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
10 #include "health-sessiond.hpp"
11 #include "manage-consumer.hpp"
12 #include "testpoint.hpp"
14 #include "ust-consumer.hpp"
17 #include <common/pipe.hpp>
18 #include <common/utils.hpp>
24 struct thread_notifiers
{
25 struct lttng_pipe
*quit_pipe
;
26 struct consumer_data
*consumer_data
;
28 int initialization_result
;
32 static void mark_thread_as_ready(struct thread_notifiers
*notifiers
)
34 DBG("Marking consumer management thread as ready");
35 notifiers
->initialization_result
= 0;
36 sem_post(¬ifiers
->ready
);
39 static void mark_thread_intialization_as_failed(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
= nullptr;
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
, 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
= lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
184 consumer_data
->metadata_fd
= lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
185 if (consumer_data
->cmd_sock
< 0 || consumer_data
->metadata_fd
< 0) {
186 PERROR("consumer connect cmd socket");
187 mark_thread_intialization_as_failed(notifiers
);
191 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
193 /* Create metadata socket lock. */
194 consumer_data
->metadata_sock
.lock
= zmalloc
<pthread_mutex_t
>();
195 if (consumer_data
->metadata_sock
.lock
== nullptr) {
196 PERROR("zmalloc pthread mutex");
197 mark_thread_intialization_as_failed(notifiers
);
200 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, nullptr);
202 DBG("Consumer command socket ready (fd: %d)", consumer_data
->cmd_sock
);
203 DBG("Consumer metadata socket ready (fd: %d)", consumer_data
->metadata_fd
);
206 * Remove the consumerd error sock since we've established a connection.
208 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
210 mark_thread_intialization_as_failed(notifiers
);
214 /* Add new accepted error socket. */
215 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
217 mark_thread_intialization_as_failed(notifiers
);
221 /* Add metadata socket that is successfully connected. */
222 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
, LPOLLIN
| LPOLLRDHUP
);
224 mark_thread_intialization_as_failed(notifiers
);
228 health_code_update();
231 * Transfer the write-end of the channel monitoring pipe to the consumer
232 * by issuing a SET_CHANNEL_MONITOR_PIPE command.
234 cmd_socket_wrapper
= consumer_allocate_socket(&consumer_data
->cmd_sock
);
235 if (!cmd_socket_wrapper
) {
236 mark_thread_intialization_as_failed(notifiers
);
239 cmd_socket_wrapper
->lock
= &consumer_data
->lock
;
241 pthread_mutex_lock(cmd_socket_wrapper
->lock
);
242 ret
= consumer_init(cmd_socket_wrapper
, the_sessiond_uuid
);
244 ERR("Failed to send sessiond uuid to consumer daemon");
245 mark_thread_intialization_as_failed(notifiers
);
246 pthread_mutex_unlock(cmd_socket_wrapper
->lock
);
249 pthread_mutex_unlock(cmd_socket_wrapper
->lock
);
251 ret
= consumer_send_channel_monitor_pipe(cmd_socket_wrapper
,
252 consumer_data
->channel_monitor_pipe
);
254 mark_thread_intialization_as_failed(notifiers
);
258 /* Discard the socket wrapper as it is no longer needed. */
259 consumer_destroy_socket(cmd_socket_wrapper
);
260 cmd_socket_wrapper
= nullptr;
262 /* The thread is completely initialized, signal that it is ready. */
263 mark_thread_as_ready(notifiers
);
265 /* Infinite blocking call, waiting for transmission */
267 health_code_update();
269 /* Exit the thread because the thread quit pipe has been triggered. */
271 /* Not a health error. */
277 ret
= lttng_poll_wait(&events
, -1);
285 for (i
= 0; i
< nb_fd
; i
++) {
286 /* Fetch once the poll data */
287 const auto revents
= LTTNG_POLL_GETEV(&events
, i
);
288 const auto pollfd
= LTTNG_POLL_GETFD(&events
, i
);
290 health_code_update();
293 * Thread quit pipe has been triggered, flag that we should stop
294 * but continue the current loop to handle potential data from
297 if (pollfd
== thread_quit_pipe_fd
) {
299 } else if (pollfd
== sock
) {
300 /* Event on the consumerd socket */
301 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
) &&
302 !(revents
& LPOLLIN
)) {
303 ERR("consumer err socket second poll error");
306 health_code_update();
307 /* Wait for any kconsumerd error */
308 ret
= lttcomm_recv_unix_sock(
309 sock
, &code
, sizeof(enum lttcomm_return_code
));
311 ERR("consumer closed the command socket");
315 ERR("consumer return code : %s",
316 lttcomm_get_readable_code((lttcomm_return_code
) -code
));
319 } else if (pollfd
== consumer_data
->metadata_fd
) {
320 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
) &&
321 !(revents
& LPOLLIN
)) {
322 ERR("consumer err metadata socket second poll error");
325 /* UST metadata requests */
326 ret
= ust_consumer_metadata_request(&consumer_data
->metadata_sock
);
328 ERR("Handling metadata request");
332 /* No need for an else branch all FDs are tested prior. */
334 health_code_update();
340 * We lock here because we are about to close the sockets and some other
341 * thread might be using them so get exclusive access which will abort all
342 * other consumer command by other threads.
344 pthread_mutex_lock(&consumer_data
->lock
);
346 /* Immediately set the consumerd state to stopped */
347 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
348 uatomic_set(&the_kernel_consumerd_state
, CONSUMER_ERROR
);
349 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
350 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
351 uatomic_set(&the_ust_consumerd_state
, CONSUMER_ERROR
);
353 /* Code flow error... */
357 if (consumer_data
->err_sock
>= 0) {
358 ret
= close(consumer_data
->err_sock
);
362 consumer_data
->err_sock
= -1;
364 if (consumer_data
->cmd_sock
>= 0) {
365 ret
= close(consumer_data
->cmd_sock
);
369 consumer_data
->cmd_sock
= -1;
371 if (consumer_data
->metadata_sock
.fd_ptr
&& *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
372 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
384 unlink(consumer_data
->err_unix_sock_path
);
385 unlink(consumer_data
->cmd_unix_sock_path
);
386 pthread_mutex_unlock(&consumer_data
->lock
);
388 /* Cleanup metadata socket mutex. */
389 if (consumer_data
->metadata_sock
.lock
) {
390 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
391 free(consumer_data
->metadata_sock
.lock
);
393 lttng_poll_clean(&events
);
395 if (cmd_socket_wrapper
) {
396 consumer_destroy_socket(cmd_socket_wrapper
);
401 ERR("Health error occurred in %s", __func__
);
403 health_unregister(the_health_sessiond
);
404 DBG("consumer thread cleanup completed");
406 rcu_thread_offline();
407 rcu_unregister_thread();
412 static bool shutdown_consumer_management_thread(void *data
)
414 struct thread_notifiers
*notifiers
= (thread_notifiers
*) data
;
415 const int write_fd
= lttng_pipe_get_writefd(notifiers
->quit_pipe
);
417 return notify_thread_pipe(write_fd
) == 1;
420 static void cleanup_consumer_management_thread(void *data
)
422 struct thread_notifiers
*notifiers
= (thread_notifiers
*) data
;
424 lttng_pipe_destroy(notifiers
->quit_pipe
);
428 bool launch_consumer_management_thread(struct consumer_data
*consumer_data
)
430 struct lttng_pipe
*quit_pipe
;
431 struct thread_notifiers
*notifiers
= nullptr;
432 struct lttng_thread
*thread
;
434 notifiers
= zmalloc
<thread_notifiers
>();
439 quit_pipe
= lttng_pipe_open(FD_CLOEXEC
);
443 notifiers
->quit_pipe
= quit_pipe
;
444 notifiers
->consumer_data
= consumer_data
;
445 sem_init(¬ifiers
->ready
, 0, 0);
447 thread
= lttng_thread_create("Consumer management",
448 thread_consumer_management
,
449 shutdown_consumer_management_thread
,
450 cleanup_consumer_management_thread
,
455 wait_until_thread_is_ready(notifiers
);
456 lttng_thread_put(thread
);
457 return notifiers
->initialization_result
== 0;
459 cleanup_consumer_management_thread(notifiers
);