2 * Copyright (C) 2020 Jérémie Galarneau <jeremie.galarneau@efficios.com>
4 * SPDX-License-Identifier: GPL-2.0-only
8 #include "action-executor.h"
10 #include "health-sessiond.h"
11 #include "lttng-sessiond.h"
12 #include "notification-thread-internal.h"
15 #include <common/macros.h>
16 #include <common/optional.h>
17 #include <lttng/action/action-internal.h>
18 #include <lttng/action/group.h>
19 #include <lttng/action/notify.h>
20 #include <lttng/action/rotate-session.h>
21 #include <lttng/action/snapshot-session.h>
22 #include <lttng/action/start-session.h>
23 #include <lttng/action/stop-session.h>
24 #include <lttng/condition/evaluation.h>
25 #include <lttng/condition/event-rule-internal.h>
26 #include <lttng/lttng-error.h>
27 #include <lttng/trigger/trigger-internal.h>
31 #include <urcu/list.h>
33 #define THREAD_NAME "Action Executor"
34 #define MAX_QUEUED_WORK_COUNT 8192
36 struct action_work_item
{
38 struct lttng_trigger
*trigger
;
39 struct lttng_evaluation
*evaluation
;
40 struct notification_client_list
*client_list
;
41 LTTNG_OPTIONAL(struct lttng_credentials
) object_creds
;
42 struct cds_list_head list_node
;
45 struct action_executor
{
46 struct lttng_thread
*thread
;
47 struct notification_thread_handle
*notification_thread_handle
;
49 uint64_t pending_count
;
50 struct cds_list_head list
;
55 uint64_t next_work_item_id
;
59 * Only return non-zero on a fatal error that should shut down the action
62 typedef int (*action_executor_handler
)(struct action_executor
*executor
,
63 const struct action_work_item
*,
64 const struct lttng_action
*action
);
66 static int action_executor_notify_handler(struct action_executor
*executor
,
67 const struct action_work_item
*,
68 const struct lttng_action
*);
69 static int action_executor_start_session_handler(struct action_executor
*executor
,
70 const struct action_work_item
*,
71 const struct lttng_action
*);
72 static int action_executor_stop_session_handler(struct action_executor
*executor
,
73 const struct action_work_item
*,
74 const struct lttng_action
*);
75 static int action_executor_rotate_session_handler(struct action_executor
*executor
,
76 const struct action_work_item
*,
77 const struct lttng_action
*);
78 static int action_executor_snapshot_session_handler(struct action_executor
*executor
,
79 const struct action_work_item
*,
80 const struct lttng_action
*);
81 static int action_executor_group_handler(struct action_executor
*executor
,
82 const struct action_work_item
*,
83 const struct lttng_action
*);
84 static int action_executor_generic_handler(struct action_executor
*executor
,
85 const struct action_work_item
*,
86 const struct lttng_action
*);
88 static const action_executor_handler action_executors
[] = {
89 [LTTNG_ACTION_TYPE_NOTIFY
] = action_executor_notify_handler
,
90 [LTTNG_ACTION_TYPE_START_SESSION
] = action_executor_start_session_handler
,
91 [LTTNG_ACTION_TYPE_STOP_SESSION
] = action_executor_stop_session_handler
,
92 [LTTNG_ACTION_TYPE_ROTATE_SESSION
] = action_executor_rotate_session_handler
,
93 [LTTNG_ACTION_TYPE_SNAPSHOT_SESSION
] = action_executor_snapshot_session_handler
,
94 [LTTNG_ACTION_TYPE_GROUP
] = action_executor_group_handler
,
