2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
36 #include <sys/types.h>
38 #include <urcu/uatomic.h>
41 #include <common/common.h>
42 #include <common/compat/socket.h>
43 #include <common/compat/getenv.h>
44 #include <common/defaults.h>
45 #include <common/kernel-consumer/kernel-consumer.h>
46 #include <common/futex.h>
47 #include <common/relayd/relayd.h>
48 #include <common/utils.h>
49 #include <common/daemonize.h>
50 #include <common/config/session-config.h>
52 #include "lttng-sessiond.h"
53 #include "buffer-registry.h"
60 #include "kernel-consumer.h"
64 #include "ust-consumer.h"
67 #include "health-sessiond.h"
68 #include "testpoint.h"
69 #include "ust-thread.h"
70 #include "agent-thread.h"
72 #include "load-session-thread.h"
75 #include "ht-cleanup.h"
77 #define CONSUMERD_FILE "lttng-consumerd"
80 static const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
81 static int tracing_group_name_override
;
82 static char *opt_pidfile
;
83 static int opt_sig_parent
;
84 static int opt_verbose_consumer
;
85 static int opt_daemon
, opt_background
;
86 static int opt_no_kernel
;
87 static char *opt_load_session_path
;
88 static pid_t ppid
; /* Parent PID for --sig-parent option */
89 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
91 static int lockfile_fd
= -1;
93 /* Set to 1 when a SIGUSR1 signal is received. */
94 static int recv_child_signal
;
97 * Consumer daemon specific control data. Every value not initialized here is
98 * set to 0 by the static definition.
100 static struct consumer_data kconsumer_data
= {
101 .type
= LTTNG_CONSUMER_KERNEL
,
102 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
103 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
106 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
107 .lock
= PTHREAD_MUTEX_INITIALIZER
,
108 .cond
= PTHREAD_COND_INITIALIZER
,
109 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
111 static struct consumer_data ustconsumer64_data
= {
112 .type
= LTTNG_CONSUMER64_UST
,
113 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
114 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
117 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
118 .lock
= PTHREAD_MUTEX_INITIALIZER
,
119 .cond
= PTHREAD_COND_INITIALIZER
,
120 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
122 static struct consumer_data ustconsumer32_data
= {
123 .type
= LTTNG_CONSUMER32_UST
,
124 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
125 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
128 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
129 .lock
= PTHREAD_MUTEX_INITIALIZER
,
130 .cond
= PTHREAD_COND_INITIALIZER
,
131 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
134 /* Command line options */
135 static const struct option long_options
[] = {
136 { "client-sock", required_argument
, 0, 'c' },
137 { "apps-sock", required_argument
, 0, 'a' },
138 { "kconsumerd-cmd-sock", required_argument
, 0, '\0' },
139 { "kconsumerd-err-sock", required_argument
, 0, '\0' },
140 { "ustconsumerd32-cmd-sock", required_argument
, 0, '\0' },
141 { "ustconsumerd32-err-sock", required_argument
, 0, '\0' },
142 { "ustconsumerd64-cmd-sock", required_argument
, 0, '\0' },
143 { "ustconsumerd64-err-sock", required_argument
, 0, '\0' },
144 { "consumerd32-path", required_argument
, 0, '\0' },
145 { "consumerd32-libdir", required_argument
, 0, '\0' },
146 { "consumerd64-path", required_argument
, 0, '\0' },
147 { "consumerd64-libdir", required_argument
, 0, '\0' },
148 { "daemonize", no_argument
, 0, 'd' },
149 { "background", no_argument
, 0, 'b' },
150 { "sig-parent", no_argument
, 0, 'S' },
151 { "help", no_argument
, 0, 'h' },
152 { "group", required_argument
, 0, 'g' },
153 { "version", no_argument
, 0, 'V' },
154 { "quiet", no_argument
, 0, 'q' },
155 { "verbose", no_argument
, 0, 'v' },
156 { "verbose-consumer", no_argument
, 0, '\0' },
157 { "no-kernel", no_argument
, 0, '\0' },
158 { "pidfile", required_argument
, 0, 'p' },
159 { "agent-tcp-port", required_argument
, 0, '\0' },
160 { "config", required_argument
, 0, 'f' },
161 { "load", required_argument
, 0, 'l' },
162 { "kmod-probes", required_argument
, 0, '\0' },
163 { "extra-kmod-probes", required_argument
, 0, '\0' },
167 /* Command line options to ignore from configuration file */
168 static const char *config_ignore_options
[] = { "help", "version", "config" };
170 /* Shared between threads */
171 static int dispatch_thread_exit
;
173 /* Global application Unix socket path */
174 static char apps_unix_sock_path
[PATH_MAX
];
175 /* Global client Unix socket path */
176 static char client_unix_sock_path
[PATH_MAX
];
177 /* global wait shm path for UST */
178 static char wait_shm_path
[PATH_MAX
];
179 /* Global health check unix path */
180 static char health_unix_sock_path
[PATH_MAX
];
182 /* Sockets and FDs */
183 static int client_sock
= -1;
184 static int apps_sock
= -1;
185 int kernel_tracer_fd
= -1;
186 static int kernel_poll_pipe
[2] = { -1, -1 };
189 * Quit pipe for all threads. This permits a single cancellation point
190 * for all threads when receiving an event on the pipe.
192 static int thread_quit_pipe
[2] = { -1, -1 };
195 * This pipe is used to inform the thread managing application communication
196 * that a command is queued and ready to be processed.
198 static int apps_cmd_pipe
[2] = { -1, -1 };
200 int apps_cmd_notify_pipe
[2] = { -1, -1 };
202 /* Pthread, Mutexes and Semaphores */
203 static pthread_t apps_thread
;
204 static pthread_t apps_notify_thread
;
205 static pthread_t reg_apps_thread
;
206 static pthread_t client_thread
;
207 static pthread_t kernel_thread
;
208 static pthread_t dispatch_thread
;
209 static pthread_t health_thread
;
210 static pthread_t ht_cleanup_thread
;
211 static pthread_t agent_reg_thread
;
212 static pthread_t load_session_thread
;
215 * UST registration command queue. This queue is tied with a futex and uses a N
216 * wakers / 1 waiter implemented and detailed in futex.c/.h
218 * The thread_registration_apps and thread_dispatch_ust_registration uses this
219 * queue along with the wait/wake scheme. The thread_manage_apps receives down
220 * the line new application socket and monitors it for any I/O error or clean
221 * close that triggers an unregistration of the application.
223 static struct ust_cmd_queue ust_cmd_queue
;
226 * Pointer initialized before thread creation.
228 * This points to the tracing session list containing the session count and a
229 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
230 * MUST NOT be taken if you call a public function in session.c.
232 * The lock is nested inside the structure: session_list_ptr->lock. Please use
233 * session_lock_list and session_unlock_list for lock acquisition.
235 static struct ltt_session_list
*session_list_ptr
;
237 int ust_consumerd64_fd
= -1;
238 int ust_consumerd32_fd
= -1;
240 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
241 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
242 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
243 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
244 static int consumerd32_bin_override
;
245 static int consumerd64_bin_override
;
246 static int consumerd32_libdir_override
;
247 static int consumerd64_libdir_override
;
249 static const char *module_proc_lttng
= "/proc/lttng";
252 * Consumer daemon state which is changed when spawning it, killing it or in
253 * case of a fatal error.
255 enum consumerd_state
{
256 CONSUMER_STARTED
= 1,
257 CONSUMER_STOPPED
= 2,
262 * This consumer daemon state is used to validate if a client command will be
263 * able to reach the consumer. If not, the client is informed. For instance,
264 * doing a "lttng start" when the consumer state is set to ERROR will return an
265 * error to the client.
267 * The following example shows a possible race condition of this scheme:
269 * consumer thread error happens
271 * client cmd checks state -> still OK
272 * consumer thread exit, sets error
273 * client cmd try to talk to consumer
276 * However, since the consumer is a different daemon, we have no way of making
277 * sure the command will reach it safely even with this state flag. This is why
278 * we consider that up to the state validation during command processing, the
279 * command is safe. After that, we can not guarantee the correctness of the
280 * client request vis-a-vis the consumer.
282 static enum consumerd_state ust_consumerd_state
;
283 static enum consumerd_state kernel_consumerd_state
;
286 * Socket timeout for receiving and sending in seconds.
288 static int app_socket_timeout
;
290 /* Set in main() with the current page size. */
293 /* Application health monitoring */
294 struct health_app
*health_sessiond
;
296 /* Agent TCP port for registration. Used by the agent thread. */
297 unsigned int agent_tcp_port
= DEFAULT_AGENT_TCP_PORT
;
299 /* Am I root or not. */
300 int is_root
; /* Set to 1 if the daemon is running as root */
302 const char * const config_section_name
= "sessiond";
304 /* Load session thread information to operate. */
305 struct load_session_thread_data
*load_info
;
307 /* Global hash tables */
308 struct lttng_ht
*agent_apps_ht_by_sock
= NULL
;
311 * Whether sessiond is ready for commands/health check requests.
312 * NR_LTTNG_SESSIOND_READY must match the number of calls to
313 * sessiond_notify_ready().
315 #define NR_LTTNG_SESSIOND_READY 3
316 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
318 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
320 return (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) ? 1 : 0;
323 /* Notify parents that we are ready for cmd and health check */
325 void sessiond_notify_ready(void)
327 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
329 * Notify parent pid that we are ready to accept command
330 * for client side. This ppid is the one from the
331 * external process that spawned us.
333 if (opt_sig_parent
) {
338 * Notify the parent of the fork() process that we are
341 if (opt_daemon
|| opt_background
) {
342 kill(child_ppid
, SIGUSR1
);
348 void setup_consumerd_path(void)
350 const char *bin
, *libdir
;
353 * Allow INSTALL_BIN_PATH to be used as a target path for the
354 * native architecture size consumer if CONFIG_CONSUMER*_PATH
355 * has not been defined.
357 #if (CAA_BITS_PER_LONG == 32)
358 if (!consumerd32_bin
[0]) {
359 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
361 if (!consumerd32_libdir
[0]) {
362 consumerd32_libdir
= INSTALL_LIB_PATH
;
364 #elif (CAA_BITS_PER_LONG == 64)
365 if (!consumerd64_bin
[0]) {
366 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
368 if (!consumerd64_libdir
[0]) {
369 consumerd64_libdir
= INSTALL_LIB_PATH
;
372 #error "Unknown bitness"
376 * runtime env. var. overrides the build default.
378 bin
= lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
380 consumerd32_bin
= bin
;
382 bin
= lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
384 consumerd64_bin
= bin
;
386 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
388 consumerd32_libdir
= libdir
;
390 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
392 consumerd64_libdir
= libdir
;
397 int __sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
,
404 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
410 ret
= lttng_poll_add(events
, a_pipe
[0], LPOLLIN
| LPOLLERR
);
422 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
424 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
426 return __sessiond_set_thread_pollset(events
, size
, thread_quit_pipe
);
430 * Init thread quit pipe.
432 * Return -1 on error or 0 if all pipes are created.
434 static int __init_thread_quit_pipe(int *a_pipe
)
440 PERROR("thread quit pipe");
444 for (i
= 0; i
< 2; i
++) {
445 ret
= fcntl(a_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
456 static int init_thread_quit_pipe(void)
458 return __init_thread_quit_pipe(thread_quit_pipe
);
462 * Stop all threads by closing the thread quit pipe.
464 static void stop_threads(void)
468 /* Stopping all threads */
469 DBG("Terminating all threads");
470 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
472 ERR("write error on thread quit pipe");
475 /* Dispatch thread */
476 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
477 futex_nto1_wake(&ust_cmd_queue
.futex
);
481 * Close every consumer sockets.
483 static void close_consumer_sockets(void)
487 if (kconsumer_data
.err_sock
>= 0) {
488 ret
= close(kconsumer_data
.err_sock
);
490 PERROR("kernel consumer err_sock close");
493 if (ustconsumer32_data
.err_sock
>= 0) {
494 ret
= close(ustconsumer32_data
.err_sock
);
496 PERROR("UST consumerd32 err_sock close");
499 if (ustconsumer64_data
.err_sock
>= 0) {
500 ret
= close(ustconsumer64_data
.err_sock
);
502 PERROR("UST consumerd64 err_sock close");
505 if (kconsumer_data
.cmd_sock
>= 0) {
506 ret
= close(kconsumer_data
.cmd_sock
);
508 PERROR("kernel consumer cmd_sock close");
511 if (ustconsumer32_data
.cmd_sock
>= 0) {
512 ret
= close(ustconsumer32_data
.cmd_sock
);
514 PERROR("UST consumerd32 cmd_sock close");
517 if (ustconsumer64_data
.cmd_sock
>= 0) {
518 ret
= close(ustconsumer64_data
.cmd_sock
);
520 PERROR("UST consumerd64 cmd_sock close");
526 * Generate the full lock file path using the rundir.
528 * Return the snprintf() return value thus a negative value is an error.
530 static int generate_lock_file_path(char *path
, size_t len
)
537 /* Build lockfile path from rundir. */
538 ret
= snprintf(path
, len
, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE
, rundir
);
540 PERROR("snprintf lockfile path");
547 * Wait on consumer process termination.
549 * Need to be called with the consumer data lock held or from a context
550 * ensuring no concurrent access to data (e.g: cleanup).
552 static void wait_consumer(struct consumer_data
*consumer_data
)
557 if (consumer_data
->pid
<= 0) {
561 DBG("Waiting for complete teardown of consumerd (PID: %d)",
563 ret
= waitpid(consumer_data
->pid
, &status
, 0);
565 PERROR("consumerd waitpid pid: %d", consumer_data
->pid
)
567 if (!WIFEXITED(status
)) {
568 ERR("consumerd termination with error: %d",
571 consumer_data
->pid
= 0;
575 * Cleanup the session daemon's data structures.
577 static void sessiond_cleanup(void)
580 struct ltt_session
*sess
, *stmp
;
583 DBG("Cleanup sessiond");
586 * Close the thread quit pipe. It has already done its job,
587 * since we are now called.
589 utils_close_pipe(thread_quit_pipe
);
592 * If opt_pidfile is undefined, the default file will be wiped when
593 * removing the rundir.
