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-internal.h>
20 #include <lttng/action/notify.h>
21 #include <lttng/action/rotate-session.h>
22 #include <lttng/action/snapshot-session.h>
23 #include <lttng/action/start-session.h>
24 #include <lttng/action/stop-session.h>
25 #include <lttng/condition/evaluation.h>
26 #include <lttng/condition/on-event-internal.h>
27 #include <lttng/lttng-error.h>
28 #include <lttng/trigger/trigger-internal.h>
32 #include <urcu/list.h>
34 #define THREAD_NAME "Action Executor"
35 #define MAX_QUEUED_WORK_COUNT 8192
37 struct action_work_item
{
39 struct lttng_trigger
*trigger
;
40 struct lttng_evaluation
*evaluation
;
41 struct notification_client_list
*client_list
;
42 LTTNG_OPTIONAL(struct lttng_credentials
) object_creds
;
43 struct cds_list_head list_node
;
46 struct action_executor
{
47 struct lttng_thread
*thread
;
48 struct notification_thread_handle
*notification_thread_handle
;
50 uint64_t pending_count
;
51 struct cds_list_head list
;
56 uint64_t next_work_item_id
;
60 * Only return non-zero on a fatal error that should shut down the action
63 typedef int (*action_executor_handler
)(struct action_executor
*executor
,
64 const struct action_work_item
*,
65 const struct lttng_action
*action
);
67 static int action_executor_notify_handler(struct action_executor
*executor
,
68 const struct action_work_item
*,
69 const struct lttng_action
*);
70 static int action_executor_start_session_handler(struct action_executor
*executor
,
71 const struct action_work_item
*,
72 const struct lttng_action
*);
73 static int action_executor_stop_session_handler(struct action_executor
*executor
,
74 const struct action_work_item
*,
75 const struct lttng_action
*);
76 static int action_executor_rotate_session_handler(struct action_executor
*executor
,
77 const struct action_work_item
*,
78 const struct lttng_action
*);
79 static int action_executor_snapshot_session_handler(struct action_executor
*executor
,
80 const struct action_work_item
*,
81 const struct lttng_action
*);
82 static int action_executor_group_handler(struct action_executor
*executor
,
83 const struct action_work_item
*,
84 const struct lttng_action
*);
85 static int action_executor_generic_handler(struct action_executor
*executor
,
86 const struct action_work_item
*,
87 const struct lttng_action
*);
89 static const action_executor_handler action_executors
[] = {
90 [LTTNG_ACTION_TYPE_NOTIFY
] = action_executor_notify_handler
,
91 [LTTNG_ACTION_TYPE_START_SESSION
] = action_executor_start_session_handler
,
92 [LTTNG_ACTION_TYPE_STOP_SESSION
] = action_executor_stop_session_handler
,
93 [LTTNG_ACTION_TYPE_ROTATE_SESSION
] = action_executor_rotate_session_handler
,
94 [LTTNG_ACTION_TYPE_SNAPSHOT_SESSION
] = action_executor_snapshot_session_handler
,
95 [LTTNG_ACTION_TYPE_GROUP
] = action_executor_group_handler
,
98 static const char *action_type_names
[] = {
99 [LTTNG_ACTION_TYPE_NOTIFY
] = "Notify",
100 [LTTNG_ACTION_TYPE_START_SESSION
] = "Start session",
101 [LTTNG_ACTION_TYPE_STOP_SESSION
