Commit | Line | Data |
---|---|---|
3bd1e081 | 1 | /* |
a4eb26f0 | 2 | * Copyright (C) 2011 EfficiOS Inc. |
ab5be9fa MJ |
3 | * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
4 | * Copyright (C) 2012 David Goulet <dgoulet@efficios.com> | |
3bd1e081 | 5 | * |
ab5be9fa | 6 | * SPDX-License-Identifier: GPL-2.0-only |
3bd1e081 | 7 | * |
3bd1e081 MD |
8 | */ |
9 | ||
bdc8d1bb | 10 | #include "common/index/ctf-index.h" |
37c7ffca | 11 | #include <stdint.h> |
6c1c0768 | 12 | #define _LGPL_SOURCE |
3bd1e081 | 13 | #include <assert.h> |
3bd1e081 MD |
14 | #include <poll.h> |
15 | #include <pthread.h> | |
16 | #include <stdlib.h> | |
17 | #include <string.h> | |
18 | #include <sys/mman.h> | |
19 | #include <sys/socket.h> | |
20 | #include <sys/types.h> | |
21 | #include <unistd.h> | |
77c7c900 | 22 | #include <inttypes.h> |
331744e3 | 23 | #include <signal.h> |
3bd1e081 | 24 | |
51a9e1c7 | 25 | #include <bin/lttng-consumerd/health-consumerd.h> |
990570ed | 26 | #include <common/common.h> |
fb3a43a9 | 27 | #include <common/utils.h> |
d2956687 | 28 | #include <common/time.h> |
fb3a43a9 | 29 | #include <common/compat/poll.h> |
f263b7fd | 30 | #include <common/compat/endian.h> |
309167d2 | 31 | #include <common/index/index.h> |
10a8a223 | 32 | #include <common/kernel-ctl/kernel-ctl.h> |
00e2e675 | 33 | #include <common/sessiond-comm/relayd.h> |
10a8a223 DG |
34 | #include <common/sessiond-comm/sessiond-comm.h> |
35 | #include <common/kernel-consumer/kernel-consumer.h> | |
00e2e675 | 36 | #include <common/relayd/relayd.h> |
10a8a223 | 37 | #include <common/ust-consumer/ust-consumer.h> |
c8fea79c JR |
38 | #include <common/consumer/consumer-timer.h> |
39 | #include <common/consumer/consumer.h> | |
40 | #include <common/consumer/consumer-stream.h> | |
41 | #include <common/consumer/consumer-testpoint.h> | |
42 | #include <common/align.h> | |
5feafd41 | 43 | #include <common/consumer/consumer-metadata-cache.h> |
d2956687 JG |
44 | #include <common/trace-chunk.h> |
45 | #include <common/trace-chunk-registry.h> | |
46 | #include <common/string-utils/format.h> | |
c35f9726 | 47 | #include <common/dynamic-array.h> |
3bd1e081 MD |
48 | |
49 | struct lttng_consumer_global_data consumer_data = { | |
3bd1e081 MD |
50 | .stream_count = 0, |
51 | .need_update = 1, | |
52 | .type = LTTNG_CONSUMER_UNKNOWN, | |
53 | }; | |
54 | ||
d8ef542d MD |
55 | enum consumer_channel_action { |
56 | CONSUMER_CHANNEL_ADD, | |
a0cbdd2e | 57 | CONSUMER_CHANNEL_DEL, |
d8ef542d MD |
58 | CONSUMER_CHANNEL_QUIT, |
59 | }; | |
60 | ||
61 | struct consumer_channel_msg { | |
62 | enum consumer_channel_action action; | |
a0cbdd2e MD |
63 | struct lttng_consumer_channel *chan; /* add */ |
64 | uint64_t key; /* del */ | |
d8ef542d MD |
65 | }; |
66 | ||
fc0c52e9 JG |
67 | enum open_packet_status { |
68 | OPEN_PACKET_STATUS_OPENED, | |
69 | OPEN_PACKET_STATUS_NO_SPACE, | |
70 | OPEN_PACKET_STATUS_ERROR, | |
71 | }; | |
72 | ||
80957876 | 73 | /* Flag used to temporarily pause data consumption from testpoints. */ |
cf0bcb51 JG |
74 | int data_consumption_paused; |
75 | ||
3bd1e081 MD |
76 | /* |
77 | * Flag to inform the polling thread to quit when all fd hung up. Updated by | |
78 | * the consumer_thread_receive_fds when it notices that all fds has hung up. | |
79 | * Also updated by the signal handler (consumer_should_exit()). Read by the | |
80 | * polling threads. | |
81 | */ | |
10211f5c | 82 | int consumer_quit; |
3bd1e081 | 83 | |
43c34bc3 | 84 | /* |
43c34bc3 DG |
85 | * Global hash table containing respectively metadata and data streams. The |
86 | * stream element in this ht should only be updated by the metadata poll thread | |
87 | * for the metadata and the data poll thread for the data. | |
88 | */ | |
40dc48e0 DG |
89 | static struct lttng_ht *metadata_ht; |
90 | static struct lttng_ht *data_ht; | |
43c34bc3 | 91 | |
5da88b0f MD |
92 | static const char *get_consumer_domain(void) |
93 | { | |
94 | switch (consumer_data.type) { | |
95 | case LTTNG_CONSUMER_KERNEL: | |
96 | return DEFAULT_KERNEL_TRACE_DIR; | |
97 | case LTTNG_CONSUMER64_UST: | |
98 | /* Fall-through. */ | |
99 | case LTTNG_CONSUMER32_UST: | |
100 | return DEFAULT_UST_TRACE_DIR; | |
101 | default: | |
102 | abort(); | |
103 | } | |
104 | } | |
105 | ||
acdb9057 DG |
106 | /* |
107 | * Notify a thread lttng pipe to poll back again. This usually means that some | |
108 | * global state has changed so we just send back the thread in a poll wait | |
109 | * call. | |
110 | */ | |
111 | static void notify_thread_lttng_pipe(struct lttng_pipe *pipe) | |
112 | { | |
113 | struct lttng_consumer_stream *null_stream = NULL; | |
114 | ||
115 | assert(pipe); | |
116 | ||
117 | (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream)); | |
118 | } | |
119 | ||
5c635c72 MD |
120 | static void notify_health_quit_pipe(int *pipe) |
121 | { | |
6cd525e8 | 122 | ssize_t ret; |
5c635c72 | 123 | |
6cd525e8 MD |
124 | ret = lttng_write(pipe[1], "4", 1); |
125 | if (ret < 1) { | |
5c635c72 MD |
126 | PERROR("write consumer health quit"); |
127 | } | |
128 | } | |
129 | ||
d8ef542d MD |
130 | static void notify_channel_pipe(struct lttng_consumer_local_data *ctx, |
131 | struct lttng_consumer_channel *chan, | |
a0cbdd2e | 132 | uint64_t key, |
d8ef542d MD |
133 | enum consumer_channel_action action) |
134 | { | |
135 | struct consumer_channel_msg msg; | |
6cd525e8 | 136 | ssize_t ret; |
d8ef542d | 137 | |
e56251fc DG |
138 | memset(&msg, 0, sizeof(msg)); |
139 | ||
d8ef542d MD |
140 | msg.action = action; |
141 | msg.chan = chan; | |
f21dae48 | 142 | msg.key = key; |
6cd525e8 MD |
143 | ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg)); |
144 | if (ret < sizeof(msg)) { | |
145 | PERROR("notify_channel_pipe write error"); | |
146 | } | |
d8ef542d MD |
147 | } |
148 | ||
a0cbdd2e MD |
149 | void notify_thread_del_channel(struct lttng_consumer_local_data *ctx, |
150 | uint64_t key) | |
151 | { | |
152 | notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL); | |
153 | } | |
154 | ||
d8ef542d MD |
155 | static int read_channel_pipe(struct lttng_consumer_local_data *ctx, |
156 | struct lttng_consumer_channel **chan, | |
a0cbdd2e | 157 | uint64_t *key, |
d8ef542d MD |
158 | enum consumer_channel_action *action) |
159 | { | |
160 | struct consumer_channel_msg msg; | |
6cd525e8 | 161 | ssize_t ret; |
d8ef542d | 162 | |
6cd525e8 MD |
163 | ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg)); |
164 | if (ret < sizeof(msg)) { | |
165 | ret = -1; | |
166 | goto error; | |
d8ef542d | 167 | } |
6cd525e8 MD |
168 | *action = msg.action; |
169 | *chan = msg.chan; | |
170 | *key = msg.key; | |
171 | error: | |
172 | return (int) ret; | |
d8ef542d MD |
173 | } |
174 | ||
212d67a2 DG |
175 | /* |
176 | * Cleanup the stream list of a channel. Those streams are not yet globally | |
177 | * visible | |
178 | */ | |
179 | static void clean_channel_stream_list(struct lttng_consumer_channel *channel) | |
180 | { | |
181 | struct lttng_consumer_stream *stream, *stmp; | |
182 | ||
183 | assert(channel); | |
184 | ||
185 | /* Delete streams that might have been left in the stream list. */ | |
186 | cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head, | |
187 | send_node) { | |
188 | cds_list_del(&stream->send_node); | |
189 | /* | |
190 | * Once a stream is added to this list, the buffers were created so we | |
191 | * have a guarantee that this call will succeed. Setting the monitor | |
192 | * mode to 0 so we don't lock nor try to delete the stream from the | |
193 | * global hash table. | |
194 | */ | |
195 | stream->monitor = 0; | |
196 | consumer_stream_destroy(stream, NULL); | |
197 | } | |
198 | } | |
199 | ||
3bd1e081 MD |
200 | /* |
201 | * Find a stream. The consumer_data.lock must be locked during this | |
202 | * call. | |
203 | */ | |
d88aee68 | 204 | static struct lttng_consumer_stream *find_stream(uint64_t key, |
8389e4f8 | 205 | struct lttng_ht *ht) |
3bd1e081 | 206 | { |
e4421fec | 207 | struct lttng_ht_iter iter; |
d88aee68 | 208 | struct lttng_ht_node_u64 *node; |
e4421fec | 209 | struct lttng_consumer_stream *stream = NULL; |
3bd1e081 | 210 | |
8389e4f8 DG |
211 | assert(ht); |
212 | ||
d88aee68 DG |
213 | /* -1ULL keys are lookup failures */ |
214 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 215 | return NULL; |
7a57cf92 | 216 | } |
e4421fec | 217 | |
6065ceec DG |
218 | rcu_read_lock(); |
219 | ||
d88aee68 DG |
220 | lttng_ht_lookup(ht, &key, &iter); |
221 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
222 | if (node != NULL) { |
223 | stream = caa_container_of(node, struct lttng_consumer_stream, node); | |
3bd1e081 | 224 | } |
e4421fec | 225 | |
6065ceec DG |
226 | rcu_read_unlock(); |
227 | ||
e4421fec | 228 | return stream; |
3bd1e081 MD |
229 | } |
230 | ||
da009f2c | 231 | static void steal_stream_key(uint64_t key, struct lttng_ht *ht) |
7ad0a0cb MD |
232 | { |
233 | struct lttng_consumer_stream *stream; | |
234 | ||
04253271 | 235 | rcu_read_lock(); |
ffe60014 | 236 | stream = find_stream(key, ht); |
04253271 | 237 | if (stream) { |
da009f2c | 238 | stream->key = (uint64_t) -1ULL; |
04253271 MD |
239 | /* |
240 | * We don't want the lookup to match, but we still need | |
241 | * to iterate on this stream when iterating over the hash table. Just | |
242 | * change the node key. | |
243 | */ | |
da009f2c | 244 | stream->node.key = (uint64_t) -1ULL; |
04253271 MD |
245 | } |
246 | rcu_read_unlock(); | |
7ad0a0cb MD |
247 | } |
248 | ||
d56db448 DG |
249 | /* |
250 | * Return a channel object for the given key. | |
251 | * | |
252 | * RCU read side lock MUST be acquired before calling this function and | |
253 | * protects the channel ptr. | |
254 | */ | |
d88aee68 | 255 | struct lttng_consumer_channel *consumer_find_channel(uint64_t key) |
3bd1e081 | 256 | { |
e4421fec | 257 | struct lttng_ht_iter iter; |
d88aee68 | 258 | struct lttng_ht_node_u64 *node; |
e4421fec | 259 | struct lttng_consumer_channel *channel = NULL; |
3bd1e081 | 260 | |
d88aee68 DG |
261 | /* -1ULL keys are lookup failures */ |
262 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 263 | return NULL; |
7a57cf92 | 264 | } |
e4421fec | 265 | |
d88aee68 DG |
266 | lttng_ht_lookup(consumer_data.channel_ht, &key, &iter); |
267 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
268 | if (node != NULL) { |
269 | channel = caa_container_of(node, struct lttng_consumer_channel, node); | |
3bd1e081 | 270 | } |
e4421fec DG |
271 | |
272 | return channel; | |
3bd1e081 MD |
273 | } |
274 | ||
b5a6470f DG |
275 | /* |
276 | * There is a possibility that the consumer does not have enough time between | |
277 | * the close of the channel on the session daemon and the cleanup in here thus | |
278 | * once we have a channel add with an existing key, we know for sure that this | |
279 | * channel will eventually get cleaned up by all streams being closed. | |
280 | * | |
281 | * This function just nullifies the already existing channel key. | |
282 | */ | |
283 | static void steal_channel_key(uint64_t key) | |
284 | { | |
285 | struct lttng_consumer_channel *channel; | |
286 | ||
287 | rcu_read_lock(); | |
288 | channel = consumer_find_channel(key); | |
289 | if (channel) { | |
290 | channel->key = (uint64_t) -1ULL; | |
291 | /* | |
292 | * We don't want the lookup to match, but we still need to iterate on | |
293 | * this channel when iterating over the hash table. Just change the | |
294 | * node key. | |
295 | */ | |
296 | channel->node.key = (uint64_t) -1ULL; | |
297 | } | |
298 | rcu_read_unlock(); | |
299 | } | |
300 | ||
ffe60014 | 301 | static void free_channel_rcu(struct rcu_head *head) |
702b1ea4 | 302 | { |
d88aee68 DG |
303 | struct lttng_ht_node_u64 *node = |
304 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
ffe60014 DG |
305 | struct lttng_consumer_channel *channel = |
306 | caa_container_of(node, struct lttng_consumer_channel, node); | |
702b1ea4 | 307 | |
b83e03c4 MD |
308 | switch (consumer_data.type) { |
309 | case LTTNG_CONSUMER_KERNEL: | |
310 | break; | |
311 | case LTTNG_CONSUMER32_UST: | |
312 | case LTTNG_CONSUMER64_UST: | |
313 | lttng_ustconsumer_free_channel(channel); | |
314 | break; | |
315 | default: | |
316 | ERR("Unknown consumer_data type"); | |
317 | abort(); | |
318 | } | |
ffe60014 | 319 | free(channel); |
702b1ea4 MD |
320 | } |
321 | ||
00e2e675 DG |
322 | /* |
323 | * RCU protected relayd socket pair free. | |
324 | */ | |
ffe60014 | 325 | static void free_relayd_rcu(struct rcu_head *head) |
00e2e675 | 326 | { |
d88aee68 DG |
327 | struct lttng_ht_node_u64 *node = |
328 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
00e2e675 DG |
329 | struct consumer_relayd_sock_pair *relayd = |
330 | caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
331 | ||
8994307f DG |
332 | /* |
333 | * Close all sockets. This is done in the call RCU since we don't want the | |
334 | * socket fds to be reassigned thus potentially creating bad state of the | |
335 | * relayd object. | |
336 | * | |
337 | * We do not have to lock the control socket mutex here since at this stage | |
338 | * there is no one referencing to this relayd object. | |
339 | */ | |
340 | (void) relayd_close(&relayd->control_sock); | |
341 | (void) relayd_close(&relayd->data_sock); | |
342 | ||
3a84e2f3 | 343 | pthread_mutex_destroy(&relayd->ctrl_sock_mutex); |
00e2e675 DG |
344 | free(relayd); |
345 | } | |
346 | ||
347 | /* | |
348 | * Destroy and free relayd socket pair object. | |
00e2e675 | 349 | */ |
51230d70 | 350 | void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
351 | { |
352 | int ret; | |
353 | struct lttng_ht_iter iter; | |
354 | ||
173af62f DG |
355 | if (relayd == NULL) { |
356 | return; | |
357 | } | |
358 | ||
00e2e675 DG |
359 | DBG("Consumer destroy and close relayd socket pair"); |
360 | ||
361 | iter.iter.node = &relayd->node.node; | |
362 | ret = lttng_ht_del(consumer_data.relayd_ht, &iter); | |
173af62f | 363 | if (ret != 0) { |
8994307f | 364 | /* We assume the relayd is being or is destroyed */ |
173af62f DG |
365 | return; |
366 | } | |
00e2e675 | 367 | |
00e2e675 | 368 | /* RCU free() call */ |
ffe60014 DG |
369 | call_rcu(&relayd->node.head, free_relayd_rcu); |
370 | } | |
371 | ||
372 | /* | |
373 | * Remove a channel from the global list protected by a mutex. This function is | |
374 | * also responsible for freeing its data structures. | |
375 | */ | |
376 | void consumer_del_channel(struct lttng_consumer_channel *channel) | |
377 | { | |
ffe60014 DG |
378 | struct lttng_ht_iter iter; |
379 | ||
d88aee68 | 380 | DBG("Consumer delete channel key %" PRIu64, channel->key); |
ffe60014 DG |
381 | |
382 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 383 | pthread_mutex_lock(&channel->lock); |
ffe60014 | 384 | |
212d67a2 DG |
385 | /* Destroy streams that might have been left in the stream list. */ |
386 | clean_channel_stream_list(channel); | |
51e762e5 | 387 | |
d3e2ba59 JD |
388 | if (channel->live_timer_enabled == 1) { |
389 | consumer_timer_live_stop(channel); | |
390 | } | |
e9404c27 JG |
391 | if (channel->monitor_timer_enabled == 1) { |
392 | consumer_timer_monitor_stop(channel); | |
393 | } | |
d3e2ba59 | 394 | |
ffe60014 DG |
395 | switch (consumer_data.type) { |
396 | case LTTNG_CONSUMER_KERNEL: | |
397 | break; | |
398 | case LTTNG_CONSUMER32_UST: | |
399 | case LTTNG_CONSUMER64_UST: | |
400 | lttng_ustconsumer_del_channel(channel); | |
401 | break; | |
402 | default: | |
403 | ERR("Unknown consumer_data type"); | |
404 | assert(0); | |
405 | goto end; | |
406 | } | |
407 | ||
d2956687 JG |
408 | lttng_trace_chunk_put(channel->trace_chunk); |
409 | channel->trace_chunk = NULL; | |
5c3892a6 | 410 | |
d2956687 JG |
411 | if (channel->is_published) { |
412 | int ret; | |
413 | ||
414 | rcu_read_lock(); | |
415 | iter.iter.node = &channel->node.node; | |
416 | ret = lttng_ht_del(consumer_data.channel_ht, &iter); | |
417 | assert(!ret); | |
ffe60014 | 418 | |
d2956687 JG |
419 | iter.iter.node = &channel->channels_by_session_id_ht_node.node; |
420 | ret = lttng_ht_del(consumer_data.channels_by_session_id_ht, | |
421 | &iter); | |
422 | assert(!ret); | |
423 | rcu_read_unlock(); | |
424 | } | |
425 | ||
b6921a17 JG |
426 | channel->is_deleted = true; |
427 | call_rcu(&channel->node.head, free_channel_rcu); | |
ffe60014 | 428 | end: |
a9838785 | 429 | pthread_mutex_unlock(&channel->lock); |
ffe60014 | 430 | pthread_mutex_unlock(&consumer_data.lock); |
00e2e675 DG |
431 | } |
432 | ||
228b5bf7 DG |
433 | /* |
434 | * Iterate over the relayd hash table and destroy each element. Finally, | |
435 | * destroy the whole hash table. | |
436 | */ | |
437 | static void cleanup_relayd_ht(void) | |
438 | { | |
439 | struct lttng_ht_iter iter; | |
440 | struct consumer_relayd_sock_pair *relayd; | |
441 | ||
442 | rcu_read_lock(); | |
443 | ||
444 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
445 | node.node) { | |
51230d70 | 446 | consumer_destroy_relayd(relayd); |
228b5bf7 DG |
447 | } |
448 | ||
228b5bf7 | 449 | rcu_read_unlock(); |
36b588ed MD |
450 | |
451 | lttng_ht_destroy(consumer_data.relayd_ht); | |
228b5bf7 DG |
452 | } |
453 | ||
8994307f DG |
454 | /* |
455 | * Update the end point status of all streams having the given network sequence | |
456 | * index (relayd index). | |
457 | * | |
458 | * It's atomically set without having the stream mutex locked which is fine | |
459 | * because we handle the write/read race with a pipe wakeup for each thread. | |
460 | */ | |
da009f2c | 461 | static void update_endpoint_status_by_netidx(uint64_t net_seq_idx, |
8994307f DG |
462 | enum consumer_endpoint_status status) |
463 | { | |
464 | struct lttng_ht_iter iter; | |
465 | struct lttng_consumer_stream *stream; | |
466 | ||
da009f2c | 467 | DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx); |
8994307f DG |
468 | |
469 | rcu_read_lock(); | |
470 | ||
471 | /* Let's begin with metadata */ | |
472 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
473 | if (stream->net_seq_idx == net_seq_idx) { | |
474 | uatomic_set(&stream->endpoint_status, status); | |
475 | DBG("Delete flag set to metadata stream %d", stream->wait_fd); | |
476 | } | |
477 | } | |
478 | ||
479 | /* Follow up by the data streams */ | |
480 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
481 | if (stream->net_seq_idx == net_seq_idx) { | |
482 | uatomic_set(&stream->endpoint_status, status); | |
483 | DBG("Delete flag set to data stream %d", stream->wait_fd); | |
484 | } | |
485 | } | |
486 | rcu_read_unlock(); | |
487 | } | |
488 | ||
489 | /* | |
490 | * Cleanup a relayd object by flagging every associated streams for deletion, | |
491 | * destroying the object meaning removing it from the relayd hash table, | |
492 | * closing the sockets and freeing the memory in a RCU call. | |
493 | * | |
494 | * If a local data context is available, notify the threads that the streams' | |
495 | * state have changed. | |
496 | */ | |
9276e5c8 | 497 | void lttng_consumer_cleanup_relayd(struct consumer_relayd_sock_pair *relayd) |
8994307f | 498 | { |
da009f2c | 499 | uint64_t netidx; |
8994307f DG |
500 | |
501 | assert(relayd); | |
502 | ||
9276e5c8 | 503 | DBG("Cleaning up relayd object ID %"PRIu64, relayd->net_seq_idx); |
9617607b | 504 | |
8994307f DG |
505 | /* Save the net sequence index before destroying the object */ |
506 | netidx = relayd->net_seq_idx; | |
507 | ||
508 | /* | |
509 | * Delete the relayd from the relayd hash table, close the sockets and free | |
510 | * the object in a RCU call. | |
511 | */ | |
51230d70 | 512 | consumer_destroy_relayd(relayd); |
8994307f DG |
513 | |
514 | /* Set inactive endpoint to all streams */ | |
515 | update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE); | |