596 ret
= remove(opt_pidfile
);
598 PERROR("remove pidfile %s", opt_pidfile
);
602 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
605 snprintf(path
, PATH_MAX
,
607 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
608 DBG("Removing %s", path
);
611 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
612 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
);
613 DBG("Removing %s", path
);
617 snprintf(path
, PATH_MAX
,
618 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
620 DBG("Removing %s", path
);
623 snprintf(path
, PATH_MAX
,
624 DEFAULT_KCONSUMERD_PATH
,
626 DBG("Removing directory %s", path
);
629 /* ust consumerd 32 */
630 snprintf(path
, PATH_MAX
,
631 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
633 DBG("Removing %s", path
);
636 snprintf(path
, PATH_MAX
,
637 DEFAULT_USTCONSUMERD32_PATH
,
639 DBG("Removing directory %s", path
);
642 /* ust consumerd 64 */
643 snprintf(path
, PATH_MAX
,
644 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
646 DBG("Removing %s", path
);
649 snprintf(path
, PATH_MAX
,
650 DEFAULT_USTCONSUMERD64_PATH
,
652 DBG("Removing directory %s", path
);
655 DBG("Cleaning up all sessions");
657 /* Destroy session list mutex */
658 if (session_list_ptr
!= NULL
) {
659 pthread_mutex_destroy(&session_list_ptr
->lock
);
661 /* Cleanup ALL session */
662 cds_list_for_each_entry_safe(sess
, stmp
,
663 &session_list_ptr
->head
, list
) {
664 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
668 wait_consumer(&kconsumer_data
);
669 wait_consumer(&ustconsumer64_data
);
670 wait_consumer(&ustconsumer32_data
);
672 DBG("Cleaning up all agent apps");
673 agent_app_ht_clean();
675 DBG("Closing all UST sockets");
676 ust_app_clean_list();
677 buffer_reg_destroy_registries();
679 if (is_root
&& !opt_no_kernel
) {
680 DBG2("Closing kernel fd");
681 if (kernel_tracer_fd
>= 0) {
682 ret
= close(kernel_tracer_fd
);
687 DBG("Unloading kernel modules");
688 modprobe_remove_lttng_all();
692 close_consumer_sockets();
695 load_session_destroy_data(load_info
);
700 * Cleanup lock file by deleting it and finaly closing it which will
701 * release the file system lock.
703 if (lockfile_fd
>= 0) {
704 char lockfile_path
[PATH_MAX
];
706 ret
= generate_lock_file_path(lockfile_path
,
707 sizeof(lockfile_path
));
709 ret
= remove(lockfile_path
);
711 PERROR("remove lock file");
713 ret
= close(lockfile_fd
);
715 PERROR("close lock file");
721 * We do NOT rmdir rundir because there are other processes
722 * using it, for instance lttng-relayd, which can start in
723 * parallel with this teardown.
730 * Cleanup the daemon's option data structures.
732 static void sessiond_cleanup_options(void)
734 DBG("Cleaning up options");
737 * If the override option is set, the pointer points to a *non* const
738 * thus freeing it even though the variable type is set to const.
740 if (tracing_group_name_override
) {
741 free((void *) tracing_group_name
);
743 if (consumerd32_bin_override
) {
744 free((void *) consumerd32_bin
);
746 if (consumerd64_bin_override
) {
747 free((void *) consumerd64_bin
);
749 if (consumerd32_libdir_override
) {
750 free((void *) consumerd32_libdir
);
752 if (consumerd64_libdir_override
) {
753 free((void *) consumerd64_libdir
);
757 free(opt_load_session_path
);
758 free(kmod_probes_list
);
759 free(kmod_extra_probes_list
);
761 run_as_destroy_worker();
764 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
765 "Matthew, BEET driven development works!%c[%dm",
766 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
771 * Send data on a unix socket using the liblttsessiondcomm API.
773 * Return lttcomm error code.
775 static int send_unix_sock(int sock
, void *buf
, size_t len
)
777 /* Check valid length */
782 return lttcomm_send_unix_sock(sock
, buf
, len
);
786 * Free memory of a command context structure.
788 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
790 DBG("Clean command context structure");
792 if ((*cmd_ctx
)->llm
) {
793 free((*cmd_ctx
)->llm
);
795 if ((*cmd_ctx
)->lsm
) {
796 free((*cmd_ctx
)->lsm
);
804 * Notify UST applications using the shm mmap futex.
806 static int notify_ust_apps(int active
)
810 DBG("Notifying applications of session daemon state: %d", active
);
812 /* See shm.c for this call implying mmap, shm and futex calls */
813 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
814 if (wait_shm_mmap
== NULL
) {
818 /* Wake waiting process */
819 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
821 /* Apps notified successfully */
829 * Setup the outgoing data buffer for the response (llm) by allocating the
830 * right amount of memory and copying the original information from the lsm
833 * Return 0 on success, negative value on error.
835 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
,
836 const void *payload_buf
, size_t payload_len
,
837 const void *cmd_header_buf
, size_t cmd_header_len
)
840 const size_t header_len
= sizeof(struct lttcomm_lttng_msg
);
841 const size_t cmd_header_offset
= header_len
;
842 const size_t payload_offset
= cmd_header_offset
+ cmd_header_len
;
843 const size_t total_msg_size
= header_len
+ cmd_header_len
+ payload_len
;
845 cmd_ctx
->llm
= zmalloc(total_msg_size
);
847 if (cmd_ctx
->llm
== NULL
) {
853 /* Copy common data */
854 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
855 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
856 cmd_ctx
->llm
->cmd_header_size
= cmd_header_len
;
857 cmd_ctx
->llm
->data_size
= payload_len
;
858 cmd_ctx
->lttng_msg_size
= total_msg_size
;
860 /* Copy command header */
861 if (cmd_header_len
) {
862 memcpy(((uint8_t *) cmd_ctx
->llm
) + cmd_header_offset
, cmd_header_buf
,
868 memcpy(((uint8_t *) cmd_ctx
->llm
) + payload_offset
, payload_buf
,
877 * Version of setup_lttng_msg() without command header.
879 static int setup_lttng_msg_no_cmd_header(struct command_ctx
*cmd_ctx
,
880 void *payload_buf
, size_t payload_len
)
882 return setup_lttng_msg(cmd_ctx
, payload_buf
, payload_len
, NULL
, 0);
885 * Update the kernel poll set of all channel fd available over all tracing
886 * session. Add the wakeup pipe at the end of the set.
888 static int update_kernel_poll(struct lttng_poll_event
*events
)
891 struct ltt_session
*session
;
892 struct ltt_kernel_channel
*channel
;
894 DBG("Updating kernel poll set");
897 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
898 session_lock(session
);
899 if (session
->kernel_session
== NULL
) {
900 session_unlock(session
);
904 cds_list_for_each_entry(channel
,
905 &session
->kernel_session
->channel_list
.head
, list
) {
906 /* Add channel fd to the kernel poll set */
907 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
909 session_unlock(session
);
912 DBG("Channel fd %d added to kernel set", channel
->fd
);
914 session_unlock(session
);
916 session_unlock_list();
921 session_unlock_list();
926 * Find the channel fd from 'fd' over all tracing session. When found, check
927 * for new channel stream and send those stream fds to the kernel consumer.
929 * Useful for CPU hotplug feature.
931 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
934 struct ltt_session
*session
;
935 struct ltt_kernel_session
*ksess
;
936 struct ltt_kernel_channel
*channel
;
938 DBG("Updating kernel streams for channel fd %d", fd
);
941 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
942 session_lock(session
);
943 if (session
->kernel_session
== NULL
) {
944 session_unlock(session
);
947 ksess
= session
->kernel_session
;
949 cds_list_for_each_entry(channel
,
950 &ksess
->channel_list
.head
, list
) {
951 struct lttng_ht_iter iter
;
952 struct consumer_socket
*socket
;
954 if (channel
->fd
!= fd
) {
957 DBG("Channel found, updating kernel streams");
958 ret
= kernel_open_channel_stream(channel
);
962 /* Update the stream global counter */
963 ksess
->stream_count_global
+= ret
;
966 * Have we already sent fds to the consumer? If yes, it
967 * means that tracing is started so it is safe to send
968 * our updated stream fds.
970 if (ksess
->consumer_fds_sent
!= 1
971 || ksess
->consumer
== NULL
) {
977 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
978 &iter
.iter
, socket
, node
.node
) {
979 pthread_mutex_lock(socket
->lock
);
980 ret
= kernel_consumer_send_channel_stream(socket
,
982 session
->output_traces
? 1 : 0);
983 pthread_mutex_unlock(socket
->lock
);
991 session_unlock(session
);
993 session_unlock_list();
997 session_unlock(session
);
998 session_unlock_list();
1003 * For each tracing session, update newly registered apps. The session list
1004 * lock MUST be acquired before calling this.
1006 static void update_ust_app(int app_sock
)
1008 struct ltt_session
*sess
, *stmp
;
1010 /* Consumer is in an ERROR state. Stop any application update. */
1011 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
1012 /* Stop the update process since the consumer is dead. */
1016 /* For all tracing session(s) */
1017 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
1018 struct ust_app
*app
;
1021 if (!sess
->ust_session
) {
1022 goto unlock_session
;
1026 assert(app_sock
>= 0);
1027 app
= ust_app_find_by_sock(app_sock
);
1030 * Application can be unregistered before so
1031 * this is possible hence simply stopping the
1034 DBG3("UST app update failed to find app sock %d",
1038 ust_app_global_update(sess
->ust_session
, app
);
1042 session_unlock(sess
);
1047 * This thread manage event coming from the kernel.
1049 * Features supported in this thread:
1052 static void *thread_manage_kernel(void *data
)
1054 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
1055 uint32_t revents
, nb_fd
;
1057 struct lttng_poll_event events
;
1059 DBG("[thread] Thread manage kernel started");
1061 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
1064 * This first step of the while is to clean this structure which could free
1065 * non NULL pointers so initialize it before the loop.
1067 lttng_poll_init(&events
);
1069 if (testpoint(sessiond_thread_manage_kernel
)) {
1070 goto error_testpoint
;
1073 health_code_update();
1075 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
1076 goto error_testpoint
;
1080 health_code_update();
1082 if (update_poll_flag
== 1) {
1083 /* Clean events object. We are about to populate it again. */
1084 lttng_poll_clean(&events
);
1086 ret
= sessiond_set_thread_pollset(&events
, 2);
1088 goto error_poll_create
;
1091 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1096 /* This will add the available kernel channel if any. */
1097 ret
= update_kernel_poll(&events
);
1101 update_poll_flag
= 0;
1104 DBG("Thread kernel polling");
1106 /* Poll infinite value of time */
1108 health_poll_entry();
1109 ret
= lttng_poll_wait(&events
, -1);
1110 DBG("Thread kernel return from poll on %d fds",
1111 LTTNG_POLL_GETNB(&events
));
1115 * Restart interrupted system call.
1117 if (errno
== EINTR
) {
1121 } else if (ret
== 0) {
1122 /* Should not happen since timeout is infinite */
1123 ERR("Return value of poll is 0 with an infinite timeout.\n"
1124 "This should not have happened! Continuing...");
1130 for (i
= 0; i
< nb_fd
; i
++) {
1131 /* Fetch once the poll data */
1132 revents
= LTTNG_POLL_GETEV(&events
, i
);
1133 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1135 health_code_update();
1138 /* No activity for this FD (poll implementation). */
1142 /* Thread quit pipe has been closed. Killing thread. */
1143 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1149 /* Check for data on kernel pipe */
1150 if (revents
& LPOLLIN
) {
1151 if (pollfd
== kernel_poll_pipe
[0]) {
1152 (void) lttng_read(kernel_poll_pipe
[0],
1155 * Ret value is useless here, if this pipe gets any actions an
1156 * update is required anyway.
1158 update_poll_flag
= 1;
1162 * New CPU detected by the kernel. Adding kernel stream to
1163 * kernel session and updating the kernel consumer
1165 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1171 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1172 update_poll_flag
= 1;
1175 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1183 lttng_poll_clean(&events
);
1186 utils_close_pipe(kernel_poll_pipe
);
1187 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1190 ERR("Health error occurred in %s", __func__
);
1191 WARN("Kernel thread died unexpectedly. "
1192 "Kernel tracing can continue but CPU hotplug is disabled.");
1194 health_unregister(health_sessiond
);
1195 DBG("Kernel thread dying");
1200 * Signal pthread condition of the consumer data that the thread.
1202 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1204 pthread_mutex_lock(&data
->cond_mutex
);
1207 * The state is set before signaling. It can be any value, it's the waiter
1208 * job to correctly interpret this condition variable associated to the
1209 * consumer pthread_cond.
1211 * A value of 0 means that the corresponding thread of the consumer data
1212 * was not started. 1 indicates that the thread has started and is ready
1213 * for action. A negative value means that there was an error during the
1216 data
->consumer_thread_is_ready
= state
;
1217 (void) pthread_cond_signal(&data
->cond
);
1219 pthread_mutex_unlock(&data
->cond_mutex
);
1223 * This thread manage the consumer error sent back to the session daemon.
1225 static void *thread_manage_consumer(void *data
)
1227 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1228 uint32_t revents
, nb_fd
;
1229 enum lttcomm_return_code code
;
1230 struct lttng_poll_event events
;
1231 struct consumer_data
*consumer_data
= data
;
1233 DBG("[thread] Manage consumer started");
1235 rcu_register_thread();
1236 rcu_thread_online();
1238 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1240 health_code_update();
1243 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1244 * metadata_sock. Nothing more will be added to this poll set.
1246 ret
= sessiond_set_thread_pollset(&events
, 3);
1252 * The error socket here is already in a listening state which was done
1253 * just before spawning this thread to avoid a race between the consumer
1254 * daemon exec trying to connect and the listen() call.
1256 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1261 health_code_update();
1263 /* Infinite blocking call, waiting for transmission */
1265 health_poll_entry();
1267 if (testpoint(sessiond_thread_manage_consumer
)) {
1271 ret
= lttng_poll_wait(&events
, -1);
1275 * Restart interrupted system call.
1277 if (errno
== EINTR
) {
1285 for (i
= 0; i
< nb_fd
; i
++) {
1286 /* Fetch once the poll data */
1287 revents
= LTTNG_POLL_GETEV(&events
, i
);
1288 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1290 health_code_update();
1293 /* No activity for this FD (poll implementation). */
1297 /* Thread quit pipe has been closed. Killing thread. */
1298 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1304 /* Event on the registration socket */
1305 if (pollfd
== consumer_data
->err_sock
) {
1306 if (revents
& LPOLLIN
) {
1308 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1309 ERR("consumer err socket poll error");
1312 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1318 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1324 * Set the CLOEXEC flag. Return code is useless because either way, the
1327 (void) utils_set_fd_cloexec(sock
);
1329 health_code_update();
1331 DBG2("Receiving code from consumer err_sock");
1333 /* Getting status code from kconsumerd */
1334 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1335 sizeof(enum lttcomm_return_code
));
1340 health_code_update();
1341 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1342 /* Connect both socket, command and metadata. */
1343 consumer_data
->cmd_sock
=
1344 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1345 consumer_data
->metadata_fd
=
1346 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1347 if (consumer_data
->cmd_sock
< 0
1348 || consumer_data
->metadata_fd
< 0) {
1349 PERROR("consumer connect cmd socket");
1350 /* On error, signal condition and quit. */
1351 signal_consumer_condition(consumer_data
, -1);
1354 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1355 /* Create metadata socket lock. */
1356 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1357 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1358 PERROR("zmalloc pthread mutex");
1362 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1364 signal_consumer_condition(consumer_data
, 1);
1365 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1366 DBG("Consumer metadata socket ready (fd: %d)",
1367 consumer_data
->metadata_fd
);
1369 ERR("consumer error when waiting for SOCK_READY : %s",
1370 lttcomm_get_readable_code(-code
));
1374 /* Remove the consumerd error sock since we've established a connexion */
1375 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1380 /* Add new accepted error socket. */
1381 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1386 /* Add metadata socket that is successfully connected. */
1387 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1388 LPOLLIN
| LPOLLRDHUP
);
1393 health_code_update();
1395 /* Infinite blocking call, waiting for transmission */
1398 health_code_update();
1400 /* Exit the thread because the thread quit pipe has been triggered. */
1402 /* Not a health error. */
1407 health_poll_entry();
1408 ret
= lttng_poll_wait(&events
, -1);
1412 * Restart interrupted system call.
