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
, err
= -1, should_quit
= 0;
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 auto thread_quit_pipe_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
, thread_quit_pipe_fd
, LPOLLIN
);
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 const auto revents
= LTTNG_POLL_GETEV(&events
, i
);
125 const auto pollfd
= LTTNG_POLL_GETFD(&events
, i
);
127 health_code_update();
129 /* Activity on thread quit pipe, exiting. */
130 if (pollfd
== thread_quit_pipe_fd
) {
131 DBG("Activity on thread quit pipe");
133 mark_thread_intialization_as_failed(notifiers
);
135 } else if (pollfd
== consumer_data
->err_sock
) {
136 /* Event on the registration socket */
137 if (revents
& LPOLLIN
) {
139 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
140 ERR("consumer err socket poll error");
141 mark_thread_intialization_as_failed(notifiers
);
144 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
145 mark_thread_intialization_as_failed(notifiers
);
151 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
153 mark_thread_intialization_as_failed(notifiers
);
158 * Set the CLOEXEC flag. Return code is useless because either way, the
161 (void) utils_set_fd_cloexec(sock
);
163 health_code_update();
165 DBG2("Receiving code from consumer err_sock");
167 /* Getting status code from kconsumerd */
168 ret
= lttcomm_recv_unix_sock(sock
, &code
,
169 sizeof(enum lttcomm_return_code
));
171 mark_thread_intialization_as_failed(notifiers
);
175 health_code_update();
176 if (code
!= LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
177 ERR("consumer error when waiting for SOCK_READY : %s",
178 lttcomm_get_readable_code((lttcomm_return_code
) -code
));
179 mark_thread_intialization_as_failed(notifiers
);
183 /* Connect both command and metadata sockets. */
184 consumer_data
->cmd_sock
=
185 lttcomm_connect_unix_sock(
186 consumer_data
->cmd_unix_sock_path
);
187 consumer_data
->metadata_fd
=
188 lttcomm_connect_unix_sock(
189 consumer_data
->cmd_unix_sock_path
);
190 if (consumer_data
->cmd_sock
< 0 || consumer_data
->metadata_fd
< 0) {
191 PERROR("consumer connect cmd socket");
192 mark_thread_intialization_as_failed(notifiers
);
196 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
198 /* Create metadata socket lock. */
199 consumer_data
->metadata_sock
.lock
= zmalloc
<pthread_mutex_t
>();
200 if (consumer_data
->metadata_sock
.lock
== NULL
) {
201 PERROR("zmalloc pthread mutex");
202 mark_thread_intialization_as_failed(notifiers
);
205 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
207 DBG("Consumer command socket ready (fd: %d)", consumer_data
->cmd_sock
);
208 DBG("Consumer metadata socket ready (fd: %d)",
209 consumer_data
->metadata_fd
);
212 * Remove the consumerd error sock since we've established a connection.
214 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
216 mark_thread_intialization_as_failed(notifiers
);
220 /* Add new accepted error socket. */
221 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
223 mark_thread_intialization_as_failed(notifiers
);
227 /* Add metadata socket that is successfully connected. */
228 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
229 LPOLLIN
| LPOLLRDHUP
);
231 mark_thread_intialization_as_failed(notifiers
);
235 health_code_update();
238 * Transfer the write-end of the channel monitoring pipe to the consumer
239 * by issuing a SET_CHANNEL_MONITOR_PIPE command.