516 | ||
517 | /* | |
518 | * With a local data context, notify the threads that the streams' state | |
519 | * have changed. The write() action on the pipe acts as an "implicit" | |
520 | * memory barrier ordering the updates of the end point status from the | |
521 | * read of this status which happens AFTER receiving this notify. | |
522 | */ | |
9276e5c8 JR |
523 | notify_thread_lttng_pipe(relayd->ctx->consumer_data_pipe); |
524 | notify_thread_lttng_pipe(relayd->ctx->consumer_metadata_pipe); | |
8994307f DG |
525 | } |
526 | ||
a6ba4fe1 DG |
527 | /* |
528 | * Flag a relayd socket pair for destruction. Destroy it if the refcount | |
529 | * reaches zero. | |
530 | * | |
531 | * RCU read side lock MUST be aquired before calling this function. | |
532 | */ | |
533 | void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) | |
534 | { | |
535 | assert(relayd); | |
536 | ||
537 | /* Set destroy flag for this object */ | |
538 | uatomic_set(&relayd->destroy_flag, 1); | |
539 | ||
540 | /* Destroy the relayd if refcount is 0 */ | |
541 | if (uatomic_read(&relayd->refcount) == 0) { | |
51230d70 | 542 | consumer_destroy_relayd(relayd); |
a6ba4fe1 DG |
543 | } |
544 | } | |
545 | ||
3bd1e081 | 546 | /* |
1d1a276c DG |
547 | * Completly destroy stream from every visiable data structure and the given |
548 | * hash table if one. | |
549 | * | |
550 | * One this call returns, the stream object is not longer usable nor visible. | |
3bd1e081 | 551 | */ |
e316aad5 DG |
552 | void consumer_del_stream(struct lttng_consumer_stream *stream, |
553 | struct lttng_ht *ht) | |
3bd1e081 | 554 | { |
1d1a276c | 555 | consumer_stream_destroy(stream, ht); |
3bd1e081 MD |
556 | } |
557 | ||
5ab66908 MD |
558 | /* |
559 | * XXX naming of del vs destroy is all mixed up. | |
560 | */ | |
561 | void consumer_del_stream_for_data(struct lttng_consumer_stream *stream) | |
562 | { | |
563 | consumer_stream_destroy(stream, data_ht); | |
564 | } | |
565 | ||
566 | void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream) | |
567 | { | |
568 | consumer_stream_destroy(stream, metadata_ht); | |
569 | } | |
570 | ||
d9a2e16e JD |
571 | void consumer_stream_update_channel_attributes( |
572 | struct lttng_consumer_stream *stream, | |
573 | struct lttng_consumer_channel *channel) | |
574 | { | |
575 | stream->channel_read_only_attributes.tracefile_size = | |
576 | channel->tracefile_size; | |
d9a2e16e JD |
577 | } |
578 | ||
3bd1e081 MD |
579 | /* |
580 | * Add a stream to the global list protected by a mutex. | |
581 | */ | |
66d583dc | 582 | void consumer_add_data_stream(struct lttng_consumer_stream *stream) |
3bd1e081 | 583 | { |
5ab66908 | 584 | struct lttng_ht *ht = data_ht; |
3bd1e081 | 585 | |
e316aad5 | 586 | assert(stream); |
43c34bc3 | 587 | assert(ht); |
c77fc10a | 588 | |
d88aee68 | 589 | DBG3("Adding consumer stream %" PRIu64, stream->key); |
e316aad5 DG |
590 | |
591 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 592 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 593 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 594 | pthread_mutex_lock(&stream->lock); |
b0b335c8 | 595 | rcu_read_lock(); |
e316aad5 | 596 | |
43c34bc3 | 597 | /* Steal stream identifier to avoid having streams with the same key */ |
ffe60014 | 598 | steal_stream_key(stream->key, ht); |
43c34bc3 | 599 | |
d88aee68 | 600 | lttng_ht_add_unique_u64(ht, &stream->node); |
00e2e675 | 601 | |
d8ef542d MD |
602 | lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, |
603 | &stream->node_channel_id); | |
604 | ||
ca22feea DG |
605 | /* |
606 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
607 | * the key since the HT does not use it and we allow to add redundant keys | |
608 | * into this table. | |
609 | */ | |
d88aee68 | 610 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 611 | |
e316aad5 | 612 | /* |
ffe60014 DG |
613 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
614 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
615 | * causes the count to become 0 also causes a stream to be added. The |
616 | * channel deletion will thus be triggered by the following removal of this | |
617 | * stream. | |
618 | */ | |
ffe60014 | 619 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
620 | /* Increment refcount before decrementing nb_init_stream_left */ |
621 | cmm_smp_wmb(); | |
ffe60014 | 622 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
623 | } |
624 | ||
625 | /* Update consumer data once the node is inserted. */ | |
3bd1e081 MD |
626 | consumer_data.stream_count++; |
627 | consumer_data.need_update = 1; | |
628 | ||
e316aad5 | 629 | rcu_read_unlock(); |
2e818a6a | 630 | pthread_mutex_unlock(&stream->lock); |
ec6ea7d0 | 631 | pthread_mutex_unlock(&stream->chan->timer_lock); |
a9838785 | 632 | pthread_mutex_unlock(&stream->chan->lock); |
3bd1e081 | 633 | pthread_mutex_unlock(&consumer_data.lock); |
3bd1e081 MD |
634 | } |
635 | ||
00e2e675 | 636 | /* |
3f8e211f DG |
637 | * Add relayd socket to global consumer data hashtable. RCU read side lock MUST |
638 | * be acquired before calling this. | |
00e2e675 | 639 | */ |
d09e1200 | 640 | static int add_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
641 | { |
642 | int ret = 0; | |
d88aee68 | 643 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
644 | struct lttng_ht_iter iter; |
645 | ||
ffe60014 | 646 | assert(relayd); |
00e2e675 | 647 | |
00e2e675 | 648 | lttng_ht_lookup(consumer_data.relayd_ht, |
d88aee68 DG |
649 | &relayd->net_seq_idx, &iter); |
650 | node = lttng_ht_iter_get_node_u64(&iter); | |
00e2e675 | 651 | if (node != NULL) { |
00e2e675 DG |
652 | goto end; |
653 | } | |
d88aee68 | 654 | lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node); |
00e2e675 | 655 | |
00e2e675 DG |
656 | end: |
657 | return ret; | |
658 | } | |
659 | ||
660 | /* | |
661 | * Allocate and return a consumer relayd socket. | |
662 | */ | |
027a694f | 663 | static struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( |
da009f2c | 664 | uint64_t net_seq_idx) |
00e2e675 DG |
665 | { |
666 | struct consumer_relayd_sock_pair *obj = NULL; | |
667 | ||
da009f2c MD |
668 | /* net sequence index of -1 is a failure */ |
669 | if (net_seq_idx == (uint64_t) -1ULL) { | |
00e2e675 DG |
670 | goto error; |
671 | } | |
672 | ||
673 | obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); | |
674 | if (obj == NULL) { | |
675 | PERROR("zmalloc relayd sock"); | |
676 | goto error; | |
677 | } | |
678 | ||
679 | obj->net_seq_idx = net_seq_idx; | |
680 | obj->refcount = 0; | |
173af62f | 681 | obj->destroy_flag = 0; |
f96e4545 MD |
682 | obj->control_sock.sock.fd = -1; |
683 | obj->data_sock.sock.fd = -1; | |
d88aee68 | 684 | lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx); |
00e2e675 DG |
685 | pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); |
686 | ||
687 | error: | |
688 | return obj; | |
689 | } | |
690 | ||
691 | /* | |
692 | * Find a relayd socket pair in the global consumer data. | |
693 | * | |
694 | * Return the object if found else NULL. | |
b0b335c8 MD |
695 | * RCU read-side lock must be held across this call and while using the |
696 | * returned object. | |
00e2e675 | 697 | */ |
d88aee68 | 698 | struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key) |
00e2e675 DG |
699 | { |
700 | struct lttng_ht_iter iter; | |
d88aee68 | 701 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
702 | struct consumer_relayd_sock_pair *relayd = NULL; |
703 | ||
704 | /* Negative keys are lookup failures */ | |
d88aee68 | 705 | if (key == (uint64_t) -1ULL) { |
00e2e675 DG |
706 | goto error; |
707 | } | |
708 | ||
d88aee68 | 709 | lttng_ht_lookup(consumer_data.relayd_ht, &key, |
00e2e675 | 710 | &iter); |
d88aee68 | 711 | node = lttng_ht_iter_get_node_u64(&iter); |
00e2e675 DG |
712 | if (node != NULL) { |
713 | relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
714 | } | |
715 | ||
00e2e675 DG |
716 | error: |
717 | return relayd; | |
718 | } | |
719 | ||
10a50311 JD |
720 | /* |
721 | * Find a relayd and send the stream | |
722 | * | |
723 | * Returns 0 on success, < 0 on error | |
724 | */ | |
725 | int consumer_send_relayd_stream(struct lttng_consumer_stream *stream, | |
726 | char *path) | |
727 | { | |
728 | int ret = 0; | |
729 | struct consumer_relayd_sock_pair *relayd; | |
730 | ||
731 | assert(stream); | |
732 | assert(stream->net_seq_idx != -1ULL); | |
733 | assert(path); | |
734 | ||
735 | /* The stream is not metadata. Get relayd reference if exists. */ | |
736 | rcu_read_lock(); | |
737 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
738 | if (relayd != NULL) { | |
739 | /* Add stream on the relayd */ | |
740 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
741 | ret = relayd_add_stream(&relayd->control_sock, stream->name, | |
5da88b0f | 742 | get_consumer_domain(), path, &stream->relayd_stream_id, |
d2956687 JG |
743 | stream->chan->tracefile_size, |
744 | stream->chan->tracefile_count, | |
745 | stream->trace_chunk); | |
10a50311 JD |
746 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
747 | if (ret < 0) { | |
9276e5c8 JR |
748 | ERR("Relayd add stream failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
749 | lttng_consumer_cleanup_relayd(relayd); | |
10a50311 JD |
750 | goto end; |
751 | } | |
1c20f0e2 | 752 | |
10a50311 | 753 | uatomic_inc(&relayd->refcount); |
d01178b6 | 754 | stream->sent_to_relayd = 1; |
10a50311 JD |
755 | } else { |
756 | ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.", | |
757 | stream->key, stream->net_seq_idx); | |
758 | ret = -1; | |
759 | goto end; | |
760 | } | |
761 | ||
762 | DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64, | |
763 | stream->name, stream->key, stream->net_seq_idx); | |
764 | ||
765 | end: | |
766 | rcu_read_unlock(); | |
767 | return ret; | |
768 | } | |
769 | ||
a4baae1b JD |
770 | /* |
771 | * Find a relayd and send the streams sent message | |
772 | * | |
773 | * Returns 0 on success, < 0 on error | |
774 | */ | |
775 | int consumer_send_relayd_streams_sent(uint64_t net_seq_idx) | |
776 | { | |
777 | int ret = 0; | |
778 | struct consumer_relayd_sock_pair *relayd; | |
779 | ||
780 | assert(net_seq_idx != -1ULL); | |
781 | ||
782 | /* The stream is not metadata. Get relayd reference if exists. */ | |
783 | rcu_read_lock(); | |
784 | relayd = consumer_find_relayd(net_seq_idx); | |
785 | if (relayd != NULL) { | |
786 | /* Add stream on the relayd */ | |
787 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
788 | ret = relayd_streams_sent(&relayd->control_sock); | |
789 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
790 | if (ret < 0) { | |
9276e5c8 JR |
791 | ERR("Relayd streams sent failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
792 | lttng_consumer_cleanup_relayd(relayd); | |
a4baae1b JD |
793 | goto end; |
794 | } | |
795 | } else { | |
796 | ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.", | |
797 | net_seq_idx); | |
798 | ret = -1; | |
799 | goto end; | |
800 | } | |
801 | ||
802 | ret = 0; | |
803 | DBG("All streams sent relayd id %" PRIu64, net_seq_idx); | |
804 | ||
805 | end: | |
806 | rcu_read_unlock(); | |
807 | return ret; | |
808 | } | |
809 | ||
10a50311 JD |
810 | /* |
811 | * Find a relayd and close the stream | |
812 | */ | |
813 | void close_relayd_stream(struct lttng_consumer_stream *stream) | |
814 | { | |
815 | struct consumer_relayd_sock_pair *relayd; | |
816 | ||
817 | /* The stream is not metadata. Get relayd reference if exists. */ | |
818 | rcu_read_lock(); | |
819 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
820 | if (relayd) { | |
821 | consumer_stream_relayd_close(stream, relayd); | |
822 | } | |
823 | rcu_read_unlock(); | |
824 | } | |
825 | ||
00e2e675 DG |
826 | /* |
827 | * Handle stream for relayd transmission if the stream applies for network | |
828 | * streaming where the net sequence index is set. | |
829 | * | |
830 | * Return destination file descriptor or negative value on error. | |
831 | */ | |
6197aea7 | 832 | static int write_relayd_stream_header(struct lttng_consumer_stream *stream, |
1d4dfdef DG |
833 | size_t data_size, unsigned long padding, |
834 | struct consumer_relayd_sock_pair *relayd) | |
00e2e675 DG |
835 | { |
836 | int outfd = -1, ret; | |
00e2e675 DG |
837 | struct lttcomm_relayd_data_hdr data_hdr; |
838 | ||
839 | /* Safety net */ | |
840 | assert(stream); | |
6197aea7 | 841 | assert(relayd); |
00e2e675 DG |
842 | |
843 | /* Reset data header */ | |
844 | memset(&data_hdr, 0, sizeof(data_hdr)); | |
845 | ||
00e2e675 DG |
846 | if (stream->metadata_flag) { |
847 | /* Caller MUST acquire the relayd control socket lock */ | |
848 | ret = relayd_send_metadata(&relayd->control_sock, data_size); | |
849 | if (ret < 0) { | |
850 | goto error; | |
851 | } | |
852 | ||
853 | /* Metadata are always sent on the control socket. */ | |
6151a90f | 854 | outfd = relayd->control_sock.sock.fd; |
00e2e675 DG |
855 | } else { |
856 | /* Set header with stream information */ | |
857 | data_hdr.stream_id = htobe64(stream->relayd_stream_id); | |
858 | data_hdr.data_size = htobe32(data_size); | |
1d4dfdef | 859 | data_hdr.padding_size = htobe32(padding); |
c35f9726 | 860 | |
39df6d9f DG |
861 | /* |
862 | * Note that net_seq_num below is assigned with the *current* value of | |
863 | * next_net_seq_num and only after that the next_net_seq_num will be | |
864 | * increment. This is why when issuing a command on the relayd using | |
865 | * this next value, 1 should always be substracted in order to compare | |
866 | * the last seen sequence number on the relayd side to the last sent. | |
867 | */ | |
3604f373 | 868 | data_hdr.net_seq_num = htobe64(stream->next_net_seq_num); |
00e2e675 DG |
869 | /* Other fields are zeroed previously */ |
870 | ||
871 | ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, | |
872 | sizeof(data_hdr)); | |
873 | if (ret < 0) { | |
874 | goto error; | |
875 | } | |
876 | ||
3604f373 DG |
877 | ++stream->next_net_seq_num; |
878 | ||
00e2e675 | 879 | /* Set to go on data socket */ |
6151a90f | 880 | outfd = relayd->data_sock.sock.fd; |
00e2e675 DG |
881 | } |
882 | ||
883 | error: | |
884 | return outfd; | |
885 | } | |
886 | ||
920f2ddb JG |
887 | /* |
888 | * Write a character on the metadata poll pipe to wake the metadata thread. | |
889 | * Returns 0 on success, -1 on error. | |
890 | */ | |
891 | int consumer_metadata_wakeup_pipe(const struct lttng_consumer_channel *channel) | |
892 | { | |
893 | int ret = 0; | |
894 | ||
895 | DBG("Waking up metadata poll thread (writing to pipe): channel name = '%s'", | |
896 | channel->name); | |
897 | if (channel->monitor && channel->metadata_stream) { | |
898 | const char dummy = 'c'; | |
899 | const ssize_t write_ret = lttng_write( | |
900 | channel->metadata_stream->ust_metadata_poll_pipe[1], | |
901 | &dummy, 1); | |
902 | ||
903 | if (write_ret < 1) { | |
904 | if (errno == EWOULDBLOCK) { | |
905 | /* | |
906 | * This is fine, the metadata poll thread | |
907 | * is having a hard time keeping-up, but | |
908 | * it will eventually wake-up and consume | |
909 | * the available data. | |
910 | */ | |
911 | ret = 0; | |
912 | } else { | |
913 | PERROR("Failed to write to UST metadata pipe while attempting to wake-up the metadata poll thread"); | |
914 | ret = -1; | |
915 | goto end; | |
916 | } | |
917 | } | |
918 | } | |
919 | ||
920 | end: | |
921 | return ret; | |
922 | } | |
923 | ||
d2956687 JG |
924 | /* |
925 | * Trigger a dump of the metadata content. Following/during the succesful | |
926 | * completion of this call, the metadata poll thread will start receiving | |
927 | * metadata packets to consume. | |
928 | * | |
929 | * The caller must hold the channel and stream locks. | |
930 | */ | |
931 | static | |
932 | int consumer_metadata_stream_dump(struct lttng_consumer_stream *stream) | |
933 | { | |
934 | int ret; | |
935 | ||
936 | ASSERT_LOCKED(stream->chan->lock); | |
937 | ASSERT_LOCKED(stream->lock); | |
938 | assert(stream->metadata_flag); | |
939 | assert(stream->chan->trace_chunk); | |
940 | ||
941 | switch (consumer_data.type) { | |
942 | case LTTNG_CONSUMER_KERNEL: | |
943 | /* | |
944 | * Reset the position of what has been read from the | |
945 | * metadata cache to 0 so we can dump it again. | |
946 | */ | |
947 | ret = kernctl_metadata_cache_dump(stream->wait_fd); | |
948 | break; | |
949 | case LTTNG_CONSUMER32_UST: | |
950 | case LTTNG_CONSUMER64_UST: | |
951 | /* | |
952 | * Reset the position pushed from the metadata cache so it | |
953 | * will write from the beginning on the next push. | |
954 | */ | |
955 | stream->ust_metadata_pushed = 0; | |
956 | ret = consumer_metadata_wakeup_pipe(stream->chan); | |
957 | break; | |
958 | default: | |
959 | ERR("Unknown consumer_data type"); | |
960 | abort(); | |
961 | } | |
962 | if (ret < 0) { | |
963 | ERR("Failed to dump the metadata cache"); | |
964 | } | |
965 | return ret; | |
966 | } | |
967 | ||
968 | static | |
969 | int lttng_consumer_channel_set_trace_chunk( | |
970 | struct lttng_consumer_channel *channel, | |
971 | struct lttng_trace_chunk *new_trace_chunk) | |
972 | { | |
d2956687 | 973 | pthread_mutex_lock(&channel->lock); |
b6921a17 JG |
974 | if (channel->is_deleted) { |
975 | /* | |
976 | * The channel has been logically deleted and should no longer | |
977 | * be used. It has released its reference to its current trace | |
978 | * chunk and should not acquire a new one. | |
979 | * | |
980 | * Return success as there is nothing for the caller to do. | |
981 | */ | |
982 | goto end; | |
983 | } | |
d2956687 JG |
984 | |
985 | /* | |
986 | * The acquisition of the reference cannot fail (barring | |
987 | * a severe internal error) since a reference to the published | |
988 | * chunk is already held by the caller. | |
989 | */ | |
990 | if (new_trace_chunk) { | |
991 | const bool acquired_reference = lttng_trace_chunk_get( | |
992 | new_trace_chunk); | |
993 | ||
994 | assert(acquired_reference); | |
995 | } | |
996 | ||
997 | lttng_trace_chunk_put(channel->trace_chunk); | |
998 | channel->trace_chunk = new_trace_chunk; | |
d2956687 JG |
999 | end: |
1000 | pthread_mutex_unlock(&channel->lock); | |
ce1aa6fe | 1001 | return 0; |
d2956687 JG |
1002 | } |
1003 | ||
3bd1e081 | 1004 | /* |
ffe60014 DG |
1005 | * Allocate and return a new lttng_consumer_channel object using the given key |
1006 | * to initialize the hash table node. | |
1007 | * | |
1008 | * On error, return NULL. | |
3bd1e081 | 1009 | */ |
886224ff | 1010 | struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, |
ffe60014 | 1011 | uint64_t session_id, |
d2956687 | 1012 | const uint64_t *chunk_id, |
ffe60014 DG |
1013 | const char *pathname, |
1014 | const char *name, | |
57a269f2 | 1015 | uint64_t relayd_id, |
1624d5b7 JD |
1016 | enum lttng_event_output output, |
1017 | uint64_t tracefile_size, | |
2bba9e53 | 1018 | uint64_t tracefile_count, |
1950109e | 1019 | uint64_t session_id_per_pid, |
ecc48a90 | 1020 | unsigned int monitor, |
d7ba1388 | 1021 | unsigned int live_timer_interval, |
ee8935c3 | 1022 | bool is_in_live_session, |
3d071855 | 1023 | const char *root_shm_path, |
d7ba1388 | 1024 | const char *shm_path) |
3bd1e081 | 1025 | { |
d2956687 JG |
1026 | struct lttng_consumer_channel *channel = NULL; |
1027 | struct lttng_trace_chunk *trace_chunk = NULL; | |
1028 | ||
1029 | if (chunk_id) { | |
1030 | trace_chunk = lttng_trace_chunk_registry_find_chunk( | |
1031 | consumer_data.chunk_registry, session_id, | |
1032 | *chunk_id); | |
1033 | if (!trace_chunk) { | |
1034 | ERR("Failed to find trace chunk reference during creation of channel"); | |
1035 | goto end; | |
1036 | } | |
1037 | } | |
3bd1e081 | 1038 | |