1414 if (errno
== EINTR
) {
1422 for (i
= 0; i
< nb_fd
; i
++) {
1423 /* Fetch once the poll data */
1424 revents
= LTTNG_POLL_GETEV(&events
, i
);
1425 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1427 health_code_update();
1430 /* No activity for this FD (poll implementation). */
1435 * Thread quit pipe has been triggered, flag that we should stop
1436 * but continue the current loop to handle potential data from
1439 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1441 if (pollfd
== sock
) {
1442 /* Event on the consumerd socket */
1443 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1444 && !(revents
& LPOLLIN
)) {
1445 ERR("consumer err socket second poll error");
1448 health_code_update();
1449 /* Wait for any kconsumerd error */
1450 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1451 sizeof(enum lttcomm_return_code
));
1453 ERR("consumer closed the command socket");
1457 ERR("consumer return code : %s",
1458 lttcomm_get_readable_code(-code
));
1461 } else if (pollfd
== consumer_data
->metadata_fd
) {
1462 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1463 && !(revents
& LPOLLIN
)) {
1464 ERR("consumer err metadata socket second poll error");
1467 /* UST metadata requests */
1468 ret
= ust_consumer_metadata_request(
1469 &consumer_data
->metadata_sock
);
1471 ERR("Handling metadata request");
1475 /* No need for an else branch all FDs are tested prior. */
1477 health_code_update();
1483 * We lock here because we are about to close the sockets and some other
1484 * thread might be using them so get exclusive access which will abort all
1485 * other consumer command by other threads.
1487 pthread_mutex_lock(&consumer_data
->lock
);
1489 /* Immediately set the consumerd state to stopped */
1490 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1491 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1492 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1493 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1494 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1496 /* Code flow error... */
1500 if (consumer_data
->err_sock
>= 0) {
1501 ret
= close(consumer_data
->err_sock
);
1505 consumer_data
->err_sock
= -1;
1507 if (consumer_data
->cmd_sock
>= 0) {
1508 ret
= close(consumer_data
->cmd_sock
);
1512 consumer_data
->cmd_sock
= -1;
1514 if (consumer_data
->metadata_sock
.fd_ptr
&&
1515 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1516 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1528 unlink(consumer_data
->err_unix_sock_path
);
1529 unlink(consumer_data
->cmd_unix_sock_path
);
1530 pthread_mutex_unlock(&consumer_data
->lock
);
1532 /* Cleanup metadata socket mutex. */
1533 if (consumer_data
->metadata_sock
.lock
) {
1534 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1535 free(consumer_data
->metadata_sock
.lock
);
1537 lttng_poll_clean(&events
);
1541 ERR("Health error occurred in %s", __func__
);
1543 health_unregister(health_sessiond
);
1544 DBG("consumer thread cleanup completed");
1546 rcu_thread_offline();
1547 rcu_unregister_thread();
1553 * This thread manage application communication.
1555 static void *thread_manage_apps(void *data
)
1557 int i
, ret
, pollfd
, err
= -1;
1559 uint32_t revents
, nb_fd
;
1560 struct lttng_poll_event events
;
1562 DBG("[thread] Manage application started");
1564 rcu_register_thread();
1565 rcu_thread_online();
1567 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1569 if (testpoint(sessiond_thread_manage_apps
)) {
1570 goto error_testpoint
;
1573 health_code_update();
1575 ret
= sessiond_set_thread_pollset(&events
, 2);
1577 goto error_poll_create
;
1580 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1585 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1589 health_code_update();
1592 DBG("Apps thread polling");
1594 /* Inifinite blocking call, waiting for transmission */
1596 health_poll_entry();
1597 ret
= lttng_poll_wait(&events
, -1);
1598 DBG("Apps thread return from poll on %d fds",
1599 LTTNG_POLL_GETNB(&events
));
1603 * Restart interrupted system call.
1605 if (errno
== EINTR
) {
1613 for (i
= 0; i
< nb_fd
; i
++) {
1614 /* Fetch once the poll data */
1615 revents
= LTTNG_POLL_GETEV(&events
, i
);
1616 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1618 health_code_update();
1621 /* No activity for this FD (poll implementation). */
1625 /* Thread quit pipe has been closed. Killing thread. */
1626 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1632 /* Inspect the apps cmd pipe */
1633 if (pollfd
== apps_cmd_pipe
[0]) {
1634 if (revents
& LPOLLIN
) {
1638 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1639 if (size_ret
< sizeof(sock
)) {
1640 PERROR("read apps cmd pipe");
1644 health_code_update();
1647 * Since this is a command socket (write then read),
1648 * we only monitor the error events of the socket.
1650 ret
= lttng_poll_add(&events
, sock
,
1651 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1656 DBG("Apps with sock %d added to poll set", sock
);
1657 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1658 ERR("Apps command pipe error");
1661 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
1666 * At this point, we know that a registered application made
1667 * the event at poll_wait.
1669 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1670 /* Removing from the poll set */
1671 ret
= lttng_poll_del(&events
, pollfd
);
1676 /* Socket closed on remote end. */
1677 ust_app_unregister(pollfd
);
1679 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1684 health_code_update();
1690 lttng_poll_clean(&events
);
1693 utils_close_pipe(apps_cmd_pipe
);
1694 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1697 * We don't clean the UST app hash table here since already registered
1698 * applications can still be controlled so let them be until the session
1699 * daemon dies or the applications stop.
1704 ERR("Health error occurred in %s", __func__
);
1706 health_unregister(health_sessiond
);
1707 DBG("Application communication apps thread cleanup complete");
1708 rcu_thread_offline();
1709 rcu_unregister_thread();
1714 * Send a socket to a thread This is called from the dispatch UST registration
1715 * thread once all sockets are set for the application.
1717 * The sock value can be invalid, we don't really care, the thread will handle
1718 * it and make the necessary cleanup if so.
1720 * On success, return 0 else a negative value being the errno message of the
1723 static int send_socket_to_thread(int fd
, int sock
)
1728 * It's possible that the FD is set as invalid with -1 concurrently just
1729 * before calling this function being a shutdown state of the thread.
1736 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1737 if (ret
< sizeof(sock
)) {
1738 PERROR("write apps pipe %d", fd
);
1745 /* All good. Don't send back the write positive ret value. */
1752 * Sanitize the wait queue of the dispatch registration thread meaning removing
1753 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1754 * notify socket is never received.
1756 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1758 int ret
, nb_fd
= 0, i
;
1759 unsigned int fd_added
= 0;
1760 struct lttng_poll_event events
;
1761 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1765 lttng_poll_init(&events
);
1767 /* Just skip everything for an empty queue. */
1768 if (!wait_queue
->count
) {
1772 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1777 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1778 &wait_queue
->head
, head
) {
1779 assert(wait_node
->app
);
1780 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1781 LPOLLHUP
| LPOLLERR
);
1794 * Poll but don't block so we can quickly identify the faulty events and
1795 * clean them afterwards from the wait queue.
1797 ret
= lttng_poll_wait(&events
, 0);
1803 for (i
= 0; i
< nb_fd
; i
++) {
1804 /* Get faulty FD. */
1805 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1806 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1809 /* No activity for this FD (poll implementation). */
1813 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1814 &wait_queue
->head
, head
) {
1815 if (pollfd
== wait_node
->app
->sock
&&
1816 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1817 cds_list_del(&wait_node
->head
);
1818 wait_queue
->count
--;
1819 ust_app_destroy(wait_node
->app
);
1822 * Silence warning of use-after-free in
1823 * cds_list_for_each_entry_safe which uses
1824 * __typeof__(*wait_node).
1829 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1836 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1840 lttng_poll_clean(&events
);
1844 lttng_poll_clean(&events
);
1846 ERR("Unable to sanitize wait queue");
1851 * Dispatch request from the registration threads to the application
1852 * communication thread.
1854 static void *thread_dispatch_ust_registration(void *data
)
1857 struct cds_wfcq_node
*node
;
1858 struct ust_command
*ust_cmd
= NULL
;
1859 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1860 struct ust_reg_wait_queue wait_queue
= {
1864 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1866 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1867 goto error_testpoint
;
1870 health_code_update();
1872 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1874 DBG("[thread] Dispatch UST command started");
1876 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1877 health_code_update();
1879 /* Atomically prepare the queue futex */
1880 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1883 struct ust_app
*app
= NULL
;
1887 * Make sure we don't have node(s) that have hung up before receiving
1888 * the notify socket. This is to clean the list in order to avoid
1889 * memory leaks from notify socket that are never seen.
1891 sanitize_wait_queue(&wait_queue
);
1893 health_code_update();
1894 /* Dequeue command for registration */
1895 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1897 DBG("Woken up but nothing in the UST command queue");
1898 /* Continue thread execution */
1902 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1904 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1905 " gid:%d sock:%d name:%s (version %d.%d)",
1906 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1907 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1908 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1909 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1911 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1912 wait_node
= zmalloc(sizeof(*wait_node
));
1914 PERROR("zmalloc wait_node dispatch");
1915 ret
= close(ust_cmd
->sock
);
1917 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1919 lttng_fd_put(LTTNG_FD_APPS
, 1);
1923 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1925 /* Create application object if socket is CMD. */
1926 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1928 if (!wait_node
->app
) {
1929 ret
= close(ust_cmd
->sock
);
1931 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1933 lttng_fd_put(LTTNG_FD_APPS
, 1);
1939 * Add application to the wait queue so we can set the notify
1940 * socket before putting this object in the global ht.
1942 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1947 * We have to continue here since we don't have the notify
1948 * socket and the application MUST be added to the hash table
1949 * only at that moment.
1954 * Look for the application in the local wait queue and set the
1955 * notify socket if found.
1957 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1958 &wait_queue
.head
, head
) {
1959 health_code_update();
1960 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1961 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1962 cds_list_del(&wait_node
->head
);
1964 app
= wait_node
->app
;
1966 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1972 * With no application at this stage the received socket is
1973 * basically useless so close it before we free the cmd data
1974 * structure for good.
1977 ret
= close(ust_cmd
->sock
);
1979 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1981 lttng_fd_put(LTTNG_FD_APPS
, 1);
1988 * @session_lock_list
1990 * Lock the global session list so from the register up to the
1991 * registration done message, no thread can see the application
1992 * and change its state.
1994 session_lock_list();
1998 * Add application to the global hash table. This needs to be
1999 * done before the update to the UST registry can locate the
2004 /* Set app version. This call will print an error if needed. */
2005 (void) ust_app_version(app
);
2007 /* Send notify socket through the notify pipe. */
2008 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
2012 session_unlock_list();
2014 * No notify thread, stop the UST tracing. However, this is
2015 * not an internal error of the this thread thus setting
2016 * the health error code to a normal exit.
2023 * Update newly registered application with the tracing
2024 * registry info already enabled information.
2026 update_ust_app(app
->sock
);
2029 * Don't care about return value. Let the manage apps threads
2030 * handle app unregistration upon socket close.
2032 (void) ust_app_register_done(app
);
2035 * Even if the application socket has been closed, send the app
2036 * to the thread and unregistration will take place at that
2039 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
2042 session_unlock_list();
2044 * No apps. thread, stop the UST tracing. However, this is
2045 * not an internal error of the this thread thus setting
2046 * the health error code to a normal exit.
2053 session_unlock_list();
2055 } while (node
!= NULL
);
2057 health_poll_entry();
2058 /* Futex wait on queue. Blocking call on futex() */
2059 futex_nto1_wait(&ust_cmd_queue
.futex
);
2062 /* Normal exit, no error */
2066 /* Clean up wait queue. */
2067 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2068 &wait_queue
.head
, head
) {
2069 cds_list_del(&wait_node
->head
);
2074 /* Empty command queue. */
2076 /* Dequeue command for registration */
2077 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2081 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2082 ret
= close(ust_cmd
->sock
);
2084 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2086 lttng_fd_put(LTTNG_FD_APPS
, 1);
2091 DBG("Dispatch thread dying");
2094 ERR("Health error occurred in %s", __func__
);
2096 health_unregister(health_sessiond
);
2101 * This thread manage application registration.
2103 static void *thread_registration_apps(void *data
)
2105 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2106 uint32_t revents
, nb_fd
;
2107 struct lttng_poll_event events
;
2109 * Get allocated in this thread, enqueued to a global queue, dequeued and
2110 * freed in the manage apps thread.
2112 struct ust_command
*ust_cmd
= NULL
;
2114 DBG("[thread] Manage application registration started");
2116 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2118 if (testpoint(sessiond_thread_registration_apps
)) {
2119 goto error_testpoint
;
2122 ret
= lttcomm_listen_unix_sock(apps_sock
);
2128 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2129 * more will be added to this poll set.
2131 ret
= sessiond_set_thread_pollset(&events
, 2);
2133 goto error_create_poll
;
2136 /* Add the application registration socket */
2137 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2139 goto error_poll_add
;
2142 /* Notify all applications to register */
2143 ret
= notify_ust_apps(1);
2145 ERR("Failed to notify applications or create the wait shared memory.\n"
2146 "Execution continues but there might be problem for already\n"
2147 "running applications that wishes to register.");
2151 DBG("Accepting application registration");
2153 /* Inifinite blocking call, waiting for transmission */
2155 health_poll_entry();
2156 ret
= lttng_poll_wait(&events
, -1);
2160 * Restart interrupted system call.
2162 if (errno
== EINTR
) {
2170 for (i
= 0; i
< nb_fd
; i
++) {
2171 health_code_update();
2173 /* Fetch once the poll data */
2174 revents
= LTTNG_POLL_GETEV(&events
, i
);
2175 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2178 /* No activity for this FD (poll implementation). */
2182 /* Thread quit pipe has been closed. Killing thread. */
2183 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2189 /* Event on the registration socket */
2190 if (pollfd
== apps_sock
) {
2191 if (revents
& LPOLLIN
) {
2192 sock
= lttcomm_accept_unix_sock(apps_sock
);
2198 * Set socket timeout for both receiving and ending.
