2 * Copyright (C) 2011 EfficiOS Inc.
3 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * Copyright (C) 2013 Jérémie Galarneau <jeremie.galarneau@efficios.com>
6 * SPDX-License-Identifier: GPL-2.0-only
12 #include <common/pipe.hpp>
13 #include <common/utils.hpp>
15 #include "manage-consumer.hpp"
16 #include "testpoint.hpp"
17 #include "health-sessiond.hpp"
20 #include "ust-consumer.hpp"
23 struct thread_notifiers
{
24 struct lttng_pipe
*quit_pipe
;
25 struct consumer_data
*consumer_data
;
27 int initialization_result
;
31 static void mark_thread_as_ready(struct thread_notifiers
*notifiers
)
33 DBG("Marking consumer management thread as ready");
34 notifiers
->initialization_result
= 0;
35 sem_post(¬ifiers
->ready
);
38 static void mark_thread_intialization_as_failed(
39 struct thread_notifiers
*notifiers
)
41 ERR("Consumer management thread entering error state");
42 notifiers
->initialization_result
= -1;
43 sem_post(¬ifiers
->ready
);
46 static void wait_until_thread_is_ready(struct thread_notifiers
*notifiers
)
48 DBG("Waiting for consumer management thread to be ready");
49 sem_wait(¬ifiers
->ready
);
50 DBG("Consumer management thread is ready");
54 * This thread manage the consumer error sent back to the session daemon.
56 static void *thread_consumer_management(void *data
)
58 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
59 uint32_t revents
, nb_fd
;
60 enum lttcomm_return_code code
;
61 struct lttng_poll_event events
;
62 struct thread_notifiers
*notifiers
= (thread_notifiers
*) data
;
63 struct consumer_data
*consumer_data
= notifiers
->consumer_data
;
64 const int quit_pipe_read_fd
= lttng_pipe_get_readfd(notifiers
->quit_pipe
);
65 struct consumer_socket
*cmd_socket_wrapper
= NULL
;
67 DBG("[thread] Manage consumer started");
69 rcu_register_thread();
72 health_register(the_health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
77 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
78 * metadata_sock. Nothing more will be added to this poll set.
80 ret
= lttng_poll_create(&events
, 3, LTTNG_CLOEXEC
);
82 mark_thread_intialization_as_failed(notifiers
);
86 ret
= lttng_poll_add(&events
, quit_pipe_read_fd
, LPOLLIN
| LPOLLERR
);
88 mark_thread_intialization_as_failed(notifiers
);
93 * The error socket here is already in a listening state which was done
94 * just before spawning this thread to avoid a race between the consumer
95 * daemon exec trying to connect and the listen() call.
97 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
99 mark_thread_intialization_as_failed(notifiers
);
103 health_code_update();
105 /* Infinite blocking call, waiting for transmission */
108 if (testpoint(sessiond_thread_manage_consumer
)) {
109 mark_thread_intialization_as_failed(notifiers
);
113 ret
= lttng_poll_wait(&events
, -1);
116 mark_thread_intialization_as_failed(notifiers
);
122 for (i
= 0; i
< nb_fd
; i
++) {
123 /* Fetch once the poll data */
124 revents
= LTTNG_POLL_GETEV(&events
, i
);
125 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
127 health_code_update();
129 /* Thread quit pipe has been closed. Killing thread. */
130 if (pollfd
== quit_pipe_read_fd
) {
132 mark_thread_intialization_as_failed(notifiers
);
134 } else if (pollfd
== consumer_data
->err_sock
) {
135 /* Event on the registration socket */
136 if (revents
& LPOLLIN
) {
138 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
139 ERR("consumer err socket poll error");
140 mark_thread_intialization_as_failed(notifiers
);
143 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
144 mark_thread_intialization_as_failed(notifiers
);
150 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
152 mark_thread_intialization_as_failed(notifiers
);
157 * Set the CLOEXEC flag. Return code is useless because either way, the
160 (void) utils_set_fd_cloexec(sock
);
162 health_code_update();
164 DBG2("Receiving code from consumer err_sock");
166 /* Getting status code from kconsumerd */
167 ret
= lttcomm_recv_unix_sock(sock
, &code
,
168 sizeof(enum lttcomm_return_code
));
170 mark_thread_intialization_as_failed(notifiers
);
174 health_code_update();
175 if (code
!= LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
176 ERR("consumer error when waiting for SOCK_READY : %s",
177 lttcomm_get_readable_code((lttcomm_return_code
) -code
));
178 mark_thread_intialization_as_failed(notifiers
);