] = "Stop session",
102 [LTTNG_ACTION_TYPE_ROTATE_SESSION
] = "Rotate session",
103 [LTTNG_ACTION_TYPE_SNAPSHOT_SESSION
] = "Snapshot session",
104 [LTTNG_ACTION_TYPE_GROUP
] = "Group",
107 static const char *get_action_name(const struct lttng_action
*action
)
109 const enum lttng_action_type action_type
= lttng_action_get_type(action
);
111 assert(action_type
!= LTTNG_ACTION_TYPE_UNKNOWN
);
113 return action_type_names
[action_type
];
116 /* Check if this trigger allowed to interect with a given session. */
117 static bool is_trigger_allowed_for_session(const struct lttng_trigger
*trigger
,
118 struct ltt_session
*session
)
120 bool is_allowed
= false;
121 const struct lttng_credentials session_creds
= {
122 .uid
= LTTNG_OPTIONAL_INIT_VALUE(session
->uid
),
123 .gid
= LTTNG_OPTIONAL_INIT_VALUE(session
->gid
),
125 /* Can never be NULL. */
126 const struct lttng_credentials
*trigger_creds
=
127 lttng_trigger_get_credentials(trigger
);
129 is_allowed
= (lttng_credentials_is_equal_uid(trigger_creds
, &session_creds
)) ||
130 (lttng_credentials_get_uid(trigger_creds
) == 0);
132 WARN("Trigger is not allowed to interact with session `%s`: session uid = %ld, session gid = %ld, trigger uid = %ld",
134 (long int) session
->uid
,
135 (long int) session
->gid
,
136 (long int) lttng_credentials_get_uid(trigger_creds
));
142 static const char *get_trigger_name(const struct lttng_trigger
*trigger
)
144 const char *trigger_name
;
145 enum lttng_trigger_status trigger_status
;
147 trigger_status
= lttng_trigger_get_name(trigger
, &trigger_name
);
148 assert(trigger_status
== LTTNG_TRIGGER_STATUS_OK
);
153 static int client_handle_transmission_status(
154 struct notification_client
*client
,
155 enum client_transmission_status status
,
159 struct action_executor
*executor
= user_data
;
160 bool update_communication
= true;
163 case CLIENT_TRANSMISSION_STATUS_COMPLETE
:
164 DBG("Successfully sent full notification to client, client_id = %" PRIu64
,
166 update_communication
= false;
168 case CLIENT_TRANSMISSION_STATUS_QUEUED
:
169 DBG("Queued notification in client outgoing buffer, client_id = %" PRIu64
,
172 case CLIENT_TRANSMISSION_STATUS_FAIL
:
173 DBG("Communication error occurred while sending notification to client, client_id = %" PRIu64
,
177 ERR("Fatal error encoutered while sending notification to client, client_id = %" PRIu64
,
183 if (!update_communication
) {
187 /* Safe to read client's id without locking as it is immutable. */
188 ret
= notification_thread_client_communication_update(
189 executor
->notification_thread_handle
, client
->id
,
195 static int action_executor_notify_handler(struct action_executor
*executor
,
196 const struct action_work_item
*work_item
,
197 const struct lttng_action
*action
)
199 return notification_client_list_send_evaluation(work_item
->client_list
,
200 lttng_trigger_get_const_condition(work_item
->trigger
),
201 work_item
->evaluation
,
202 lttng_trigger_get_credentials(work_item
->trigger
),
203 work_item
->object_creds
.is_set
?