276b26d1 | 1039 | channel = zmalloc(sizeof(*channel)); |
3bd1e081 | 1040 | if (channel == NULL) { |
7a57cf92 | 1041 | PERROR("malloc struct lttng_consumer_channel"); |
3bd1e081 MD |
1042 | goto end; |
1043 | } | |
ffe60014 DG |
1044 | |
1045 | channel->key = key; | |
3bd1e081 | 1046 | channel->refcount = 0; |
ffe60014 | 1047 | channel->session_id = session_id; |
1950109e | 1048 | channel->session_id_per_pid = session_id_per_pid; |
ffe60014 | 1049 | channel->relayd_id = relayd_id; |
1624d5b7 JD |
1050 | channel->tracefile_size = tracefile_size; |
1051 | channel->tracefile_count = tracefile_count; | |
2bba9e53 | 1052 | channel->monitor = monitor; |
ecc48a90 | 1053 | channel->live_timer_interval = live_timer_interval; |
ee8935c3 | 1054 | channel->is_live = is_in_live_session; |
a9838785 | 1055 | pthread_mutex_init(&channel->lock, NULL); |
ec6ea7d0 | 1056 | pthread_mutex_init(&channel->timer_lock, NULL); |
ffe60014 | 1057 | |
0c759fc9 DG |
1058 | switch (output) { |
1059 | case LTTNG_EVENT_SPLICE: | |
1060 | channel->output = CONSUMER_CHANNEL_SPLICE; | |
1061 | break; | |
1062 | case LTTNG_EVENT_MMAP: | |
1063 | channel->output = CONSUMER_CHANNEL_MMAP; | |
1064 | break; | |
1065 | default: | |
1066 | assert(0); | |
1067 | free(channel); | |
1068 | channel = NULL; | |
1069 | goto end; | |
1070 | } | |
1071 | ||
07b86b52 JD |
1072 | /* |
1073 | * In monitor mode, the streams associated with the channel will be put in | |
1074 | * a special list ONLY owned by this channel. So, the refcount is set to 1 | |
1075 | * here meaning that the channel itself has streams that are referenced. | |
1076 | * | |
1077 | * On a channel deletion, once the channel is no longer visible, the | |
1078 | * refcount is decremented and checked for a zero value to delete it. With | |
1079 | * streams in no monitor mode, it will now be safe to destroy the channel. | |
1080 | */ | |
1081 | if (!channel->monitor) { | |
1082 | channel->refcount = 1; | |
1083 | } | |
1084 | ||
ffe60014 DG |
1085 | strncpy(channel->pathname, pathname, sizeof(channel->pathname)); |
1086 | channel->pathname[sizeof(channel->pathname) - 1] = '\0'; | |
1087 | ||
1088 | strncpy(channel->name, name, sizeof(channel->name)); | |
1089 | channel->name[sizeof(channel->name) - 1] = '\0'; | |
1090 | ||
3d071855 MD |
1091 | if (root_shm_path) { |
1092 | strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path)); | |
1093 | channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0'; | |
1094 | } | |
d7ba1388 MD |
1095 | if (shm_path) { |
1096 | strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path)); | |
1097 | channel->shm_path[sizeof(channel->shm_path) - 1] = '\0'; | |
1098 | } | |
1099 | ||
d88aee68 | 1100 | lttng_ht_node_init_u64(&channel->node, channel->key); |
5c3892a6 JG |
1101 | lttng_ht_node_init_u64(&channel->channels_by_session_id_ht_node, |
1102 | channel->session_id); | |
d8ef542d MD |
1103 | |
1104 | channel->wait_fd = -1; | |
ffe60014 DG |
1105 | CDS_INIT_LIST_HEAD(&channel->streams.head); |
1106 | ||
d2956687 JG |
1107 | if (trace_chunk) { |
1108 | int ret = lttng_consumer_channel_set_trace_chunk(channel, | |
1109 | trace_chunk); | |
1110 | if (ret) { | |
1111 | goto error; | |
1112 | } | |
1113 | } | |
1114 | ||
62a7b8ed | 1115 | DBG("Allocated channel (key %" PRIu64 ")", channel->key); |
3bd1e081 | 1116 | |
3bd1e081 | 1117 | end: |
d2956687 | 1118 | lttng_trace_chunk_put(trace_chunk); |
3bd1e081 | 1119 | return channel; |
d2956687 JG |
1120 | error: |
1121 | consumer_del_channel(channel); | |
1122 | channel = NULL; | |
1123 | goto end; | |
3bd1e081 MD |
1124 | } |
1125 | ||
1126 | /* | |
1127 | * Add a channel to the global list protected by a mutex. | |
821fffb2 | 1128 | * |
b5a6470f | 1129 | * Always return 0 indicating success. |
3bd1e081 | 1130 | */ |
d8ef542d MD |
1131 | int consumer_add_channel(struct lttng_consumer_channel *channel, |
1132 | struct lttng_consumer_local_data *ctx) | |
3bd1e081 | 1133 | { |
3bd1e081 | 1134 | pthread_mutex_lock(&consumer_data.lock); |
a9838785 | 1135 | pthread_mutex_lock(&channel->lock); |
ec6ea7d0 | 1136 | pthread_mutex_lock(&channel->timer_lock); |
c77fc10a | 1137 | |
b5a6470f DG |
1138 | /* |
1139 | * This gives us a guarantee that the channel we are about to add to the | |
1140 | * channel hash table will be unique. See this function comment on the why | |
1141 | * we need to steel the channel key at this stage. | |
1142 | */ | |
1143 | steal_channel_key(channel->key); | |
c77fc10a | 1144 | |
b5a6470f | 1145 | rcu_read_lock(); |
d88aee68 | 1146 | lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node); |
5c3892a6 JG |
1147 | lttng_ht_add_u64(consumer_data.channels_by_session_id_ht, |
1148 | &channel->channels_by_session_id_ht_node); | |
6065ceec | 1149 | rcu_read_unlock(); |
d2956687 | 1150 | channel->is_published = true; |
b5a6470f | 1151 | |
ec6ea7d0 | 1152 | pthread_mutex_unlock(&channel->timer_lock); |
a9838785 | 1153 | pthread_mutex_unlock(&channel->lock); |
3bd1e081 | 1154 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 1155 | |
b5a6470f | 1156 | if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) { |
a0cbdd2e | 1157 | notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD); |
d8ef542d | 1158 | } |
b5a6470f DG |
1159 | |
1160 | return 0; | |
3bd1e081 MD |
1161 | } |
1162 | ||
1163 | /* | |
1164 | * Allocate the pollfd structure and the local view of the out fds to avoid | |
1165 | * doing a lookup in the linked list and concurrency issues when writing is | |
1166 | * needed. Called with consumer_data.lock held. | |
1167 | * | |
1168 | * Returns the number of fds in the structures. | |
1169 | */ | |
ffe60014 DG |
1170 | static int update_poll_array(struct lttng_consumer_local_data *ctx, |
1171 | struct pollfd **pollfd, struct lttng_consumer_stream **local_stream, | |
9a2fcf78 | 1172 | struct lttng_ht *ht, int *nb_inactive_fd) |
3bd1e081 | 1173 | { |
3bd1e081 | 1174 | int i = 0; |
e4421fec DG |
1175 | struct lttng_ht_iter iter; |
1176 | struct lttng_consumer_stream *stream; | |
3bd1e081 | 1177 | |
ffe60014 DG |
1178 | assert(ctx); |
1179 | assert(ht); | |
1180 | assert(pollfd); | |
1181 | assert(local_stream); | |
1182 | ||
3bd1e081 | 1183 | DBG("Updating poll fd array"); |
9a2fcf78 | 1184 | *nb_inactive_fd = 0; |
481d6c57 | 1185 | rcu_read_lock(); |
43c34bc3 | 1186 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
8994307f DG |
1187 | /* |
1188 | * Only active streams with an active end point can be added to the | |
1189 | * poll set and local stream storage of the thread. | |
1190 | * | |
1191 | * There is a potential race here for endpoint_status to be updated | |
1192 | * just after the check. However, this is OK since the stream(s) will | |
1193 | * be deleted once the thread is notified that the end point state has | |
1194 | * changed where this function will be called back again. | |
9a2fcf78 JD |
1195 | * |
1196 | * We track the number of inactive FDs because they still need to be | |
1197 | * closed by the polling thread after a wakeup on the data_pipe or | |
1198 | * metadata_pipe. | |
8994307f | 1199 | */ |
d2956687 | 1200 | if (stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { |
9a2fcf78 | 1201 | (*nb_inactive_fd)++; |
3bd1e081 MD |
1202 | continue; |
1203 | } | |
7972aab2 DG |
1204 | /* |
1205 | * This clobbers way too much the debug output. Uncomment that if you | |
1206 | * need it for debugging purposes. | |
7972aab2 | 1207 | */ |
e4421fec | 1208 | (*pollfd)[i].fd = stream->wait_fd; |
3bd1e081 | 1209 | (*pollfd)[i].events = POLLIN | POLLPRI; |
e4421fec | 1210 | local_stream[i] = stream; |
3bd1e081 MD |
1211 | i++; |
1212 | } | |
481d6c57 | 1213 | rcu_read_unlock(); |
3bd1e081 MD |
1214 | |
1215 | /* | |
50f8ae69 | 1216 | * Insert the consumer_data_pipe at the end of the array and don't |
3bd1e081 MD |
1217 | * increment i so nb_fd is the number of real FD. |
1218 | */ | |
acdb9057 | 1219 | (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe); |
509bb1cf | 1220 | (*pollfd)[i].events = POLLIN | POLLPRI; |
02b3d176 DG |
1221 | |
1222 | (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe); | |
1223 | (*pollfd)[i + 1].events = POLLIN | POLLPRI; | |
3bd1e081 MD |
1224 | return i; |
1225 | } | |
1226 | ||
1227 | /* | |
84382d49 MD |
1228 | * Poll on the should_quit pipe and the command socket return -1 on |
1229 | * error, 1 if should exit, 0 if data is available on the command socket | |
3bd1e081 MD |
1230 | */ |
1231 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) | |
1232 | { | |
1233 | int num_rdy; | |
1234 | ||
88f2b785 | 1235 | restart: |
3bd1e081 MD |
1236 | num_rdy = poll(consumer_sockpoll, 2, -1); |
1237 | if (num_rdy == -1) { | |
88f2b785 MD |
1238 | /* |
1239 | * Restart interrupted system call. | |
1240 | */ | |
1241 | if (errno == EINTR) { | |
1242 | goto restart; | |
1243 | } | |
7a57cf92 | 1244 | PERROR("Poll error"); |
84382d49 | 1245 | return -1; |
3bd1e081 | 1246 | } |
509bb1cf | 1247 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
3bd1e081 | 1248 | DBG("consumer_should_quit wake up"); |
84382d49 | 1249 | return 1; |
3bd1e081 MD |
1250 | } |
1251 | return 0; | |
3bd1e081 MD |
1252 | } |
1253 | ||
1254 | /* | |
1255 | * Set the error socket. | |
1256 | */ | |
ffe60014 DG |
1257 | void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx, |
1258 | int sock) | |
3bd1e081 MD |
1259 | { |
1260 | ctx->consumer_error_socket = sock; | |
1261 | } | |
1262 | ||
1263 | /* | |
1264 | * Set the command socket path. | |
1265 | */ | |
3bd1e081 MD |
1266 | void lttng_consumer_set_command_sock_path( |
1267 | struct lttng_consumer_local_data *ctx, char *sock) | |
1268 | { | |
1269 | ctx->consumer_command_sock_path = sock; | |
1270 | } | |
1271 | ||
1272 | /* | |
1273 | * Send return code to the session daemon. | |
1274 | * If the socket is not defined, we return 0, it is not a fatal error | |
1275 | */ | |
ffe60014 | 1276 | int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd) |
3bd1e081 MD |
1277 | { |
1278 | if (ctx->consumer_error_socket > 0) { | |
1279 | return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd, | |
1280 | sizeof(enum lttcomm_sessiond_command)); | |
1281 | } | |
1282 | ||
1283 | return 0; | |
1284 | } | |
1285 | ||
1286 | /* | |
228b5bf7 DG |
1287 | * Close all the tracefiles and stream fds and MUST be called when all |
1288 | * instances are destroyed i.e. when all threads were joined and are ended. | |
3bd1e081 MD |
1289 | */ |
1290 | void lttng_consumer_cleanup(void) | |
1291 | { | |
e4421fec | 1292 | struct lttng_ht_iter iter; |
ffe60014 | 1293 | struct lttng_consumer_channel *channel; |
e10aec8f | 1294 | unsigned int trace_chunks_left; |
6065ceec DG |
1295 | |
1296 | rcu_read_lock(); | |
3bd1e081 | 1297 | |
ffe60014 DG |
1298 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel, |
1299 | node.node) { | |
702b1ea4 | 1300 | consumer_del_channel(channel); |
3bd1e081 | 1301 | } |
6065ceec DG |
1302 | |
1303 | rcu_read_unlock(); | |
d6ce1df2 | 1304 | |
d6ce1df2 | 1305 | lttng_ht_destroy(consumer_data.channel_ht); |
5c3892a6 | 1306 | lttng_ht_destroy(consumer_data.channels_by_session_id_ht); |
228b5bf7 DG |
1307 | |
1308 | cleanup_relayd_ht(); | |
1309 | ||
d8ef542d MD |
1310 | lttng_ht_destroy(consumer_data.stream_per_chan_id_ht); |
1311 | ||
228b5bf7 DG |
1312 | /* |
1313 | * This HT contains streams that are freed by either the metadata thread or | |
1314 | * the data thread so we do *nothing* on the hash table and simply destroy | |
1315 | * it. | |
1316 | */ | |
1317 | lttng_ht_destroy(consumer_data.stream_list_ht); | |
28cc88f3 | 1318 | |
e10aec8f MD |
1319 | /* |
1320 | * Trace chunks in the registry may still exist if the session | |
1321 | * daemon has encountered an internal error and could not | |
1322 | * tear down its sessions and/or trace chunks properly. | |
1323 | * | |
1324 | * Release the session daemon's implicit reference to any remaining | |
1325 | * trace chunk and print an error if any trace chunk was found. Note | |
1326 | * that there are _no_ legitimate cases for trace chunks to be left, | |
1327 | * it is a leak. However, it can happen following a crash of the | |
1328 | * session daemon and not emptying the registry would cause an assertion | |
1329 | * to hit. | |
1330 | */ | |
1331 | trace_chunks_left = lttng_trace_chunk_registry_put_each_chunk( | |
1332 | consumer_data.chunk_registry); | |
1333 | if (trace_chunks_left) { | |
1334 | ERR("%u trace chunks are leaked by lttng-consumerd. " | |
1335 | "This can be caused by an internal error of the session daemon.", | |
1336 | trace_chunks_left); | |
1337 | } | |
1338 | /* Run all callbacks freeing each chunk. */ | |
1339 | rcu_barrier(); | |
28cc88f3 | 1340 | lttng_trace_chunk_registry_destroy(consumer_data.chunk_registry); |
3bd1e081 MD |
1341 | } |
1342 | ||
1343 | /* | |
1344 | * Called from signal handler. | |
1345 | */ | |
1346 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) | |
1347 | { | |
6cd525e8 MD |
1348 | ssize_t ret; |
1349 | ||
10211f5c | 1350 | CMM_STORE_SHARED(consumer_quit, 1); |
6cd525e8 MD |
1351 | ret = lttng_write(ctx->consumer_should_quit[1], "4", 1); |
1352 | if (ret < 1) { | |
7a57cf92 | 1353 | PERROR("write consumer quit"); |
3bd1e081 | 1354 | } |
ab1027f4 DG |
1355 | |
1356 | DBG("Consumer flag that it should quit"); | |
3bd1e081 MD |
1357 | } |
1358 | ||
5199ffc4 JG |
1359 | |
1360 | /* | |
1361 | * Flush pending writes to trace output disk file. | |
1362 | */ | |
1363 | static | |
00e2e675 DG |
1364 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
1365 | off_t orig_offset) | |
3bd1e081 | 1366 | { |
c7a78aab | 1367 | int ret; |
3bd1e081 MD |
1368 | int outfd = stream->out_fd; |
1369 | ||
1370 | /* | |
1371 | * This does a blocking write-and-wait on any page that belongs to the | |
1372 | * subbuffer prior to the one we just wrote. | |
1373 | * Don't care about error values, as these are just hints and ways to | |
1374 | * limit the amount of page cache used. | |
1375 | */ | |
ffe60014 | 1376 | if (orig_offset < stream->max_sb_size) { |
3bd1e081 MD |
1377 | return; |
1378 | } | |
ffe60014 DG |
1379 | lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size, |
1380 | stream->max_sb_size, | |
3bd1e081 MD |
1381 | SYNC_FILE_RANGE_WAIT_BEFORE |
1382 | | SYNC_FILE_RANGE_WRITE | |
1383 | | SYNC_FILE_RANGE_WAIT_AFTER); | |
1384 | /* | |
1385 | * Give hints to the kernel about how we access the file: | |
1386 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after | |
1387 | * we write it. | |
1388 | * | |
1389 | * We need to call fadvise again after the file grows because the | |
1390 | * kernel does not seem to apply fadvise to non-existing parts of the | |
1391 | * file. | |
1392 | * | |
1393 | * Call fadvise _after_ having waited for the page writeback to | |
1394 | * complete because the dirty page writeback semantic is not well | |
1395 | * defined. So it can be expected to lead to lower throughput in | |
1396 | * streaming. | |
1397 | */ | |
c7a78aab | 1398 | ret = posix_fadvise(outfd, orig_offset - stream->max_sb_size, |
ffe60014 | 1399 | stream->max_sb_size, POSIX_FADV_DONTNEED); |
a0d0e127 | 1400 | if (ret && ret != -ENOSYS) { |
a74a5f4a JG |
1401 | errno = ret; |
1402 | PERROR("posix_fadvise on fd %i", outfd); | |
c7a78aab | 1403 | } |
3bd1e081 MD |
1404 | } |
1405 | ||
1406 | /* | |
1407 | * Initialise the necessary environnement : | |
1408 | * - create a new context | |
1409 | * - create the poll_pipe | |
1410 | * - create the should_quit pipe (for signal handler) | |
1411 | * - create the thread pipe (for splice) | |
1412 | * | |
1413 | * Takes a function pointer as argument, this function is called when data is | |
1414 | * available on a buffer. This function is responsible to do the | |
1415 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the | |
1416 | * buffer configuration and then kernctl_put_next_subbuf at the end. | |
1417 | * | |
1418 | * Returns a pointer to the new context or NULL on error. | |
1419 | */ | |
1420 | struct lttng_consumer_local_data *lttng_consumer_create( | |
1421 | enum lttng_consumer_type type, | |
4078b776 | 1422 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
bdc8d1bb | 1423 | struct lttng_consumer_local_data *ctx, bool locked_by_caller), |
3bd1e081 MD |
1424 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
1425 | int (*recv_stream)(struct lttng_consumer_stream *stream), | |
30319bcb | 1426 | int (*update_stream)(uint64_t stream_key, uint32_t state)) |
3bd1e081 | 1427 | { |
d8ef542d | 1428 | int ret; |
3bd1e081 MD |
1429 | struct lttng_consumer_local_data *ctx; |
1430 | ||
1431 | assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || | |
1432 | consumer_data.type == type); | |
1433 | consumer_data.type = type; | |
1434 | ||
effcf122 | 1435 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
3bd1e081 | 1436 | if (ctx == NULL) { |
7a57cf92 | 1437 | PERROR("allocating context"); |
3bd1e081 MD |
1438 | goto error; |
1439 | } | |
1440 | ||
1441 | ctx->consumer_error_socket = -1; | |
331744e3 | 1442 | ctx->consumer_metadata_socket = -1; |
75d83e50 | 1443 | pthread_mutex_init(&ctx->metadata_socket_lock, NULL); |
3bd1e081 MD |
1444 | /* assign the callbacks */ |
1445 | ctx->on_buffer_ready = buffer_ready; | |
1446 | ctx->on_recv_channel = recv_channel; | |
1447 | ctx->on_recv_stream = recv_stream; | |
1448 | ctx->on_update_stream = update_stream; | |
1449 | ||
acdb9057 DG |
1450 | ctx->consumer_data_pipe = lttng_pipe_open(0); |
1451 | if (!ctx->consumer_data_pipe) { | |
3bd1e081 MD |
1452 | goto error_poll_pipe; |
1453 | } | |
1454 | ||
02b3d176 DG |
1455 | ctx->consumer_wakeup_pipe = lttng_pipe_open(0); |
1456 | if (!ctx->consumer_wakeup_pipe) { | |
1457 | goto error_wakeup_pipe; | |
1458 | } | |
1459 | ||
3bd1e081 MD |
1460 | ret = pipe(ctx->consumer_should_quit); |
1461 | if (ret < 0) { | |
7a57cf92 | 1462 | PERROR("Error creating recv pipe"); |
3bd1e081 MD |
1463 | goto error_quit_pipe; |
1464 | } | |
1465 | ||
d8ef542d MD |
1466 | ret = pipe(ctx->consumer_channel_pipe); |
1467 | if (ret < 0) { | |
1468 | PERROR("Error creating channel pipe"); | |
1469 | goto error_channel_pipe; | |
1470 | } | |
1471 | ||
13886d2d DG |
1472 | ctx->consumer_metadata_pipe = lttng_pipe_open(0); |
1473 | if (!ctx->consumer_metadata_pipe) { | |
fb3a43a9 DG |
1474 | goto error_metadata_pipe; |
1475 | } | |
3bd1e081 | 1476 | |
e9404c27 JG |
1477 | ctx->channel_monitor_pipe = -1; |
1478 | ||
fb3a43a9 | 1479 | return ctx; |
3bd1e081 | 1480 | |
fb3a43a9 | 1481 | error_metadata_pipe: |
d8ef542d MD |
1482 | utils_close_pipe(ctx->consumer_channel_pipe); |
1483 | error_channel_pipe: | |
d8ef542d | 1484 | utils_close_pipe(ctx->consumer_should_quit); |
3bd1e081 | 1485 | error_quit_pipe: |
02b3d176 DG |
1486 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
1487 | error_wakeup_pipe: | |
acdb9057 | 1488 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
3bd1e081 MD |
1489 | error_poll_pipe: |
1490 | free(ctx); | |
1491 | error: | |
1492 | return NULL; | |
1493 | } | |
1494 | ||
282dadbc MD |
1495 | /* |
1496 | * Iterate over all streams of the hashtable and free them properly. | |
1497 | */ | |
1498 | static void destroy_data_stream_ht(struct lttng_ht *ht) | |
1499 | { | |
1500 | struct lttng_ht_iter iter; | |
1501 | struct lttng_consumer_stream *stream; | |
1502 | ||
1503 | if (ht == NULL) { | |
1504 | return; | |
1505 | } | |
1506 | ||
1507 | rcu_read_lock(); | |
1508 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1509 | /* | |
1510 | * Ignore return value since we are currently cleaning up so any error | |
1511 | * can't be handled. | |
1512 | */ | |
1513 | (void) consumer_del_stream(stream, ht); | |
1514 | } | |
1515 | rcu_read_unlock(); | |
1516 | ||