2199 * app_socket_timeout is in seconds, whereas
2200 * lttcomm_setsockopt_rcv_timeout and
2201 * lttcomm_setsockopt_snd_timeout expect msec as
2204 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2205 app_socket_timeout
* 1000);
2206 (void) lttcomm_setsockopt_snd_timeout(sock
,
2207 app_socket_timeout
* 1000);
2210 * Set the CLOEXEC flag. Return code is useless because
2211 * either way, the show must go on.
2213 (void) utils_set_fd_cloexec(sock
);
2215 /* Create UST registration command for enqueuing */
2216 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2217 if (ust_cmd
== NULL
) {
2218 PERROR("ust command zmalloc");
2227 * Using message-based transmissions to ensure we don't
2228 * have to deal with partially received messages.
2230 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2232 ERR("Exhausted file descriptors allowed for applications.");
2242 health_code_update();
2243 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2246 /* Close socket of the application. */
2251 lttng_fd_put(LTTNG_FD_APPS
, 1);
2255 health_code_update();
2257 ust_cmd
->sock
= sock
;
2260 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2261 " gid:%d sock:%d name:%s (version %d.%d)",
2262 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2263 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2264 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2265 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2268 * Lock free enqueue the registration request. The red pill
2269 * has been taken! This apps will be part of the *system*.
2271 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2274 * Wake the registration queue futex. Implicit memory
2275 * barrier with the exchange in cds_wfcq_enqueue.
2277 futex_nto1_wake(&ust_cmd_queue
.futex
);
2278 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2279 ERR("Register apps socket poll error");
2282 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2291 /* Notify that the registration thread is gone */
2294 if (apps_sock
>= 0) {
2295 ret
= close(apps_sock
);
2305 lttng_fd_put(LTTNG_FD_APPS
, 1);
2307 unlink(apps_unix_sock_path
);
2310 lttng_poll_clean(&events
);
2314 DBG("UST Registration thread cleanup complete");
2317 ERR("Health error occurred in %s", __func__
);
2319 health_unregister(health_sessiond
);
2325 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2326 * exec or it will fails.
2328 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2331 struct timespec timeout
;
2334 * Make sure we set the readiness flag to 0 because we are NOT ready.
2335 * This access to consumer_thread_is_ready does not need to be
2336 * protected by consumer_data.cond_mutex (yet) since the consumer
2337 * management thread has not been started at this point.
2339 consumer_data
->consumer_thread_is_ready
= 0;
2341 /* Setup pthread condition */
2342 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2345 PERROR("pthread_condattr_init consumer data");
2350 * Set the monotonic clock in order to make sure we DO NOT jump in time
2351 * between the clock_gettime() call and the timedwait call. See bug #324
2352 * for a more details and how we noticed it.
2354 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2357 PERROR("pthread_condattr_setclock consumer data");
2361 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2364 PERROR("pthread_cond_init consumer data");
2368 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2372 PERROR("pthread_create consumer");
2377 /* We are about to wait on a pthread condition */
2378 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2380 /* Get time for sem_timedwait absolute timeout */
2381 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2383 * Set the timeout for the condition timed wait even if the clock gettime
2384 * call fails since we might loop on that call and we want to avoid to
2385 * increment the timeout too many times.
2387 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2390 * The following loop COULD be skipped in some conditions so this is why we
2391 * set ret to 0 in order to make sure at least one round of the loop is
2397 * Loop until the condition is reached or when a timeout is reached. Note
2398 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2399 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2400 * possible. This loop does not take any chances and works with both of
2403 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2404 if (clock_ret
< 0) {
2405 PERROR("clock_gettime spawn consumer");
2406 /* Infinite wait for the consumerd thread to be ready */
2407 ret
= pthread_cond_wait(&consumer_data
->cond
,
2408 &consumer_data
->cond_mutex
);
2410 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2411 &consumer_data
->cond_mutex
, &timeout
);
2415 /* Release the pthread condition */
2416 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2420 if (ret
== ETIMEDOUT
) {
2424 * Call has timed out so we kill the kconsumerd_thread and return
2427 ERR("Condition timed out. The consumer thread was never ready."
2429 pth_ret
= pthread_cancel(consumer_data
->thread
);
2431 PERROR("pthread_cancel consumer thread");
2434 PERROR("pthread_cond_wait failed consumer thread");
2436 /* Caller is expecting a negative value on failure. */
2441 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2442 if (consumer_data
->pid
== 0) {
2443 ERR("Consumerd did not start");
2444 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2447 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2456 * Join consumer thread
2458 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2462 /* Consumer pid must be a real one. */
2463 if (consumer_data
->pid
> 0) {
2465 ret
= kill(consumer_data
->pid
, SIGTERM
);
2467 PERROR("Error killing consumer daemon");
2470 return pthread_join(consumer_data
->thread
, &status
);
2477 * Fork and exec a consumer daemon (consumerd).
2479 * Return pid if successful else -1.
2481 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2485 const char *consumer_to_use
;
2486 const char *verbosity
;
2489 DBG("Spawning consumerd");
2496 if (opt_verbose_consumer
) {
2497 verbosity
= "--verbose";
2498 } else if (lttng_opt_quiet
) {
2499 verbosity
= "--quiet";
2504 switch (consumer_data
->type
) {
2505 case LTTNG_CONSUMER_KERNEL
:
2507 * Find out which consumerd to execute. We will first try the
2508 * 64-bit path, then the sessiond's installation directory, and
2509 * fallback on the 32-bit one,
2511 DBG3("Looking for a kernel consumer at these locations:");
2512 DBG3(" 1) %s", consumerd64_bin
);
2513 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2514 DBG3(" 3) %s", consumerd32_bin
);
2515 if (stat(consumerd64_bin
, &st
) == 0) {
2516 DBG3("Found location #1");
2517 consumer_to_use
= consumerd64_bin
;
2518 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2519 DBG3("Found location #2");
2520 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2521 } else if (stat(consumerd32_bin
, &st
) == 0) {
2522 DBG3("Found location #3");
2523 consumer_to_use
= consumerd32_bin
;
2525 DBG("Could not find any valid consumerd executable");
2529 DBG("Using kernel consumer at: %s", consumer_to_use
);
2530 ret
= execl(consumer_to_use
,
2531 "lttng-consumerd", verbosity
, "-k",
2532 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2533 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2534 "--group", tracing_group_name
,
2537 case LTTNG_CONSUMER64_UST
:
2539 char *tmpnew
= NULL
;
2541 if (consumerd64_libdir
[0] != '\0') {
2545 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2549 tmplen
= strlen("LD_LIBRARY_PATH=")
2550 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2551 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2556 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2557 strcat(tmpnew
, consumerd64_libdir
);
2558 if (tmp
[0] != '\0') {
2559 strcat(tmpnew
, ":");
2560 strcat(tmpnew
, tmp
);
2562 ret
= putenv(tmpnew
);
2569 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2570 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2571 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2572 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2573 "--group", tracing_group_name
,
2575 if (consumerd64_libdir
[0] != '\0') {
2580 case LTTNG_CONSUMER32_UST
:
2582 char *tmpnew
= NULL
;
2584 if (consumerd32_libdir
[0] != '\0') {
2588 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2592 tmplen
= strlen("LD_LIBRARY_PATH=")
2593 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2594 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2599 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2600 strcat(tmpnew
, consumerd32_libdir
);
2601 if (tmp
[0] != '\0') {
2602 strcat(tmpnew
, ":");
2603 strcat(tmpnew
, tmp
);
2605 ret
= putenv(tmpnew
);
2612 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2613 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2614 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2615 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2616 "--group", tracing_group_name
,
2618 if (consumerd32_libdir
[0] != '\0') {
2624 PERROR("unknown consumer type");
2628 PERROR("Consumer execl()");
2630 /* Reaching this point, we got a failure on our execl(). */
2632 } else if (pid
> 0) {
2635 PERROR("start consumer fork");
2643 * Spawn the consumerd daemon and session daemon thread.
2645 static int start_consumerd(struct consumer_data
*consumer_data
)
2650 * Set the listen() state on the socket since there is a possible race
2651 * between the exec() of the consumer daemon and this call if place in the
2652 * consumer thread. See bug #366 for more details.
2654 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2659 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2660 if (consumer_data
->pid
!= 0) {
2661 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2665 ret
= spawn_consumerd(consumer_data
);
2667 ERR("Spawning consumerd failed");
2668 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2672 /* Setting up the consumer_data pid */
2673 consumer_data
->pid
= ret
;
2674 DBG2("Consumer pid %d", consumer_data
->pid
);
2675 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2677 DBG2("Spawning consumer control thread");
2678 ret
= spawn_consumer_thread(consumer_data
);
2680 ERR("Fatal error spawning consumer control thread");
2688 /* Cleanup already created sockets on error. */
2689 if (consumer_data
->err_sock
>= 0) {
2692 err
= close(consumer_data
->err_sock
);
2694 PERROR("close consumer data error socket");
2701 * Setup necessary data for kernel tracer action.
2703 static int init_kernel_tracer(void)
2707 /* Modprobe lttng kernel modules */
2708 ret
= modprobe_lttng_control();
2713 /* Open debugfs lttng */
2714 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2715 if (kernel_tracer_fd
< 0) {
2716 DBG("Failed to open %s", module_proc_lttng
);
2721 /* Validate kernel version */
2722 ret
= kernel_validate_version(kernel_tracer_fd
);
2727 ret
= modprobe_lttng_data();
2732 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2736 modprobe_remove_lttng_control();
2737 ret
= close(kernel_tracer_fd
);
2741 kernel_tracer_fd
= -1;
2742 return LTTNG_ERR_KERN_VERSION
;
2745 ret
= close(kernel_tracer_fd
);
2751 modprobe_remove_lttng_control();
2754 WARN("No kernel tracer available");
2755 kernel_tracer_fd
= -1;
2757 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2759 return LTTNG_ERR_KERN_NA
;
2765 * Copy consumer output from the tracing session to the domain session. The
2766 * function also applies the right modification on a per domain basis for the
2767 * trace files destination directory.
2769 * Should *NOT* be called with RCU read-side lock held.
2771 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2774 const char *dir_name
;
2775 struct consumer_output
*consumer
;
2778 assert(session
->consumer
);
2781 case LTTNG_DOMAIN_KERNEL
:
2782 DBG3("Copying tracing session consumer output in kernel session");
2784 * XXX: We should audit the session creation and what this function
2785 * does "extra" in order to avoid a destroy since this function is used
2786 * in the domain session creation (kernel and ust) only. Same for UST
2789 if (session
->kernel_session
->consumer
) {
2790 consumer_output_put(session
->kernel_session
->consumer
);
2792 session
->kernel_session
->consumer
=
2793 consumer_copy_output(session
->consumer
);
2794 /* Ease our life a bit for the next part */
2795 consumer
= session
->kernel_session
->consumer
;
2796 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2798 case LTTNG_DOMAIN_JUL
:
2799 case LTTNG_DOMAIN_LOG4J
:
2800 case LTTNG_DOMAIN_PYTHON
:
2801 case LTTNG_DOMAIN_UST
:
2802 DBG3("Copying tracing session consumer output in UST session");
2803 if (session
->ust_session
->consumer
) {
2804 consumer_output_put(session
->ust_session
->consumer
);
2806 session
->ust_session
->consumer
=
2807 consumer_copy_output(session
->consumer
);
2808 /* Ease our life a bit for the next part */
2809 consumer
= session
->ust_session
->consumer
;
2810 dir_name
= DEFAULT_UST_TRACE_DIR
;
2813 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2817 /* Append correct directory to subdir */
2818 strncat(consumer
->subdir
, dir_name
,
2819 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2820 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2829 * Create an UST session and add it to the session ust list.
2831 * Should *NOT* be called with RCU read-side lock held.
2833 static int create_ust_session(struct ltt_session
*session
,
2834 struct lttng_domain
*domain
)
2837 struct ltt_ust_session
*lus
= NULL
;
2841 assert(session
->consumer
);
2843 switch (domain
->type
) {
2844 case LTTNG_DOMAIN_JUL
:
2845 case LTTNG_DOMAIN_LOG4J
:
2846 case LTTNG_DOMAIN_PYTHON
:
2847 case LTTNG_DOMAIN_UST
:
2850 ERR("Unknown UST domain on create session %d", domain
->type
);
2851 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2855 DBG("Creating UST session");
2857 lus
= trace_ust_create_session(session
->id
);
2859 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2863 lus
->uid
= session
->uid
;
2864 lus
->gid
= session
->gid
;
2865 lus
->output_traces
= session
->output_traces
;
2866 lus
->snapshot_mode
= session
->snapshot_mode
;
2867 lus
->live_timer_interval
= session
->live_timer
;
2868 session
->ust_session
= lus
;
2869 if (session
->shm_path
[0]) {
2870 strncpy(lus
->root_shm_path
, session
->shm_path
,
2871 sizeof(lus
->root_shm_path
));
2872 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2873 strncpy(lus
->shm_path
, session
->shm_path
,
2874 sizeof(lus
->shm_path
));
2875 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2876 strncat(lus
->shm_path
, "/ust",
2877 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2879 /* Copy session output to the newly created UST session */
2880 ret
= copy_session_consumer(domain
->type
, session
);
2881 if (ret
!= LTTNG_OK
) {
2889 session
->ust_session
= NULL
;
2894 * Create a kernel tracer session then create the default channel.
2896 static int create_kernel_session(struct ltt_session
*session
)
2900 DBG("Creating kernel session");
2902 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2904 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2908 /* Code flow safety */
2909 assert(session
->kernel_session
);
2911 /* Copy session output to the newly created Kernel session */
2912 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2913 if (ret
!= LTTNG_OK
) {
2917 /* Create directory(ies) on local filesystem. */
2918 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2919 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2920 ret
= run_as_mkdir_recursive(
2921 session
->kernel_session
->consumer
->dst
.trace_path
,
2922 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2924 if (errno
!= EEXIST
) {
2925 ERR("Trace directory creation error");
2931 session
->kernel_session
->uid
= session
->uid
;
2932 session
->kernel_session
->gid
= session
->gid
;
2933 session
->kernel_session
->output_traces
= session
->output_traces
;
2934 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2939 trace_kernel_destroy_session(session
->kernel_session
);
2940 session
->kernel_session
= NULL
;
2945 * Count number of session permitted by uid/gid.
2947 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2950 struct ltt_session
*session
;
2952 DBG("Counting number of available session for UID %d GID %d",
2954 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2956 * Only list the sessions the user can control.
2958 if (!session_access_ok(session
, uid
, gid
)) {
2967 * Process the command requested by the lttng client within the command
2968 * context structure. This function make sure that the return structure (llm)
2969 * is set and ready for transmission before returning.
2971 * Return any error encountered or 0 for success.
2973 * "sock" is only used for special-case var. len data.
2975 * Should *NOT* be called with RCU read-side lock held.