97 static const char *action_type_names
[] = {
98 [LTTNG_ACTION_TYPE_NOTIFY
] = "Notify",
99 [LTTNG_ACTION_TYPE_START_SESSION
] = "Start session",
100 [LTTNG_ACTION_TYPE_STOP_SESSION
] = "Stop session",
101 [LTTNG_ACTION_TYPE_ROTATE_SESSION
] = "Rotate session",
102 [LTTNG_ACTION_TYPE_SNAPSHOT_SESSION
] = "Snapshot session",
103 [LTTNG_ACTION_TYPE_GROUP
] = "Group",
106 static const char *get_action_name(const struct lttng_action
*action
)
108 const enum lttng_action_type action_type
= lttng_action_get_type(action
);
110 assert(action_type
!= LTTNG_ACTION_TYPE_UNKNOWN
);
112 return action_type_names
[action_type
];
115 /* Check if this trigger allowed to interect with a given session. */
116 static bool is_trigger_allowed_for_session(const struct lttng_trigger
*trigger
,
117 struct ltt_session
*session
)
119 bool is_allowed
= false;
120 const struct lttng_credentials session_creds
= {
121 .uid
= LTTNG_OPTIONAL_INIT_VALUE(session
->uid
),
122 .gid
= LTTNG_OPTIONAL_INIT_VALUE(session
->gid
),
124 /* Can never be NULL. */
125 const struct lttng_credentials
*trigger_creds
=
126 lttng_trigger_get_credentials(trigger
);
128 is_allowed
= (lttng_credentials_is_equal_uid(trigger_creds
, &session_creds
)) ||
129 (lttng_credentials_get_uid(trigger_creds
) == 0);
131 WARN("Trigger is not allowed to interact with session `%s`: session uid = %ld, session gid = %ld, trigger uid = %ld",
133 (long int) session
->uid
,
134 (long int) session
->gid
,
135 (long int) lttng_credentials_get_uid(trigger_creds
));
141 static const char *get_trigger_name(const struct lttng_trigger
*trigger
)
143 const char *trigger_name
;
144 enum lttng_trigger_status trigger_status
;
146 trigger_status
= lttng_trigger_get_name(trigger
, &trigger_name
);
147 assert(trigger_status
== LTTNG_TRIGGER_STATUS_OK
);
152 static int client_handle_transmission_status(
153 struct notification_client
*client
,
154 enum client_transmission_status status
,
158 struct action_executor
*executor
= user_data
;
159 bool update_communication
= true;
162 case CLIENT_TRANSMISSION_STATUS_COMPLETE
:
163 DBG("Successfully sent full notification to client, client_id = %" PRIu64
,
165 update_communication
= false;
167 case CLIENT_TRANSMISSION_STATUS_QUEUED
:
168 DBG("Queued notification in client outgoing buffer, client_id = %" PRIu64
,
171 case CLIENT_TRANSMISSION_STATUS_FAIL
:
172 DBG("Communication error occurred while sending notification to client, client_id = %" PRIu64
,
176 ERR("Fatal error encoutered while sending notification to client, client_id = %" PRIu64
,
182 if (!update_communication
) {
186 /* Safe to read client's id without locking as it is immutable. */
187 ret
= notification_thread_client_communication_update(
188 executor
->notification_thread_handle
, client
->id
,
194 static int action_executor_notify_handler(struct action_executor
*executor
,
195 const struct action_work_item
*work_item
,
196 const struct lttng_action
*action
)
198 return notification_client_list_send_evaluation(work_item
->client_list
,
199 lttng_trigger_get_const_condition(work_item
->trigger
),
200 work_item
->evaluation
,
201 lttng_trigger_get_credentials(work_item
->trigger
),
202 work_item
->object_creds
.is_set
?