204 &(work_item
->object_creds
.value
) :
206 client_handle_transmission_status
, executor
);
209 static int action_executor_start_session_handler(struct action_executor
*executor
,
210 const struct action_work_item
*work_item
,
211 const struct lttng_action
*action
)
214 const char *session_name
;
215 enum lttng_action_status action_status
;
216 struct ltt_session
*session
;
217 enum lttng_error_code cmd_ret
;
219 action_status
= lttng_action_start_session_get_session_name(
220 action
, &session_name
);
221 if (action_status
!= LTTNG_ACTION_STATUS_OK
) {
222 ERR("Failed to get session name from `%s` action",
223 get_action_name(action
));
229 session
= session_find_by_name(session_name
);
231 DBG("Failed to find session `%s` by name while executing `%s` action of trigger `%s`",
232 session_name
, get_action_name(action
),
233 get_trigger_name(work_item
->trigger
));
234 goto error_unlock_list
;
237 session_lock(session
);
238 if (!is_trigger_allowed_for_session(work_item
->trigger
, session
)) {
239 goto error_dispose_session
;
242 cmd_ret
= cmd_start_trace(session
);
245 DBG("Successfully started session `%s` on behalf of trigger `%s`",
246 session_name
, get_trigger_name(work_item
->trigger
));
248 case LTTNG_ERR_TRACE_ALREADY_STARTED
:
249 DBG("Attempted to start session `%s` on behalf of trigger `%s` but it was already started",
250 session_name
, get_trigger_name(work_item
->trigger
));
253 WARN("Failed to start session `%s` on behalf of trigger `%s`: %s",
254 session_name
, get_trigger_name(work_item
->trigger
),
255 lttng_strerror(-cmd_ret
));
259 error_dispose_session
:
260 session_unlock(session
);
261 session_put(session
);
263 session_unlock_list();
268 static int action_executor_stop_session_handler(struct action_executor
*executor
,
269 const struct action_work_item
*work_item
,
270 const struct lttng_action
*action
)
273 const char *session_name
;
274 enum lttng_action_status action_status
;
275 struct ltt_session
*session
;
276 enum lttng_error_code cmd_ret
;
278 action_status
= lttng_action_stop_session_get_session_name(
279 action
, &session_name
);
280 if (action_status
!= LTTNG_ACTION_STATUS_OK
) {
281 ERR("Failed to get session name from `%s` action",
282 get_action_name(action
));
288 session
= session_find_by_name(session_name
);
290 DBG("Failed to find session `%s` by name while executing `%s` action of trigger `%s`",
291 session_name
, get_action_name(action
),
292 get_trigger_name(work_item
->trigger
));
293 goto error_unlock_list
;
296 session_lock(session
);
297 if (!is_trigger_allowed_for_session(work_item
->trigger
, session
)) {
298 goto error_dispose_session
;
301 cmd_ret
= cmd_stop_trace(session
);
304 DBG("Successfully stopped session `%s` on behalf of trigger `%s`",
305 session_name
, get_trigger_name(work_item
->trigger
));
307 case LTTNG_ERR_TRACE_ALREADY_STOPPED
:
308 DBG("Attempted to stop session `%s` on behalf of trigger `%s` but it was already stopped",
309 session_name
, get_trigger_name(work_item
->trigger
));
312 WARN("Failed to stop session `%s` on behalf of trigger `%s`: %s",
313 session_name
, get_trigger_name(work_item
->trigger
),
314 lttng_strerror(-cmd_ret
));
318 error_dispose_session
:
319 session_unlock(session
);
320 session_put(session
);
322 session_unlock_list();
327 static int action_executor_rotate_session_handler(struct action_executor
*executor
,
328 const struct action_work_item
*work_item
,
329 const struct lttng_action
*action
)
332 const char *session_name
;
333 enum lttng_action_status action_status
;
334 struct ltt_session
*session
;
335 enum lttng_error_code cmd_ret
;
337 action_status
= lttng_action_rotate_session_get_session_name(
338 action
, &session_name