2977 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2981 int need_tracing_session
= 1;
2984 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2986 assert(!rcu_read_ongoing());
2990 switch (cmd_ctx
->lsm
->cmd_type
) {
2991 case LTTNG_CREATE_SESSION
:
2992 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2993 case LTTNG_CREATE_SESSION_LIVE
:
2994 case LTTNG_DESTROY_SESSION
:
2995 case LTTNG_LIST_SESSIONS
:
2996 case LTTNG_LIST_DOMAINS
:
2997 case LTTNG_START_TRACE
:
2998 case LTTNG_STOP_TRACE
:
2999 case LTTNG_DATA_PENDING
:
3000 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3001 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3002 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3003 case LTTNG_SNAPSHOT_RECORD
:
3004 case LTTNG_SAVE_SESSION
:
3005 case LTTNG_SET_SESSION_SHM_PATH
:
3006 case LTTNG_METADATA_REGENERATE
:
3013 if (opt_no_kernel
&& need_domain
3014 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
3016 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3018 ret
= LTTNG_ERR_KERN_NA
;
3023 /* Deny register consumer if we already have a spawned consumer. */
3024 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
3025 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3026 if (kconsumer_data
.pid
> 0) {
3027 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3028 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3031 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3035 * Check for command that don't needs to allocate a returned payload. We do
3036 * this here so we don't have to make the call for no payload at each
3039 switch(cmd_ctx
->lsm
->cmd_type
) {
3040 case LTTNG_LIST_SESSIONS
:
3041 case LTTNG_LIST_TRACEPOINTS
:
3042 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3043 case LTTNG_LIST_DOMAINS
:
3044 case LTTNG_LIST_CHANNELS
:
3045 case LTTNG_LIST_EVENTS
:
3046 case LTTNG_LIST_SYSCALLS
:
3047 case LTTNG_LIST_TRACKER_PIDS
:
3048 case LTTNG_DATA_PENDING
:
3051 /* Setup lttng message with no payload */
3052 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0);
3054 /* This label does not try to unlock the session */
3055 goto init_setup_error
;
3059 /* Commands that DO NOT need a session. */
3060 switch (cmd_ctx
->lsm
->cmd_type
) {
3061 case LTTNG_CREATE_SESSION
:
3062 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3063 case LTTNG_CREATE_SESSION_LIVE
:
3064 case LTTNG_CALIBRATE
:
3065 case LTTNG_LIST_SESSIONS
:
3066 case LTTNG_LIST_TRACEPOINTS
:
3067 case LTTNG_LIST_SYSCALLS
:
3068 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3069 case LTTNG_SAVE_SESSION
:
3070 need_tracing_session
= 0;
3073 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3075 * We keep the session list lock across _all_ commands
3076 * for now, because the per-session lock does not
3077 * handle teardown properly.
3079 session_lock_list();
3080 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3081 if (cmd_ctx
->session
== NULL
) {
3082 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3085 /* Acquire lock for the session */
3086 session_lock(cmd_ctx
->session
);
3092 * Commands that need a valid session but should NOT create one if none
3093 * exists. Instead of creating one and destroying it when the command is
3094 * handled, process that right before so we save some round trip in useless
3097 switch (cmd_ctx
->lsm
->cmd_type
) {
3098 case LTTNG_DISABLE_CHANNEL
:
3099 case LTTNG_DISABLE_EVENT
:
3100 switch (cmd_ctx
->lsm
->domain
.type
) {
3101 case LTTNG_DOMAIN_KERNEL
:
3102 if (!cmd_ctx
->session
->kernel_session
) {
3103 ret
= LTTNG_ERR_NO_CHANNEL
;
3107 case LTTNG_DOMAIN_JUL
:
3108 case LTTNG_DOMAIN_LOG4J
:
3109 case LTTNG_DOMAIN_PYTHON
:
3110 case LTTNG_DOMAIN_UST
:
3111 if (!cmd_ctx
->session
->ust_session
) {
3112 ret
= LTTNG_ERR_NO_CHANNEL
;
3117 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3129 * Check domain type for specific "pre-action".
3131 switch (cmd_ctx
->lsm
->domain
.type
) {
3132 case LTTNG_DOMAIN_KERNEL
:
3134 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3138 /* Kernel tracer check */
3139 if (kernel_tracer_fd
== -1) {
3140 /* Basically, load kernel tracer modules */
3141 ret
= init_kernel_tracer();
3147 /* Consumer is in an ERROR state. Report back to client */
3148 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3149 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3153 /* Need a session for kernel command */
3154 if (need_tracing_session
) {
3155 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3156 ret
= create_kernel_session(cmd_ctx
->session
);
3158 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3163 /* Start the kernel consumer daemon */
3164 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3165 if (kconsumer_data
.pid
== 0 &&
3166 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3167 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3168 ret
= start_consumerd(&kconsumer_data
);
3170 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3173 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3175 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3179 * The consumer was just spawned so we need to add the socket to
3180 * the consumer output of the session if exist.
3182 ret
= consumer_create_socket(&kconsumer_data
,
3183 cmd_ctx
->session
->kernel_session
->consumer
);
3190 case LTTNG_DOMAIN_JUL
:
3191 case LTTNG_DOMAIN_LOG4J
:
3192 case LTTNG_DOMAIN_PYTHON
:
3193 case LTTNG_DOMAIN_UST
:
3195 if (!ust_app_supported()) {
3196 ret
= LTTNG_ERR_NO_UST
;
3199 /* Consumer is in an ERROR state. Report back to client */
3200 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3201 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3205 if (need_tracing_session
) {
3206 /* Create UST session if none exist. */
3207 if (cmd_ctx
->session
->ust_session
== NULL
) {
3208 ret
= create_ust_session(cmd_ctx
->session
,
3209 &cmd_ctx
->lsm
->domain
);
3210 if (ret
!= LTTNG_OK
) {
3215 /* Start the UST consumer daemons */
3217 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3218 if (consumerd64_bin
[0] != '\0' &&
3219 ustconsumer64_data
.pid
== 0 &&
3220 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3221 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3222 ret
= start_consumerd(&ustconsumer64_data
);
3224 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3225 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3229 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3230 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3232 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3236 * Setup socket for consumer 64 bit. No need for atomic access
3237 * since it was set above and can ONLY be set in this thread.
3239 ret
= consumer_create_socket(&ustconsumer64_data
,
3240 cmd_ctx
->session
->ust_session
->consumer
);
3246 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3247 if (consumerd32_bin
[0] != '\0' &&
3248 ustconsumer32_data
.pid
== 0 &&
3249 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3250 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3251 ret
= start_consumerd(&ustconsumer32_data
);
3253 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3254 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3258 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3259 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3261 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3265 * Setup socket for consumer 64 bit. No need for atomic access
3266 * since it was set above and can ONLY be set in this thread.
3268 ret
= consumer_create_socket(&ustconsumer32_data
,
3269 cmd_ctx
->session
->ust_session
->consumer
);
3281 /* Validate consumer daemon state when start/stop trace command */
3282 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3283 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3284 switch (cmd_ctx
->lsm
->domain
.type
) {
3285 case LTTNG_DOMAIN_NONE
:
3287 case LTTNG_DOMAIN_JUL
:
3288 case LTTNG_DOMAIN_LOG4J
:
3289 case LTTNG_DOMAIN_PYTHON
:
3290 case LTTNG_DOMAIN_UST
:
3291 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3292 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3296 case LTTNG_DOMAIN_KERNEL
:
3297 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3298 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3303 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3309 * Check that the UID or GID match that of the tracing session.
3310 * The root user can interact with all sessions.
3312 if (need_tracing_session
) {
3313 if (!session_access_ok(cmd_ctx
->session
,
3314 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3315 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3316 ret
= LTTNG_ERR_EPERM
;
3322 * Send relayd information to consumer as soon as we have a domain and a
3325 if (cmd_ctx
->session
&& need_domain
) {
3327 * Setup relayd if not done yet. If the relayd information was already
3328 * sent to the consumer, this call will gracefully return.
3330 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3331 if (ret
!= LTTNG_OK
) {
3336 /* Process by command type */
3337 switch (cmd_ctx
->lsm
->cmd_type
) {
3338 case LTTNG_ADD_CONTEXT
:
3341 * An LTTNG_ADD_CONTEXT command might have a supplementary
3342 * payload if the context being added is an application context.
3344 if (cmd_ctx
->lsm
->u
.context
.ctx
.ctx
==
3345 LTTNG_EVENT_CONTEXT_APP_CONTEXT
) {
3346 char *provider_name
= NULL
, *context_name
= NULL
;
3347 size_t provider_name_len
=
3348 cmd_ctx
->lsm
->u
.context
.provider_name_len
;
3349 size_t context_name_len
=
3350 cmd_ctx
->lsm
->u
.context
.context_name_len
;
3352 if (provider_name_len
== 0 || context_name_len
== 0) {
3354 * Application provider and context names MUST
3357 ret
= -LTTNG_ERR_INVALID
;
3361 provider_name
= zmalloc(provider_name_len
+ 1);
3362 if (!provider_name
) {
3363 ret
= -LTTNG_ERR_NOMEM
;
3366 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
=
3369 context_name
= zmalloc(context_name_len
+ 1);
3370 if (!context_name
) {
3371 ret
= -LTTNG_ERR_NOMEM
;
3372 goto error_add_context
;
3374 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
=
3377 ret
= lttcomm_recv_unix_sock(sock
, provider_name
,
3380 goto error_add_context
;
3383 ret
= lttcomm_recv_unix_sock(sock
, context_name
,
3386 goto error_add_context
;
3391 * cmd_add_context assumes ownership of the provider and context
3394 ret
= cmd_add_context(cmd_ctx
->session
,
3395 cmd_ctx
->lsm
->domain
.type
,
3396 cmd_ctx
->lsm
->u
.context
.channel_name
,
3397 &cmd_ctx
->lsm
->u
.context
.ctx
,
3398 kernel_poll_pipe
[1]);
3400 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
= NULL
;
3401 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
= NULL
;
3403 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
);
3404 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
);
3410 case LTTNG_DISABLE_CHANNEL
:
3412 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3413 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3416 case LTTNG_DISABLE_EVENT
:
3420 * FIXME: handle filter; for now we just receive the filter's
3421 * bytecode along with the filter expression which are sent by
3422 * liblttng-ctl and discard them.
3424 * This fixes an issue where the client may block while sending
3425 * the filter payload and encounter an error because the session
3426 * daemon closes the socket without ever handling this data.
3428 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3429 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3432 char data
[LTTNG_FILTER_MAX_LEN
];
3434 DBG("Discarding disable event command payload of size %zu", count
);
3436 ret
= lttcomm_recv_unix_sock(sock
, data
,
3437 count
> sizeof(data
) ? sizeof(data
) : count
);
3442 count
-= (size_t) ret
;
3445 /* FIXME: passing packed structure to non-packed pointer */
3446 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3447 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3448 &cmd_ctx
->lsm
->u
.disable
.event
);
3451 case LTTNG_ENABLE_CHANNEL
:
3453 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3454 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3457 case LTTNG_TRACK_PID
:
3459 ret
= cmd_track_pid(cmd_ctx
->session
,
3460 cmd_ctx
->lsm
->domain
.type
,
3461 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3464 case LTTNG_UNTRACK_PID
:
3466 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3467 cmd_ctx
->lsm
->domain
.type
,
3468 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3471 case LTTNG_ENABLE_EVENT
:
3473 struct lttng_event_exclusion
*exclusion
= NULL
;
3474 struct lttng_filter_bytecode
*bytecode
= NULL
;
3475 char *filter_expression
= NULL
;
3477 /* Handle exclusion events and receive it from the client. */
3478 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3479 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3481 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3482 (count
* LTTNG_SYMBOL_NAME_LEN
));
3484 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3488 DBG("Receiving var len exclusion event list from client ...");
3489 exclusion
->count
= count
;
3490 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3491 count
* LTTNG_SYMBOL_NAME_LEN
);
3493 DBG("Nothing recv() from client var len data... continuing");
3496 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3501 /* Get filter expression from client. */
3502 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3503 size_t expression_len
=
3504 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3506 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3507 ret
= LTTNG_ERR_FILTER_INVAL
;
3512 filter_expression
= zmalloc(expression_len
);
3513 if (!filter_expression
) {
3515 ret
= LTTNG_ERR_FILTER_NOMEM
;
3519 /* Receive var. len. data */
3520 DBG("Receiving var len filter's expression from client ...");
3521 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3524 DBG("Nothing recv() from client car len data... continuing");
3526 free(filter_expression
);
3528 ret
= LTTNG_ERR_FILTER_INVAL
;
3533 /* Handle filter and get bytecode from client. */
3534 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3535 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3537 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3538 ret
= LTTNG_ERR_FILTER_INVAL
;
3539 free(filter_expression
);
3544 bytecode
= zmalloc(bytecode_len
);
3546 free(filter_expression
);
3548 ret
= LTTNG_ERR_FILTER_NOMEM
;
3552 /* Receive var. len. data */
3553 DBG("Receiving var len filter's bytecode from client ...");
3554 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3556 DBG("Nothing recv() from client car len data... continuing");
3558 free(filter_expression
);
3561 ret
= LTTNG_ERR_FILTER_INVAL
;
3565 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3566 free(filter_expression
);
3569 ret
= LTTNG_ERR_FILTER_INVAL
;
3574 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3575 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3576 &cmd_ctx
->lsm
->u
.enable
.event
,
3577 filter_expression
, bytecode
, exclusion
,
3578 kernel_poll_pipe
[1]);
3581 case LTTNG_LIST_TRACEPOINTS
:
3583 struct lttng_event
*events
;
3586 session_lock_list();
3587 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3588 session_unlock_list();
3589 if (nb_events
< 0) {
3590 /* Return value is a negative lttng_error_code. */
3596 * Setup lttng message with payload size set to the event list size in
3597 * bytes and then copy list into the llm payload.
3599 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3600 sizeof(struct lttng_event
) * nb_events
);
3610 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3612 struct lttng_event_field
*fields
;
3615 session_lock_list();
3616 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3618 session_unlock_list();
3619 if (nb_fields
< 0) {
3620 /* Return value is a negative lttng_error_code. */
3626 * Setup lttng message with payload size set to the event list size in
3627 * bytes and then copy list into the llm payload.
3629 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, fields
,
3630 sizeof(struct lttng_event_field
) * nb_fields
);
3640 case LTTNG_LIST_SYSCALLS
:
3642 struct lttng_event
*events
;
3645 nb_events
= cmd_list_syscalls(&events
);
3646 if (nb_events
< 0) {
3647 /* Return value is a negative lttng_error_code. */
3653 * Setup lttng message with payload size set to the event list size in
3654 * bytes and then copy list into the llm payload.
3656 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3657 sizeof(struct lttng_event
) * nb_events
);
3667 case LTTNG_LIST_TRACKER_PIDS
:
3669 int32_t *pids
= NULL
;
3672 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3673 cmd_ctx
->lsm
->domain
.type
, &pids
);
3675 /* Return value is a negative lttng_error_code. */
3681 * Setup lttng message with payload size set to the event list size in
3682 * bytes and then copy list into the llm payload.