1517 | lttng_ht_destroy(ht); | |
1518 | } | |
1519 | ||
1520 | /* | |
1521 | * Iterate over all streams of the metadata hashtable and free them | |
1522 | * properly. | |
1523 | */ | |
1524 | static void destroy_metadata_stream_ht(struct lttng_ht *ht) | |
1525 | { | |
1526 | struct lttng_ht_iter iter; | |
1527 | struct lttng_consumer_stream *stream; | |
1528 | ||
1529 | if (ht == NULL) { | |
1530 | return; | |
1531 | } | |
1532 | ||
1533 | rcu_read_lock(); | |
1534 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1535 | /* | |
1536 | * Ignore return value since we are currently cleaning up so any error | |
1537 | * can't be handled. | |
1538 | */ | |
1539 | (void) consumer_del_metadata_stream(stream, ht); | |
1540 | } | |
1541 | rcu_read_unlock(); | |
1542 | ||
1543 | lttng_ht_destroy(ht); | |
1544 | } | |
1545 | ||
3bd1e081 MD |
1546 | /* |
1547 | * Close all fds associated with the instance and free the context. | |
1548 | */ | |
1549 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) | |
1550 | { | |
4c462e79 MD |
1551 | int ret; |
1552 | ||
ab1027f4 DG |
1553 | DBG("Consumer destroying it. Closing everything."); |
1554 | ||
4f2e75b9 DG |
1555 | if (!ctx) { |
1556 | return; | |
1557 | } | |
1558 | ||
282dadbc MD |
1559 | destroy_data_stream_ht(data_ht); |
1560 | destroy_metadata_stream_ht(metadata_ht); | |
1561 | ||
4c462e79 MD |
1562 | ret = close(ctx->consumer_error_socket); |
1563 | if (ret) { | |
1564 | PERROR("close"); | |
1565 | } | |
331744e3 JD |
1566 | ret = close(ctx->consumer_metadata_socket); |
1567 | if (ret) { | |
1568 | PERROR("close"); | |
1569 | } | |
d8ef542d | 1570 | utils_close_pipe(ctx->consumer_channel_pipe); |
acdb9057 | 1571 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
13886d2d | 1572 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
02b3d176 | 1573 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
d8ef542d | 1574 | utils_close_pipe(ctx->consumer_should_quit); |
fb3a43a9 | 1575 | |
3bd1e081 MD |
1576 | unlink(ctx->consumer_command_sock_path); |
1577 | free(ctx); | |
1578 | } | |
1579 | ||
6197aea7 DG |
1580 | /* |
1581 | * Write the metadata stream id on the specified file descriptor. | |
1582 | */ | |
1583 | static int write_relayd_metadata_id(int fd, | |
1584 | struct lttng_consumer_stream *stream, | |
239f61af | 1585 | unsigned long padding) |
6197aea7 | 1586 | { |
6cd525e8 | 1587 | ssize_t ret; |
1d4dfdef | 1588 | struct lttcomm_relayd_metadata_payload hdr; |
6197aea7 | 1589 | |
1d4dfdef DG |
1590 | hdr.stream_id = htobe64(stream->relayd_stream_id); |
1591 | hdr.padding_size = htobe32(padding); | |
6cd525e8 MD |
1592 | ret = lttng_write(fd, (void *) &hdr, sizeof(hdr)); |
1593 | if (ret < sizeof(hdr)) { | |
d7b75ec8 | 1594 | /* |
6f04ed72 | 1595 | * This error means that the fd's end is closed so ignore the PERROR |
d7b75ec8 DG |
1596 | * not to clubber the error output since this can happen in a normal |
1597 | * code path. | |
1598 | */ | |
1599 | if (errno != EPIPE) { | |
1600 | PERROR("write metadata stream id"); | |
1601 | } | |
1602 | DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno); | |
534d2592 DG |
1603 | /* |
1604 | * Set ret to a negative value because if ret != sizeof(hdr), we don't | |
1605 | * handle writting the missing part so report that as an error and | |
1606 | * don't lie to the caller. | |
1607 | */ | |
1608 | ret = -1; | |
6197aea7 DG |
1609 | goto end; |
1610 | } | |
1d4dfdef DG |
1611 | DBG("Metadata stream id %" PRIu64 " with padding %lu written before data", |
1612 | stream->relayd_stream_id, padding); | |
6197aea7 DG |
1613 | |
1614 | end: | |
6cd525e8 | 1615 | return (int) ret; |
6197aea7 DG |
1616 | } |
1617 | ||
3bd1e081 | 1618 | /* |
09e26845 DG |
1619 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
1620 | * core function for writing trace buffers to either the local filesystem or | |
1621 | * the network. | |
1622 | * | |
d2956687 | 1623 | * It must be called with the stream and the channel lock held. |
79d4ffb7 | 1624 | * |
09e26845 | 1625 | * Careful review MUST be put if any changes occur! |
3bd1e081 MD |
1626 | * |
1627 | * Returns the number of bytes written | |
1628 | */ | |
4078b776 | 1629 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
276cd1d7 | 1630 | struct lttng_consumer_stream *stream, |
a587bd64 | 1631 | const struct lttng_buffer_view *buffer, |
bdc8d1bb | 1632 | unsigned long padding) |
3bd1e081 | 1633 | { |
994ab360 | 1634 | ssize_t ret = 0; |
f02e1e8a DG |
1635 | off_t orig_offset = stream->out_fd_offset; |
1636 | /* Default is on the disk */ | |
1637 | int outfd = stream->out_fd; | |
f02e1e8a | 1638 | struct consumer_relayd_sock_pair *relayd = NULL; |
8994307f | 1639 | unsigned int relayd_hang_up = 0; |
a587bd64 JG |
1640 | const size_t subbuf_content_size = buffer->size - padding; |
1641 | size_t write_len; | |
f02e1e8a DG |
1642 | |
1643 | /* RCU lock for the relayd pointer */ | |
1644 | rcu_read_lock(); | |
7fd975c5 | 1645 | assert(stream->net_seq_idx != (uint64_t) -1ULL || |
948411cd | 1646 | stream->trace_chunk); |
d2956687 | 1647 | |
f02e1e8a | 1648 | /* Flag that the current stream if set for network streaming. */ |
da009f2c | 1649 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1650 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1651 | if (relayd == NULL) { | |
56591bac | 1652 | ret = -EPIPE; |
f02e1e8a DG |
1653 | goto end; |
1654 | } | |
1655 | } | |
1656 | ||
f02e1e8a DG |
1657 | /* Handle stream on the relayd if the output is on the network */ |
1658 | if (relayd) { | |
a587bd64 | 1659 | unsigned long netlen = subbuf_content_size; |
f02e1e8a DG |
1660 | |
1661 | /* | |
1662 | * Lock the control socket for the complete duration of the function | |
1663 | * since from this point on we will use the socket. | |
1664 | */ | |
1665 | if (stream->metadata_flag) { | |
1666 | /* Metadata requires the control socket. */ | |
1667 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
93ec662e JD |
1668 | if (stream->reset_metadata_flag) { |
1669 | ret = relayd_reset_metadata(&relayd->control_sock, | |
1670 | stream->relayd_stream_id, | |
1671 | stream->metadata_version); | |
1672 | if (ret < 0) { | |
1673 | relayd_hang_up = 1; | |
1674 | goto write_error; | |
1675 | } | |
1676 | stream->reset_metadata_flag = 0; | |
1677 | } | |
1d4dfdef | 1678 | netlen += sizeof(struct lttcomm_relayd_metadata_payload); |
f02e1e8a DG |
1679 | } |
1680 | ||
1d4dfdef | 1681 | ret = write_relayd_stream_header(stream, netlen, padding, relayd); |
994ab360 DG |
1682 | if (ret < 0) { |
1683 | relayd_hang_up = 1; | |
1684 | goto write_error; | |
1685 | } | |
1686 | /* Use the returned socket. */ | |
1687 | outfd = ret; | |
f02e1e8a | 1688 | |
994ab360 DG |
1689 | /* Write metadata stream id before payload */ |
1690 | if (stream->metadata_flag) { | |
239f61af | 1691 | ret = write_relayd_metadata_id(outfd, stream, padding); |
994ab360 | 1692 | if (ret < 0) { |
8994307f DG |
1693 | relayd_hang_up = 1; |
1694 | goto write_error; | |
1695 | } | |
f02e1e8a | 1696 | } |
1624d5b7 | 1697 | |
a587bd64 JG |
1698 | write_len = subbuf_content_size; |
1699 | } else { | |
1700 | /* No streaming; we have to write the full padding. */ | |
93ec662e JD |
1701 | if (stream->metadata_flag && stream->reset_metadata_flag) { |
1702 | ret = utils_truncate_stream_file(stream->out_fd, 0); | |
1703 | if (ret < 0) { | |
1704 | ERR("Reset metadata file"); | |
1705 | goto end; | |
1706 | } | |
1707 | stream->reset_metadata_flag = 0; | |
1708 | } | |
1709 | ||
1624d5b7 JD |
1710 | /* |
1711 | * Check if we need to change the tracefile before writing the packet. | |
1712 | */ | |
1713 | if (stream->chan->tracefile_size > 0 && | |
a587bd64 | 1714 | (stream->tracefile_size_current + buffer->size) > |
1624d5b7 | 1715 | stream->chan->tracefile_size) { |
d2956687 JG |
1716 | ret = consumer_stream_rotate_output_files(stream); |
1717 | if (ret) { | |
1624d5b7 JD |
1718 | goto end; |
1719 | } | |
309167d2 | 1720 | outfd = stream->out_fd; |
a1ae300f | 1721 | orig_offset = 0; |
1624d5b7 | 1722 | } |
a587bd64 | 1723 | stream->tracefile_size_current += buffer->size; |
a587bd64 | 1724 | write_len = buffer->size; |
f02e1e8a DG |
1725 | } |
1726 | ||
d02b8372 DG |
1727 | /* |
1728 | * This call guarantee that len or less is returned. It's impossible to | |
1729 | * receive a ret value that is bigger than len. | |
1730 | */ | |
a587bd64 | 1731 | ret = lttng_write(outfd, buffer->data, write_len); |
63bacd79 | 1732 | DBG("Consumer mmap write() ret %zd (len %zu)", ret, write_len); |
a587bd64 | 1733 | if (ret < 0 || ((size_t) ret != write_len)) { |
d02b8372 DG |
1734 | /* |
1735 | * Report error to caller if nothing was written else at least send the | |
1736 | * amount written. | |
1737 | */ | |
1738 | if (ret < 0) { | |
994ab360 | 1739 | ret = -errno; |
f02e1e8a | 1740 | } |
994ab360 | 1741 | relayd_hang_up = 1; |
f02e1e8a | 1742 | |
d02b8372 | 1743 | /* Socket operation failed. We consider the relayd dead */ |
fcf0f774 | 1744 | if (errno == EPIPE) { |
d02b8372 DG |
1745 | /* |
1746 | * This is possible if the fd is closed on the other side | |
1747 | * (outfd) or any write problem. It can be verbose a bit for a | |
1748 | * normal execution if for instance the relayd is stopped | |
1749 | * abruptly. This can happen so set this to a DBG statement. | |
1750 | */ | |
1751 | DBG("Consumer mmap write detected relayd hang up"); | |
994ab360 DG |
1752 | } else { |
1753 | /* Unhandled error, print it and stop function right now. */ | |
a587bd64 JG |
1754 | PERROR("Error in write mmap (ret %zd != write_len %zu)", ret, |
1755 | write_len); | |
f02e1e8a | 1756 | } |
994ab360 | 1757 | goto write_error; |
d02b8372 DG |
1758 | } |
1759 | stream->output_written += ret; | |
d02b8372 DG |
1760 | |
1761 | /* This call is useless on a socket so better save a syscall. */ | |
1762 | if (!relayd) { | |
1763 | /* This won't block, but will start writeout asynchronously */ | |
a587bd64 | 1764 | lttng_sync_file_range(outfd, stream->out_fd_offset, write_len, |
d02b8372 | 1765 | SYNC_FILE_RANGE_WRITE); |
a587bd64 | 1766 | stream->out_fd_offset += write_len; |
f5dbe415 | 1767 | lttng_consumer_sync_trace_file(stream, orig_offset); |
f02e1e8a | 1768 | } |
f02e1e8a | 1769 | |
8994307f DG |
1770 | write_error: |
1771 | /* | |
1772 | * This is a special case that the relayd has closed its socket. Let's | |
1773 | * cleanup the relayd object and all associated streams. | |
1774 | */ | |
1775 | if (relayd && relayd_hang_up) { | |
9276e5c8 JR |
1776 | ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
1777 | lttng_consumer_cleanup_relayd(relayd); | |
8994307f DG |
1778 | } |
1779 | ||
f02e1e8a DG |
1780 | end: |
1781 | /* Unlock only if ctrl socket used */ | |
1782 | if (relayd && stream->metadata_flag) { | |
1783 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1784 | } | |
1785 | ||
1786 | rcu_read_unlock(); | |
994ab360 | 1787 | return ret; |
3bd1e081 MD |
1788 | } |
1789 | ||
1790 | /* | |
1791 | * Splice the data from the ring buffer to the tracefile. | |
1792 | * | |
79d4ffb7 DG |
1793 | * It must be called with the stream lock held. |
1794 | * | |
3bd1e081 MD |
1795 | * Returns the number of bytes spliced. |
1796 | */ | |
4078b776 | 1797 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
3bd1e081 | 1798 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1799 | struct lttng_consumer_stream *stream, unsigned long len, |
bdc8d1bb | 1800 | unsigned long padding) |
3bd1e081 | 1801 | { |
f02e1e8a DG |
1802 | ssize_t ret = 0, written = 0, ret_splice = 0; |
1803 | loff_t offset = 0; | |
1804 | off_t orig_offset = stream->out_fd_offset; | |
1805 | int fd = stream->wait_fd; | |
1806 | /* Default is on the disk */ | |
1807 | int outfd = stream->out_fd; | |
f02e1e8a | 1808 | struct consumer_relayd_sock_pair *relayd = NULL; |
fb3a43a9 | 1809 | int *splice_pipe; |
8994307f | 1810 | unsigned int relayd_hang_up = 0; |
f02e1e8a | 1811 | |
3bd1e081 MD |
1812 | switch (consumer_data.type) { |
1813 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a | 1814 | break; |
7753dea8 MD |
1815 | case LTTNG_CONSUMER32_UST: |
1816 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a | 1817 | /* Not supported for user space tracing */ |
3bd1e081 MD |
1818 | return -ENOSYS; |
1819 | default: | |
1820 | ERR("Unknown consumer_data type"); | |
1821 | assert(0); | |
3bd1e081 MD |
1822 | } |
1823 | ||
f02e1e8a DG |
1824 | /* RCU lock for the relayd pointer */ |
1825 | rcu_read_lock(); | |
1826 | ||
1827 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1828 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1829 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1830 | if (relayd == NULL) { | |
ad0b0d23 | 1831 | written = -ret; |
f02e1e8a DG |
1832 | goto end; |
1833 | } | |
1834 | } | |
a2361a61 | 1835 | splice_pipe = stream->splice_pipe; |
fb3a43a9 | 1836 | |
f02e1e8a | 1837 | /* Write metadata stream id before payload */ |
1d4dfdef | 1838 | if (relayd) { |
ad0b0d23 | 1839 | unsigned long total_len = len; |
f02e1e8a | 1840 | |
1d4dfdef DG |
1841 | if (stream->metadata_flag) { |
1842 | /* | |
1843 | * Lock the control socket for the complete duration of the function | |
1844 | * since from this point on we will use the socket. | |
1845 | */ | |
1846 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1847 | ||
93ec662e JD |
1848 | if (stream->reset_metadata_flag) { |
1849 | ret = relayd_reset_metadata(&relayd->control_sock, | |
1850 | stream->relayd_stream_id, | |
1851 | stream->metadata_version); | |
1852 | if (ret < 0) { | |
1853 | relayd_hang_up = 1; | |
1854 | goto write_error; | |
1855 | } | |
1856 | stream->reset_metadata_flag = 0; | |
1857 | } | |
239f61af | 1858 | ret = write_relayd_metadata_id(splice_pipe[1], stream, |
1d4dfdef DG |
1859 | padding); |
1860 | if (ret < 0) { | |
1861 | written = ret; | |
ad0b0d23 DG |
1862 | relayd_hang_up = 1; |
1863 | goto write_error; | |
1d4dfdef DG |
1864 | } |
1865 | ||
1866 | total_len += sizeof(struct lttcomm_relayd_metadata_payload); | |
1867 | } | |
1868 | ||
1869 | ret = write_relayd_stream_header(stream, total_len, padding, relayd); | |
ad0b0d23 DG |
1870 | if (ret < 0) { |
1871 | written = ret; | |
1872 | relayd_hang_up = 1; | |
1873 | goto write_error; | |
f02e1e8a | 1874 | } |
ad0b0d23 DG |
1875 | /* Use the returned socket. */ |
1876 | outfd = ret; | |
1d4dfdef DG |
1877 | } else { |
1878 | /* No streaming, we have to set the len with the full padding */ | |
1879 | len += padding; | |
1624d5b7 | 1880 | |
93ec662e JD |
1881 | if (stream->metadata_flag && stream->reset_metadata_flag) { |
1882 | ret = utils_truncate_stream_file(stream->out_fd, 0); | |
1883 | if (ret < 0) { | |
1884 | ERR("Reset metadata file"); | |
1885 | goto end; | |
1886 | } | |
1887 | stream->reset_metadata_flag = 0; | |
1888 | } | |
1624d5b7 JD |
1889 | /* |
1890 | * Check if we need to change the tracefile before writing the packet. | |
1891 | */ | |
1892 | if (stream->chan->tracefile_size > 0 && | |
1893 | (stream->tracefile_size_current + len) > | |
1894 | stream->chan->tracefile_size) { | |
d2956687 | 1895 | ret = consumer_stream_rotate_output_files(stream); |
1624d5b7 | 1896 | if (ret < 0) { |
ad0b0d23 | 1897 | written = ret; |
1624d5b7 JD |
1898 | goto end; |
1899 | } | |
309167d2 | 1900 | outfd = stream->out_fd; |
a1ae300f | 1901 | orig_offset = 0; |
1624d5b7 JD |
1902 | } |
1903 | stream->tracefile_size_current += len; | |
f02e1e8a DG |
1904 | } |
1905 | ||
1906 | while (len > 0) { | |
1d4dfdef DG |
1907 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)", |
1908 | (unsigned long)offset, len, fd, splice_pipe[1]); | |
fb3a43a9 | 1909 | ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len, |
f02e1e8a DG |
1910 | SPLICE_F_MOVE | SPLICE_F_MORE); |
1911 | DBG("splice chan to pipe, ret %zd", ret_splice); | |
1912 | if (ret_splice < 0) { | |
d02b8372 | 1913 | ret = errno; |
ad0b0d23 | 1914 | written = -ret; |
d02b8372 | 1915 | PERROR("Error in relay splice"); |
f02e1e8a DG |
1916 | goto splice_error; |
1917 | } | |
1918 | ||
1919 | /* Handle stream on the relayd if the output is on the network */ | |
ad0b0d23 DG |
1920 | if (relayd && stream->metadata_flag) { |
1921 | size_t metadata_payload_size = | |
1922 | sizeof(struct lttcomm_relayd_metadata_payload); | |
1923 | ||
1924 | /* Update counter to fit the spliced data */ | |
1925 | ret_splice += metadata_payload_size; | |
1926 | len += metadata_payload_size; | |
1927 | /* | |
1928 | * We do this so the return value can match the len passed as | |
1929 | * argument to this function. | |
1930 | */ | |
1931 | written -= metadata_payload_size; | |
f02e1e8a DG |
1932 | } |
1933 | ||
1934 | /* Splice data out */ | |
fb3a43a9 | 1935 | ret_splice = splice(splice_pipe[0], NULL, outfd, NULL, |
f02e1e8a | 1936 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); |
a2361a61 JD |
1937 | DBG("Consumer splice pipe to file (out_fd: %d), ret %zd", |
1938 | outfd, ret_splice); | |
f02e1e8a | 1939 | if (ret_splice < 0) { |
d02b8372 | 1940 | ret = errno; |
ad0b0d23 DG |
1941 | written = -ret; |
1942 | relayd_hang_up = 1; | |
1943 | goto write_error; | |
f02e1e8a | 1944 | } else if (ret_splice > len) { |
d02b8372 DG |
1945 | /* |
1946 | * We don't expect this code path to be executed but you never know | |
1947 | * so this is an extra protection agains a buggy splice(). | |
1948 | */ | |
f02e1e8a | 1949 | ret = errno; |
ad0b0d23 | 1950 | written += ret_splice; |
d02b8372 DG |
1951 | PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice, |
1952 | len); | |
f02e1e8a | 1953 | goto splice_error; |
d02b8372 DG |
1954 | } else { |
1955 | /* All good, update current len and continue. */ | |
1956 | len -= ret_splice; | |
f02e1e8a | 1957 | } |
f02e1e8a DG |
1958 | |
1959 | /* This call is useless on a socket so better save a syscall. */ | |
1960 | if (!relayd) { | |
1961 | /* This won't block, but will start writeout asynchronously */ | |
1962 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, | |
1963 | SYNC_FILE_RANGE_WRITE); | |
1964 | stream->out_fd_offset += ret_splice; | |
1965 | } | |
e5d1a9b3 | 1966 | stream->output_written += ret_splice; |
f02e1e8a DG |
1967 | written += ret_splice; |
1968 | } | |
f5dbe415 JG |
1969 | if (!relayd) { |
1970 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1971 | } | |
f02e1e8a DG |
1972 | goto end; |
1973 | ||
8994307f DG |
1974 | write_error: |
1975 | /* | |
1976 | * This is a special case that the relayd has closed its socket. Let's | |
1977 | * cleanup the relayd object and all associated streams. | |
1978 | */ | |
1979 | if (relayd && relayd_hang_up) { | |
9276e5c8 JR |
1980 | ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
1981 | lttng_consumer_cleanup_relayd(relayd); | |
8994307f DG |
1982 | /* Skip splice error so the consumer does not fail */ |
1983 | goto end; | |
1984 | } | |
1985 | ||
f02e1e8a DG |
1986 | splice_error: |
1987 | /* send the appropriate error description to sessiond */ | |
1988 | switch (ret) { | |
f02e1e8a | 1989 | case EINVAL: |
f73fabfd | 1990 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); |
f02e1e8a DG |
1991 | break; |
1992 | case ENOMEM: | |
f73fabfd | 1993 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); |
f02e1e8a DG |
1994 | break; |
1995 | case ESPIPE: | |
f73fabfd | 1996 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); |
f02e1e8a DG |
1997 | break; |
1998 | } | |
1999 | ||
2000 | end: | |
2001 | if (relayd && stream->metadata_flag) { | |
2002 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
2003 | } | |
2004 | ||
2005 | rcu_read_unlock(); | |
2006 | return written; | |
3bd1e081 MD |
2007 | } |
2008 | ||
15055ce5 JD |
2009 | /* |
2010 | * Sample the snapshot positions for a specific fd | |