);
339 if (action_status
!= LTTNG_ACTION_STATUS_OK
) {
340 ERR("Failed to get session name from `%s` action",
341 get_action_name(action
));
347 session
= session_find_by_name(session_name
);
349 DBG("Failed to find session `%s` by name while executing `%s` action of trigger `%s`",
350 session_name
, get_action_name(action
),
351 get_trigger_name(work_item
->trigger
));
352 goto error_unlock_list
;
355 session_lock(session
);
356 if (!is_trigger_allowed_for_session(work_item
->trigger
, session
)) {
357 goto error_dispose_session
;
360 cmd_ret
= cmd_rotate_session(session
, NULL
, false,
361 LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED
);
364 DBG("Successfully started rotation of session `%s` on behalf of trigger `%s`",
365 session_name
, get_trigger_name(work_item
->trigger
));
367 case LTTNG_ERR_ROTATION_PENDING
:
368 DBG("Attempted to start a rotation of session `%s` on behalf of trigger `%s` but a rotation is already ongoing",
369 session_name
, get_trigger_name(work_item
->trigger
));
371 case LTTNG_ERR_ROTATION_MULTIPLE_AFTER_STOP
:
372 case LTTNG_ERR_ROTATION_AFTER_STOP_CLEAR
:
373 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",
374 session_name
, get_trigger_name(work_item
->trigger
));
377 WARN("Failed to start a rotation of session `%s` on behalf of trigger `%s`: %s",
378 session_name
, get_trigger_name(work_item
->trigger
),
379 lttng_strerror(-cmd_ret
));
383 error_dispose_session
:
384 session_unlock(session
);
385 session_put(session
);
387 session_unlock_list();
392 static int action_executor_snapshot_session_handler(struct action_executor
*executor
,
393 const struct action_work_item
*work_item
,
394 const struct lttng_action
*action
)
397 const char *session_name
;
398 enum lttng_action_status action_status
;
399 struct ltt_session
*session
;
400 const struct lttng_snapshot_output default_snapshot_output
= {
401 .max_size
= UINT64_MAX
,
403 const struct lttng_snapshot_output
*snapshot_output
=
404 &default_snapshot_output
;
405 enum lttng_error_code cmd_ret
;
407 action_status
= lttng_action_snapshot_session_get_session_name(
408 action
, &session_name
);
409 if (action_status
!= LTTNG_ACTION_STATUS_OK
) {
410 ERR("Failed to get session name from `%s` action",
411 get_action_name(action
));
416 action_status
= lttng_action_snapshot_session_get_output(
417 action
, &snapshot_output
);
418 if (action_status
!= LTTNG_ACTION_STATUS_OK
&&
419 action_status
!= LTTNG_ACTION_STATUS_UNSET
) {
420 ERR("Failed to get output from `%s` action",
421 get_action_name(action
));
427 session
= session_find_by_name(session_name
);
429 DBG("Failed to find session `%s` by name while executing `%s` action of trigger `%s`",
430 session_name
, get_action_name(action
),
431 get_trigger_name(work_item
->trigger
));
432 goto error_unlock_list
;
436 session_lock(session
);
437 if (!is_trigger_allowed_for_session(work_item
->trigger
, session
)) {
438 goto error_dispose_session
;
441 cmd_ret
= cmd_snapshot_record(session
, snapshot_output
, 0);
444 DBG("Successfully recorded snapshot of session `%s` on behalf of trigger `%s`",
445 session_name
, get_trigger_name(work_item
->trigger
));
448 WARN("Failed to record snapshot of session `%s` on behalf of trigger `%s`: %s",
449 session_name
, get_trigger_name(work_item
->trigger
),
450 lttng_strerror(-cmd_ret
));
454 error_dispose_session
:
455 session_unlock(session
);
456 session_put(session
);
458 session_unlock_list();
463 static int action_executor_group_handler(struct action_executor
*executor
,
464 const struct action_work_item
*work_item
,
465 const struct lttng_action
*action_group
)
468 unsigned int i
, count
;
469 enum lttng_action_status action_status