3684 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, pids
,
3685 sizeof(int32_t) * nr_pids
);
3695 case LTTNG_SET_CONSUMER_URI
:
3698 struct lttng_uri
*uris
;
3700 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3701 len
= nb_uri
* sizeof(struct lttng_uri
);
3704 ret
= LTTNG_ERR_INVALID
;
3708 uris
= zmalloc(len
);
3710 ret
= LTTNG_ERR_FATAL
;
3714 /* Receive variable len data */
3715 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3716 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3718 DBG("No URIs received from client... continuing");
3720 ret
= LTTNG_ERR_SESSION_FAIL
;
3725 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3727 if (ret
!= LTTNG_OK
) {
3734 case LTTNG_START_TRACE
:
3736 ret
= cmd_start_trace(cmd_ctx
->session
);
3739 case LTTNG_STOP_TRACE
:
3741 ret
= cmd_stop_trace(cmd_ctx
->session
);
3744 case LTTNG_CREATE_SESSION
:
3747 struct lttng_uri
*uris
= NULL
;
3749 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3750 len
= nb_uri
* sizeof(struct lttng_uri
);
3753 uris
= zmalloc(len
);
3755 ret
= LTTNG_ERR_FATAL
;
3759 /* Receive variable len data */
3760 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3761 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3763 DBG("No URIs received from client... continuing");
3765 ret
= LTTNG_ERR_SESSION_FAIL
;
3770 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3771 DBG("Creating session with ONE network URI is a bad call");
3772 ret
= LTTNG_ERR_SESSION_FAIL
;
3778 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3779 &cmd_ctx
->creds
, 0);
3785 case LTTNG_DESTROY_SESSION
:
3787 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3789 /* Set session to NULL so we do not unlock it after free. */
3790 cmd_ctx
->session
= NULL
;
3793 case LTTNG_LIST_DOMAINS
:
3796 struct lttng_domain
*domains
= NULL
;
3798 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3800 /* Return value is a negative lttng_error_code. */
3805 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, domains
,
3806 nb_dom
* sizeof(struct lttng_domain
));
3816 case LTTNG_LIST_CHANNELS
:
3818 ssize_t payload_size
;
3819 struct lttng_channel
*channels
= NULL
;
3821 payload_size
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3822 cmd_ctx
->session
, &channels
);
3823 if (payload_size
< 0) {
3824 /* Return value is a negative lttng_error_code. */
3825 ret
= -payload_size
;
3829 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, channels
,
3840 case LTTNG_LIST_EVENTS
:
3843 struct lttng_event
*events
= NULL
;
3844 struct lttcomm_event_command_header cmd_header
;
3847 memset(&cmd_header
, 0, sizeof(cmd_header
));
3848 /* Extended infos are included at the end of events */
3849 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
,
3850 cmd_ctx
->session
, cmd_ctx
->lsm
->u
.list
.channel_name
,
3851 &events
, &total_size
);
3854 /* Return value is a negative lttng_error_code. */
3859 cmd_header
.nb_events
= nb_event
;
3860 ret
= setup_lttng_msg(cmd_ctx
, events
, total_size
,
3861 &cmd_header
, sizeof(cmd_header
));
3871 case LTTNG_LIST_SESSIONS
:
3873 unsigned int nr_sessions
;
3874 void *sessions_payload
;
3877 session_lock_list();
3878 nr_sessions
= lttng_sessions_count(
3879 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3880 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3881 payload_len
= sizeof(struct lttng_session
) * nr_sessions
;
3882 sessions_payload
= zmalloc(payload_len
);
3884 if (!sessions_payload
) {
3885 session_unlock_list();
3890 cmd_list_lttng_sessions(sessions_payload
,
3891 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3892 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3893 session_unlock_list();
3895 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, sessions_payload
,
3897 free(sessions_payload
);
3906 case LTTNG_CALIBRATE
:
3908 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3909 &cmd_ctx
->lsm
->u
.calibrate
);
3912 case LTTNG_REGISTER_CONSUMER
:
3914 struct consumer_data
*cdata
;
3916 switch (cmd_ctx
->lsm
->domain
.type
) {
3917 case LTTNG_DOMAIN_KERNEL
:
3918 cdata
= &kconsumer_data
;
3921 ret
= LTTNG_ERR_UND
;
3925 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3926 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3929 case LTTNG_DATA_PENDING
:
3932 uint8_t pending_ret_byte
;
3934 pending_ret
= cmd_data_pending(cmd_ctx
->session
);
3939 * This function may returns 0 or 1 to indicate whether or not
3940 * there is data pending. In case of error, it should return an
3941 * LTTNG_ERR code. However, some code paths may still return
3942 * a nondescript error code, which we handle by returning an
3945 if (pending_ret
== 0 || pending_ret
== 1) {
3947 * ret will be set to LTTNG_OK at the end of
3950 } else if (pending_ret
< 0) {
3951 ret
= LTTNG_ERR_UNK
;
3958 pending_ret_byte
= (uint8_t) pending_ret
;
3960 /* 1 byte to return whether or not data is pending */
3961 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
,
3962 &pending_ret_byte
, 1);
3971 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3973 struct lttcomm_lttng_output_id reply
;
3975 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3976 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3977 if (ret
!= LTTNG_OK
) {
3981 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &reply
,
3987 /* Copy output list into message payload */
3991 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3993 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3994 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3997 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
4000 struct lttng_snapshot_output
*outputs
= NULL
;
4002 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
4003 if (nb_output
< 0) {
4008 assert((nb_output
> 0 && outputs
) || nb_output
== 0);
4009 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, outputs
,
4010 nb_output
* sizeof(struct lttng_snapshot_output
));
4020 case LTTNG_SNAPSHOT_RECORD
:
4022 ret
= cmd_snapshot_record(cmd_ctx
->session
,
4023 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
4024 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
4027 case LTTNG_CREATE_SESSION_SNAPSHOT
:
4030 struct lttng_uri
*uris
= NULL
;
4032 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4033 len
= nb_uri
* sizeof(struct lttng_uri
);
4036 uris
= zmalloc(len
);
4038 ret
= LTTNG_ERR_FATAL
;
4042 /* Receive variable len data */
4043 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4044 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4046 DBG("No URIs received from client... continuing");
4048 ret
= LTTNG_ERR_SESSION_FAIL
;
4053 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4054 DBG("Creating session with ONE network URI is a bad call");
4055 ret
= LTTNG_ERR_SESSION_FAIL
;
4061 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
4062 nb_uri
, &cmd_ctx
->creds
);
4066 case LTTNG_CREATE_SESSION_LIVE
:
4069 struct lttng_uri
*uris
= NULL
;
4071 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4072 len
= nb_uri
* sizeof(struct lttng_uri
);
4075 uris
= zmalloc(len
);
4077 ret
= LTTNG_ERR_FATAL
;
4081 /* Receive variable len data */
4082 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4083 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4085 DBG("No URIs received from client... continuing");
4087 ret
= LTTNG_ERR_SESSION_FAIL
;
4092 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4093 DBG("Creating session with ONE network URI is a bad call");
4094 ret
= LTTNG_ERR_SESSION_FAIL
;
4100 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4101 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4105 case LTTNG_SAVE_SESSION
:
4107 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4111 case LTTNG_SET_SESSION_SHM_PATH
:
4113 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4114 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4117 case LTTNG_METADATA_REGENERATE
:
4119 ret
= cmd_metadata_regenerate(cmd_ctx
->session
);
4123 ret
= LTTNG_ERR_UND
;
4128 if (cmd_ctx
->llm
== NULL
) {
4129 DBG("Missing llm structure. Allocating one.");
4130 if (setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0) < 0) {
4134 /* Set return code */
4135 cmd_ctx
->llm
->ret_code
= ret
;
4137 if (cmd_ctx
->session
) {
4138 session_unlock(cmd_ctx
->session
);
4140 if (need_tracing_session
) {
4141 session_unlock_list();
4144 assert(!rcu_read_ongoing());
4149 * Thread managing health check socket.
4151 static void *thread_manage_health(void *data
)
4153 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4154 uint32_t revents
, nb_fd
;
4155 struct lttng_poll_event events
;
4156 struct health_comm_msg msg
;
4157 struct health_comm_reply reply
;
4159 DBG("[thread] Manage health check started");
4161 rcu_register_thread();
4163 /* We might hit an error path before this is created. */
4164 lttng_poll_init(&events
);
4166 /* Create unix socket */
4167 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
4169 ERR("Unable to create health check Unix socket");
4174 /* lttng health client socket path permissions */
4175 ret
= chown(health_unix_sock_path
, 0,
4176 utils_get_group_id(tracing_group_name
));
4178 ERR("Unable to set group on %s", health_unix_sock_path
);
4183 ret
= chmod(health_unix_sock_path
,
4184 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4186 ERR("Unable to set permissions on %s", health_unix_sock_path
);
4193 * Set the CLOEXEC flag. Return code is useless because either way, the
4196 (void) utils_set_fd_cloexec(sock
);
4198 ret
= lttcomm_listen_unix_sock(sock
);
4204 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4205 * more will be added to this poll set.
4207 ret
= sessiond_set_thread_pollset(&events
, 2);
4212 /* Add the application registration socket */
4213 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4218 sessiond_notify_ready();
4221 DBG("Health check ready");
4223 /* Inifinite blocking call, waiting for transmission */
4225 ret
= lttng_poll_wait(&events
, -1);
4228 * Restart interrupted system call.
4230 if (errno
== EINTR
) {
4238 for (i
= 0; i
< nb_fd
; i
++) {
4239 /* Fetch once the poll data */
4240 revents
= LTTNG_POLL_GETEV(&events
, i
);
4241 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4244 /* No activity for this FD (poll implementation). */
4248 /* Thread quit pipe has been closed. Killing thread. */
4249 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4255 /* Event on the registration socket */
4256 if (pollfd
== sock
) {
4257 if (revents
& LPOLLIN
) {
4259 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4260 ERR("Health socket poll error");
4263 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4269 new_sock
= lttcomm_accept_unix_sock(sock
);
4275 * Set the CLOEXEC flag. Return code is useless because either way, the
4278 (void) utils_set_fd_cloexec(new_sock
);
4280 DBG("Receiving data from client for health...");
4281 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4283 DBG("Nothing recv() from client... continuing");
4284 ret
= close(new_sock
);
4292 rcu_thread_online();
4294 memset(&reply
, 0, sizeof(reply
));
4295 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4297 * health_check_state returns 0 if health is
4300 if (!health_check_state(health_sessiond
, i
)) {
4301 reply
.ret_code
|= 1ULL << i
;
4305 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4307 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4309 ERR("Failed to send health data back to client");
4312 /* End of transmission */
4313 ret
= close(new_sock
);
4323 ERR("Health error occurred in %s", __func__
);
4325 DBG("Health check thread dying");
4326 unlink(health_unix_sock_path
);
4334 lttng_poll_clean(&events
);
4336 rcu_unregister_thread();
4341 * This thread manage all clients request using the unix client socket for
4344 static void *thread_manage_clients(void *data
)
4346 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4348 uint32_t revents
, nb_fd
;
4349 struct command_ctx
*cmd_ctx
= NULL
;
4350 struct lttng_poll_event events
;
4352 DBG("[thread] Manage client started");
4354 rcu_register_thread();
4356 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4358 health_code_update();
4360 ret
= lttcomm_listen_unix_sock(client_sock
);
4366 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4367 * more will be added to this poll set.
4369 ret
= sessiond_set_thread_pollset(&events
, 2);
4371 goto error_create_poll
;
4374 /* Add the application registration socket */
4375 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4380 sessiond_notify_ready();
4381 ret
= sem_post(&load_info
->message_thread_ready
);
4383 PERROR("sem_post message_thread_ready");
4387 /* This testpoint is after we signal readiness to the parent. */
4388 if (testpoint(sessiond_thread_manage_clients
)) {
4392 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4396 health_code_update();
4399 DBG("Accepting client command ...");
4401 /* Inifinite blocking call, waiting for transmission */
4403 health_poll_entry();
4404 ret
= lttng_poll_wait(&events
, -1);
4408 * Restart interrupted system call.
4410 if (errno
== EINTR
) {
4418 for (i
= 0; i
< nb_fd
; i
++) {
4419 /* Fetch once the poll data */
4420 revents
= LTTNG_POLL_GETEV(&events
, i
);
4421 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4423 health_code_update();
4426 /* No activity for this FD (poll implementation). */
4430 /* Thread quit pipe has been closed. Killing thread. */
4431 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4437 /* Event on the registration socket */
4438 if (pollfd
== client_sock
) {
4439 if (revents
& LPOLLIN
) {
4441 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4442 ERR("Client socket poll error");
4445 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4451 DBG("Wait for client response");
4453 health_code_update();
4455 sock
= lttcomm_accept_unix_sock(client_sock
);
4461 * Set the CLOEXEC flag. Return code is useless because either way, the
4464 (void) utils_set_fd_cloexec(sock
);
4466 /* Set socket option for credentials retrieval */
4467 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4472 /* Allocate context command to process the client request */
4473 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4474 if (cmd_ctx
== NULL
) {
4475 PERROR("zmalloc cmd_ctx");
4479 /* Allocate data buffer for reception */
4480 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4481 if (cmd_ctx
->lsm
== NULL
) {
4482 PERROR("zmalloc cmd_ctx->lsm");
4486 cmd_ctx
->llm
= NULL
;
4487 cmd_ctx
->session
= NULL
;
4489 health_code_update();
4492 * Data is received from the lttng client. The struct
4493 * lttcomm_session_msg (lsm) contains the command and data request of
4496 DBG("Receiving data from client ...");
4497 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4498 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4500 DBG("Nothing recv() from client... continuing");
4506 clean_command_ctx(&cmd_ctx
);
4510 health_code_update();
4512 // TODO: Validate cmd_ctx including sanity check for
4513 // security purpose.
4515 rcu_thread_online();
4517 * This function dispatch the work to the kernel or userspace tracer
4518 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4519 * informations for the client. The command context struct contains
4520 * everything this function may needs.
4522 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4523 rcu_thread_offline();
4531 * TODO: Inform client somehow of the fatal error. At
4532 * this point, ret < 0 means that a zmalloc failed
4533 * (ENOMEM). Error detected but still accept
4534 * command, unless a socket error has been
4537 clean_command_ctx(&cmd_ctx
);
4541 health_code_update();
4543 DBG("Sending response (size: %d, retcode: %s (%d))",
4544 cmd_ctx
->lttng_msg_size
,
4545 lttng_strerror(-cmd_ctx
->llm
->ret_code
),
4546 cmd_ctx
->llm
->ret_code
);
4547 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4549 ERR("Failed to send data back to client");
4552 /* End of transmission */
4559 clean_command_ctx(&cmd_ctx
);
4561 health_code_update();
4573 lttng_poll_clean(&events
);
4574 clean_command_ctx(&cmd_ctx
);
4578 unlink(client_unix_sock_path
);
4579 if (client_sock
>= 0) {
4580 ret
= close(client_sock
);
4588 ERR("Health error occurred in %s", __func__
);
4591 health_unregister(health_sessiond
);
4593 DBG("Client thread dying");
4595 rcu_unregister_thread();
4598 * Since we are creating the consumer threads, we own them, so we need
4599 * to join them before our thread exits.