203 &(work_item
->object_creds
.value
) :
205 client_handle_transmission_status
, executor
);
208 static int action_executor_start_session_handler(struct action_executor
*executor
,
209 const struct action_work_item
*work_item
,
210 const struct lttng_action
*action
)
213 const char *session_name
;
214 enum lttng_action_status action_status
;
215 struct ltt_session
*session
;
216 enum lttng_error_code cmd_ret
;
218 action_status
= lttng_action_start_session_get_session_name(
219 action
, &session_name
);
220 if (action_status
!= LTTNG_ACTION_STATUS_OK
) {
221 ERR("Failed to get session name from `%s` action",
222 get_action_name(action
));
228 session
= session_find_by_name(session_name
);
230 DBG("Failed to find session `%s` by name while executing `%s` action of trigger `%s`",
231 session_name
, get_action_name(action
),
232 get_trigger_name(work_item
->trigger
));
233 goto error_unlock_list
;
236 session_lock(session
);
237 if (!is_trigger_allowed_for_session(work_item
->trigger
, session
)) {
238 goto error_dispose_session
;
241 cmd_ret
= cmd_start_trace(session
);
244 DBG("Successfully started session `%s` on behalf of trigger `%s`",
245 session_name
, get_trigger_name(work_item
->trigger
));
247 case LTTNG_ERR_TRACE_ALREADY_STARTED
:
248 DBG("Attempted to start session `%s` on behalf of trigger `%s` but it was already started",
249 session_name
, get_trigger_name(work_item
->trigger
));
252 WARN("Failed to start session `%s` on behalf of trigger `%s`: %s",
253 session_name
, get_trigger_name(work_item
->trigger
),
254 lttng_strerror(-cmd_ret
));
258 error_dispose_session
:
259 session_unlock(session
);
260 session_put(session
);
262 session_unlock_list();
267 static int action_executor_stop_session_handler(struct action_executor
*executor
,
268 const struct action_work_item
*work_item
,
269 const struct lttng_action
*action
)
272 const char *session_name
;
273 enum lttng_action_status action_status
;
274 struct ltt_session
*session
;
275 enum lttng_error_code cmd_ret
;
277 action_status
= lttng_action_stop_session_get_session_name(
278 action
, &session_name
);
279 if (action_status
!= LTTNG_ACTION_STATUS_OK
) {
280 ERR("Failed to get session name from `%s` action",
281 get_action_name(action
));
287 session
= session_find_by_name(session_name
);
289 DBG("Failed to find session `%s` by name while executing `%s` action of trigger `%s`",
290 session_name
, get_action_name(action
),
291 get_trigger_name(work_item
->trigger
));
292 goto error_unlock_list
;
295 session_lock(session
);
296 if (!is_trigger_allowed_for_session(work_item
->trigger
, session
)) {
297 goto error_dispose_session
;
300 cmd_ret
= cmd_stop_trace(session
);
303 DBG("Successfully stopped session `%s` on behalf of trigger `%s`",
304 session_name
, get_trigger_name(work_item
->trigger
));
306 case LTTNG_ERR_TRACE_ALREADY_STOPPED
:
307 DBG("Attempted to stop session `%s` on behalf of trigger `%s` but it was already stopped",
308 session_name
, get_trigger_name(work_item
->trigger
));
311 WARN("Failed to stop session `%s` on behalf of trigger `%s`: %s",
312 session_name
, get_trigger_name(work_item
->trigger
),
313 lttng_strerror(-cmd_ret
));
317 error_dispose_session
:
318 session_unlock(session
);
319 session_put(session
);
321 session_unlock_list();
326 static int action_executor_rotate_session_handler(struct action_executor
*executor
,
327 const struct action_work_item
*work_item
,
328 const struct lttng_action
*action
)
331 const char *session_name
;
332 enum lttng_action_status action_status
;
333 struct ltt_session
*session
;
334 enum lttng_error_code cmd_ret
;
336 action_status
= lttng_action_rotate_session_get_session_name(
337 action
, &session_name
);
338 if (action_status
!= LTTNG_ACTION_STATUS_OK
) {
339 ERR("Failed to get session name from `%s` action",
340 get_action_name(action
));
346 session
= session_find_by_name(session_name
);
348 DBG("Failed to find session `%s` by name while executing `%s` action of trigger `%s`",
349 session_name
, get_action_name(action
),
350 get_trigger_name(work_item
->trigger
));
351 goto error_unlock_list
;
354 session_lock(session
);
355 if (!is_trigger_allowed_for_session(work_item
->trigger
, session
)) {
356 goto error_dispose_session
;
359 cmd_ret
= cmd_rotate_session(session
, NULL
, false,
360 LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED
);
363 DBG("Successfully started rotation of session `%s` on behalf of trigger `%s`",
364 session_name
, get_trigger_name(work_item
->trigger
));
366 case LTTNG_ERR_ROTATION_PENDING
:
367 DBG("Attempted to start a rotation of session `%s` on behalf of trigger `%s` but a rotation is already ongoing",
368 session_name
, get_trigger_name(work_item
->trigger
));
370 case LTTNG_ERR_ROTATION_MULTIPLE_AFTER_STOP
:
371 case LTTNG_ERR_ROTATION_AFTER_STOP_CLEAR
:
372 DBG("Attempted to start a rotation of session `%s` on behalf of trigger `%s` but a rotation has already been completed since the last stop or clear",
373 session_name
, get_trigger_name(work_item
->trigger
));
376 WARN("Failed to start a rotation of session `%s` on behalf of trigger `%s`: %s",
377 session_name
, get_trigger_name(work_item
->trigger
),
378 lttng_strerror(-cmd_ret
));
382 error_dispose_session
:
383 session_unlock(session
);
384 session_put(session
);
386 session_unlock_list();
391 static int action_executor_snapshot_session_handler(struct action_executor
*executor
,
392 const struct action_work_item
*work_item
,
393 const struct lttng_action
*action
)
396 const char *session_name
;
397 enum lttng_action_status action_status
;
398 struct ltt_session
*session
;
399 const struct lttng_snapshot_output default_snapshot_output
= {
400 .max_size
= UINT64_MAX
,
402 const struct lttng_snapshot_output
*snapshot_output
=
403 &default_snapshot_output
;
404 enum lttng_error_code cmd_ret
;
406 action_status
= lttng_action_snapshot_session_get_session_name(
407 action
, &session_name
);
408 if (action_status
!= LTTNG_ACTION_STATUS_OK
) {
409 ERR("Failed to get session name from `%s` action",
410 get_action_name(action
));
415 action_status
= lttng_action_snapshot_session_get_output(
416 action
, &snapshot_output
);
417 if (action_status
!= LTTNG_ACTION_STATUS_OK
&&
418 action_status
!= LTTNG_ACTION_STATUS_UNSET
) {
419 ERR("Failed to get output from `%s` action",
420 get_action_name(action
));
426 session
= session_find_by_name(session_name
);
428 DBG("Failed to find session `%s` by name while executing `%s` action of trigger `%p`",
429 session_name
, get_action_name(action
),
431 goto error_unlock_list
;
435 session_lock(session
);
436 if (!is_trigger_allowed_for_session(work_item
->trigger
, session
)) {
437 goto error_dispose_session
;
440 cmd_ret
= cmd_snapshot_record(session
, snapshot_output
, 0);
443 DBG("Successfully recorded snapshot of session `%s` on behalf of trigger `%s`",
444 session_name
, get_trigger_name(work_item
->trigger
));
447 WARN("Failed to record snapshot of session `%s` on behalf of trigger `%s`: %s",
448 session_name
, get_trigger_name(work_item
->trigger
),
449 lttng_strerror(-cmd_ret
));
453 error_dispose_session
:
454 session_unlock(session
);
455 session_put(session
);
457 session_unlock_list();
462 static int action_executor_group_handler(struct action_executor
*executor
,
463 const struct action_work_item
*work_item
,
464 const struct lttng_action
*action_group
)
467 unsigned int i
, count
;
468 enum lttng_action_status action_status
;
470 action_status
= lttng_action_group_get_count(action_group
, &count
);
471 if (action_status
!= LTTNG_ACTION_STATUS_OK
) {
473 ERR("Failed to get count of action in action group");
478 DBG("Action group has %u action%s", count
, count
!= 1 ? "s" : "");
479 for (i
= 0; i
< count
; i
++) {
480 const struct lttng_action
*action
=
481 lttng_action_group_get_at_index(
484 ret
= action_executor_generic_handler(
485 executor
, work_item
, action
);
487 ERR("Stopping the execution of the action group of trigger `%s` following a fatal error",
488 get_trigger_name(work_item
->trigger
));
496 static int action_executor_generic_handler(struct action_executor
*executor
,
497 const struct action_work_item
*work_item
,
498 const struct lttng_action
*action
)
500 const enum lttng_action_type action_type
= lttng_action_get_type(action
);
502 assert(action_type
!= LTTNG_ACTION_TYPE_UNKNOWN
);
504 DBG("Executing action `%s` of trigger `%p` action work item %" PRIu64
,
505 get_action_name(action
),
506 get_trigger_name(work_item
->trigger
),
509 return action_executors
[action_type
](
510 executor
, work_item
, action
);
513 static int action_work_item_execute(struct action_executor
*executor
,
514 struct action_work_item
*work_item
)
517 const struct lttng_action
*action
=
518 lttng_trigger_get_const_action(work_item
->trigger
);
520 DBG("Starting execution of action work item %" PRIu64
" of trigger `%s`",
521 work_item
->id
, get_trigger_name(work_item
->trigger
));
522 ret
= action_executor_generic_handler(executor
, work_item
, action
);
523 DBG("Completed execution of action work item %" PRIu64
" of trigger `%s`",
524 work_item
->id
, get_trigger_name(work_item
->trigger
));
528 static void action_work_item_destroy(struct action_work_item
*work_item
)
530 lttng_trigger_put(work_item
->trigger
);
531 lttng_evaluation_destroy(work_item
->evaluation
);
532 notification_client_list_put(work_item
->client_list
);
536 static void *action_executor_thread(void *_data
)
538 struct action_executor
*executor
= _data
;
542 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_ACTION_EXECUTOR
);
544 rcu_register_thread();
547 DBG("Entering work execution loop");
548 pthread_mutex_lock(&executor
->work
.lock
);
549 while (!executor
->should_quit
) {
551 struct action_work_item
*work_item
;
553 health_code_update();
554 if (executor
->work
.pending_count
== 0) {
556 DBG("No work items enqueued, entering wait");
557 pthread_cond_wait(&executor
->work
.cond
,
558 &executor
->work
.lock
);
559 DBG("Woke-up from wait");
564 /* Pop item from front of the list with work lock held. */
565 work_item
= cds_list_first_entry(&executor
->work
.list
,
566 struct action_work_item
, list_node
);
567 cds_list_del(&work_item
->list_node
);
568 executor
->work
.pending_count
--;
571 * Work can be performed without holding the work lock,
572 * allowing new items to be queued.
574 pthread_mutex_unlock(&executor
->work
.lock
);
575 ret
= action_work_item_execute(executor
, work_item
);
576 action_work_item_destroy(work_item
);
582 health_code_update();
583 pthread_mutex_lock(&executor
->work
.lock
);
586 if (executor
->should_quit
) {
587 pthread_mutex_unlock(&executor
->work
.lock
);
589 DBG("Left work execution loop");
591 health_code_update();
593 rcu_thread_offline();
594 rcu_unregister_thread();
595 health_unregister(health_sessiond
);
600 static bool shutdown_action_executor_thread(void *_data
)
602 struct action_executor
*executor
= _data
;
604 pthread_mutex_lock(&executor
->work
.lock
);
605 executor
->should_quit
= true;
606 pthread_cond_signal(&executor
->work
.cond
);
607 pthread_mutex_unlock(&executor
->work
.lock
);