182 /* Connect both command and metadata sockets. */
183 consumer_data
->cmd_sock
=
184 lttcomm_connect_unix_sock(
185 consumer_data
->cmd_unix_sock_path
);
186 consumer_data
->metadata_fd
=
187 lttcomm_connect_unix_sock(
188 consumer_data
->cmd_unix_sock_path
);
189 if (consumer_data
->cmd_sock
< 0 || consumer_data
->metadata_fd
< 0) {
190 PERROR("consumer connect cmd socket");
191 mark_thread_intialization_as_failed(notifiers
);
195 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
197 /* Create metadata socket lock. */
198 consumer_data
->metadata_sock
.lock
= zmalloc
<pthread_mutex_t
>();
199 if (consumer_data
->metadata_sock
.lock
== NULL
) {
200 PERROR("zmalloc pthread mutex");
201 mark_thread_intialization_as_failed(notifiers
);
204 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
206 DBG("Consumer command socket ready (fd: %d)", consumer_data
->cmd_sock
);
207 DBG("Consumer metadata socket ready (fd: %d)",
208 consumer_data
->metadata_fd
);
211 * Remove the consumerd error sock since we've established a connection.
213 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
215 mark_thread_intialization_as_failed(notifiers
);
219 /* Add new accepted error socket. */
220 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
222 mark_thread_intialization_as_failed(notifiers
);
226 /* Add metadata socket that is successfully connected. */
227 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
228 LPOLLIN
| LPOLLRDHUP
);
230 mark_thread_intialization_as_failed(notifiers
);
234 health_code_update();
237 * Transfer the write-end of the channel monitoring pipe to the consumer
238 * by issuing a SET_CHANNEL_MONITOR_PIPE command.
240 cmd_socket_wrapper
= consumer_allocate_socket(&consumer_data
->cmd_sock
);
241 if (!cmd_socket_wrapper
) {
242 mark_thread_intialization_as_failed(notifiers
);
245 cmd_socket_wrapper
->lock
= &consumer_data
->lock
;
247 pthread_mutex_lock(cmd_socket_wrapper
->lock
);
248 ret
= consumer_init(cmd_socket_wrapper
, the_sessiond_uuid
);
250 ERR("Failed to send sessiond uuid to consumer daemon");
251 mark_thread_intialization_as_failed(notifiers
);
252 pthread_mutex_unlock(cmd_socket_wrapper
->lock
);
255 pthread_mutex_unlock(cmd_socket_wrapper
->lock
);
257 ret
= consumer_send_channel_monitor_pipe(cmd_socket_wrapper
,
258 consumer_data
->channel_monitor_pipe
);
260 mark_thread_intialization_as_failed(notifiers
);
264 /* Discard the socket wrapper as it is no longer needed. */
265 consumer_destroy_socket(cmd_socket_wrapper
);
266 cmd_socket_wrapper
= NULL
;
268 /* The thread is completely initialized, signal that it is ready. */
269 mark_thread_as_ready(notifiers
);
271 /* Infinite blocking call, waiting for transmission */
273 health_code_update();
275 /* Exit the thread because the thread quit pipe has been triggered. */
277 /* Not a health error. */
283 ret
= lttng_poll_wait(&events
, -1);
291 for (i
= 0; i
< nb_fd
; i
++) {
292 /* Fetch once the poll data */
293 revents
= LTTNG_POLL_GETEV(&events
, i
);
294 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
296 health_code_update();
299 * Thread quit pipe has been triggered, flag that we should stop
300 * but continue the current loop to handle potential data from
303 if (pollfd
== quit_pipe_read_fd
) {
305 } else if (pollfd
== sock
) {
306 /* Event on the consumerd socket */
307 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
308 && !(revents
& LPOLLIN
)) {
309 ERR("consumer err socket second poll error");
312 health_code_update();
313 /* Wait for any kconsumerd error */
314 ret
= lttcomm_recv_unix_sock(sock
, &code
,
315 sizeof(enum lttcomm_return_code
));
317 ERR("consumer closed the command socket");
321 ERR("consumer return code : %s",
322 lttcomm_get_readable_code((lttcomm_return_code
) -code
));
325 } else if (pollfd
== consumer_data
->metadata_fd
) {
326 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
327 && !(revents
& LPOLLIN
)) {
328 ERR("consumer err metadata socket second poll error");
331 /* UST metadata requests */
332 ret
= ust_consumer_metadata_request(
333 &consumer_data
->metadata_sock
);
335 ERR("Handling metadata request");
339 /* No need for an else branch all FDs are tested prior. */
341 health_code_update();
347 * We lock here because we are about to close the sockets and some other
348 * thread might be using them so get exclusive access which will abort all
349 * other consumer command by other threads.