4601 ret
= join_consumer_thread(&kconsumer_data
);
4604 PERROR("join_consumer");
4607 ret
= join_consumer_thread(&ustconsumer32_data
);
4610 PERROR("join_consumer ust32");
4613 ret
= join_consumer_thread(&ustconsumer64_data
);
4616 PERROR("join_consumer ust64");
4621 static int string_match(const char *str1
, const char *str2
)
4623 return (str1
&& str2
) && !strcmp(str1
, str2
);
4627 * Take an option from the getopt output and set it in the right variable to be
4630 * Return 0 on success else a negative value.
4632 static int set_option(int opt
, const char *arg
, const char *optname
)
4636 if (string_match(optname
, "client-sock") || opt
== 'c') {
4637 if (!arg
|| *arg
== '\0') {
4641 if (lttng_is_setuid_setgid()) {
4642 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4643 "-c, --client-sock");
4645 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4647 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4648 if (!arg
|| *arg
== '\0') {
4652 if (lttng_is_setuid_setgid()) {
4653 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4656 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4658 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4660 } else if (string_match(optname
, "background") || opt
== 'b') {
4662 } else if (string_match(optname
, "group") || opt
== 'g') {
4663 if (!arg
|| *arg
== '\0') {
4667 if (lttng_is_setuid_setgid()) {
4668 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4672 * If the override option is set, the pointer points to a
4673 * *non* const thus freeing it even though the variable type is
4676 if (tracing_group_name_override
) {
4677 free((void *) tracing_group_name
);
4679 tracing_group_name
= strdup(arg
);
4680 if (!tracing_group_name
) {
4684 tracing_group_name_override
= 1;
4686 } else if (string_match(optname
, "help") || opt
== 'h') {
4687 ret
= utils_show_man_page(8, "lttng-sessiond");
4689 ERR("Cannot view man page lttng-sessiond(8)");
4692 exit(ret
? EXIT_FAILURE
: EXIT_SUCCESS
);
4693 } else if (string_match(optname
, "version") || opt
== 'V') {
4694 fprintf(stdout
, "%s\n", VERSION
);
4696 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4698 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4699 if (!arg
|| *arg
== '\0') {
4703 if (lttng_is_setuid_setgid()) {
4704 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4705 "--kconsumerd-err-sock");
4707 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4709 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4710 if (!arg
|| *arg
== '\0') {
4714 if (lttng_is_setuid_setgid()) {
4715 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4716 "--kconsumerd-cmd-sock");
4718 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4720 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4721 if (!arg
|| *arg
== '\0') {
4725 if (lttng_is_setuid_setgid()) {
4726 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4727 "--ustconsumerd64-err-sock");
4729 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4731 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4732 if (!arg
|| *arg
== '\0') {
4736 if (lttng_is_setuid_setgid()) {
4737 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4738 "--ustconsumerd64-cmd-sock");
4740 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4742 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4743 if (!arg
|| *arg
== '\0') {
4747 if (lttng_is_setuid_setgid()) {
4748 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4749 "--ustconsumerd32-err-sock");
4751 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4753 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4754 if (!arg
|| *arg
== '\0') {
4758 if (lttng_is_setuid_setgid()) {
4759 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4760 "--ustconsumerd32-cmd-sock");
4762 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4764 } else if (string_match(optname
, "no-kernel")) {
4766 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4767 lttng_opt_quiet
= 1;
4768 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4769 /* Verbose level can increase using multiple -v */
4771 /* Value obtained from config file */
4772 lttng_opt_verbose
= config_parse_value(arg
);
4774 /* -v used on command line */
4775 lttng_opt_verbose
++;
4777 /* Clamp value to [0, 3] */
4778 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4779 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4780 } else if (string_match(optname
, "verbose-consumer")) {
4782 opt_verbose_consumer
= config_parse_value(arg
);
4784 opt_verbose_consumer
++;
4786 } else if (string_match(optname
, "consumerd32-path")) {
4787 if (!arg
|| *arg
== '\0') {
4791 if (lttng_is_setuid_setgid()) {
4792 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4793 "--consumerd32-path");
4795 if (consumerd32_bin_override
) {
4796 free((void *) consumerd32_bin
);
4798 consumerd32_bin
= strdup(arg
);
4799 if (!consumerd32_bin
) {
4803 consumerd32_bin_override
= 1;
4805 } else if (string_match(optname
, "consumerd32-libdir")) {
4806 if (!arg
|| *arg
== '\0') {
4810 if (lttng_is_setuid_setgid()) {
4811 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4812 "--consumerd32-libdir");
4814 if (consumerd32_libdir_override
) {
4815 free((void *) consumerd32_libdir
);
4817 consumerd32_libdir
= strdup(arg
);
4818 if (!consumerd32_libdir
) {
4822 consumerd32_libdir_override
= 1;
4824 } else if (string_match(optname
, "consumerd64-path")) {
4825 if (!arg
|| *arg
== '\0') {
4829 if (lttng_is_setuid_setgid()) {
4830 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4831 "--consumerd64-path");
4833 if (consumerd64_bin_override
) {
4834 free((void *) consumerd64_bin
);
4836 consumerd64_bin
= strdup(arg
);
4837 if (!consumerd64_bin
) {
4841 consumerd64_bin_override
= 1;
4843 } else if (string_match(optname
, "consumerd64-libdir")) {
4844 if (!arg
|| *arg
== '\0') {
4848 if (lttng_is_setuid_setgid()) {
4849 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4850 "--consumerd64-libdir");
4852 if (consumerd64_libdir_override
) {
4853 free((void *) consumerd64_libdir
);
4855 consumerd64_libdir
= strdup(arg
);
4856 if (!consumerd64_libdir
) {
4860 consumerd64_libdir_override
= 1;
4862 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4863 if (!arg
|| *arg
== '\0') {
4867 if (lttng_is_setuid_setgid()) {
4868 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4872 opt_pidfile
= strdup(arg
);
4878 } else if (string_match(optname
, "agent-tcp-port")) {
4879 if (!arg
|| *arg
== '\0') {
4883 if (lttng_is_setuid_setgid()) {
4884 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4885 "--agent-tcp-port");
4890 v
= strtoul(arg
, NULL
, 0);
4891 if (errno
!= 0 || !isdigit(arg
[0])) {
4892 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4895 if (v
== 0 || v
>= 65535) {
4896 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4899 agent_tcp_port
= (uint32_t) v
;
4900 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4902 } else if (string_match(optname
, "load") || opt
== 'l') {
4903 if (!arg
|| *arg
== '\0') {
4907 if (lttng_is_setuid_setgid()) {
4908 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4911 free(opt_load_session_path
);
4912 opt_load_session_path
= strdup(arg
);
4913 if (!opt_load_session_path
) {
4918 } else if (string_match(optname
, "kmod-probes")) {
4919 if (!arg
|| *arg
== '\0') {
4923 if (lttng_is_setuid_setgid()) {
4924 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4927 free(kmod_probes_list
);
4928 kmod_probes_list
= strdup(arg
);
4929 if (!kmod_probes_list
) {
4934 } else if (string_match(optname
, "extra-kmod-probes")) {
4935 if (!arg
|| *arg
== '\0') {
4939 if (lttng_is_setuid_setgid()) {
4940 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4941 "--extra-kmod-probes");
4943 free(kmod_extra_probes_list
);
4944 kmod_extra_probes_list
= strdup(arg
);
4945 if (!kmod_extra_probes_list
) {
4950 } else if (string_match(optname
, "config") || opt
== 'f') {
4951 /* This is handled in set_options() thus silent skip. */
4954 /* Unknown option or other error.
4955 * Error is printed by getopt, just return */
4960 if (ret
== -EINVAL
) {
4961 const char *opt_name
= "unknown";
4964 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
4966 if (opt
== long_options
[i
].val
) {
4967 opt_name
= long_options
[i
].name
;
4972 WARN("Invalid argument provided for option \"%s\", using default value.",
4980 * config_entry_handler_cb used to handle options read from a config file.
4981 * See config_entry_handler_cb comment in common/config/session-config.h for the
4982 * return value conventions.
4984 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4988 if (!entry
|| !entry
->name
|| !entry
->value
) {
4993 /* Check if the option is to be ignored */
4994 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4995 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
5000 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
5003 /* Ignore if not fully matched. */
5004 if (strcmp(entry
->name
, long_options
[i
].name
)) {
5009 * If the option takes no argument on the command line, we have to
5010 * check if the value is "true". We support non-zero numeric values,
5013 if (!long_options
[i
].has_arg
) {
5014 ret
= config_parse_value(entry
->value
);
5017 WARN("Invalid configuration value \"%s\" for option %s",
5018 entry
->value
, entry
->name
);
5020 /* False, skip boolean config option. */
5025 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
5029 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
5036 * daemon configuration loading and argument parsing
5038 static int set_options(int argc
, char **argv
)
5040 int ret
= 0, c
= 0, option_index
= 0;
5041 int orig_optopt
= optopt
, orig_optind
= optind
;
5043 const char *config_path
= NULL
;
5045 optstring
= utils_generate_optstring(long_options
,
5046 sizeof(long_options
) / sizeof(struct option
));
5052 /* Check for the --config option */
5053 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
5054 &option_index
)) != -1) {
5058 } else if (c
!= 'f') {
5059 /* if not equal to --config option. */
5063 if (lttng_is_setuid_setgid()) {
5064 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5067 config_path
= utils_expand_path(optarg
);
5069 ERR("Failed to resolve path: %s", optarg
);
5074 ret
= config_get_section_entries(config_path
, config_section_name
,
5075 config_entry_handler
, NULL
);
5078 ERR("Invalid configuration option at line %i", ret
);
5084 /* Reset getopt's global state */
5085 optopt
= orig_optopt
;
5086 optind
= orig_optind
;
5090 * getopt_long() will not set option_index if it encounters a
5093 c
= getopt_long(argc
, argv
, optstring
, long_options
,
5100 * Pass NULL as the long option name if popt left the index
5103 ret
= set_option(c
, optarg
,
5104 option_index
< 0 ? NULL
:
5105 long_options
[option_index
].name
);
5117 * Creates the two needed socket by the daemon.
5118 * apps_sock - The communication socket for all UST apps.
5119 * client_sock - The communication of the cli tool (lttng).
5121 static int init_daemon_socket(void)
5126 old_umask
= umask(0);
5128 /* Create client tool unix socket */
5129 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
5130 if (client_sock
< 0) {
5131 ERR("Create unix sock failed: %s", client_unix_sock_path
);
5136 /* Set the cloexec flag */
5137 ret
= utils_set_fd_cloexec(client_sock
);
5139 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5140 "Continuing but note that the consumer daemon will have a "
5141 "reference to this socket on exec()", client_sock
);
5144 /* File permission MUST be 660 */
5145 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5147 ERR("Set file permissions failed: %s", client_unix_sock_path
);
5152 /* Create the application unix socket */
5153 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
5154 if (apps_sock
< 0) {
5155 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
5160 /* Set the cloexec flag */
5161 ret
= utils_set_fd_cloexec(apps_sock
);
5163 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5164 "Continuing but note that the consumer daemon will have a "
5165 "reference to this socket on exec()", apps_sock
);
5168 /* File permission MUST be 666 */
5169 ret
= chmod(apps_unix_sock_path
,
5170 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5172 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
5177 DBG3("Session daemon client socket %d and application socket %d created",
5178 client_sock
, apps_sock
);
5186 * Check if the global socket is available, and if a daemon is answering at the
5187 * other side. If yes, error is returned.
5189 static int check_existing_daemon(void)
5191 /* Is there anybody out there ? */
5192 if (lttng_session_daemon_alive()) {
5200 * Set the tracing group gid onto the client socket.
5202 * Race window between mkdir and chown is OK because we are going from more
5203 * permissive (root.root) to less permissive (root.tracing).
5205 static int set_permissions(char *rundir
)
5210 gid
= utils_get_group_id(tracing_group_name
);
5212 /* Set lttng run dir */
5213 ret
= chown(rundir
, 0, gid
);
5215 ERR("Unable to set group on %s", rundir
);
5220 * Ensure all applications and tracing group can search the run
5221 * dir. Allow everyone to read the directory, since it does not
5222 * buy us anything to hide its content.
5224 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5226 ERR("Unable to set permissions on %s", rundir
);
5230 /* lttng client socket path */
5231 ret
= chown(client_unix_sock_path
, 0, gid
);
5233 ERR("Unable to set group on %s", client_unix_sock_path
);
5237 /* kconsumer error socket path */
5238 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5240 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5244 /* 64-bit ustconsumer error socket path */
5245 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5247 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5251 /* 32-bit ustconsumer compat32 error socket path */
5252 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5254 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5258 DBG("All permissions are set");
5264 * Create the lttng run directory needed for all global sockets and pipe.
5266 static int create_lttng_rundir(const char *rundir
)
5270 DBG3("Creating LTTng run directory: %s", rundir
);
5272 ret
= mkdir(rundir
, S_IRWXU
);
5274 if (errno
!= EEXIST
) {
5275 ERR("Unable to create %s", rundir
);
5287 * Setup sockets and directory needed by the kconsumerd communication with the
5290 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5294 char path
[PATH_MAX
];
5296 switch (consumer_data
->type
) {
5297 case LTTNG_CONSUMER_KERNEL
:
5298 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5300 case LTTNG_CONSUMER64_UST
:
5301 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5303 case LTTNG_CONSUMER32_UST
:
5304 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5307 ERR("Consumer type unknown");
5312 DBG2("Creating consumer directory: %s", path
);
5314 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5316 if (errno
!= EEXIST
) {
5318 ERR("Failed to create %s", path
);
5324 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5326 ERR("Unable to set group on %s", path
);
5332 /* Create the kconsumerd error unix socket */
5333 consumer_data
->err_sock
=
5334 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5335 if (consumer_data
->err_sock
< 0) {
5336 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5342 * Set the CLOEXEC flag. Return code is useless because either way, the
5345 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5347 PERROR("utils_set_fd_cloexec");
5348 /* continue anyway */
5351 /* File permission MUST be 660 */
5352 ret
= chmod(consumer_data
->err_unix_sock_path
,
5353 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5355 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5365 * Signal handler for the daemon
5367 * Simply stop all worker threads, leaving main() return gracefully after
5368 * joining all threads and calling cleanup().