2011 | * | |
2012 | * Returns 0 on success, < 0 on error | |
2013 | */ | |
2014 | int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream *stream) | |
2015 | { | |
2016 | switch (consumer_data.type) { | |
2017 | case LTTNG_CONSUMER_KERNEL: | |
2018 | return lttng_kconsumer_sample_snapshot_positions(stream); | |
2019 | case LTTNG_CONSUMER32_UST: | |
2020 | case LTTNG_CONSUMER64_UST: | |
2021 | return lttng_ustconsumer_sample_snapshot_positions(stream); | |
2022 | default: | |
2023 | ERR("Unknown consumer_data type"); | |
2024 | assert(0); | |
2025 | return -ENOSYS; | |
2026 | } | |
2027 | } | |
3bd1e081 MD |
2028 | /* |
2029 | * Take a snapshot for a specific fd | |
2030 | * | |
2031 | * Returns 0 on success, < 0 on error | |
2032 | */ | |
ffe60014 | 2033 | int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) |
3bd1e081 MD |
2034 | { |
2035 | switch (consumer_data.type) { | |
2036 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 2037 | return lttng_kconsumer_take_snapshot(stream); |
7753dea8 MD |
2038 | case LTTNG_CONSUMER32_UST: |
2039 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 2040 | return lttng_ustconsumer_take_snapshot(stream); |
3bd1e081 MD |
2041 | default: |
2042 | ERR("Unknown consumer_data type"); | |
2043 | assert(0); | |
2044 | return -ENOSYS; | |
2045 | } | |
3bd1e081 MD |
2046 | } |
2047 | ||
2048 | /* | |
2049 | * Get the produced position | |
2050 | * | |
2051 | * Returns 0 on success, < 0 on error | |
2052 | */ | |
ffe60014 | 2053 | int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, |
3bd1e081 MD |
2054 | unsigned long *pos) |
2055 | { | |
2056 | switch (consumer_data.type) { | |
2057 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 2058 | return lttng_kconsumer_get_produced_snapshot(stream, pos); |
7753dea8 MD |
2059 | case LTTNG_CONSUMER32_UST: |
2060 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 2061 | return lttng_ustconsumer_get_produced_snapshot(stream, pos); |
3bd1e081 MD |
2062 | default: |
2063 | ERR("Unknown consumer_data type"); | |
2064 | assert(0); | |
2065 | return -ENOSYS; | |
2066 | } | |
2067 | } | |
2068 | ||
15055ce5 JD |
2069 | /* |
2070 | * Get the consumed position (free-running counter position in bytes). | |
2071 | * | |
2072 | * Returns 0 on success, < 0 on error | |
2073 | */ | |
2074 | int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream *stream, | |
2075 | unsigned long *pos) | |
2076 | { | |
2077 | switch (consumer_data.type) { | |
2078 | case LTTNG_CONSUMER_KERNEL: | |
2079 | return lttng_kconsumer_get_consumed_snapshot(stream, pos); | |
2080 | case LTTNG_CONSUMER32_UST: | |
2081 | case LTTNG_CONSUMER64_UST: | |
2082 | return lttng_ustconsumer_get_consumed_snapshot(stream, pos); | |
2083 | default: | |
2084 | ERR("Unknown consumer_data type"); | |
2085 | assert(0); | |
2086 | return -ENOSYS; | |
2087 | } | |
2088 | } | |
2089 | ||
3bd1e081 MD |
2090 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, |
2091 | int sock, struct pollfd *consumer_sockpoll) | |
2092 | { | |
2093 | switch (consumer_data.type) { | |
2094 | case LTTNG_CONSUMER_KERNEL: | |
2095 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
7753dea8 MD |
2096 | case LTTNG_CONSUMER32_UST: |
2097 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
2098 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
2099 | default: | |
2100 | ERR("Unknown consumer_data type"); | |
2101 | assert(0); | |
2102 | return -ENOSYS; | |
2103 | } | |
2104 | } | |
2105 | ||
1f8d1c14 | 2106 | static |
6d574024 | 2107 | void lttng_consumer_close_all_metadata(void) |
d88aee68 DG |
2108 | { |
2109 | switch (consumer_data.type) { | |
2110 | case LTTNG_CONSUMER_KERNEL: | |
2111 | /* | |
2112 | * The Kernel consumer has a different metadata scheme so we don't | |
2113 | * close anything because the stream will be closed by the session | |
2114 | * daemon. | |
2115 | */ | |
2116 | break; | |
2117 | case LTTNG_CONSUMER32_UST: | |
2118 | case LTTNG_CONSUMER64_UST: | |
2119 | /* | |
2120 | * Close all metadata streams. The metadata hash table is passed and | |
2121 | * this call iterates over it by closing all wakeup fd. This is safe | |
2122 | * because at this point we are sure that the metadata producer is | |
2123 | * either dead or blocked. | |
2124 | */ | |
6d574024 | 2125 | lttng_ustconsumer_close_all_metadata(metadata_ht); |
d88aee68 DG |
2126 | break; |
2127 | default: | |
2128 | ERR("Unknown consumer_data type"); | |
2129 | assert(0); | |
2130 | } | |
2131 | } | |
2132 | ||
fb3a43a9 DG |
2133 | /* |
2134 | * Clean up a metadata stream and free its memory. | |
2135 | */ | |
e316aad5 DG |
2136 | void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, |
2137 | struct lttng_ht *ht) | |
fb3a43a9 | 2138 | { |
a6ef8ee6 JG |
2139 | struct lttng_consumer_channel *channel = NULL; |
2140 | bool free_channel = false; | |
fb3a43a9 DG |
2141 | |
2142 | assert(stream); | |
2143 | /* | |
2144 | * This call should NEVER receive regular stream. It must always be | |
2145 | * metadata stream and this is crucial for data structure synchronization. | |
2146 | */ | |
2147 | assert(stream->metadata_flag); | |
2148 | ||
e316aad5 DG |
2149 | DBG3("Consumer delete metadata stream %d", stream->wait_fd); |
2150 | ||
74251bb8 | 2151 | pthread_mutex_lock(&consumer_data.lock); |
a6ef8ee6 JG |
2152 | /* |
2153 | * Note that this assumes that a stream's channel is never changed and | |
2154 | * that the stream's lock doesn't need to be taken to sample its | |
2155 | * channel. | |
2156 | */ | |
2157 | channel = stream->chan; | |
2158 | pthread_mutex_lock(&channel->lock); | |
3dad2c0f | 2159 | pthread_mutex_lock(&stream->lock); |
a6ef8ee6 | 2160 | if (channel->metadata_cache) { |
081424af | 2161 | /* Only applicable to userspace consumers. */ |
a6ef8ee6 | 2162 | pthread_mutex_lock(&channel->metadata_cache->lock); |
081424af | 2163 | } |
8994307f | 2164 | |
6d574024 DG |
2165 | /* Remove any reference to that stream. */ |
2166 | consumer_stream_delete(stream, ht); | |
ca22feea | 2167 | |
6d574024 DG |
2168 | /* Close down everything including the relayd if one. */ |
2169 | consumer_stream_close(stream); | |
2170 | /* Destroy tracer buffers of the stream. */ | |
2171 | consumer_stream_destroy_buffers(stream); | |
fb3a43a9 DG |
2172 | |
2173 | /* Atomically decrement channel refcount since other threads can use it. */ | |
a6ef8ee6 JG |
2174 | if (!uatomic_sub_return(&channel->refcount, 1) |
2175 | && !uatomic_read(&channel->nb_init_stream_left)) { | |
c30aaa51 | 2176 | /* Go for channel deletion! */ |
a6ef8ee6 | 2177 | free_channel = true; |
fb3a43a9 | 2178 | } |
a6ef8ee6 | 2179 | stream->chan = NULL; |
fb3a43a9 | 2180 | |
73811ecc DG |
2181 | /* |
2182 | * Nullify the stream reference so it is not used after deletion. The | |
6d574024 DG |
2183 | * channel lock MUST be acquired before being able to check for a NULL |
2184 | * pointer value. | |
73811ecc | 2185 | */ |
a6ef8ee6 | 2186 | channel->metadata_stream = NULL; |
73811ecc | 2187 | |
a6ef8ee6 JG |
2188 | if (channel->metadata_cache) { |
2189 | pthread_mutex_unlock(&channel->metadata_cache->lock); | |
081424af | 2190 | } |
3dad2c0f | 2191 | pthread_mutex_unlock(&stream->lock); |
a6ef8ee6 | 2192 | pthread_mutex_unlock(&channel->lock); |
74251bb8 | 2193 | pthread_mutex_unlock(&consumer_data.lock); |
e316aad5 | 2194 | |
a6ef8ee6 JG |
2195 | if (free_channel) { |
2196 | consumer_del_channel(channel); | |
e316aad5 DG |
2197 | } |
2198 | ||
d2956687 JG |
2199 | lttng_trace_chunk_put(stream->trace_chunk); |
2200 | stream->trace_chunk = NULL; | |
6d574024 | 2201 | consumer_stream_free(stream); |
fb3a43a9 DG |
2202 | } |
2203 | ||
2204 | /* | |
2205 | * Action done with the metadata stream when adding it to the consumer internal | |
2206 | * data structures to handle it. | |
2207 | */ | |
66d583dc | 2208 | void consumer_add_metadata_stream(struct lttng_consumer_stream *stream) |
fb3a43a9 | 2209 | { |
5ab66908 | 2210 | struct lttng_ht *ht = metadata_ht; |
76082088 | 2211 | struct lttng_ht_iter iter; |
d88aee68 | 2212 | struct lttng_ht_node_u64 *node; |
fb3a43a9 | 2213 | |
e316aad5 DG |
2214 | assert(stream); |
2215 | assert(ht); | |
2216 | ||
d88aee68 | 2217 | DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key); |
e316aad5 DG |
2218 | |
2219 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 2220 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 2221 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 2222 | pthread_mutex_lock(&stream->lock); |
e316aad5 | 2223 | |
e316aad5 DG |
2224 | /* |
2225 | * From here, refcounts are updated so be _careful_ when returning an error | |
2226 | * after this point. | |
2227 | */ | |
2228 | ||
fb3a43a9 | 2229 | rcu_read_lock(); |
76082088 DG |
2230 | |
2231 | /* | |
2232 | * Lookup the stream just to make sure it does not exist in our internal | |
2233 | * state. This should NEVER happen. | |
2234 | */ | |
d88aee68 DG |
2235 | lttng_ht_lookup(ht, &stream->key, &iter); |
2236 | node = lttng_ht_iter_get_node_u64(&iter); | |
76082088 DG |
2237 | assert(!node); |
2238 | ||
e316aad5 | 2239 | /* |
ffe60014 DG |
2240 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
2241 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
2242 | * causes the count to become 0 also causes a stream to be added. The |
2243 | * channel deletion will thus be triggered by the following removal of this | |
2244 | * stream. | |
2245 | */ | |
ffe60014 | 2246 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
2247 | /* Increment refcount before decrementing nb_init_stream_left */ |
2248 | cmm_smp_wmb(); | |
ffe60014 | 2249 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
2250 | } |
2251 | ||
d88aee68 | 2252 | lttng_ht_add_unique_u64(ht, &stream->node); |
ca22feea | 2253 | |
446156b4 | 2254 | lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, |
d8ef542d MD |
2255 | &stream->node_channel_id); |
2256 | ||
ca22feea DG |
2257 | /* |
2258 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
2259 | * the key since the HT does not use it and we allow to add redundant keys | |
2260 | * into this table. | |
2261 | */ | |
d88aee68 | 2262 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 2263 | |
fb3a43a9 | 2264 | rcu_read_unlock(); |
e316aad5 | 2265 | |
2e818a6a | 2266 | pthread_mutex_unlock(&stream->lock); |
a9838785 | 2267 | pthread_mutex_unlock(&stream->chan->lock); |
ec6ea7d0 | 2268 | pthread_mutex_unlock(&stream->chan->timer_lock); |
e316aad5 | 2269 | pthread_mutex_unlock(&consumer_data.lock); |
fb3a43a9 DG |
2270 | } |
2271 | ||
8994307f DG |
2272 | /* |
2273 | * Delete data stream that are flagged for deletion (endpoint_status). | |
2274 | */ | |
2275 | static void validate_endpoint_status_data_stream(void) | |
2276 | { | |
2277 | struct lttng_ht_iter iter; | |
2278 | struct lttng_consumer_stream *stream; | |
2279 | ||
2280 | DBG("Consumer delete flagged data stream"); | |
2281 | ||
2282 | rcu_read_lock(); | |
2283 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
2284 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2285 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2286 | continue; |
2287 | } | |
2288 | /* Delete it right now */ | |
2289 | consumer_del_stream(stream, data_ht); | |
2290 | } | |
2291 | rcu_read_unlock(); | |
2292 | } | |
2293 | ||
2294 | /* | |
2295 | * Delete metadata stream that are flagged for deletion (endpoint_status). | |
2296 | */ | |
2297 | static void validate_endpoint_status_metadata_stream( | |
2298 | struct lttng_poll_event *pollset) | |
2299 | { | |
2300 | struct lttng_ht_iter iter; | |
2301 | struct lttng_consumer_stream *stream; | |
2302 | ||
2303 | DBG("Consumer delete flagged metadata stream"); | |
2304 | ||
2305 | assert(pollset); | |
2306 | ||
2307 | rcu_read_lock(); | |
2308 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
2309 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2310 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2311 | continue; |
2312 | } | |
2313 | /* | |
2314 | * Remove from pollset so the metadata thread can continue without | |
2315 | * blocking on a deleted stream. | |
2316 | */ | |
2317 | lttng_poll_del(pollset, stream->wait_fd); | |
2318 | ||
2319 | /* Delete it right now */ | |
2320 | consumer_del_metadata_stream(stream, metadata_ht); | |
2321 | } | |
2322 | rcu_read_unlock(); | |
2323 | } | |
2324 | ||
fb3a43a9 DG |
2325 | /* |
2326 | * Thread polls on metadata file descriptor and write them on disk or on the | |
2327 | * network. | |
2328 | */ | |
7d980def | 2329 | void *consumer_thread_metadata_poll(void *data) |
fb3a43a9 | 2330 | { |
1fc79fb4 | 2331 | int ret, i, pollfd, err = -1; |
fb3a43a9 | 2332 | uint32_t revents, nb_fd; |
e316aad5 | 2333 | struct lttng_consumer_stream *stream = NULL; |
fb3a43a9 | 2334 | struct lttng_ht_iter iter; |
d88aee68 | 2335 | struct lttng_ht_node_u64 *node; |
fb3a43a9 DG |
2336 | struct lttng_poll_event events; |
2337 | struct lttng_consumer_local_data *ctx = data; | |
2338 | ssize_t len; | |
2339 | ||
2340 | rcu_register_thread(); | |
2341 | ||
1fc79fb4 MD |
2342 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA); |
2343 | ||
2d57de81 MD |
2344 | if (testpoint(consumerd_thread_metadata)) { |
2345 | goto error_testpoint; | |
2346 | } | |
2347 | ||
9ce5646a MD |
2348 | health_code_update(); |
2349 | ||
fb3a43a9 DG |
2350 | DBG("Thread metadata poll started"); |
2351 | ||
fb3a43a9 DG |
2352 | /* Size is set to 1 for the consumer_metadata pipe */ |
2353 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2354 | if (ret < 0) { | |
2355 | ERR("Poll set creation failed"); | |
d8ef542d | 2356 | goto end_poll; |
fb3a43a9 DG |
2357 | } |
2358 | ||
13886d2d DG |
2359 | ret = lttng_poll_add(&events, |
2360 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN); | |
fb3a43a9 DG |
2361 | if (ret < 0) { |
2362 | goto end; | |
2363 | } | |
2364 | ||
2365 | /* Main loop */ | |
2366 | DBG("Metadata main loop started"); | |
2367 | ||
2368 | while (1) { | |
fb3a43a9 | 2369 | restart: |
7fa2082e | 2370 | health_code_update(); |
9ce5646a | 2371 | health_poll_entry(); |
7fa2082e | 2372 | DBG("Metadata poll wait"); |
fb3a43a9 | 2373 | ret = lttng_poll_wait(&events, -1); |
7fa2082e MD |
2374 | DBG("Metadata poll return from wait with %d fd(s)", |
2375 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2376 | health_poll_exit(); |
40063ead | 2377 | DBG("Metadata event caught in thread"); |
fb3a43a9 DG |
2378 | if (ret < 0) { |
2379 | if (errno == EINTR) { | |
40063ead | 2380 | ERR("Poll EINTR caught"); |
fb3a43a9 DG |
2381 | goto restart; |
2382 | } | |
d9607cd7 MD |
2383 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2384 | err = 0; /* All is OK */ | |
2385 | } | |
2386 | goto end; | |
fb3a43a9 DG |
2387 | } |
2388 | ||
0d9c5d77 DG |
2389 | nb_fd = ret; |
2390 | ||
e316aad5 | 2391 | /* From here, the event is a metadata wait fd */ |
fb3a43a9 | 2392 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2393 | health_code_update(); |
2394 | ||
fb3a43a9 DG |
2395 | revents = LTTNG_POLL_GETEV(&events, i); |
2396 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2397 | ||
13886d2d | 2398 | if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) { |
03e43155 | 2399 | if (revents & LPOLLIN) { |
13886d2d DG |
2400 | ssize_t pipe_len; |
2401 | ||
2402 | pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe, | |
2403 | &stream, sizeof(stream)); | |
6cd525e8 | 2404 | if (pipe_len < sizeof(stream)) { |
03e43155 MD |
2405 | if (pipe_len < 0) { |
2406 | PERROR("read metadata stream"); | |
2407 | } | |
fb3a43a9 | 2408 | /* |
03e43155 MD |
2409 | * Remove the pipe from the poll set and continue the loop |
2410 | * since their might be data to consume. | |
fb3a43a9 | 2411 | */ |
03e43155 MD |
2412 | lttng_poll_del(&events, |
2413 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2414 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
fb3a43a9 DG |
2415 | continue; |
2416 | } | |
2417 | ||
8994307f DG |
2418 | /* A NULL stream means that the state has changed. */ |
2419 | if (stream == NULL) { | |
2420 | /* Check for deleted streams. */ | |
2421 | validate_endpoint_status_metadata_stream(&events); | |
3714380f | 2422 | goto restart; |
8994307f DG |
2423 | } |
2424 | ||
fb3a43a9 DG |
2425 | DBG("Adding metadata stream %d to poll set", |
2426 | stream->wait_fd); | |
2427 | ||
fb3a43a9 DG |
2428 | /* Add metadata stream to the global poll events list */ |
2429 | lttng_poll_add(&events, stream->wait_fd, | |
6d574024 | 2430 | LPOLLIN | LPOLLPRI | LPOLLHUP); |
03e43155 MD |
2431 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
2432 | DBG("Metadata thread pipe hung up"); | |
2433 | /* | |
2434 | * Remove the pipe from the poll set and continue the loop | |
2435 | * since their might be data to consume. | |
2436 | */ | |
2437 | lttng_poll_del(&events, | |
2438 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2439 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
2440 | continue; | |
2441 | } else { | |
2442 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
2443 | goto end; | |
fb3a43a9 DG |
2444 | } |
2445 | ||
e316aad5 | 2446 | /* Handle other stream */ |
fb3a43a9 DG |
2447 | continue; |
2448 | } | |
2449 | ||
d09e1200 | 2450 | rcu_read_lock(); |
d88aee68 DG |
2451 | { |
2452 | uint64_t tmp_id = (uint64_t) pollfd; | |
2453 | ||
2454 | lttng_ht_lookup(metadata_ht, &tmp_id, &iter); | |
2455 | } | |
2456 | node = lttng_ht_iter_get_node_u64(&iter); | |
e316aad5 | 2457 | assert(node); |
fb3a43a9 DG |
2458 | |
2459 | stream = caa_container_of(node, struct lttng_consumer_stream, | |
58b1f425 | 2460 | node); |
fb3a43a9 | 2461 | |
03e43155 MD |
2462 | if (revents & (LPOLLIN | LPOLLPRI)) { |
2463 | /* Get the data out of the metadata file descriptor */ | |
2464 | DBG("Metadata available on fd %d", pollfd); | |
2465 | assert(stream->wait_fd == pollfd); | |
2466 | ||
2467 | do { | |
2468 | health_code_update(); | |
2469 | ||
bdc8d1bb | 2470 | len = ctx->on_buffer_ready(stream, ctx, false); |
03e43155 MD |
2471 | /* |
2472 | * We don't check the return value here since if we get | |
83f4233d | 2473 | * a negative len, it means an error occurred thus we |
03e43155 MD |
2474 | * simply remove it from the poll set and free the |
2475 | * stream. | |
2476 | */ | |
2477 | } while (len > 0); | |
2478 | ||
2479 | /* It's ok to have an unavailable sub-buffer */ | |
2480 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { | |
2481 | /* Clean up stream from consumer and free it. */ | |
2482 | lttng_poll_del(&events, stream->wait_fd); | |
2483 | consumer_del_metadata_stream(stream, metadata_ht); | |
2484 | } | |
2485 | } else if (revents & (LPOLLERR | LPOLLHUP)) { | |
e316aad5 | 2486 | DBG("Metadata fd %d is hup|err.", pollfd); |
fb3a43a9 DG |
2487 | if (!stream->hangup_flush_done |
2488 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2489 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2490 | DBG("Attempting to flush and consume the UST buffers"); | |
2491 | lttng_ustconsumer_on_stream_hangup(stream); | |
2492 | ||
2493 | /* We just flushed the stream now read it. */ | |
4bb94b75 | 2494 | do { |
9ce5646a MD |
2495 | health_code_update(); |
2496 | ||
bdc8d1bb | 2497 | len = ctx->on_buffer_ready(stream, ctx, false); |
4bb94b75 DG |
2498 | /* |
2499 | * We don't check the return value here since if we get | |
83f4233d | 2500 | * a negative len, it means an error occurred thus we |
4bb94b75 DG |
2501 | * simply remove it from the poll set and free the |
2502 | * stream. | |
2503 | */ | |
2504 | } while (len > 0); | |
fb3a43a9 DG |
2505 | } |
2506 | ||
fb3a43a9 | 2507 | lttng_poll_del(&events, stream->wait_fd); |
e316aad5 DG |
2508 | /* |
2509 | * This call update the channel states, closes file descriptors | |
2510 | * and securely free the stream. | |
2511 | */ | |
2512 | consumer_del_metadata_stream(stream, metadata_ht); | |
03e43155 MD |
2513 | } else { |
2514 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
6f2f1a70 | 2515 | rcu_read_unlock(); |
03e43155 | 2516 | goto end; |
fb3a43a9 | 2517 | } |
e316aad5 | 2518 | /* Release RCU lock for the stream looked up */ |
d09e1200 | 2519 | rcu_read_unlock(); |
fb3a43a9 DG |
2520 | } |
2521 | } | |
2522 | ||
1fc79fb4 MD |
2523 | /* All is OK */ |
2524 | err = 0; | |
fb3a43a9 DG |
2525 | end: |
2526 | DBG("Metadata poll thread exiting"); | |
fb3a43a9 | 2527 | |
d8ef542d MD |
2528 | lttng_poll_clean(&events); |
2529 | end_poll: | |
2d57de81 | 2530 | error_testpoint: |
1fc79fb4 MD |
2531 | if (err) { |
2532 | health_error(); | |
2533 | ERR("Health error occurred in %s", __func__); | |
2534 | } | |
2535 | health_unregister(health_consumerd); | |
fb3a43a9 DG |
2536 | rcu_unregister_thread(); |
2537 | return NULL; | |
2538 | } | |
2539 | ||
3bd1e081 | 2540 | /* |
e4421fec | 2541 | * This thread polls the fds in the set to consume the data and write |
3bd1e081 MD |
2542 | * it to tracefile if necessary. |
2543 | */ | |
7d980def | 2544 | void *consumer_thread_data_poll(void *data) |
3bd1e081 | 2545 | { |
1fc79fb4 | 2546 | int num_rdy, num_hup, high_prio, ret, i, err = -1; |
3bd1e081 MD |
2547 | struct pollfd *pollfd = NULL; |
2548 | /* local view of the streams */ | |
c869f647 | 2549 | struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; |
3bd1e081 | 2550 | /* local view of consumer_data.fds_count */ |
8bdcc002 JG |
2551 | int nb_fd = 0; |
2552 | /* 2 for the consumer_data_pipe and wake up pipe */ | |
2553 | const int nb_pipes_fd = 2; | |
9a2fcf78 JD |
2554 | /* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */ |
2555 | int nb_inactive_fd = 0; | |
3bd1e081 | 2556 | struct lttng_consumer_local_data *ctx = data; |
00e2e675 | 2557 | ssize_t len; |
3bd1e081 | 2558 | |
e7b994a3 DG |
2559 | rcu_register_thread(); |
2560 | ||
1fc79fb4 MD |
2561 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA); |
2562 | ||
2d57de81 MD |
2563 | if (testpoint(consumerd_thread_data)) { |
2564 | goto error_testpoint; | |
2565 | } | |
2566 | ||
9ce5646a MD |
2567 | health_code_update(); |
2568 | ||
4df6c8cb MD |
2569 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream *)); |
2570 | if (local_stream == NULL) { | |
2571 | PERROR("local_stream malloc"); | |
2572 | goto end; | |
2573 | } | |
3bd1e081 MD |
2574 | |
2575 | while (1) { | |
9ce5646a MD |
2576 | health_code_update(); |
2577 | ||
3bd1e081 MD |
2578 | high_prio = 0; |
2579 | num_hup = 0; | |
2580 | ||
2581 | /* | |
e4421fec | 2582 | * the fds set has been updated, we need to update our |
3bd1e081 MD |
2583 | * local array as well |
2584 | */ | |
2585 | pthread_mutex_lock(&consumer_data.lock); | |
2586 | if (consumer_data.need_update) { | |
0e428499 DG |
2587 | free(pollfd); |
2588 | pollfd = NULL; | |
2589 | ||
2590 | free(local_stream); | |
2591 | local_stream = NULL; | |
3bd1e081 | 2592 | |
8bdcc002 | 2593 | /* Allocate for all fds */ |
261de637 | 2594 | pollfd = zmalloc((consumer_data.stream_count + nb_pipes_fd) * sizeof(struct pollfd)); |
3bd1e081 | 2595 | if (pollfd == NULL) { |
7a57cf92 | 2596 | PERROR("pollfd malloc"); |
3bd1e081 MD |
2597 | pthread_mutex_unlock(&consumer_data.lock); |
2598 | goto end; | |
2599 | } | |
2600 | ||
261de637 | 2601 | local_stream = zmalloc((consumer_data.stream_count + nb_pipes_fd) * |
747f8642 | 2602 | sizeof(struct lttng_consumer_stream *)); |
3bd1e081 | 2603 | if (local_stream == NULL) { |
7a57cf92 | 2604 | PERROR("local_stream malloc"); |
3bd1e081 MD |
2605 | pthread_mutex_unlock(&consumer_data.lock); |
2606 | goto end; | |
2607 | } | |
ffe60014 | 2608 | ret = update_poll_array(ctx, &pollfd, local_stream, |
9a2fcf78 | 2609 | data_ht, &nb_inactive_fd); |
3bd1e081 MD |
2610 | if (ret < 0) { |
2611 | ERR("Error in allocating pollfd or local_outfds"); | |
f73fabfd | 2612 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2613 | pthread_mutex_unlock(&consumer_data.lock); |
2614 | goto end; | |
2615 | } | |
2616 | nb_fd = ret; | |
2617 | consumer_data.need_update = 0; | |
2618 | } | |
2619 | pthread_mutex_unlock(&consumer_data.lock); | |
2620 | ||
4078b776 | 2621 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
9a2fcf78 JD |
2622 | if (nb_fd == 0 && nb_inactive_fd == 0 && |
2623 | CMM_LOAD_SHARED(consumer_quit) == 1) { | |
1fc79fb4 | 2624 | err = 0; /* All is OK */ |
4078b776 MD |
2625 | goto end; |
2626 | } | |
3bd1e081 | 2627 | /* poll on the array of fds */ |
88f2b785 | 2628 | restart: |
261de637 | 2629 | DBG("polling on %d fd", nb_fd + nb_pipes_fd); |
cf0bcb51 JG |
2630 | if (testpoint(consumerd_thread_data_poll)) { |
2631 | goto end; | |
2632 | } | |
9ce5646a | 2633 | health_poll_entry(); |
261de637 | 2634 | num_rdy = poll(pollfd, nb_fd + nb_pipes_fd, -1); |
9ce5646a | 2635 | health_poll_exit(); |
3bd1e081 MD |
2636 | DBG("poll num_rdy : %d", num_rdy); |
2637 | if (num_rdy == -1) { | |
88f2b785 MD |
2638 | /* |
2639 | * Restart interrupted system call. | |
2640 | */ | |
2641 | if (errno == EINTR) { | |
2642 | goto restart; | |
2643 | } | |
7a57cf92 | 2644 | PERROR("Poll error"); |
f73fabfd | 2645 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2646 | goto end; |
2647 | } else if (num_rdy == 0) { | |
2648 | DBG("Polling thread timed out"); | |
2649 | goto end; | |
2650 | } | |
2651 | ||
80957876 JG |
2652 | if (caa_unlikely(data_consumption_paused)) { |
2653 | DBG("Data consumption paused, sleeping..."); | |
2654 | sleep(1); | |
2655 | goto restart; | |
2656 | } | |
2657 | ||
3bd1e081 | 2658 | /* |
50f8ae69 | 2659 | * If the consumer_data_pipe triggered poll go directly to the |
00e2e675 DG |
2660 | * beginning of the loop to update the array. We want to prioritize |
2661 | * array update over low-priority reads. | |
3bd1e081 | 2662 | */ |
509bb1cf | 2663 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
ab30f567 | 2664 | ssize_t pipe_readlen; |
04fdd819 | 2665 | |
50f8ae69 | 2666 | DBG("consumer_data_pipe wake up"); |
acdb9057 DG |
2667 | pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe, |
2668 | &new_stream, sizeof(new_stream)); | |
6cd525e8 MD |
2669 | if (pipe_readlen < sizeof(new_stream)) { |
2670 | PERROR("Consumer data pipe"); | |
23f5f35d DG |
2671 | /* Continue so we can at least handle the current stream(s). */ |
2672 | continue; | |
2673 | } | |
c869f647 DG |
2674 | |
2675 | /* | |
2676 | * If the stream is NULL, just ignore it. It's also possible that | |
2677 | * the sessiond poll thread changed the consumer_quit state and is | |
2678 | * waking us up to test it. | |
2679 | */ | |
2680 | if (new_stream == NULL) { | |
8994307f | 2681 | validate_endpoint_status_data_stream(); |
c869f647 DG |
2682 | continue; |
2683 | } | |
2684 | ||
c869f647 | 2685 | /* Continue to update the local streams and handle prio ones */ |
3bd1e081 MD |
2686 | continue; |
2687 | } | |
2688 | ||
02b3d176 DG |
2689 | /* Handle wakeup pipe. */ |
2690 | if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) { | |
2691 | char dummy; | |
2692 | ssize_t pipe_readlen; | |
2693 | ||
2694 | pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy, | |
2695 | sizeof(dummy)); | |
2696 | if (pipe_readlen < 0) { | |
2697 | PERROR("Consumer data wakeup pipe"); | |
2698 | } | |
2699 | /* We've been awakened to handle stream(s). */ | |
2700 | ctx->has_wakeup = 0; | |
2701 | } | |
2702 | ||
3bd1e081 MD |
2703 | /* Take care of high priority channels first. */ |
2704 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2705 | health_code_update(); |
2706 | ||
9617607b DG |
2707 | if (local_stream[i] == NULL) { |
2708 | continue; | |
2709 | } | |
fb3a43a9 | 2710 | if (pollfd[i].revents & POLLPRI) { |
d41f73b7 MD |
2711 | DBG("Urgent read on fd %d", pollfd[i].fd); |
2712 | high_prio = 1; | |
bdc8d1bb | 2713 | len = ctx->on_buffer_ready(local_stream[i], ctx, false); |
d41f73b7 | 2714 | /* it's ok to have an unavailable sub-buffer */ |
b64403e3 | 2715 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2716 | /* Clean the stream and free it. */ |
2717 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2718 | local_stream[i] = NULL; |
4078b776 MD |
2719 | } else if (len > 0) { |
2720 | local_stream[i]->data_read = 1; | |
d41f73b7 | 2721 | } |
3bd1e081 MD |
2722 | } |
2723 | } | |
2724 | ||
4078b776 MD |
2725 | /* |
2726 | * If we read high prio channel in this loop, try again | |
2727 | * for more high prio data. | |
2728 | */ | |
2729 | if (high_prio) { | |
3bd1e081 MD |
2730 | continue; |
2731 | } | |
2732 | ||
2733 | /* Take care of low priority channels. */ | |
4078b776 | 2734 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2735 | health_code_update(); |
2736 | ||
9617607b DG |
2737 | if (local_stream[i] == NULL) { |
2738 | continue; | |
2739 | } | |
4078b776 | 2740 | if ((pollfd[i].revents & POLLIN) || |
02b3d176 DG |
2741 | local_stream[i]->hangup_flush_done || |
2742 | local_stream[i]->has_data) { | |
4078b776 | 2743 | DBG("Normal read on fd %d", pollfd[i].fd); |
bdc8d1bb | 2744 | len = ctx->on_buffer_ready(local_stream[i], ctx, false); |
4078b776 | 2745 | /* it's ok to have an unavailable sub-buffer */ |
b64403e3 | 2746 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2747 | /* Clean the stream and free it. */ |
2748 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2749 | local_stream[i] = NULL; |
4078b776 MD |
2750 | } else if (len > 0) { |
2751 | local_stream[i]->data_read = 1; | |
2752 | } | |
2753 | } | |
2754 | } | |
2755 | ||
2756 | /* Handle hangup and errors */ | |
2757 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2758 | health_code_update(); |
2759 | ||
9617607b DG |
2760 | if (local_stream[i] == NULL) { |
2761 | continue; | |
2762 | } | |
4078b776 MD |
2763 | if (!local_stream[i]->hangup_flush_done |
2764 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) | |
2765 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2766 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2767 | DBG("fd %d is hup|err|nval. Attempting flush and read.", | |
9617607b | 2768 | pollfd[i].fd); |
4078b776 MD |
2769 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); |
2770 | /* Attempt read again, for the data we just flushed. */ | |
2771 | local_stream[i]->data_read = 1; | |
2772 | } | |
2773 | /* | |
2774 | * If the poll flag is HUP/ERR/NVAL and we have | |
2775 | * read no data in this pass, we can remove the | |
2776 | * stream from its hash table. | |
2777 | */ | |
2778 | if ((pollfd[i].revents & POLLHUP)) { | |
2779 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); | |
2780 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2781 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2782 | local_stream[i] = NULL; |
4078b776 MD |
2783 | num_hup++; |
2784 | } | |
2785 | } else if (pollfd[i].revents & POLLERR) { | |
2786 | ERR("Error returned in polling fd %d.", pollfd[i].fd); | |
2787 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2788 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2789 | local_stream[i] = NULL; |
4078b776 MD |
2790 | num_hup++; |
2791 | } | |
2792 | } else if (pollfd[i].revents & POLLNVAL) { | |
2793 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); | |
2794 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2795 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2796 | local_stream[i] = NULL; |
4078b776 | 2797 | num_hup++; |
3bd1e081 MD |
2798 | } |
2799 | } | |
9617607b DG |
2800 | if (local_stream[i] != NULL) { |
2801 | local_stream[i]->data_read = 0; | |
2802 | } | |
3bd1e081 MD |
2803 | } |
2804 | } | |
1fc79fb4 MD |
2805 | /* All is OK */ |
2806 | err = 0; | |
3bd1e081 MD |
2807 | end: |
2808 | DBG("polling thread exiting"); | |
0e428499 DG |
2809 | free(pollfd); |
2810 | free(local_stream); | |
fb3a43a9 DG |
2811 | |
2812 | /* | |
2813 | * Close the write side of the pipe so epoll_wait() in | |
7d980def DG |
2814 | * consumer_thread_metadata_poll can catch it. The thread is monitoring the |
2815 | * read side of the pipe. If we close them both, epoll_wait strangely does | |
2816 | * not return and could create a endless wait period if the pipe is the | |
2817 | * only tracked fd in the poll set. The thread will take care of closing | |
2818 | * the read side. | |
fb3a43a9 | 2819 | */ |
13886d2d | 2820 | (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe); |
fb3a43a9 | 2821 | |
2d57de81 | 2822 | error_testpoint: |
1fc79fb4 MD |
2823 | if (err) { |
2824 | health_error(); | |
2825 | ERR("Health error occurred in %s", __func__); | |
2826 | } | |
2827 | health_unregister(health_consumerd); | |
2828 | ||
e7b994a3 | 2829 | rcu_unregister_thread(); |
3bd1e081 MD |
2830 | return NULL; |
2831 | } | |
2832 | ||
d8ef542d MD |
2833 | /* |
2834 | * Close wake-up end of each stream belonging to the channel. This will | |
2835 | * allow the poll() on the stream read-side to detect when the | |
2836 | * write-side (application) finally closes them. | |
2837 | */ | |
2838 | static | |
2839 | void consumer_close_channel_streams(struct lttng_consumer_channel *channel) | |
2840 | { | |
2841 | struct lttng_ht *ht; | |
2842 | struct lttng_consumer_stream *stream; | |
2843 | struct lttng_ht_iter iter; | |
2844 | ||
2845 | ht = consumer_data.stream_per_chan_id_ht; | |
2846 | ||
2847 | rcu_read_lock(); | |
2848 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
2849 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
2850 | ht->match_fct, &channel->key, | |
2851 | &iter.iter, stream, node_channel_id.node) { | |
f2ad556d MD |
2852 | /* |
2853 | * Protect against teardown with mutex. | |
2854 | */ | |
2855 | pthread_mutex_lock(&stream->lock); | |
2856 | if (cds_lfht_is_node_deleted(&stream->node.node)) { | |
2857 | goto next; | |
2858 | } | |
d8ef542d MD |
2859 | switch (consumer_data.type) { |
2860 | case LTTNG_CONSUMER_KERNEL: | |
2861 | break; | |
2862 | case LTTNG_CONSUMER32_UST: | |
2863 | case LTTNG_CONSUMER64_UST: | |
b4a650f3 DG |
2864 | if (stream->metadata_flag) { |
2865 | /* Safe and protected by the stream lock. */ | |
2866 | lttng_ustconsumer_close_metadata(stream->chan); | |
2867 | } else { | |
2868 | /* | |
2869 | * Note: a mutex is taken internally within | |
2870 | * liblttng-ust-ctl to protect timer wakeup_fd | |
2871 | * use from concurrent close. | |
2872 | */ | |
2873 | lttng_ustconsumer_close_stream_wakeup(stream); | |
2874 | } | |
d8ef542d MD |
2875 | break; |
2876 | default: | |
2877 | ERR("Unknown consumer_data type"); | |
2878 | assert(0); | |
2879 | } | |
f2ad556d MD |
2880 | next: |
2881 | pthread_mutex_unlock(&stream->lock); | |
d8ef542d MD |
2882 | } |
2883 | rcu_read_unlock(); | |
2884 | } | |
2885 | ||
2886 | static void destroy_channel_ht(struct lttng_ht *ht) | |
2887 | { | |
2888 | struct lttng_ht_iter iter; | |
2889 | struct lttng_consumer_channel *channel; | |
2890 | int ret; | |
2891 | ||
2892 | if (ht == NULL) { | |
2893 | return; | |
2894 | } | |
2895 | ||
2896 | rcu_read_lock(); | |
2897 | cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) { | |
2898 | ret = lttng_ht_del(ht, &iter); | |
2899 | assert(ret != 0); | |
2900 | } | |
2901 | rcu_read_unlock(); | |
2902 | ||
2903 | lttng_ht_destroy(ht); | |
2904 | } | |
2905 | ||
2906 | /* | |
2907 | * This thread polls the channel fds to detect when they are being | |
2908 | * closed. It closes all related streams if the channel is detected as | |
2909 | * closed. It is currently only used as a shim layer for UST because the | |
2910 | * consumerd needs to keep the per-stream wakeup end of pipes open for | |
2911 | * periodical flush. | |
2912 | */ | |
2913 | void *consumer_thread_channel_poll(void *data) | |
2914 | { | |
1fc79fb4 | 2915 | int ret, i, pollfd, err = -1; |
d8ef542d MD |
2916 | uint32_t revents, nb_fd; |
2917 | struct lttng_consumer_channel *chan = NULL; | |
2918 | struct lttng_ht_iter iter; | |
2919 | struct lttng_ht_node_u64 *node; | |
2920 | struct lttng_poll_event events; | |
2921 | struct lttng_consumer_local_data *ctx = data; | |
2922 | struct lttng_ht *channel_ht; | |
2923 | ||
2924 | rcu_register_thread(); | |
2925 | ||
1fc79fb4 MD |
2926 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL); |
2927 | ||
2d57de81 MD |
2928 | if (testpoint(consumerd_thread_channel)) { |
2929 | goto error_testpoint; | |
2930 | } | |
2931 | ||
9ce5646a MD |
2932 | health_code_update(); |
2933 | ||
d8ef542d MD |
2934 | channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
2935 | if (!channel_ht) { | |
2936 | /* ENOMEM at this point. Better to bail out. */ | |
2937 | goto end_ht; | |
2938 | } | |
2939 | ||
2940 | DBG("Thread channel poll started"); | |
2941 | ||
2942 | /* Size is set to 1 for the consumer_channel pipe */ | |
2943 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2944 | if (ret < 0) { | |
2945 | ERR("Poll set creation failed"); | |
2946 | goto end_poll; | |
2947 | } | |
2948 | ||
2949 | ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN); | |
2950 | if (ret < 0) { | |
2951 | goto end; | |
2952 | } | |
2953 | ||
2954 | /* Main loop */ | |
2955 | DBG("Channel main loop started"); | |
2956 | ||
2957 | while (1) { | |
d8ef542d | 2958 | restart: |
7fa2082e MD |
2959 | health_code_update(); |
2960 | DBG("Channel poll wait"); | |
9ce5646a | 2961 | health_poll_entry(); |
d8ef542d | 2962 | ret = lttng_poll_wait(&events, -1); |
7fa2082e MD |
2963 | DBG("Channel poll return from wait with %d fd(s)", |
2964 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2965 | health_poll_exit(); |
40063ead | 2966 | DBG("Channel event caught in thread"); |
d8ef542d MD |
2967 | if (ret < 0) { |
2968 | if (errno == EINTR) { | |
40063ead | 2969 | ERR("Poll EINTR caught"); |
d8ef542d MD |
2970 | goto restart; |
2971 | } | |
d9607cd7 MD |
2972 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2973 | err = 0; /* All is OK */ | |
2974 | } | |
d8ef542d MD |
2975 | goto end; |
2976 | } | |
2977 | ||
2978 | nb_fd = ret; | |
2979 | ||
2980 | /* From here, the event is a channel wait fd */ | |
2981 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2982 | health_code_update(); |
2983 | ||
d8ef542d MD |
2984 | revents = LTTNG_POLL_GETEV(&events, i); |
2985 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2986 | ||
d8ef542d | 2987 | if (pollfd == ctx->consumer_channel_pipe[0]) { |
03e43155 | 2988 | if (revents & LPOLLIN) { |
d8ef542d | 2989 | enum consumer_channel_action action; |
a0cbdd2e | 2990 | uint64_t key; |
d8ef542d | 2991 | |
a0cbdd2e | 2992 | ret = read_channel_pipe(ctx, &chan, &key, &action); |
d8ef542d | 2993 | if (ret <= 0) { |
03e43155 MD |
2994 | if (ret < 0) { |
2995 | ERR("Error reading channel pipe"); | |
2996 | } | |
2997 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
d8ef542d MD |
2998 | continue; |
2999 | } | |
3000 | ||
3001 | switch (action) { | |
3002 | case CONSUMER_CHANNEL_ADD: | |
3003 | DBG("Adding channel %d to poll set", | |
3004 | chan->wait_fd); | |
3005 | ||
3006 | lttng_ht_node_init_u64(&chan->wait_fd_node, | |
3007 | chan->wait_fd); | |
c7260a81 | 3008 | rcu_read_lock(); |
d8ef542d MD |
3009 | lttng_ht_add_unique_u64(channel_ht, |
3010 | &chan->wait_fd_node); | |
c7260a81 | 3011 | rcu_read_unlock(); |
d8ef542d MD |
3012 | /* Add channel to the global poll events list */ |
3013 | lttng_poll_add(&events, chan->wait_fd, | |
03e43155 | 3014 | LPOLLERR | LPOLLHUP); |
d8ef542d | 3015 | break; |
a0cbdd2e MD |
3016 | case CONSUMER_CHANNEL_DEL: |
3017 | { | |
b4a650f3 DG |
3018 | /* |
3019 | * This command should never be called if the channel | |
3020 | * has streams monitored by either the data or metadata | |
3021 | * thread. The consumer only notify this thread with a | |