351 pthread_mutex_lock(&consumer_data
->lock
);
353 /* Immediately set the consumerd state to stopped */
354 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
355 uatomic_set(&the_kernel_consumerd_state
, CONSUMER_ERROR
);
356 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
357 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
358 uatomic_set(&the_ust_consumerd_state
, CONSUMER_ERROR
);
360 /* Code flow error... */
364 if (consumer_data
->err_sock
>= 0) {
365 ret
= close(consumer_data
->err_sock
);
369 consumer_data
->err_sock
= -1;
371 if (consumer_data
->cmd_sock
>= 0) {
372 ret
= close(consumer_data
->cmd_sock
);
376 consumer_data
->cmd_sock
= -1;
378 if (consumer_data
->metadata_sock
.fd_ptr
&&
379 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
380 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
392 unlink(consumer_data
->err_unix_sock_path
);
393 unlink(consumer_data
->cmd_unix_sock_path
);
394 pthread_mutex_unlock(&consumer_data
->lock
);
396 /* Cleanup metadata socket mutex. */
397 if (consumer_data
->metadata_sock
.lock
) {
398 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
399 free(consumer_data
->metadata_sock
.lock
);
401 lttng_poll_clean(&events
);
403 if (cmd_socket_wrapper
) {
404 consumer_destroy_socket(cmd_socket_wrapper
);
409 ERR("Health error occurred in %s", __func__
);
411 health_unregister(the_health_sessiond
);
412 DBG("consumer thread cleanup completed");
414 rcu_thread_offline();
415 rcu_unregister_thread();
420 static bool shutdown_consumer_management_thread(void *data
)
422 struct thread_notifiers
*notifiers
= (thread_notifiers
*) data
;
423 const int write_fd
= lttng_pipe_get_writefd(notifiers
->quit_pipe
);
425 return notify_thread_pipe(write_fd
) == 1;
428 static void cleanup_consumer_management_thread(void *data
)
430 struct thread_notifiers
*notifiers
= (thread_notifiers
*) data
;
432 lttng_pipe_destroy(notifiers
->quit_pipe
);
436 bool launch_consumer_management_thread(struct consumer_data
*consumer_data
)
438 struct lttng_pipe
*quit_pipe
;
439 struct thread_notifiers
*notifiers
= NULL
;
440 struct lttng_thread
*thread
;
442 notifiers
= zmalloc
<thread_notifiers
>();
447 quit_pipe
= lttng_pipe_open(FD_CLOEXEC
);
451 notifiers
->quit_pipe
= quit_pipe
;
452 notifiers
->consumer_data
= consumer_data
;
453 sem_init(¬ifiers
->ready
, 0, 0);
455 thread
= lttng_thread_create("Consumer management",
456 thread_consumer_management
,
457 shutdown_consumer_management_thread
,
458 cleanup_consumer_management_thread
,
463 wait_until_thread_is_ready(notifiers
);
464 lttng_thread_put(thread
);
465 if (notifiers
->initialization_result
) {
470 cleanup_consumer_management_thread(notifiers
);