5370 static void sighandler(int sig
)
5374 DBG("SIGPIPE caught");
5377 DBG("SIGINT caught");
5381 DBG("SIGTERM caught");
5385 CMM_STORE_SHARED(recv_child_signal
, 1);
5393 * Setup signal handler for :
5394 * SIGINT, SIGTERM, SIGPIPE
5396 static int set_signal_handler(void)
5399 struct sigaction sa
;
5402 if ((ret
= sigemptyset(&sigset
)) < 0) {
5403 PERROR("sigemptyset");
5407 sa
.sa_handler
= sighandler
;
5408 sa
.sa_mask
= sigset
;
5410 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5411 PERROR("sigaction");
5415 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5416 PERROR("sigaction");
5420 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5421 PERROR("sigaction");
5425 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5426 PERROR("sigaction");
5430 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5436 * Set open files limit to unlimited. This daemon can open a large number of
5437 * file descriptors in order to consume multiple kernel traces.
5439 static void set_ulimit(void)
5444 /* The kernel does not allow an infinite limit for open files */
5445 lim
.rlim_cur
= 65535;
5446 lim
.rlim_max
= 65535;
5448 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5450 PERROR("failed to set open files limit");
5455 * Write pidfile using the rundir and opt_pidfile.
5457 static int write_pidfile(void)
5460 char pidfile_path
[PATH_MAX
];
5465 if (lttng_strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
))) {
5470 /* Build pidfile path from rundir and opt_pidfile. */
5471 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5472 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5474 PERROR("snprintf pidfile path");
5480 * Create pid file in rundir.
5482 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5488 * Create lockfile using the rundir and return its fd.
5490 static int create_lockfile(void)
5493 char lockfile_path
[PATH_MAX
];
5495 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5500 ret
= utils_create_lock_file(lockfile_path
);
5506 * Write agent TCP port using the rundir.
5508 static int write_agent_port(void)
5511 char path
[PATH_MAX
];
5515 ret
= snprintf(path
, sizeof(path
), "%s/"
5516 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5518 PERROR("snprintf agent port path");
5523 * Create TCP agent port file in rundir.
5525 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5534 int main(int argc
, char **argv
)
5536 int ret
= 0, retval
= 0;
5538 const char *home_path
, *env_app_timeout
;
5540 init_kernel_workarounds();
5542 rcu_register_thread();
5544 if (set_signal_handler()) {
5546 goto exit_set_signal_handler
;
5549 setup_consumerd_path();
5551 page_size
= sysconf(_SC_PAGESIZE
);
5552 if (page_size
< 0) {
5553 PERROR("sysconf _SC_PAGESIZE");
5554 page_size
= LONG_MAX
;
5555 WARN("Fallback page size to %ld", page_size
);
5559 * Parse arguments and load the daemon configuration file.
5561 * We have an exit_options exit path to free memory reserved by
5562 * set_options. This is needed because the rest of sessiond_cleanup()
5563 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5564 * depends on set_options.
5567 if (set_options(argc
, argv
)) {
5573 if (opt_daemon
|| opt_background
) {
5576 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5584 * We are in the child. Make sure all other file descriptors are
5585 * closed, in case we are called with more opened file
5586 * descriptors than the standard ones.
5588 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5593 if (run_as_create_worker(argv
[0]) < 0) {
5594 goto exit_create_run_as_worker_cleanup
;
5598 * Starting from here, we can create threads. This needs to be after
5599 * lttng_daemonize due to RCU.
5603 * Initialize the health check subsystem. This call should set the
5604 * appropriate time values.
5606 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5607 if (!health_sessiond
) {
5608 PERROR("health_app_create error");
5610 goto exit_health_sessiond_cleanup
;
5613 /* Create thread to clean up RCU hash tables */
5614 if (init_ht_cleanup_thread(&ht_cleanup_thread
)) {
5616 goto exit_ht_cleanup
;
5619 /* Create thread quit pipe */
5620 if (init_thread_quit_pipe()) {
5622 goto exit_init_data
;
5625 /* Check if daemon is UID = 0 */
5626 is_root
= !getuid();
5629 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5632 goto exit_init_data
;
5635 /* Create global run dir with root access */
5636 if (create_lttng_rundir(rundir
)) {
5638 goto exit_init_data
;
5641 if (strlen(apps_unix_sock_path
) == 0) {
5642 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5643 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5646 goto exit_init_data
;
5650 if (strlen(client_unix_sock_path
) == 0) {
5651 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5652 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5655 goto exit_init_data
;
5659 /* Set global SHM for ust */
5660 if (strlen(wait_shm_path
) == 0) {
5661 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5662 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5665 goto exit_init_data
;
5669 if (strlen(health_unix_sock_path
) == 0) {
5670 ret
= snprintf(health_unix_sock_path
,
5671 sizeof(health_unix_sock_path
),
5672 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5675 goto exit_init_data
;
5679 /* Setup kernel consumerd path */
5680 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5681 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5684 goto exit_init_data
;
5686 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5687 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5690 goto exit_init_data
;
5693 DBG2("Kernel consumer err path: %s",
5694 kconsumer_data
.err_unix_sock_path
);
5695 DBG2("Kernel consumer cmd path: %s",
5696 kconsumer_data
.cmd_unix_sock_path
);
5698 home_path
= utils_get_home_dir();
5699 if (home_path
== NULL
) {
5700 /* TODO: Add --socket PATH option */
5701 ERR("Can't get HOME directory for sockets creation.");
5703 goto exit_init_data
;
5707 * Create rundir from home path. This will create something like
5710 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5713 goto exit_init_data
;
5716 if (create_lttng_rundir(rundir
)) {
5718 goto exit_init_data
;
5721 if (strlen(apps_unix_sock_path
) == 0) {
5722 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5723 DEFAULT_HOME_APPS_UNIX_SOCK
,
5727 goto exit_init_data
;
5731 /* Set the cli tool unix socket path */
5732 if (strlen(client_unix_sock_path
) == 0) {
5733 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5734 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5738 goto exit_init_data
;
5742 /* Set global SHM for ust */
5743 if (strlen(wait_shm_path
) == 0) {
5744 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5745 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5749 goto exit_init_data
;
5753 /* Set health check Unix path */
5754 if (strlen(health_unix_sock_path
) == 0) {
5755 ret
= snprintf(health_unix_sock_path
,
5756 sizeof(health_unix_sock_path
),
5757 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5761 goto exit_init_data
;
5766 lockfile_fd
= create_lockfile();
5767 if (lockfile_fd
< 0) {
5769 goto exit_init_data
;
5772 /* Set consumer initial state */
5773 kernel_consumerd_state
= CONSUMER_STOPPED
;
5774 ust_consumerd_state
= CONSUMER_STOPPED
;
5776 DBG("Client socket path %s", client_unix_sock_path
);
5777 DBG("Application socket path %s", apps_unix_sock_path
);
5778 DBG("Application wait path %s", wait_shm_path
);
5779 DBG("LTTng run directory path: %s", rundir
);
5781 /* 32 bits consumerd path setup */
5782 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5783 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5785 PERROR("snprintf 32-bit consumer error socket path");
5787 goto exit_init_data
;
5789 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5790 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5792 PERROR("snprintf 32-bit consumer command socket path");
5794 goto exit_init_data
;
5797 DBG2("UST consumer 32 bits err path: %s",
5798 ustconsumer32_data
.err_unix_sock_path
);
5799 DBG2("UST consumer 32 bits cmd path: %s",
5800 ustconsumer32_data
.cmd_unix_sock_path
);
5802 /* 64 bits consumerd path setup */
5803 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5804 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5806 PERROR("snprintf 64-bit consumer error socket path");
5808 goto exit_init_data
;
5810 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5811 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5813 PERROR("snprintf 64-bit consumer command socket path");
5815 goto exit_init_data
;
5818 DBG2("UST consumer 64 bits err path: %s",
5819 ustconsumer64_data
.err_unix_sock_path
);
5820 DBG2("UST consumer 64 bits cmd path: %s",
5821 ustconsumer64_data
.cmd_unix_sock_path
);
5824 * See if daemon already exist.
5826 if (check_existing_daemon()) {
5827 ERR("Already running daemon.\n");
5829 * We do not goto exit because we must not cleanup()
5830 * because a daemon is already running.
5833 goto exit_init_data
;
5837 * Init UST app hash table. Alloc hash table before this point since
5838 * cleanup() can get called after that point.
5840 if (ust_app_ht_alloc()) {
5841 ERR("Failed to allocate UST app hash table");
5843 goto exit_init_data
;
5847 * Initialize agent app hash table. We allocate the hash table here
5848 * since cleanup() can get called after this point.
5850 if (agent_app_ht_alloc()) {
5851 ERR("Failed to allocate Agent app hash table");
5853 goto exit_init_data
;
5857 * These actions must be executed as root. We do that *after* setting up
5858 * the sockets path because we MUST make the check for another daemon using
5859 * those paths *before* trying to set the kernel consumer sockets and init
5863 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
5865 goto exit_init_data
;
5868 /* Setup kernel tracer */
5869 if (!opt_no_kernel
) {
5870 init_kernel_tracer();
5871 if (kernel_tracer_fd
>= 0) {
5872 ret
= syscall_init_table();
5874 ERR("Unable to populate syscall table. "
5875 "Syscall tracing won't work "
5876 "for this session daemon.");
5881 /* Set ulimit for open files */
5884 /* init lttng_fd tracking must be done after set_ulimit. */
5887 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
5889 goto exit_init_data
;
5892 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
5894 goto exit_init_data
;
5897 /* Setup the needed unix socket */
5898 if (init_daemon_socket()) {
5900 goto exit_init_data
;
5903 /* Set credentials to socket */
5904 if (is_root
&& set_permissions(rundir
)) {
5906 goto exit_init_data
;
5909 /* Get parent pid if -S, --sig-parent is specified. */
5910 if (opt_sig_parent
) {
5914 /* Setup the kernel pipe for waking up the kernel thread */
5915 if (is_root
&& !opt_no_kernel
) {
5916 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5918 goto exit_init_data
;
5922 /* Setup the thread apps communication pipe. */
5923 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
5925 goto exit_init_data
;
5928 /* Setup the thread apps notify communication pipe. */
5929 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
5931 goto exit_init_data
;
5934 /* Initialize global buffer per UID and PID registry. */
5935 buffer_reg_init_uid_registry();
5936 buffer_reg_init_pid_registry();
5938 /* Init UST command queue. */
5939 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
5942 * Get session list pointer. This pointer MUST NOT be free'd. This list
5943 * is statically declared in session.c
5945 session_list_ptr
= session_get_list();
5949 /* Check for the application socket timeout env variable. */
5950 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5951 if (env_app_timeout
) {
5952 app_socket_timeout
= atoi(env_app_timeout
);
5954 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5957 ret
= write_pidfile();
5959 ERR("Error in write_pidfile");
5961 goto exit_init_data
;
5963 ret
= write_agent_port();
5965 ERR("Error in write_agent_port");
5967 goto exit_init_data
;
5970 /* Initialize communication library */
5972 /* Initialize TCP timeout values */
5973 lttcomm_inet_init();
5975 if (load_session_init_data(&load_info
) < 0) {
5977 goto exit_init_data
;
5979 load_info
->path
= opt_load_session_path
;
5981 /* Create health-check thread */
5982 ret
= pthread_create(&health_thread
, NULL
,
5983 thread_manage_health
, (void *) NULL
);
5986 PERROR("pthread_create health");
5991 /* Create thread to manage the client socket */
5992 ret
= pthread_create(&client_thread
, NULL
,
5993 thread_manage_clients
, (void *) NULL
);
5996 PERROR("pthread_create clients");
6001 /* Create thread to dispatch registration */
6002 ret
= pthread_create(&dispatch_thread
, NULL
,
6003 thread_dispatch_ust_registration
, (void *) NULL
);
6006 PERROR("pthread_create dispatch");
6011 /* Create thread to manage application registration. */
6012 ret
= pthread_create(®_apps_thread
, NULL
,
6013 thread_registration_apps
, (void *) NULL
);
6016 PERROR("pthread_create registration");
6021 /* Create thread to manage application socket */
6022 ret
= pthread_create(&apps_thread
, NULL
,
6023 thread_manage_apps
, (void *) NULL
);
6026 PERROR("pthread_create apps");
6031 /* Create thread to manage application notify socket */
6032 ret
= pthread_create(&apps_notify_thread
, NULL
,
6033 ust_thread_manage_notify
, (void *) NULL
);
6036 PERROR("pthread_create notify");
6038 goto exit_apps_notify
;
6041 /* Create agent registration thread. */
6042 ret
= pthread_create(&agent_reg_thread
, NULL
,
6043 agent_thread_manage_registration
, (void *) NULL
);
6046 PERROR("pthread_create agent");
6048 goto exit_agent_reg
;
6051 /* Don't start this thread if kernel tracing is not requested nor root */
6052 if (is_root
&& !opt_no_kernel
) {
6053 /* Create kernel thread to manage kernel event */
6054 ret
= pthread_create(&kernel_thread
, NULL
,
6055 thread_manage_kernel
, (void *) NULL
);
6058 PERROR("pthread_create kernel");
6064 /* Create session loading thread. */
6065 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
6069 PERROR("pthread_create load_session_thread");
6071 goto exit_load_session
;
6075 * This is where we start awaiting program completion (e.g. through
6076 * signal that asks threads to teardown).
6079 ret
= pthread_join(load_session_thread
, &status
);
6082 PERROR("pthread_join load_session_thread");
6087 if (is_root
&& !opt_no_kernel
) {
6088 ret
= pthread_join(kernel_thread
, &status
);
6091 PERROR("pthread_join");
6097 ret
= pthread_join(agent_reg_thread
, &status
);
6100 PERROR("pthread_join agent");
6105 ret
= pthread_join(apps_notify_thread
, &status
);
6108 PERROR("pthread_join apps notify");
6113 ret
= pthread_join(apps_thread
, &status
);
6116 PERROR("pthread_join apps");
6121 ret
= pthread_join(reg_apps_thread
, &status
);
6124 PERROR("pthread_join");
6130 * Join dispatch thread after joining reg_apps_thread to ensure
6131 * we don't leak applications in the queue.
6133 ret
= pthread_join(dispatch_thread
, &status
);
6136 PERROR("pthread_join");
6141 ret
= pthread_join(client_thread
, &status
);
6144 PERROR("pthread_join");
6149 ret
= pthread_join(health_thread
, &status
);
6152 PERROR("pthread_join health thread");
6159 * sessiond_cleanup() is called when no other thread is running, except
6160 * the ht_cleanup thread, which is needed to destroy the hash tables.
6162 rcu_thread_online();
6164 rcu_thread_offline();
6165 rcu_unregister_thread();
6168 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6169 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6170 * the queue is empty before shutting down the clean-up thread.
6174 ret
= fini_ht_cleanup_thread(&ht_cleanup_thread
);
6180 health_app_destroy(health_sessiond
);
6181 exit_health_sessiond_cleanup
:
6182 exit_create_run_as_worker_cleanup
:
6185 sessiond_cleanup_options();
6187 exit_set_signal_handler
: