&reply
, 0, sizeof(reply
));
3900 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
3902 * health_check_state returns 0 if health is
3905 if (!health_check_state(health_sessiond
, i
)) {
3906 reply
.ret_code
|= 1ULL << i
;
3910 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
3912 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3914 ERR("Failed to send health data back to client");
3917 /* End of transmission */
3918 ret
= close(new_sock
);
3928 ERR("Health error occurred in %s", __func__
);
3930 DBG("Health check thread dying");
3931 unlink(health_unix_sock_path
);
3939 lttng_poll_clean(&events
);
3941 rcu_unregister_thread();
3946 * This thread manage all clients request using the unix client socket for
3949 static void *thread_manage_clients(void *data
)
3951 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3953 uint32_t revents
, nb_fd
;
3954 struct command_ctx
*cmd_ctx
= NULL
;
3955 struct lttng_poll_event events
;
3957 DBG("[thread] Manage client started");
3959 rcu_register_thread();
3961 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
3963 health_code_update();
3965 ret
= lttcomm_listen_unix_sock(client_sock
);
3971 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3972 * more will be added to this poll set.
3974 ret
= sessiond_set_thread_pollset(&events
, 2);
3976 goto error_create_poll
;
3979 /* Add the application registration socket */
3980 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3985 sessiond_notify_ready();
3986 ret
= sem_post(&load_info
->message_thread_ready
);
3988 PERROR("sem_post message_thread_ready");
3992 /* This testpoint is after we signal readiness to the parent. */
3993 if (testpoint(sessiond_thread_manage_clients
)) {
3997 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4001 health_code_update();
4004 DBG("Accepting client command ...");
4006 /* Inifinite blocking call, waiting for transmission */
4008 health_poll_entry();
4009 ret
= lttng_poll_wait(&events
, -1);
4013 * Restart interrupted system call.
4015 if (errno
== EINTR
) {
4023 for (i
= 0; i
< nb_fd
; i
++) {
4024 /* Fetch once the poll data */
4025 revents
= LTTNG_POLL_GETEV(&events
, i
);
4026 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4028 health_code_update();
4030 /* Thread quit pipe has been closed. Killing thread. */
4031 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4037 /* Event on the registration socket */
4038 if (pollfd
== client_sock
) {
4039 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4040 ERR("Client socket poll error");
4046 DBG("Wait for client response");
4048 health_code_update();
4050 sock
= lttcomm_accept_unix_sock(client_sock
);
4056 * Set the CLOEXEC flag. Return code is useless because either way, the
4059 (void) utils_set_fd_cloexec(sock
);
4061 /* Set socket option for credentials retrieval */
4062 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4067 /* Allocate context command to process the client request */
4068 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4069 if (cmd_ctx
== NULL
) {
4070 PERROR("zmalloc cmd_ctx");
4074 /* Allocate data buffer for reception */
4075 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4076 if (cmd_ctx
->lsm
== NULL
) {
4077 PERROR("zmalloc cmd_ctx->lsm");
4081 cmd_ctx
->llm
= NULL
;
4082 cmd_ctx
->session
= NULL
;
4084 health_code_update();
4087 * Data is received from the lttng client. The struct
4088 * lttcomm_session_msg (lsm) contains the command and data request of
4091 DBG("Receiving data from client ...");
4092 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4093 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4095 DBG("Nothing recv() from client... continuing");
4101 clean_command_ctx(&cmd_ctx
);
4105 health_code_update();
4107 // TODO: Validate cmd_ctx including sanity check for
4108 // security purpose.
4110 rcu_thread_online();
4112 * This function dispatch the work to the kernel or userspace tracer
4113 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4114 * informations for the client. The command context struct contains
4115 * everything this function may needs.