3022 | * channel del. command if it receives a destroy | |
3023 | * channel command from the session daemon that send it | |
3024 | * if a command prior to the GET_CHANNEL failed. | |
3025 | */ | |
3026 | ||
c7260a81 | 3027 | rcu_read_lock(); |
a0cbdd2e MD |
3028 | chan = consumer_find_channel(key); |
3029 | if (!chan) { | |
c7260a81 | 3030 | rcu_read_unlock(); |
a0cbdd2e MD |
3031 | ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key); |
3032 | break; | |
3033 | } | |
3034 | lttng_poll_del(&events, chan->wait_fd); | |
f623cc0b | 3035 | iter.iter.node = &chan->wait_fd_node.node; |
a0cbdd2e MD |
3036 | ret = lttng_ht_del(channel_ht, &iter); |
3037 | assert(ret == 0); | |
a0cbdd2e | 3038 | |
f2a444f1 DG |
3039 | switch (consumer_data.type) { |
3040 | case LTTNG_CONSUMER_KERNEL: | |
3041 | break; | |
3042 | case LTTNG_CONSUMER32_UST: | |
3043 | case LTTNG_CONSUMER64_UST: | |
212d67a2 DG |
3044 | health_code_update(); |
3045 | /* Destroy streams that might have been left in the stream list. */ | |
3046 | clean_channel_stream_list(chan); | |
f2a444f1 DG |
3047 | break; |
3048 | default: | |
3049 | ERR("Unknown consumer_data type"); | |
3050 | assert(0); | |
3051 | } | |
3052 | ||
a0cbdd2e MD |
3053 | /* |
3054 | * Release our own refcount. Force channel deletion even if | |
3055 | * streams were not initialized. | |
3056 | */ | |
3057 | if (!uatomic_sub_return(&chan->refcount, 1)) { | |
3058 | consumer_del_channel(chan); | |
3059 | } | |
c7260a81 | 3060 | rcu_read_unlock(); |
a0cbdd2e MD |
3061 | goto restart; |
3062 | } | |
d8ef542d MD |
3063 | case CONSUMER_CHANNEL_QUIT: |
3064 | /* | |
3065 | * Remove the pipe from the poll set and continue the loop | |
3066 | * since their might be data to consume. | |
3067 | */ | |
3068 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
3069 | continue; | |
3070 | default: | |
3071 | ERR("Unknown action"); | |
3072 | break; | |
3073 | } | |
03e43155 MD |
3074 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
3075 | DBG("Channel thread pipe hung up"); | |
3076 | /* | |
3077 | * Remove the pipe from the poll set and continue the loop | |
3078 | * since their might be data to consume. | |
3079 | */ | |
3080 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
3081 | continue; | |
3082 | } else { | |
3083 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
3084 | goto end; | |
d8ef542d MD |
3085 | } |
3086 | ||
3087 | /* Handle other stream */ | |
3088 | continue; | |
3089 | } | |
3090 | ||
3091 | rcu_read_lock(); | |
3092 | { | |
3093 | uint64_t tmp_id = (uint64_t) pollfd; | |
3094 | ||
3095 | lttng_ht_lookup(channel_ht, &tmp_id, &iter); | |
3096 | } | |
3097 | node = lttng_ht_iter_get_node_u64(&iter); | |
3098 | assert(node); | |
3099 | ||
3100 | chan = caa_container_of(node, struct lttng_consumer_channel, | |
3101 | wait_fd_node); | |
3102 | ||
3103 | /* Check for error event */ | |
3104 | if (revents & (LPOLLERR | LPOLLHUP)) { | |
3105 | DBG("Channel fd %d is hup|err.", pollfd); | |
3106 | ||
3107 | lttng_poll_del(&events, chan->wait_fd); | |
3108 | ret = lttng_ht_del(channel_ht, &iter); | |
3109 | assert(ret == 0); | |
b4a650f3 DG |
3110 | |
3111 | /* | |
3112 | * This will close the wait fd for each stream associated to | |
3113 | * this channel AND monitored by the data/metadata thread thus | |
3114 | * will be clean by the right thread. | |
3115 | */ | |
d8ef542d | 3116 | consumer_close_channel_streams(chan); |
f2ad556d MD |
3117 | |
3118 | /* Release our own refcount */ | |
3119 | if (!uatomic_sub_return(&chan->refcount, 1) | |
3120 | && !uatomic_read(&chan->nb_init_stream_left)) { | |
3121 | consumer_del_channel(chan); | |
3122 | } | |
03e43155 MD |
3123 | } else { |
3124 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
3125 | rcu_read_unlock(); | |
3126 | goto end; | |
d8ef542d MD |
3127 | } |
3128 | ||
3129 | /* Release RCU lock for the channel looked up */ | |
3130 | rcu_read_unlock(); | |
3131 | } | |
3132 | } | |
3133 | ||
1fc79fb4 MD |
3134 | /* All is OK */ |
3135 | err = 0; | |
d8ef542d MD |
3136 | end: |
3137 | lttng_poll_clean(&events); | |
3138 | end_poll: | |
3139 | destroy_channel_ht(channel_ht); | |
3140 | end_ht: | |
2d57de81 | 3141 | error_testpoint: |
d8ef542d | 3142 | DBG("Channel poll thread exiting"); |
1fc79fb4 MD |
3143 | if (err) { |
3144 | health_error(); | |
3145 | ERR("Health error occurred in %s", __func__); | |
3146 | } | |
3147 | health_unregister(health_consumerd); | |
d8ef542d MD |
3148 | rcu_unregister_thread(); |
3149 | return NULL; | |
3150 | } | |
3151 | ||
331744e3 JD |
3152 | static int set_metadata_socket(struct lttng_consumer_local_data *ctx, |
3153 | struct pollfd *sockpoll, int client_socket) | |
3154 | { | |
3155 | int ret; | |
3156 | ||
3157 | assert(ctx); | |
3158 | assert(sockpoll); | |
3159 | ||
84382d49 MD |
3160 | ret = lttng_consumer_poll_socket(sockpoll); |
3161 | if (ret) { | |
331744e3 JD |
3162 | goto error; |
3163 | } | |
3164 | DBG("Metadata connection on client_socket"); | |
3165 | ||
3166 | /* Blocking call, waiting for transmission */ | |
3167 | ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket); | |
3168 | if (ctx->consumer_metadata_socket < 0) { | |
3169 | WARN("On accept metadata"); | |
3170 | ret = -1; | |
3171 | goto error; | |
3172 | } | |
3173 | ret = 0; | |
3174 | ||
3175 | error: | |
3176 | return ret; | |
3177 | } | |
3178 | ||
3bd1e081 MD |
3179 | /* |
3180 | * This thread listens on the consumerd socket and receives the file | |
3181 | * descriptors from the session daemon. | |
3182 | */ | |
7d980def | 3183 | void *consumer_thread_sessiond_poll(void *data) |
3bd1e081 | 3184 | { |
1fc79fb4 | 3185 | int sock = -1, client_socket, ret, err = -1; |
3bd1e081 MD |
3186 | /* |
3187 | * structure to poll for incoming data on communication socket avoids | |
3188 | * making blocking sockets. | |
3189 | */ | |
3190 | struct pollfd consumer_sockpoll[2]; | |
3191 | struct lttng_consumer_local_data *ctx = data; | |
3192 | ||
e7b994a3 DG |
3193 | rcu_register_thread(); |
3194 | ||
1fc79fb4 MD |
3195 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND); |
3196 | ||
2d57de81 MD |
3197 | if (testpoint(consumerd_thread_sessiond)) { |
3198 | goto error_testpoint; | |
3199 | } | |
3200 | ||
9ce5646a MD |
3201 | health_code_update(); |
3202 | ||
3bd1e081 MD |
3203 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
3204 | unlink(ctx->consumer_command_sock_path); | |
3205 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); | |
3206 | if (client_socket < 0) { | |
3207 | ERR("Cannot create command socket"); | |
3208 | goto end; | |
3209 | } | |
3210 | ||
3211 | ret = lttcomm_listen_unix_sock(client_socket); | |
3212 | if (ret < 0) { | |
3213 | goto end; | |
3214 | } | |
3215 | ||
32258573 | 3216 | DBG("Sending ready command to lttng-sessiond"); |
f73fabfd | 3217 | ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); |
3bd1e081 MD |
3218 | /* return < 0 on error, but == 0 is not fatal */ |
3219 | if (ret < 0) { | |
32258573 | 3220 | ERR("Error sending ready command to lttng-sessiond"); |
3bd1e081 MD |
3221 | goto end; |
3222 | } | |
3223 | ||
3bd1e081 MD |
3224 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ |
3225 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; | |
3226 | consumer_sockpoll[0].events = POLLIN | POLLPRI; | |
3227 | consumer_sockpoll[1].fd = client_socket; | |
3228 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3229 | ||
84382d49 MD |
3230 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3231 | if (ret) { | |
3232 | if (ret > 0) { | |
3233 | /* should exit */ | |
3234 | err = 0; | |
3235 | } | |
3bd1e081 MD |
3236 | goto end; |
3237 | } | |
3238 | DBG("Connection on client_socket"); | |
3239 | ||
3240 | /* Blocking call, waiting for transmission */ | |
3241 | sock = lttcomm_accept_unix_sock(client_socket); | |
534d2592 | 3242 | if (sock < 0) { |
3bd1e081 MD |
3243 | WARN("On accept"); |
3244 | goto end; | |
3245 | } | |
3bd1e081 | 3246 | |
331744e3 JD |
3247 | /* |
3248 | * Setup metadata socket which is the second socket connection on the | |
3249 | * command unix socket. | |
3250 | */ | |
3251 | ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket); | |
84382d49 MD |
3252 | if (ret) { |
3253 | if (ret > 0) { | |
3254 | /* should exit */ | |
3255 | err = 0; | |
3256 | } | |
331744e3 JD |
3257 | goto end; |
3258 | } | |
3259 | ||
d96f09c6 DG |
3260 | /* This socket is not useful anymore. */ |
3261 | ret = close(client_socket); | |
3262 | if (ret < 0) { | |
3263 | PERROR("close client_socket"); | |
3264 | } | |
3265 | client_socket = -1; | |
3266 | ||
3bd1e081 MD |
3267 | /* update the polling structure to poll on the established socket */ |
3268 | consumer_sockpoll[1].fd = sock; | |
3269 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3270 | ||
3271 | while (1) { | |
9ce5646a MD |
3272 | health_code_update(); |
3273 | ||
3274 | health_poll_entry(); | |
3275 | ret = lttng_consumer_poll_socket(consumer_sockpoll); | |
3276 | health_poll_exit(); | |
84382d49 MD |
3277 | if (ret) { |
3278 | if (ret > 0) { | |
3279 | /* should exit */ | |
3280 | err = 0; | |
3281 | } | |
3bd1e081 MD |
3282 | goto end; |
3283 | } | |
3284 | DBG("Incoming command on sock"); | |
3285 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
4cbc1a04 DG |
3286 | if (ret <= 0) { |
3287 | /* | |
3288 | * This could simply be a session daemon quitting. Don't output | |
3289 | * ERR() here. | |
3290 | */ | |
3291 | DBG("Communication interrupted on command socket"); | |
41ba6035 | 3292 | err = 0; |
3bd1e081 MD |
3293 | goto end; |
3294 | } | |
10211f5c | 3295 | if (CMM_LOAD_SHARED(consumer_quit)) { |
3bd1e081 | 3296 | DBG("consumer_thread_receive_fds received quit from signal"); |
1fc79fb4 | 3297 | err = 0; /* All is OK */ |
3bd1e081 MD |
3298 | goto end; |
3299 | } | |
ffe60014 | 3300 | DBG("received command on sock"); |
3bd1e081 | 3301 | } |
1fc79fb4 MD |
3302 | /* All is OK */ |
3303 | err = 0; | |
3304 | ||
3bd1e081 | 3305 | end: |
ffe60014 | 3306 | DBG("Consumer thread sessiond poll exiting"); |
3bd1e081 | 3307 | |
d88aee68 DG |
3308 | /* |
3309 | * Close metadata streams since the producer is the session daemon which | |
3310 | * just died. | |
3311 | * | |
3312 | * NOTE: for now, this only applies to the UST tracer. | |
3313 | */ | |
6d574024 | 3314 | lttng_consumer_close_all_metadata(); |
d88aee68 | 3315 | |
3bd1e081 MD |
3316 | /* |
3317 | * when all fds have hung up, the polling thread | |
3318 | * can exit cleanly | |
3319 | */ | |
10211f5c | 3320 | CMM_STORE_SHARED(consumer_quit, 1); |
3bd1e081 | 3321 | |
04fdd819 | 3322 | /* |
c869f647 | 3323 | * Notify the data poll thread to poll back again and test the |
8994307f | 3324 | * consumer_quit state that we just set so to quit gracefully. |
04fdd819 | 3325 | */ |
acdb9057 | 3326 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
c869f647 | 3327 | |
a0cbdd2e | 3328 | notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT); |
d8ef542d | 3329 | |
5c635c72 MD |
3330 | notify_health_quit_pipe(health_quit_pipe); |
3331 | ||
d96f09c6 DG |
3332 | /* Cleaning up possibly open sockets. */ |
3333 | if (sock >= 0) { | |
3334 | ret = close(sock); | |
3335 | if (ret < 0) { | |
3336 | PERROR("close sock sessiond poll"); | |
3337 | } | |
3338 | } | |
3339 | if (client_socket >= 0) { | |
38476d24 | 3340 | ret = close(client_socket); |
d96f09c6 DG |
3341 | if (ret < 0) { |
3342 | PERROR("close client_socket sessiond poll"); | |
3343 | } | |
3344 | } | |
3345 | ||
2d57de81 | 3346 | error_testpoint: |
1fc79fb4 MD |
3347 | if (err) { |
3348 | health_error(); | |
3349 | ERR("Health error occurred in %s", __func__); | |
3350 | } | |
3351 | health_unregister(health_consumerd); | |
3352 | ||
e7b994a3 | 3353 | rcu_unregister_thread(); |
3bd1e081 MD |
3354 | return NULL; |
3355 | } | |
d41f73b7 | 3356 | |
fc0c52e9 JG |
3357 | static |
3358 | int consumer_flush_buffer(struct lttng_consumer_stream *stream, | |
3359 | int producer_active) | |
3360 | { | |
3361 | int ret = 0; | |
3362 | ||
3363 | switch (consumer_data.type) { | |
3364 | case LTTNG_CONSUMER_KERNEL: | |
3365 | if (producer_active) { | |
3366 | ret = kernctl_buffer_flush(stream->wait_fd); | |
3367 | if (ret < 0) { | |
3368 | ERR("Failed to flush kernel stream"); | |
3369 | goto end; | |
3370 | } | |
3371 | } else { | |
3372 | ret = kernctl_buffer_flush_empty(stream->wait_fd); | |
3373 | if (ret < 0) { | |
3374 | /* | |
3375 | * Doing a buffer flush which does not take into | |
3376 | * account empty packets. This is not perfect, | |
3377 | * but required as a fall-back when | |
3378 | * "flush_empty" is not implemented by | |
3379 | * lttng-modules. | |
3380 | */ | |
3381 | ret = kernctl_buffer_flush(stream->wait_fd); | |
3382 | if (ret < 0) { | |
3383 | ERR("Failed to flush kernel stream"); | |
3384 | goto end; | |
3385 | } | |
3386 | } | |
3387 | } | |
3388 | break; | |
3389 | case LTTNG_CONSUMER32_UST: | |
3390 | case LTTNG_CONSUMER64_UST: | |
3391 | lttng_ustconsumer_flush_buffer(stream, producer_active); | |
3392 | break; | |
3393 | default: | |
3394 | ERR("Unknown consumer_data type"); | |
3395 | abort(); | |
3396 | } | |
3397 | ||
3398 | end: | |
3399 | return ret; | |
3400 | } | |
3401 | ||
3402 | static enum open_packet_status open_packet(struct lttng_consumer_stream *stream) | |
3403 | { | |
3404 | int ret; | |
3405 | enum open_packet_status status; | |
3406 | unsigned long produced_pos_before, produced_pos_after; | |
3407 | ||
3408 | ret = lttng_consumer_sample_snapshot_positions(stream); | |
3409 | if (ret < 0) { | |
3410 | ERR("Failed to snapshot positions before post-rotation empty packet flush: stream id = %" PRIu64 | |
3411 | ", channel name = %s, session id = %" PRIu64, | |
3412 | stream->key, stream->chan->name, | |
3413 | stream->chan->session_id); | |
3414 | status = OPEN_PACKET_STATUS_ERROR; | |
3415 | goto end; | |
3416 | } | |
3417 | ||
3418 | ret = lttng_consumer_get_produced_snapshot( | |
3419 | stream, &produced_pos_before); | |
3420 | if (ret < 0) { | |
3421 | ERR("Failed to read produced position before post-rotation empty packet flush: stream id = %" PRIu64 | |
3422 | ", channel name = %s, session id = %" PRIu64, | |
3423 | stream->key, stream->chan->name, | |
3424 | stream->chan->session_id); | |
3425 | status = OPEN_PACKET_STATUS_ERROR; | |
3426 | goto end; | |
3427 | } | |
3428 | ||
3429 | ret = consumer_flush_buffer(stream, 0); | |
3430 | if (ret) { | |
3431 | ERR("Failed to flush an empty packet at rotation point: stream id = %" PRIu64 | |
3432 | ", channel name = %s, session id = %" PRIu64, | |
3433 | stream->key, stream->chan->name, | |
3434 | stream->chan->session_id); | |
3435 | status = OPEN_PACKET_STATUS_ERROR; | |
3436 | goto end; | |
3437 | } | |
3438 | ||
3439 | ret = lttng_consumer_sample_snapshot_positions(stream); | |
3440 | if (ret < 0) { | |
3441 | ERR("Failed to snapshot positions after post-rotation empty packet flush: stream id = %" PRIu64 | |
3442 | ", channel name = %s, session id = %" PRIu64, | |
3443 | stream->key, stream->chan->name, | |
3444 | stream->chan->session_id); | |
3445 | status = OPEN_PACKET_STATUS_ERROR; | |
3446 | goto end; | |
3447 | } | |
3448 | ||
3449 | ret = lttng_consumer_get_produced_snapshot(stream, &produced_pos_after); | |
3450 | if (ret < 0) { | |
3451 | ERR("Failed to read produced position after post-rotation empty packet flush: stream id = %" PRIu64 | |
3452 | ", channel name = %s, session id = %" PRIu64, | |
3453 | stream->key, stream->chan->name, | |
3454 | stream->chan->session_id); | |
3455 | status = OPEN_PACKET_STATUS_ERROR; | |
3456 | goto end; | |
3457 | } | |
3458 | ||
3459 | /* | |
3460 | * Determine if the flush had an effect by comparing the produced | |
3461 | * positons before and after the flush. | |
3462 | */ | |
3463 | status = produced_pos_before != produced_pos_after ? | |
3464 | OPEN_PACKET_STATUS_OPENED : | |
3465 | OPEN_PACKET_STATUS_NO_SPACE; | |
1223361a JG |
3466 | if (status == OPEN_PACKET_STATUS_OPENED) { |
3467 | stream->opened_packet_in_current_trace_chunk = true; | |
3468 | } | |
fc0c52e9 JG |
3469 | end: |
3470 | return status; | |
3471 | } | |
3472 | ||
3473 | static bool stream_is_rotating_to_null_chunk( | |
3474 | const struct lttng_consumer_stream *stream) | |
3475 | { | |
3476 | bool rotating_to_null_chunk = false; | |
3477 | ||
3478 | if (stream->rotate_position == -1ULL) { | |
3479 | /* No rotation ongoing. */ | |
3480 | goto end; | |
3481 | } | |
3482 | ||
3483 | if (stream->trace_chunk == stream->chan->trace_chunk || | |
3484 | !stream->chan->trace_chunk) { | |
3485 | rotating_to_null_chunk = true; | |
3486 | } | |
3487 | end: | |
3488 | return rotating_to_null_chunk; | |
3489 | } | |
3490 | ||
4078b776 | 3491 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
bdc8d1bb JG |
3492 | struct lttng_consumer_local_data *ctx, |
3493 | bool locked_by_caller) | |
d41f73b7 | 3494 | { |
9a63f11d | 3495 | ssize_t ret, written_bytes = 0; |
9573993e | 3496 | int rotation_ret; |
bdc8d1bb | 3497 | struct stream_subbuffer subbuffer = {}; |
74251bb8 | 3498 | |
bdc8d1bb JG |
3499 | if (!locked_by_caller) { |
3500 | stream->read_subbuffer_ops.lock(stream); | |
8aaed9e7 JR |
3501 | } else { |
3502 | stream->read_subbuffer_ops.assert_locked(stream); | |
bdc8d1bb JG |
3503 | } |
3504 | ||
3505 | if (stream->read_subbuffer_ops.on_wake_up) { | |
3506 | ret = stream->read_subbuffer_ops.on_wake_up(stream); | |
3507 | if (ret) { | |
3508 | goto end; | |
3509 | } | |
94d49140 | 3510 | } |
74251bb8 | 3511 | |
9573993e JG |
3512 | /* |
3513 | * If the stream was flagged to be ready for rotation before we extract | |
3514 | * the next packet, rotate it now. | |
3515 | */ | |
3516 | if (stream->rotate_ready) { | |
3517 | DBG("Rotate stream before consuming data"); | |
3518 | ret = lttng_consumer_rotate_stream(ctx, stream); | |
3519 | if (ret < 0) { | |
3520 | ERR("Stream rotation error before consuming data"); | |
3521 | goto end; | |
3522 | } | |
3523 | } | |
3524 | ||
bdc8d1bb JG |
3525 | ret = stream->read_subbuffer_ops.get_next_subbuffer(stream, &subbuffer); |
3526 | if (ret) { | |
3527 | if (ret == -ENODATA) { | |
3528 | /* Not an error. */ | |
3529 | ret = 0; | |
fa2f196b | 3530 | goto sleep_stream; |
bdc8d1bb JG |
3531 | } |
3532 | goto end; | |
d41f73b7 | 3533 | } |
74251bb8 | 3534 | |
bdc8d1bb JG |
3535 | ret = stream->read_subbuffer_ops.pre_consume_subbuffer( |
3536 | stream, &subbuffer); | |
3537 | if (ret) { | |
3538 | goto error_put_subbuf; | |
3539 | } | |
3540 | ||
3541 | written_bytes = stream->read_subbuffer_ops.consume_subbuffer( | |
3542 | ctx, stream, &subbuffer); | |
56c86e0a FD |
3543 | if (written_bytes <= 0) { |
3544 | ERR("Error consuming subbuffer: (%zd)", written_bytes); | |
3545 | ret = (int) written_bytes; | |
3546 | goto error_put_subbuf; | |
bdc8d1bb JG |
3547 | } |
3548 | ||
3549 | ret = stream->read_subbuffer_ops.put_next_subbuffer(stream, &subbuffer); | |
3550 | if (ret) { | |
9573993e JG |
3551 | goto end; |
3552 | } | |
3553 | ||
bdc8d1bb JG |
3554 | if (stream->read_subbuffer_ops.post_consume) { |
3555 | ret = stream->read_subbuffer_ops.post_consume(stream, &subbuffer, ctx); | |
3556 | if (ret) { | |
3557 | goto end; | |
3558 | } | |
3559 | } | |
3560 | ||
9573993e JG |
3561 | /* |
3562 | * After extracting the packet, we check if the stream is now ready to | |
3563 | * be rotated and perform the action immediately. | |
3564 | * | |
3565 | * Don't overwrite `ret` as callers expect the number of bytes | |
3566 | * consumed to be returned on success. | |
3567 | */ | |
3568 | rotation_ret = lttng_consumer_stream_is_rotate_ready(stream); | |
3569 | if (rotation_ret == 1) { | |
3570 | rotation_ret = lttng_consumer_rotate_stream(ctx, stream); | |
3571 | if (rotation_ret < 0) { | |