;
471 action_status
= lttng_action_group_get_count(action_group
, &count
);
472 if (action_status
!= LTTNG_ACTION_STATUS_OK
) {
474 ERR("Failed to get count of action in action group");
479 DBG("Action group has %u action%s", count
, count
!= 1 ? "s" : "");
480 for (i
= 0; i
< count
; i
++) {
481 const struct lttng_action
*action
=
482 lttng_action_group_get_at_index(
485 ret
= action_executor_generic_handler(
486 executor
, work_item
, action
);
488 ERR("Stopping the execution of the action group of trigger `%s` following a fatal error",
489 get_trigger_name(work_item
->trigger
));
497 static int action_executor_generic_handler(struct action_executor
*executor
,
498 const struct action_work_item
*work_item
,
499 const struct lttng_action
*action
)
501 const enum lttng_action_type action_type
= lttng_action_get_type(action
);
503 assert(action_type
!= LTTNG_ACTION_TYPE_UNKNOWN
);
505 DBG("Executing action `%s` of trigger `%s` action work item %" PRIu64
,
506 get_action_name(action
),
507 get_trigger_name(work_item
->trigger
),
510 return action_executors
[action_type
](
511 executor
, work_item
, action
);
514 static int action_work_item_execute(struct action_executor
*executor
,
515 struct action_work_item
*work_item
)
518 const struct lttng_action
*action
=
519 lttng_trigger_get_const_action(work_item
->trigger
);
521 DBG("Starting execution of action work item %" PRIu64
" of trigger `%s`",
522 work_item
->id
, get_trigger_name(work_item
->trigger
));
523 ret
= action_executor_generic_handler(executor
, work_item
, action
);
524 DBG("Completed execution of action work item %" PRIu64
" of trigger `%s`",
525 work_item
->id
, get_trigger_name(work_item
->trigger
));
529 static void action_work_item_destroy(struct action_work_item
*work_item
)
531 lttng_trigger_put(work_item
->trigger
);
532 lttng_evaluation_destroy(work_item
->evaluation
);
533 notification_client_list_put(work_item
->client_list
);
537 static void *action_executor_thread(void *_data
)
539 struct action_executor
*executor
= _data
;
543 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_ACTION_EXECUTOR
);
545 rcu_register_thread();
548 DBG("Entering work execution loop");
549 pthread_mutex_lock(&executor
->work
.lock
);
550 while (!executor
->should_quit
) {
552 struct action_work_item
*work_item
;
554 health_code_update();
555 if (executor
->work
.pending_count
== 0) {
557 DBG("No work items enqueued, entering wait");
558 pthread_cond_wait(&executor
->work
.cond
,
559 &executor
->work
.lock
);
560 DBG("Woke-up from wait");
565 /* Pop item from front of the list with work lock held. */
566 work_item
= cds_list_first_entry(&executor
->work
.list
,
567 struct action_work_item
, list_node
);
568 cds_list_del(&work_item
->list_node
);
569 executor
->work
.pending_count
--;
572 * Work can be performed without holding the work lock,
573 * allowing new items to be queued.
575 pthread_mutex_unlock(&executor
->work
.lock
);
576 ret
= action_work_item_execute(executor
, work_item
);
577 action_work_item_destroy(work_item
);
583 health_code_update();
584 pthread_mutex_lock(&executor
->work
.lock
);
587 if (executor
->should_quit
) {
588 pthread_mutex_unlock(&executor
->work
.lock
);
590 DBG("Left work execution loop");
592 health_code_update();
594 rcu_thread_offline();
595 rcu_unregister_thread();
596 health_unregister(health_sessiond
);
601 static bool shutdown_action_executor_thread(void *_data
)
603 struct action_executor
*executor
= _data
;
605 pthread_mutex_lock(&executor
->work
.lock
);
606 executor
->should_quit
= true;
607 pthread_cond_signal(&executor
->work
.cond
);
608 pthread_mutex_unlock(&executor
->work
.lock
);
612 static void clean_up_action_executor_thread(void *_data
)
614 struct action_executor
*executor
= _data
;
616 assert(cds_list_empty(&executor
->work
.list
));
618 pthread_mutex_destroy(&executor
->work
.lock
);
619 pthread_cond_destroy(&executor
->work
.cond
);
623 struct action_executor
*action_executor_create(
624 struct notification_thread_handle
*handle
)
626 struct action_executor
*executor
= zmalloc(sizeof(*executor
));
632 CDS_INIT_LIST_HEAD(&executor
->work
.list
);
633 pthread_cond_init(&executor
->work
.cond
, NULL
);
634 pthread_mutex_init(&executor
->work
.lock
, NULL
);
635 executor
->notification_thread_handle
= handle
;
637 executor
->thread
= lttng_thread_create(THREAD_NAME
,
638 action_executor_thread
, shutdown_action_executor_thread
,
639 clean_up_action_executor_thread
, executor
);
644 void action_executor_destroy(struct action_executor
*executor
)
646 struct action_work_item
*work_item
, *tmp
;
648 /* TODO Wait for work list to drain? */
649 lttng_thread_shutdown(executor
->thread
);
650 pthread_mutex_lock(&executor
->work
.lock
);
651 if (executor
->work
.pending_count
!= 0) {
653 " trigger action%s still queued for execution and will be discarded",
654 executor
->work
.pending_count
,
655 executor
->work
.pending_count
== 1 ? " is" :
659 cds_list_for_each_entry_safe (
660 work_item
, tmp
, &executor
->work
.list
, list_node
) {
661 WARN("Discarding action work item %" PRIu64
662 " associated to trigger `%s`",
663 work_item
->id
, get_trigger_name(work_item
->trigger
));
664 cds_list_del(&work_item
->list_node
);
665 action_work_item_destroy(work_item
);
667 pthread_mutex_unlock(&executor
->work
.lock
);
668 lttng_thread_put(executor
->thread
);
671 /* RCU read-lock must be held by the caller. */
672 enum action_executor_status
action_executor_enqueue(
673 struct action_executor
*executor
,
674 struct lttng_trigger
*trigger
,
675 struct lttng_evaluation
*evaluation
,
676 const struct lttng_credentials
*object_creds
,
677 struct notification_client_list
*client_list
)
679 enum action_executor_status executor_status
= ACTION_EXECUTOR_STATUS_OK
;
680 const uint64_t work_item_id
= executor
->next_work_item_id
++;
681 struct action_work_item
*work_item
;
684 pthread_mutex_lock(&executor
->work
.lock
);
685 /* Check for queue overflow. */
686 if (executor
->work
.pending_count
>= MAX_QUEUED_WORK_COUNT
) {
687 /* Most likely spammy, remove if it is the case. */
688 DBG("Refusing to enqueue action for trigger `%s` as work item %" PRIu64
689 " (overflow)", get_trigger_name(trigger
), work_item_id
);
690 executor_status
= ACTION_EXECUTOR_STATUS_OVERFLOW
;
694 work_item
= zmalloc(sizeof(*work_item
));
696 PERROR("Failed to allocate action executor work item on behalf of trigger `%s`",
697 get_trigger_name(trigger
));
698 executor_status
= ACTION_EXECUTOR_STATUS_ERROR
;
702 lttng_trigger_get(trigger
);
704 const bool reference_acquired
=
705 notification_client_list_get(client_list
);
707 assert(reference_acquired
);
710 *work_item
= (typeof(*work_item
)){
713 /* Ownership transferred to the work item. */
714 .evaluation
= evaluation
,
716 .is_set
= !!object_creds
,
717 .value
= object_creds
? *object_creds
:
718 (typeof(work_item
->object_creds
.value
)) {},
720 .client_list
= client_list
,
721 .list_node
= CDS_LIST_HEAD_INIT(work_item
->list_node
),
725 cds_list_add_tail(&work_item
->list_node
, &executor
->work
.list
);
726 executor
->work
.pending_count
++;
727 DBG("Enqueued action for trigger `%s` as work item #%" PRIu64
,
728 get_trigger_name(trigger
), work_item_id
);
733 pthread_cond_signal(&executor
->work
.cond
);
735 pthread_mutex_unlock(&executor
->work
.lock
);
737 lttng_evaluation_destroy(evaluation
);
738 return executor_status
;