4117 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4118 rcu_thread_offline();
4126 * TODO: Inform client somehow of the fatal error. At
4127 * this point, ret < 0 means that a zmalloc failed
4128 * (ENOMEM). Error detected but still accept
4129 * command, unless a socket error has been
4132 clean_command_ctx(&cmd_ctx
);
4136 health_code_update();
4138 DBG("Sending response (size: %d, retcode: %s)",
4139 cmd_ctx
->lttng_msg_size
,
4140 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
4141 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4143 ERR("Failed to send data back to client");
4146 /* End of transmission */
4153 clean_command_ctx(&cmd_ctx
);
4155 health_code_update();
4167 lttng_poll_clean(&events
);
4168 clean_command_ctx(&cmd_ctx
);
4172 unlink(client_unix_sock_path
);
4173 if (client_sock
>= 0) {
4174 ret
= close(client_sock
);
4182 ERR("Health error occurred in %s", __func__
);
4185 health_unregister(health_sessiond
);
4187 DBG("Client thread dying");
4189 rcu_unregister_thread();
4195 * usage function on stderr
4197 static void usage(void)
4199 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4200 fprintf(stderr
, " -h, --help Display this usage.\n");
4201 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4202 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4203 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4204 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4205 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4206 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4207 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4208 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4209 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4210 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4211 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4212 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4213 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4214 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
4215 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4216 fprintf(stderr
, " -V, --version Show version number.\n");
4217 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4218 fprintf(stderr
, " -q, --quiet No output at all.\n");
4219 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4220 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4221 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4222 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4223 fprintf(stderr
, " --agent-tcp-port Agent registration TCP port\n");
4224 fprintf(stderr
, " -f --config PATH Load daemon configuration file\n");
4225 fprintf(stderr
, " -l --load PATH Load session configuration\n");
4226 fprintf(stderr
, " --kmod-probes Specify kernel module probes to load\n");
4227 fprintf(stderr
, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4231 * Take an option from the getopt output and set it in the right variable to be
4234 * Return 0 on success else a negative value.
4236 static int set_option(int opt
, const char *arg
, const char *optname
)
4240 if (arg
&& arg
[0] == '\0') {
4242 * This only happens if the value is read from daemon config
4243 * file. This means the option requires an argument and the
4244 * configuration file contains a line such as:
4253 fprintf(stderr
, "option %s", optname
);
4255 fprintf(stderr
, " with arg %s\n", arg
);
4259 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4262 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4272 * If the override option is set, the pointer points to a
4273 * *non* const thus freeing it even though the variable type is
4276 if (tracing_group_name_override
) {
4277 free((void *) tracing_group_name
);
4279 tracing_group_name
= strdup(arg
);
4280 if (!tracing_group_name
) {
4284 tracing_group_name_override
= 1;
4290 fprintf(stdout
, "%s\n", VERSION
);
4296 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4299 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4302 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4305 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4308 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4311 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4317 lttng_opt_quiet
= 1;
4320 /* Verbose level can increase using multiple -v */
4322 /* Value obtained from config file */
4323 lttng_opt_verbose
= config_parse_value(arg
);
4325 /* -v used on command line */
4326 lttng_opt_verbose
++;
4328 /* Clamp value to [0, 3] */
4329 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4330 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4334 opt_verbose_consumer
= config_parse_value(arg
);
4336 opt_verbose_consumer
+= 1;
4340 if (consumerd32_bin_override
) {
4341 free((void *) consumerd32_bin
);
4343 consumerd32_bin
= strdup(arg
);
4344 if (!consumerd32_bin
) {
4348 consumerd32_bin_override
= 1;
4351 if (consumerd32_libdir_override
) {
4352 free((void *) consumerd32_libdir
);
4354 consumerd32_libdir
= strdup(arg
);
4355 if (!consumerd32_libdir
) {
4359 consumerd32_libdir_override
= 1;
4362 if (consumerd64_bin_override
) {
4363 free((void *) consumerd64_bin
);
4365 consumerd64_bin
= strdup(arg
);
4366 if (!consumerd64_bin
) {
4370 consumerd64_bin_override
= 1;
4373 if (consumerd64_libdir_override
) {
4374 free((void *) consumerd64_libdir
);
4376 consumerd64_libdir
= strdup(arg
);
4377 if (!consumerd64_libdir
) {
4381 consumerd64_libdir_override
= 1;
4385 opt_pidfile
= strdup(arg
);
4391 case 'J': /* Agent TCP port. */
4396 v
= strtoul(arg
, NULL
, 0);
4397 if (errno
!= 0 || !isdigit(arg
[0])) {
4398 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4401 if (v
== 0 || v
>= 65535) {
4402 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4405 agent_tcp_port
= (uint32_t) v
;
4406 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4410 free(opt_load_session_path
);
4411 opt_load_session_path
= strdup(arg
);
4412 if (!opt_load_session_path
) {
4417 case 'P': /* probe modules list */
4418 free(kmod_probes_list
);
4419 kmod_probes_list
= strdup(arg
);
4420 if (!kmod_probes_list
) {
4426 free(kmod_extra_probes_list
);
4427 kmod_extra_probes_list
= strdup(arg
);
4428 if (!kmod_extra_probes_list
) {
4434 /* This is handled in set_options() thus silent break. */
4437 /* Unknown option or other error.
4438 * Error is printed by getopt, just return */
4443 if (ret
== -EINVAL
) {
4444 const char *opt_name
= "unknown";
4447 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
4449 if (opt
== long_options
[i
].val
) {
4450 opt_name
= long_options
[i
].name
;
4455 WARN("Invalid argument provided for option \"%s\", using default value.",
4463 * config_entry_handler_cb used to handle options read from a config file.
4464 * See config_entry_handler_cb comment in common/config/config.h for the
4465 * return value conventions.
4467 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4471 if (!entry
|| !entry
->name
|| !entry
->value
) {
4476 /* Check if the option is to be ignored */
4477 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4478 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4483 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4486 /* Ignore if not fully matched. */
4487 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4492 * If the option takes no argument on the command line, we have to
4493 * check if the value is "true". We support non-zero numeric values,
4496 if (!long_options
[i
].has_arg
) {
4497 ret
= config_parse_value(entry
->value
);
4500 WARN("Invalid configuration value \"%s\" for option %s",
4501 entry
->value
, entry
->name
);
4503 /* False, skip boolean config option. */
4508 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4512 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4519 * daemon configuration loading and argument parsing
4521 static int set_options(int argc
, char **argv
)
4523 int ret
= 0, c
= 0, option_index
= 0;
4524 int orig_optopt
= optopt
, orig_optind
= optind
;
4526 const char *config_path
= NULL
;
4528 optstring
= utils_generate_optstring(long_options
,
4529 sizeof(long_options
) / sizeof(struct option
));
4535 /* Check for the --config option */
4536 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
4537 &option_index
)) != -1) {
4541 } else if (c
!= 'f') {
4542 /* if not equal to --config option. */
4546 config_path
= utils_expand_path(optarg
);
4548 ERR("Failed to resolve path: %s", optarg
);
4552 ret
= config_get_section_entries(config_path
, config_section_name
,
4553 config_entry_handler
, NULL
);
4556 ERR("Invalid configuration option at line %i", ret
);
4562 /* Reset getopt's global state */
4563 optopt
= orig_optopt
;
4564 optind
= orig_optind
;
4566 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
4571 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
4583 * Creates the two needed socket by the daemon.
4584 * apps_sock - The communication socket for all UST apps.
4585 * client_sock - The communication of the cli tool (lttng).
4587 static int init_daemon_socket(void)
4592 old_umask
= umask(0);
4594 /* Create client tool unix socket */
4595 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4596 if (client_sock
< 0) {
4597 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4602 /* Set the cloexec flag */
4603 ret
= utils_set_fd_cloexec(client_sock
);
4605 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4606 "Continuing but note that the consumer daemon will have a "
4607 "reference to this socket on exec()", client_sock
);
4610 /* File permission MUST be 660 */
4611 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4613 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4618 /* Create the application unix socket */
4619 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4620 if (apps_sock
< 0) {
4621 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4626 /* Set the cloexec flag */
4627 ret
= utils_set_fd_cloexec(apps_sock
);
4629 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4630 "Continuing but note that the consumer daemon will have a "
4631 "reference to this socket on exec()", apps_sock
);
4634 /* File permission MUST be 666 */
4635 ret
= chmod(apps_unix_sock_path
,
4636 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4638 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4643 DBG3("Session daemon client socket %d and application socket %d created",
4644 client_sock
, apps_sock
);
4652 * Check if the global socket is available, and if a daemon is answering at the
4653 * other side. If yes, error is returned.
4655 static int check_existing_daemon(void)
4657 /* Is there anybody out there ? */
4658 if (lttng_session_daemon_alive()) {
4666 * Set the tracing group gid onto the client socket.
4668 * Race window between mkdir and chown is OK because we are going from more
4669 * permissive (root.root) to less permissive (root.tracing).
4671 static int set_permissions(char *rundir
)
4676 gid
= utils_get_group_id(tracing_group_name
);
4678 /* Set lttng run dir */
4679 ret
= chown(rundir
, 0, gid
);
4681 ERR("Unable to set group on %s", rundir
);
4686 * Ensure all applications and tracing group can search the run
4687 * dir. Allow everyone to read the directory, since it does not
4688 * buy us anything to hide its content.
4690 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
4692 ERR("Unable to set permissions on %s", rundir
);
4696 /* lttng client socket path */
4697 ret
= chown(client_unix_sock_path
, 0, gid
);
4699 ERR("Unable to set group on %s", client_unix_sock_path
);
4703 /* kconsumer error socket path */
4704 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
4706 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4710 /* 64-bit ustconsumer error socket path */
4711 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
4713 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4717 /* 32-bit ustconsumer compat32 error socket path */
4718 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
4720 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4724 DBG("All permissions are set");
4730 * Create the lttng run directory needed for all global sockets and pipe.
4732 static int create_lttng_rundir(const char *rundir
)
4736 DBG3("Creating LTTng run directory: %s", rundir
);
4738 ret
= mkdir(rundir
, S_IRWXU
);
4740 if (errno
!= EEXIST
) {
4741 ERR("Unable to create %s", rundir
);
4753 * Setup sockets and directory needed by the kconsumerd communication with the
4756 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4760 char path
[PATH_MAX
];
4762 switch (consumer_data
->type
) {
4763 case LTTNG_CONSUMER_KERNEL
:
4764 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4766 case LTTNG_CONSUMER64_UST
:
4767 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4769 case LTTNG_CONSUMER32_UST
:
4770 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4773 ERR("Consumer type unknown");
4778 DBG2("Creating consumer directory: %s", path
);
4780 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
4782 if (errno
!= EEXIST
) {
4784 ERR("Failed to create %s", path
);
4790 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
4792 ERR("Unable to set group on %s", path
);
4798 /* Create the kconsumerd error unix socket */
4799 consumer_data
->err_sock
=
4800 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4801 if (consumer_data
->err_sock
< 0) {
4802 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4808 * Set the CLOEXEC flag. Return code is useless because either way, the
4811 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4813 PERROR("utils_set_fd_cloexec");
4814 /* continue anyway */
4817 /* File permission MUST be 660 */
4818 ret
= chmod(consumer_data
->err_unix_sock_path
,
4819 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4821 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4831 * Signal handler for the daemon
4833 * Simply stop all worker threads, leaving main() return gracefully after
4834 * joining all threads and calling cleanup().
4836 static void sighandler(int sig
)
4840 DBG("SIGPIPE caught");
4843 DBG("SIGINT caught");
4847 DBG("SIGTERM caught");
4851 CMM_STORE_SHARED(recv_child_signal
, 1);
4859 * Setup signal handler for :
4860 * SIGINT, SIGTERM, SIGPIPE
4862 static int set_signal_handler(void)
4865 struct sigaction sa
;
4868 if ((ret
= sigemptyset(&sigset
)) < 0) {
4869 PERROR("sigemptyset");
4873 sa
.sa_handler
= sighandler
;
4874 sa
.sa_mask
= sigset
;
4876 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4877 PERROR("sigaction");
4881 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4882 PERROR("sigaction");
4886 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4887 PERROR("sigaction");
4891 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
4892 PERROR("sigaction");
4896 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4902 * Set open files limit to unlimited. This daemon can open a large number of
4903 * file descriptors in order to consumer multiple kernel traces.
4905 static void set_ulimit(void)
4910 /* The kernel does not allowed an infinite limit for open files */
4911 lim
.rlim_cur
= 65535;
4912 lim
.rlim_max
= 65535;
4914 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4916 PERROR("failed to set open files limit");
4921 * Write pidfile using the rundir and opt_pidfile.
4923 static void write_pidfile(void)
4926 char pidfile_path
[PATH_MAX
];
4931 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4933 /* Build pidfile path from rundir and opt_pidfile. */
4934 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4935 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4937 PERROR("snprintf pidfile path");
4943 * Create pid file in rundir. Return value is of no importance. The
4944 * execution will continue even though we are not able to write the file.
4946 (void) utils_create_pid_file(getpid(), pidfile_path
);
4953 * Create lockfile using the rundir and return its fd.
4955 static int create_lockfile(void)
4958 char lockfile_path
[PATH_MAX
];
4960 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
4965 ret
= utils_create_lock_file(lockfile_path
);
4971 * Write agent TCP port using the rundir.
4973 static void write_agent_port(void)
4976 char path
[PATH_MAX
];
4980 ret
= snprintf(path
, sizeof(path
), "%s/"
4981 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
4983 PERROR("snprintf agent port path");
4988 * Create TCP agent port file in rundir. Return value is of no importance.
4989 * The execution will continue even though we are not able to write the
4992 (void) utils_create_pid_file(agent_tcp_port
, path
);
4999 * Start the load session thread and dettach from it so the main thread can
5000 * continue. This does not return a value since whatever the outcome, the main
5001 * thread will continue.
5003 static void start_load_session_thread(void)
5007 /* Create session loading thread. */
5008 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
5011 PERROR("pthread_create load_session_thread");
5015 ret
= pthread_detach(load_session_thread
);
5017 PERROR("pthread_detach load_session_thread");
5020 /* Everything went well so don't cleanup anything. */
5023 /* The cleanup() function will destroy the load_info data. */
5030 int main(int argc
, char **argv
)
5034 const char *home_path
, *env_app_timeout
;
5036 init_kernel_workarounds();
5038 rcu_register_thread();
5040 if ((ret
= set_signal_handler()) < 0) {
5044 setup_consumerd_path();
5046 page_size
= sysconf(_SC_PAGESIZE
);
5047 if (page_size
< 0) {
5048 PERROR("sysconf _SC_PAGESIZE");
5049 page_size
= LONG_MAX
;
5050 WARN("Fallback page size to %ld", page_size
);
5053 /* Parse arguments and load the daemon configuration file */
5055 if ((ret
= set_options(argc
, argv
)) < 0) {
5060 if (opt_daemon
|| opt_background
) {
5063 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5070 * We are in the child. Make sure all other file descriptors are
5071 * closed, in case we are called with more opened file descriptors than
5072 * the standard ones.
5074 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5079 /* Create thread quit pipe */
5080 if ((ret
= init_thread_quit_pipe()) < 0) {
5084 /* Check if daemon is UID = 0 */
5085 is_root
= !getuid();
5088 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5094 /* Create global run dir with root access */
5095 ret
= create_lttng_rundir(rundir
);
5100 if (strlen(apps_unix_sock_path
) == 0) {
5101 snprintf(apps_unix_sock_path
, PATH_MAX
,
5102 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5105 if (strlen(client_unix_sock_path
) == 0) {
5106 snprintf(client_unix_sock_path
, PATH_MAX
,
5107 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5110 /* Set global SHM for ust */
5111 if (strlen(wait_shm_path
) == 0) {
5112 snprintf(wait_shm_path
, PATH_MAX
,
5113 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5116 if (strlen(health_unix_sock_path
) == 0) {
5117 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
5118 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5121 /* Setup kernel consumerd path */
5122 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5123 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5124 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5125 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5127 DBG2("Kernel consumer err path: %s",
5128 kconsumer_data
.err_unix_sock_path
);
5129 DBG2("Kernel consumer cmd path: %s",
5130 kconsumer_data
.cmd_unix_sock_path
);
5132 home_path
= utils_get_home_dir();
5133 if (home_path
== NULL
) {
5134 /* TODO: Add --socket PATH option */
5135 ERR("Can't get HOME directory for sockets creation.");
5141 * Create rundir from home path. This will create something like
5144 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5150 ret
= create_lttng_rundir(rundir
);
5155 if (strlen(apps_unix_sock_path
) == 0) {
5156 snprintf(apps_unix_sock_path
, PATH_MAX
,
5157 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
5160 /* Set the cli tool unix socket path */
5161 if (strlen(client_unix_sock_path
) == 0) {
5162 snprintf(client_unix_sock_path
, PATH_MAX
,
5163 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
5166 /* Set global SHM for ust */
5167 if (strlen(wait_shm_path
) == 0) {
5168 snprintf(wait_shm_path
, PATH_MAX
,
5169 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
5172 /* Set health check Unix path */
5173 if (strlen(health_unix_sock_path
) == 0) {
5174 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
5175 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
5179 lockfile_fd
= create_lockfile();
5180 if (lockfile_fd
< 0) {
5184 /* Set consumer initial state */
5185 kernel_consumerd_state
= CONSUMER_STOPPED
;
5186 ust_consumerd_state
= CONSUMER_STOPPED
;
5188 DBG("Client socket path %s", client_unix_sock_path
);
5189 DBG("Application socket path %s", apps_unix_sock_path
);
5190 DBG("Application wait path %s", wait_shm_path
);
5191 DBG("LTTng run directory path: %s", rundir
);
5193 /* 32 bits consumerd path setup */
5194 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5195 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5196 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5197 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5199 DBG2("UST consumer 32 bits err path: %s",
5200 ustconsumer32_data
.err_unix_sock_path
);
5201 DBG2("UST consumer 32 bits cmd path: %s",
5202 ustconsumer32_data
.cmd_unix_sock_path
);
5204 /* 64 bits consumerd path setup */
5205 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5206 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5207 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5208 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5210 DBG2("UST consumer 64 bits err path: %s",
5211 ustconsumer64_data
.err_unix_sock_path
);
5212 DBG2("UST consumer 64 bits cmd path: %s",
5213 ustconsumer64_data
.cmd_unix_sock_path
);
5216 * See if daemon already exist.
5218 if ((ret
= check_existing_daemon()) < 0) {
5219 ERR("Already running daemon.\n");
5221 * We do not goto exit because we must not cleanup()
5222 * because a daemon is already running.
5228 * Init UST app hash table. Alloc hash table before this point since
5229 * cleanup() can get called after that point.
5233 /* Initialize agent domain subsystem. */
5234 if ((ret
= agent_setup()) < 0) {
5235 /* ENOMEM at this point. */
5239 /* After this point, we can safely call cleanup() with "goto exit" */
5242 * These actions must be executed as root. We do that *after* setting up
5243 * the sockets path because we MUST make the check for another daemon using
5244 * those paths *before* trying to set the kernel consumer sockets and init
5248 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
5253 /* Setup kernel tracer */
5254 if (!opt_no_kernel
) {
5255 init_kernel_tracer();
5256 if (kernel_tracer_fd
>= 0) {
5257 ret
= syscall_init_table();
5259 ERR("Unable to populate syscall table. Syscall tracing"
5260 " won't work for this session daemon.");
5265 /* Set ulimit for open files */
5268 /* init lttng_fd tracking must be done after set_ulimit. */
5271 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
5276 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
5281 /* Setup the needed unix socket */
5282 if ((ret
= init_daemon_socket()) < 0) {
5286 /* Set credentials to socket */
5287 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
5291 /* Get parent pid if -S, --sig-parent is specified. */
5292 if (opt_sig_parent
) {
5296 /* Setup the kernel pipe for waking up the kernel thread */
5297 if (is_root
&& !opt_no_kernel
) {
5298 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
5303 /* Setup the thread ht_cleanup communication pipe. */
5304 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
5308 /* Setup the thread apps communication pipe. */
5309 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
5313 /* Setup the thread apps notify communication pipe. */
5314 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
5318 /* Initialize global buffer per UID and PID registry. */
5319 buffer_reg_init_uid_registry();
5320 buffer_reg_init_pid_registry();
5322 /* Init UST command queue. */
5323 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
5326 * Get session list pointer. This pointer MUST NOT be free(). This list is
5327 * statically declared in session.c
5329 session_list_ptr
= session_get_list();
5331 /* Set up max poll set size */
5332 lttng_poll_set_max_size();
5336 /* Check for the application socket timeout env variable. */
5337 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5338 if (env_app_timeout
) {
5339 app_socket_timeout
= atoi(env_app_timeout
);
5341 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5347 /* Initialize communication library */
5349 /* This is to get the TCP timeout value. */
5350 lttcomm_inet_init();
5352 if (load_session_init_data(&load_info
) < 0) {
5355 load_info
->path
= opt_load_session_path
;
5358 * Initialize the health check subsystem. This call should set the
5359 * appropriate time values.
5361 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5362 if (!health_sessiond
) {
5363 PERROR("health_app_create error");
5364 goto exit_health_sessiond_cleanup
;
5367 /* Create thread to clean up RCU hash tables */
5368 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5369 thread_ht_cleanup
, (void *) NULL
);
5371 PERROR("pthread_create ht_cleanup");
5372 goto exit_ht_cleanup
;
5375 /* Create health-check thread */
5376 ret
= pthread_create(&health_thread
, NULL
,
5377 thread_manage_health
, (void *) NULL
);
5379 PERROR("pthread_create health");
5383 /* Create thread to manage the client socket */
5384 ret
= pthread_create(&client_thread
, NULL
,
5385 thread_manage_clients
, (void *) NULL
);
5387 PERROR("pthread_create clients");
5391 /* Create thread to dispatch registration */
5392 ret
= pthread_create(&dispatch_thread
, NULL
,
5393 thread_dispatch_ust_registration
, (void *) NULL
);
5395 PERROR("pthread_create dispatch");
5399 /* Create thread to manage application registration. */
5400 ret
= pthread_create(®_apps_thread
, NULL
,
5401 thread_registration_apps
, (void *) NULL
);
5403 PERROR("pthread_create registration");
5407 /* Create thread to manage application socket */
5408 ret
= pthread_create(&apps_thread
, NULL
,
5409 thread_manage_apps
, (void *) NULL
);
5411 PERROR("pthread_create apps");
5415 /* Create thread to manage application notify socket */
5416 ret
= pthread_create(&apps_notify_thread
, NULL
,
5417 ust_thread_manage_notify
, (void *) NULL
);
5419 PERROR("pthread_create notify");
5420 goto exit_apps_notify
;
5423 /* Create agent registration thread. */
5424 ret
= pthread_create(&agent_reg_thread
, NULL
,
5425 agent_thread_manage_registration
, (void *) NULL
);
5427 PERROR("pthread_create agent");
5428 goto exit_agent_reg
;
5431 /* Don't start this thread if kernel tracing is not requested nor root */
5432 if (is_root
&& !opt_no_kernel
) {
5433 /* Create kernel thread to manage kernel event */
5434 ret
= pthread_create(&kernel_thread
, NULL
,
5435 thread_manage_kernel
, (void *) NULL
);
5437 PERROR("pthread_create kernel");
5442 /* Load possible session(s). */
5443 start_load_session_thread();
5445 if (is_root
&& !opt_no_kernel
) {
5446 ret
= pthread_join(kernel_thread
, &status
);
5448 PERROR("pthread_join");
5449 goto error
; /* join error, exit without cleanup */
5454 ret
= pthread_join(agent_reg_thread
, &status
);
5456 PERROR("pthread_join agent");
5457 goto error
; /* join error, exit without cleanup */
5461 ret
= pthread_join(apps_notify_thread
, &status
);
5463 PERROR("pthread_join apps notify");
5464 goto error
; /* join error, exit without cleanup */
5468 ret
= pthread_join(apps_thread
, &status
);
5470 PERROR("pthread_join apps");
5471 goto error
; /* join error, exit without cleanup */
5476 ret
= pthread_join(reg_apps_thread
, &status
);
5478 PERROR("pthread_join");
5479 goto error
; /* join error, exit without cleanup */
5483 ret
= pthread_join(dispatch_thread
, &status
);
5485 PERROR("pthread_join");
5486 goto error
; /* join error, exit without cleanup */
5490 ret
= pthread_join(client_thread
, &status
);
5492 PERROR("pthread_join");
5493 goto error
; /* join error, exit without cleanup */
5496 ret
= join_consumer_thread(&kconsumer_data
);
5498 PERROR("join_consumer");
5499 goto error
; /* join error, exit without cleanup */
5502 ret
= join_consumer_thread(&ustconsumer32_data
);
5504 PERROR("join_consumer ust32");
5505 goto error
; /* join error, exit without cleanup */
5508 ret
= join_consumer_thread(&ustconsumer64_data
);
5510 PERROR("join_consumer ust64");
5511 goto error
; /* join error, exit without cleanup */
5515 ret
= pthread_join(health_thread
, &status
);
5517 PERROR("pthread_join health thread");
5518 goto error
; /* join error, exit without cleanup */
5522 ret
= pthread_join(ht_cleanup_thread
, &status
);
5524 PERROR("pthread_join ht cleanup thread");
5525 goto error
; /* join error, exit without cleanup */
5528 health_app_destroy(health_sessiond
);
5529 exit_health_sessiond_cleanup
:
5532 * cleanup() is called when no other thread is running.
5534 rcu_thread_online();
5536 rcu_thread_offline();
5537 rcu_unregister_thread();