241 cmd_socket_wrapper
= consumer_allocate_socket(&consumer_data
->cmd_sock
);
242 if (!cmd_socket_wrapper
) {
243 mark_thread_intialization_as_failed(notifiers
);
246 cmd_socket_wrapper
->lock
= &consumer_data
->lock
;
248 pthread_mutex_lock(cmd_socket_wrapper
->lock
);
249 ret
= consumer_init(cmd_socket_wrapper
, the_sessiond_uuid
);
251 ERR("Failed to send sessiond uuid to consumer daemon");
252 mark_thread_intialization_as_failed(notifiers
);
253 pthread_mutex_unlock(cmd_socket_wrapper
->lock
);
256 pthread_mutex_unlock(cmd_socket_wrapper
->lock
);
258 ret
= consumer_send_channel_monitor_pipe(cmd_socket_wrapper
,
259 consumer_data
->channel_monitor_pipe
);
261 mark_thread_intialization_as_failed(notifiers
);
265 /* Discard the socket wrapper as it is no longer needed. */
266 consumer_destroy_socket(cmd_socket_wrapper
);
267 cmd_socket_wrapper
= NULL
;
269 /* The thread is completely initialized, signal that it is ready. */
270 mark_thread_as_ready(notifiers
);
272 /* Infinite blocking call, waiting for transmission */
274 health_code_update();
276 /* Exit the thread because the thread quit pipe has been triggered. */
278 /* Not a health error. */
284 ret
= lttng_poll_wait(&events
, -1);
292 for (i
= 0; i
< nb_fd
; i
++) {
293 /* Fetch once the poll data */
294 const auto revents
= LTTNG_POLL_GETEV(&events
, i
);
295 const auto pollfd
= LTTNG_POLL_GETFD(&events
, i
);
297 health_code_update();
300 * Thread quit pipe has been triggered, flag that we should stop
301 * but continue the current loop to handle potential data from
304 if (pollfd
== thread_quit_pipe_fd
) {
306 } else if (pollfd
== sock
) {
307 /* Event on the consumerd socket */
308 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
309 && !(revents
& LPOLLIN
)) {
310 ERR("consumer err socket second poll error");
313 health_code_update();
314 /* Wait for any kconsumerd error */
315 ret
= lttcomm_recv_unix_sock(sock
, &code
,
316 sizeof(enum lttcomm_return_code
));
318 ERR("consumer closed the command socket");
322 ERR("consumer return code : %s",
323 lttcomm_get_readable_code((lttcomm_return_code
) -code
));
326 } else if (pollfd
== consumer_data
->metadata_fd
) {
327 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
328 && !(revents
& LPOLLIN
)) {
329 ERR("consumer err metadata socket second poll error");
332 /* UST metadata requests */
333 ret
= ust_consumer_metadata_request(
334 &consumer_data
->metadata_sock
);
336 ERR("Handling metadata request");
340 /* No need for an else branch all FDs are tested prior. */
342 health_code_update();
348 * We lock here because we are about to close the sockets and some other
349 * thread might be using them so get exclusive access which will abort all
350 * other consumer command by other threads.
352 pthread_mutex_lock(&consumer_data
->lock
);
354 /* Immediately set the consumerd state to stopped */
355 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
356 uatomic_set(&the_kernel_consumerd_state
, CONSUMER_ERROR
);
357 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
358 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
359 uatomic_set(&the_ust_consumerd_state
, CONSUMER_ERROR
);
361 /* Code flow error... */
365 if (consumer_data
->err_sock
>= 0) {
366 ret
= close(consumer_data
->err_sock
);
370 consumer_data
->err_sock
= -1;
372 if (consumer_data
->cmd_sock
>= 0) {
373 ret
= close(consumer_data
->cmd_sock
);
377 consumer_data
->cmd_sock
= -1;
379 if (consumer_data
->metadata_sock
.fd_ptr
&&
380 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
381 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
393 unlink(consumer_data
->err_unix_sock_path
);
394 unlink(consumer_data
->cmd_unix_sock_path
);
395 pthread_mutex_unlock(&consumer_data
->lock
);
397 /* Cleanup metadata socket mutex. */
398 if (consumer_data
->metadata_sock
.lock
) {
399 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
400 free(consumer_data
->metadata_sock
.lock
);
402 lttng_poll_clean(&events
);
404 if (cmd_socket_wrapper
) {
405 consumer_destroy_socket(cmd_socket_wrapper
);
410 ERR("Health error occurred in %s", __func__
);
412 health_unregister(the_health_sessiond
);
413 DBG("consumer thread cleanup completed");
415 rcu_thread_offline();
416 rcu_unregister_thread();
421 static bool shutdown_consumer_management_thread(void *data
)
423 struct thread_notifiers
*notifiers
= (thread_notifiers
*) data
;
424 const int write_fd
= lttng_pipe_get_writefd(notifiers
->quit_pipe
);
426 return notify_thread_pipe(write_fd
) == 1;
429 static void cleanup_consumer_management_thread(void *data
)
431 struct thread_notifiers
*notifiers
= (thread_notifiers
*) data
;
433 lttng_pipe_destroy(notifiers
->quit_pipe
);
437 bool launch_consumer_management_thread(struct consumer_data
*consumer_data
)
439 struct lttng_pipe
*quit_pipe
;
440 struct thread_notifiers
*notifiers
= NULL
;
441 struct lttng_thread
*thread
;
443 notifiers
= zmalloc
<thread_notifiers
>();
448 quit_pipe
= lttng_pipe_open(FD_CLOEXEC
);
452 notifiers
->quit_pipe
= quit_pipe
;
453 notifiers
->consumer_data
= consumer_data
;
454 sem_init(¬ifiers
->ready
, 0, 0);
456 thread
= lttng_thread_create("Consumer management",
457 thread_consumer_management
,
458 shutdown_consumer_management_thread
,
459 cleanup_consumer_management_thread
,
464 wait_until_thread_is_ready(notifiers
);
465 lttng_thread_put(thread
);
466 if (notifiers
->initialization_result
) {
471 cleanup_consumer_management_thread(notifiers
);