611 static void clean_up_action_executor_thread(void *_data
)
613 struct action_executor
*executor
= _data
;
615 assert(cds_list_empty(&executor
->work
.list
));
617 pthread_mutex_destroy(&executor
->work
.lock
);
618 pthread_cond_destroy(&executor
->work
.cond
);
622 struct action_executor
*action_executor_create(
623 struct notification_thread_handle
*handle
)
625 struct action_executor
*executor
= zmalloc(sizeof(*executor
));
631 CDS_INIT_LIST_HEAD(&executor
->work
.list
);
632 pthread_cond_init(&executor
->work
.cond
, NULL
);
633 pthread_mutex_init(&executor
->work
.lock
, NULL
);
634 executor
->notification_thread_handle
= handle
;
636 executor
->thread
= lttng_thread_create(THREAD_NAME
,
637 action_executor_thread
, shutdown_action_executor_thread
,
638 clean_up_action_executor_thread
, executor
);
643 void action_executor_destroy(struct action_executor
*executor
)
645 struct action_work_item
*work_item
, *tmp
;
647 /* TODO Wait for work list to drain? */
648 lttng_thread_shutdown(executor
->thread
);
649 pthread_mutex_lock(&executor
->work
.lock
);
650 if (executor
->work
.pending_count
!= 0) {
652 " trigger action%s still queued for execution and will be discarded",
653 executor
->work
.pending_count
,
654 executor
->work
.pending_count
== 1 ? " is" :
658 cds_list_for_each_entry_safe (
659 work_item
, tmp
, &executor
->work
.list
, list_node
) {
660 WARN("Discarding action work item %" PRIu64
661 " associated to trigger `%s`",
662 work_item
->id
, get_trigger_name(work_item
->trigger
));
663 cds_list_del(&work_item
->list_node
);
664 action_work_item_destroy(work_item
);
666 pthread_mutex_unlock(&executor
->work
.lock
);
667 lttng_thread_put(executor
->thread
);
670 /* RCU read-lock must be held by the caller. */
671 enum action_executor_status
action_executor_enqueue(
672 struct action_executor
*executor
,
673 struct lttng_trigger
*trigger
,
674 struct lttng_evaluation
*evaluation
,
675 const struct lttng_credentials
*object_creds
,
676 struct notification_client_list
*client_list
)
678 enum action_executor_status executor_status
= ACTION_EXECUTOR_STATUS_OK
;
679 const uint64_t work_item_id
= executor
->next_work_item_id
++;
680 struct action_work_item
*work_item
;
683 pthread_mutex_lock(&executor
->work
.lock
);
684 /* Check for queue overflow. */
685 if (executor
->work
.pending_count
>= MAX_QUEUED_WORK_COUNT
) {
686 /* Most likely spammy, remove if it is the case. */
687 DBG("Refusing to enqueue action for trigger `%s` as work item %" PRIu64
688 " (overflow)", get_trigger_name(trigger
), work_item_id
);
689 executor_status
= ACTION_EXECUTOR_STATUS_OVERFLOW
;
693 work_item
= zmalloc(sizeof(*work_item
));
695 PERROR("Failed to allocate action executor work item on behalf of trigger `%p`",
696 get_trigger_name(trigger
));
697 executor_status
= ACTION_EXECUTOR_STATUS_ERROR
;
701 lttng_trigger_get(trigger
);
703 const bool reference_acquired
=
704 notification_client_list_get(client_list
);
706 assert(reference_acquired
);
709 *work_item
= (typeof(*work_item
)){
712 /* Ownership transferred to the work item. */
713 .evaluation
= evaluation
,
715 .is_set
= !!object_creds
,
716 .value
= object_creds
? *object_creds
:
717 (typeof(work_item
->object_creds
.value
)) {},
719 .client_list
= client_list
,
720 .list_node
= CDS_LIST_HEAD_INIT(work_item
->list_node
),
724 cds_list_add_tail(&work_item
->list_node
, &executor
->work
.list
);
725 executor
->work
.pending_count
++;
726 DBG("Enqueued action for trigger `%p` as work item %" PRIu64
,
727 get_trigger_name(trigger
), work_item_id
);
732 pthread_cond_signal(&executor
->work
.cond
);
734 pthread_mutex_unlock(&executor
->work
.lock
);
736 lttng_evaluation_destroy(evaluation
);
737 return executor_status
;