3572 | ret = rotation_ret; | |
3573 | ERR("Stream rotation error after consuming data"); | |
3574 | goto end; | |
3575 | } | |
3576 | } else if (rotation_ret < 0) { | |
3577 | ret = rotation_ret; | |
3578 | ERR("Failed to check if stream was ready to rotate after consuming data"); | |
3579 | goto end; | |
3580 | } | |
3581 | ||
fc0c52e9 JG |
3582 | /* |
3583 | * TODO roll into a post_consume op as this doesn't apply to metadata | |
3584 | * streams. | |
3585 | */ | |
3586 | if (!stream->opened_packet_in_current_trace_chunk && | |
3587 | stream->trace_chunk && !stream->metadata_flag && | |
3588 | !stream_is_rotating_to_null_chunk(stream)) { | |
3589 | const enum open_packet_status status = open_packet(stream); | |
3590 | ||
3591 | switch (status) { | |
3592 | case OPEN_PACKET_STATUS_OPENED: | |
3593 | DBG("Opened a packet after consuming a packet rotation: stream id = %" PRIu64 | |
3594 | ", channel name = %s, session id = %" PRIu64, | |
3595 | stream->key, stream->chan->name, | |
3596 | stream->chan->session_id); | |
fc0c52e9 JG |
3597 | break; |
3598 | case OPEN_PACKET_STATUS_NO_SPACE: | |
3599 | /* | |
3600 | * Can't open a packet as there is no space left. | |
3601 | * This means that new events were produced, resulting | |
1223361a | 3602 | * in a packet being opened, which is what we wanted |
fc0c52e9 JG |
3603 | * anyhow. |
3604 | */ | |
3605 | DBG("No space left to open a packet after consuming a packet: stream id = %" PRIu64 | |
3606 | ", channel name = %s, session id = %" PRIu64, | |
3607 | stream->key, stream->chan->name, | |
3608 | stream->chan->session_id); | |
3609 | stream->opened_packet_in_current_trace_chunk = true; | |
3610 | break; | |
3611 | case OPEN_PACKET_STATUS_ERROR: | |
3612 | /* Logged by callee. */ | |
3613 | ret = -1; | |
3614 | goto end; | |
3615 | default: | |
3616 | abort(); | |
3617 | } | |
fc0c52e9 JG |
3618 | } |
3619 | ||
8ac94c7d | 3620 | sleep_stream: |
bdc8d1bb JG |
3621 | if (stream->read_subbuffer_ops.on_sleep) { |
3622 | stream->read_subbuffer_ops.on_sleep(stream, ctx); | |
3623 | } | |
3624 | ||
3625 | ret = written_bytes; | |
9573993e | 3626 | end: |
bdc8d1bb JG |
3627 | if (!locked_by_caller) { |
3628 | stream->read_subbuffer_ops.unlock(stream); | |
94d49140 | 3629 | } |
bdc8d1bb | 3630 | |
74251bb8 | 3631 | return ret; |
bdc8d1bb JG |
3632 | error_put_subbuf: |
3633 | (void) stream->read_subbuffer_ops.put_next_subbuffer(stream, &subbuffer); | |
3634 | goto end; | |
d41f73b7 MD |
3635 | } |
3636 | ||
3637 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
3638 | { | |
3639 | switch (consumer_data.type) { | |
3640 | case LTTNG_CONSUMER_KERNEL: | |
3641 | return lttng_kconsumer_on_recv_stream(stream); | |
7753dea8 MD |
3642 | case LTTNG_CONSUMER32_UST: |
3643 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
3644 | return lttng_ustconsumer_on_recv_stream(stream); |
3645 | default: | |
3646 | ERR("Unknown consumer_data type"); | |
3647 | assert(0); | |
3648 | return -ENOSYS; | |
3649 | } | |
3650 | } | |
e4421fec DG |
3651 | |
3652 | /* | |
3653 | * Allocate and set consumer data hash tables. | |
3654 | */ | |
282dadbc | 3655 | int lttng_consumer_init(void) |
e4421fec | 3656 | { |
d88aee68 | 3657 | consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3658 | if (!consumer_data.channel_ht) { |
3659 | goto error; | |
3660 | } | |
3661 | ||
5c3892a6 JG |
3662 | consumer_data.channels_by_session_id_ht = |
3663 | lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3664 | if (!consumer_data.channels_by_session_id_ht) { | |
3665 | goto error; | |
3666 | } | |
3667 | ||
d88aee68 | 3668 | consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3669 | if (!consumer_data.relayd_ht) { |
3670 | goto error; | |
3671 | } | |
3672 | ||
d88aee68 | 3673 | consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3674 | if (!consumer_data.stream_list_ht) { |
3675 | goto error; | |
3676 | } | |
3677 | ||
d8ef542d | 3678 | consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3679 | if (!consumer_data.stream_per_chan_id_ht) { |
3680 | goto error; | |
3681 | } | |
3682 | ||
3683 | data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3684 | if (!data_ht) { | |
3685 | goto error; | |
3686 | } | |
3687 | ||
3688 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3689 | if (!metadata_ht) { | |
3690 | goto error; | |
3691 | } | |
3692 | ||
28cc88f3 JG |
3693 | consumer_data.chunk_registry = lttng_trace_chunk_registry_create(); |
3694 | if (!consumer_data.chunk_registry) { | |
3695 | goto error; | |
3696 | } | |
3697 | ||
282dadbc MD |
3698 | return 0; |
3699 | ||
3700 | error: | |
3701 | return -1; | |
e4421fec | 3702 | } |
7735ef9e DG |
3703 | |
3704 | /* | |
3705 | * Process the ADD_RELAYD command receive by a consumer. | |
3706 | * | |
3707 | * This will create a relayd socket pair and add it to the relayd hash table. | |
3708 | * The caller MUST acquire a RCU read side lock before calling it. | |
3709 | */ | |
37c7ffca JR |
3710 | void consumer_add_relayd_socket(uint64_t net_seq_idx, |
3711 | int sock_type, | |
3712 | struct lttng_consumer_local_data *ctx, | |
3713 | int sock, | |
6151a90f | 3714 | struct pollfd *consumer_sockpoll, |
37c7ffca JR |
3715 | uint64_t sessiond_id, |
3716 | uint64_t relayd_session_id, | |
3717 | uint32_t relayd_version_major, | |
3718 | uint32_t relayd_version_minor, | |
3719 | enum lttcomm_sock_proto relayd_socket_protocol) | |
7735ef9e | 3720 | { |
cd2b09ed | 3721 | int fd = -1, ret = -1, relayd_created = 0; |
0c759fc9 | 3722 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
d4298c99 | 3723 | struct consumer_relayd_sock_pair *relayd = NULL; |
7735ef9e | 3724 | |
6151a90f | 3725 | assert(ctx); |
6151a90f | 3726 | |
da009f2c | 3727 | DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx); |
7735ef9e DG |
3728 | |
3729 | /* Get relayd reference if exists. */ | |
3730 | relayd = consumer_find_relayd(net_seq_idx); | |
3731 | if (relayd == NULL) { | |
da009f2c | 3732 | assert(sock_type == LTTNG_STREAM_CONTROL); |
7735ef9e DG |
3733 | /* Not found. Allocate one. */ |
3734 | relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); | |
3735 | if (relayd == NULL) { | |
618a6a28 MD |
3736 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
3737 | goto error; | |
0d08d75e | 3738 | } else { |
30319bcb | 3739 | relayd->sessiond_session_id = sessiond_id; |
0d08d75e | 3740 | relayd_created = 1; |
7735ef9e | 3741 | } |
0d08d75e DG |
3742 | |
3743 | /* | |
3744 | * This code path MUST continue to the consumer send status message to | |
3745 | * we can notify the session daemon and continue our work without | |
3746 | * killing everything. | |
3747 | */ | |
da009f2c MD |
3748 | } else { |
3749 | /* | |
3750 | * relayd key should never be found for control socket. | |
3751 | */ | |
3752 | assert(sock_type != LTTNG_STREAM_CONTROL); | |
0d08d75e DG |
3753 | } |
3754 | ||
3755 | /* First send a status message before receiving the fds. */ | |
0c759fc9 | 3756 | ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS); |
618a6a28 | 3757 | if (ret < 0) { |
0d08d75e | 3758 | /* Somehow, the session daemon is not responding anymore. */ |
618a6a28 MD |
3759 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); |
3760 | goto error_nosignal; | |
7735ef9e DG |
3761 | } |
3762 | ||
3763 | /* Poll on consumer socket. */ | |
84382d49 MD |
3764 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3765 | if (ret) { | |
3766 | /* Needing to exit in the middle of a command: error. */ | |
0d08d75e | 3767 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
618a6a28 | 3768 | goto error_nosignal; |
7735ef9e DG |
3769 | } |
3770 | ||
3771 | /* Get relayd socket from session daemon */ | |
3772 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); | |
3773 | if (ret != sizeof(fd)) { | |
4028eeb9 | 3774 | fd = -1; /* Just in case it gets set with an invalid value. */ |
0d08d75e DG |
3775 | |
3776 | /* | |
3777 | * Failing to receive FDs might indicate a major problem such as | |
3778 | * reaching a fd limit during the receive where the kernel returns a | |
3779 | * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we | |
3780 | * don't take any chances and stop everything. | |
3781 | * | |
3782 | * XXX: Feature request #558 will fix that and avoid this possible | |
3783 | * issue when reaching the fd limit. | |
3784 | */ | |
3785 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); | |
618a6a28 | 3786 | ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD; |
f50f23d9 DG |
3787 | goto error; |
3788 | } | |
3789 | ||
7735ef9e DG |
3790 | /* Copy socket information and received FD */ |
3791 | switch (sock_type) { | |
3792 | case LTTNG_STREAM_CONTROL: | |
3793 | /* Copy received lttcomm socket */ | |
37c7ffca JR |
3794 | ret = lttcomm_populate_sock_from_open_socket( |
3795 | &relayd->control_sock.sock, fd, | |
3796 | relayd_socket_protocol); | |
7735ef9e | 3797 | |
6151a90f | 3798 | /* Assign version values. */ |
37c7ffca JR |
3799 | relayd->control_sock.major = relayd_version_major; |
3800 | relayd->control_sock.minor = relayd_version_minor; | |
c5b6f4f0 | 3801 | |
d3e2ba59 | 3802 | relayd->relayd_session_id = relayd_session_id; |
c5b6f4f0 | 3803 | |
7735ef9e DG |
3804 | break; |
3805 | case LTTNG_STREAM_DATA: | |
3806 | /* Copy received lttcomm socket */ | |
37c7ffca JR |
3807 | ret = lttcomm_populate_sock_from_open_socket( |
3808 | &relayd->data_sock.sock, fd, | |
3809 | relayd_socket_protocol); | |
6151a90f | 3810 | /* Assign version values. */ |
37c7ffca JR |
3811 | relayd->data_sock.major = relayd_version_major; |
3812 | relayd->data_sock.minor = relayd_version_minor; | |
7735ef9e DG |
3813 | break; |
3814 | default: | |
3815 | ERR("Unknown relayd socket type (%d)", sock_type); | |
618a6a28 | 3816 | ret_code = LTTCOMM_CONSUMERD_FATAL; |
7735ef9e DG |
3817 | goto error; |
3818 | } | |
3819 | ||
37c7ffca JR |
3820 | if (ret < 0) { |
3821 | ret_code = LTTCOMM_CONSUMERD_FATAL; | |
3822 | goto error; | |
3823 | } | |
3824 | ||
d88aee68 | 3825 | DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)", |
7735ef9e DG |
3826 | sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", |
3827 | relayd->net_seq_idx, fd); | |
39d9954c FD |
3828 | /* |
3829 | * We gave the ownership of the fd to the relayd structure. Set the | |
3830 | * fd to -1 so we don't call close() on it in the error path below. | |
3831 | */ | |
3832 | fd = -1; | |
7735ef9e | 3833 | |
618a6a28 MD |
3834 | /* We successfully added the socket. Send status back. */ |
3835 | ret = consumer_send_status_msg(sock, ret_code); | |
3836 | if (ret < 0) { | |
3837 | /* Somehow, the session daemon is not responding anymore. */ | |
3838 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3839 | goto error_nosignal; | |
3840 | } | |
3841 | ||
7735ef9e DG |
3842 | /* |
3843 | * Add relayd socket pair to consumer data hashtable. If object already | |
3844 | * exists or on error, the function gracefully returns. | |
3845 | */ | |
9276e5c8 | 3846 | relayd->ctx = ctx; |
d09e1200 | 3847 | add_relayd(relayd); |
7735ef9e DG |
3848 | |
3849 | /* All good! */ | |
2527bf85 | 3850 | return; |
7735ef9e DG |
3851 | |
3852 | error: | |
618a6a28 MD |
3853 | if (consumer_send_status_msg(sock, ret_code) < 0) { |
3854 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3855 | } | |
3856 | ||
3857 | error_nosignal: | |
4028eeb9 DG |
3858 | /* Close received socket if valid. */ |
3859 | if (fd >= 0) { | |
3860 | if (close(fd)) { | |
3861 | PERROR("close received socket"); | |
3862 | } | |
3863 | } | |
cd2b09ed DG |
3864 | |
3865 | if (relayd_created) { | |
cd2b09ed DG |
3866 | free(relayd); |
3867 | } | |
7735ef9e | 3868 | } |
ca22feea | 3869 | |
f7079f67 DG |
3870 | /* |
3871 | * Search for a relayd associated to the session id and return the reference. | |
3872 | * | |
3873 | * A rcu read side lock MUST be acquire before calling this function and locked | |
3874 | * until the relayd object is no longer necessary. | |
3875 | */ | |
3876 | static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id) | |
3877 | { | |
3878 | struct lttng_ht_iter iter; | |
f7079f67 | 3879 | struct consumer_relayd_sock_pair *relayd = NULL; |
f7079f67 DG |
3880 | |
3881 | /* Iterate over all relayd since they are indexed by net_seq_idx. */ | |
3882 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
3883 | node.node) { | |
18261bd1 DG |
3884 | /* |
3885 | * Check by sessiond id which is unique here where the relayd session | |
3886 | * id might not be when having multiple relayd. | |
3887 | */ | |
3888 | if (relayd->sessiond_session_id == id) { | |
f7079f67 | 3889 | /* Found the relayd. There can be only one per id. */ |
18261bd1 | 3890 | goto found; |
f7079f67 DG |
3891 | } |
3892 | } | |
3893 | ||
18261bd1 DG |
3894 | return NULL; |
3895 | ||
3896 | found: | |
f7079f67 DG |
3897 | return relayd; |
3898 | } | |
3899 | ||
ca22feea DG |
3900 | /* |
3901 | * Check if for a given session id there is still data needed to be extract | |
3902 | * from the buffers. | |
3903 | * | |
6d805429 | 3904 | * Return 1 if data is pending or else 0 meaning ready to be read. |
ca22feea | 3905 | */ |
6d805429 | 3906 | int consumer_data_pending(uint64_t id) |
ca22feea DG |
3907 | { |
3908 | int ret; | |
3909 | struct lttng_ht_iter iter; | |
3910 | struct lttng_ht *ht; | |
3911 | struct lttng_consumer_stream *stream; | |
f7079f67 | 3912 | struct consumer_relayd_sock_pair *relayd = NULL; |
6d805429 | 3913 | int (*data_pending)(struct lttng_consumer_stream *); |
ca22feea | 3914 | |
6d805429 | 3915 | DBG("Consumer data pending command on session id %" PRIu64, id); |
ca22feea | 3916 | |
6f6eda74 | 3917 | rcu_read_lock(); |
ca22feea DG |
3918 | pthread_mutex_lock(&consumer_data.lock); |
3919 | ||
3920 | switch (consumer_data.type) { | |
3921 | case LTTNG_CONSUMER_KERNEL: | |
6d805429 | 3922 | data_pending = lttng_kconsumer_data_pending; |
ca22feea DG |
3923 | break; |
3924 | case LTTNG_CONSUMER32_UST: | |
3925 | case LTTNG_CONSUMER64_UST: | |
6d805429 | 3926 | data_pending = lttng_ustconsumer_data_pending; |
ca22feea DG |
3927 | break; |
3928 | default: | |
3929 | ERR("Unknown consumer data type"); | |
3930 | assert(0); | |
3931 | } | |
3932 | ||
3933 | /* Ease our life a bit */ | |
3934 | ht = consumer_data.stream_list_ht; | |
3935 | ||
c8f59ee5 | 3936 | cds_lfht_for_each_entry_duplicate(ht->ht, |
d88aee68 DG |
3937 | ht->hash_fct(&id, lttng_ht_seed), |
3938 | ht->match_fct, &id, | |
ca22feea | 3939 | &iter.iter, stream, node_session_id.node) { |
bb586a6e | 3940 | pthread_mutex_lock(&stream->lock); |
ca22feea | 3941 | |
4e9a4686 DG |
3942 | /* |
3943 | * A removed node from the hash table indicates that the stream has | |
3944 | * been deleted thus having a guarantee that the buffers are closed | |
3945 | * on the consumer side. However, data can still be transmitted | |
3946 | * over the network so don't skip the relayd check. | |
3947 | */ | |
3948 | ret = cds_lfht_is_node_deleted(&stream->node.node); | |
3949 | if (!ret) { | |
3950 | /* Check the stream if there is data in the buffers. */ | |
6d805429 DG |
3951 | ret = data_pending(stream); |
3952 | if (ret == 1) { | |
4e9a4686 | 3953 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3954 | goto data_pending; |
4e9a4686 DG |
3955 | } |
3956 | } | |
3957 | ||
d9f0c7c7 JR |
3958 | pthread_mutex_unlock(&stream->lock); |
3959 | } | |
3960 | ||
3961 | relayd = find_relayd_by_session_id(id); | |
3962 | if (relayd) { | |
3963 | unsigned int is_data_inflight = 0; | |
3964 | ||
3965 | /* Send init command for data pending. */ | |
3966 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3967 | ret = relayd_begin_data_pending(&relayd->control_sock, | |
3968 | relayd->relayd_session_id); | |
3969 | if (ret < 0) { | |
3970 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3971 | /* Communication error thus the relayd so no data pending. */ | |
3972 | goto data_not_pending; | |
3973 | } | |
3974 | ||
3975 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
3976 | ht->hash_fct(&id, lttng_ht_seed), | |
3977 | ht->match_fct, &id, | |
3978 | &iter.iter, stream, node_session_id.node) { | |
c8f59ee5 | 3979 | if (stream->metadata_flag) { |
ad7051c0 DG |
3980 | ret = relayd_quiescent_control(&relayd->control_sock, |
3981 | stream->relayd_stream_id); | |
c8f59ee5 | 3982 | } else { |
6d805429 | 3983 | ret = relayd_data_pending(&relayd->control_sock, |
39df6d9f DG |
3984 | stream->relayd_stream_id, |
3985 | stream->next_net_seq_num - 1); | |
c8f59ee5 | 3986 | } |
d9f0c7c7 JR |
3987 | |
3988 | if (ret == 1) { | |
3989 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3990 | goto data_pending; | |
3991 | } else if (ret < 0) { | |
9276e5c8 JR |
3992 | ERR("Relayd data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
3993 | lttng_consumer_cleanup_relayd(relayd); | |
3994 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
9276e5c8 JR |
3995 | goto data_not_pending; |
3996 | } | |
c8f59ee5 | 3997 | } |
f7079f67 | 3998 | |
d9f0c7c7 | 3999 | /* Send end command for data pending. */ |
f7079f67 DG |
4000 | ret = relayd_end_data_pending(&relayd->control_sock, |
4001 | relayd->relayd_session_id, &is_data_inflight); | |
4002 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
bdd88757 | 4003 | if (ret < 0) { |
9276e5c8 JR |
4004 | ERR("Relayd end data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
4005 | lttng_consumer_cleanup_relayd(relayd); | |
f7079f67 DG |
4006 | goto data_not_pending; |
4007 | } | |
bdd88757 DG |
4008 | if (is_data_inflight) { |
4009 | goto data_pending; | |
4010 | } | |
f7079f67 DG |
4011 | } |
4012 | ||
ca22feea | 4013 | /* |
f7079f67 DG |
4014 | * Finding _no_ node in the hash table and no inflight data means that the |
4015 | * stream(s) have been removed thus data is guaranteed to be available for | |
4016 | * analysis from the trace files. | |
ca22feea DG |
4017 | */ |
4018 | ||
f7079f67 | 4019 | data_not_pending: |
ca22feea DG |
4020 | /* Data is available to be read by a viewer. */ |
4021 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 4022 | rcu_read_unlock(); |
6d805429 | 4023 | return 0; |
ca22feea | 4024 | |
f7079f67 | 4025 | data_pending: |
ca22feea DG |
4026 | /* Data is still being extracted from buffers. */ |
4027 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 4028 | rcu_read_unlock(); |
6d805429 | 4029 | return 1; |
ca22feea | 4030 | } |
f50f23d9 DG |
4031 | |
4032 | /* | |
4033 | * Send a ret code status message to the sessiond daemon. | |
4034 | * | |
4035 | * Return the sendmsg() return value. | |
4036 | */ | |
4037 | int consumer_send_status_msg(int sock, int ret_code) | |
4038 | { | |
4039 | struct lttcomm_consumer_status_msg msg; | |
4040 | ||
53efb85a | 4041 | memset(&msg, 0, sizeof(msg)); |
f50f23d9 DG |
4042 | msg.ret_code = ret_code; |
4043 | ||
4044 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
4045 | } | |
ffe60014 DG |
4046 | |
4047 | /* | |
4048 | * Send a channel status message to the sessiond daemon. | |
4049 | * | |
4050 | * Return the sendmsg() return value. | |
4051 | */ | |
4052 | int consumer_send_status_channel(int sock, | |
4053 | struct lttng_consumer_channel *channel) | |
4054 | { | |
4055 | struct lttcomm_consumer_status_channel msg; | |
4056 | ||
4057 | assert(sock >= 0); | |
4058 | ||
53efb85a | 4059 | memset(&msg, 0, sizeof(msg)); |
ffe60014 | 4060 | if (!channel) { |
0c759fc9 | 4061 | msg.ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL; |
ffe60014 | 4062 | } else { |
0c759fc9 | 4063 | msg.ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
ffe60014 DG |
4064 | msg.key = channel->key; |
4065 | msg.stream_count = channel->streams.count; | |
4066 | } | |
4067 | ||
4068 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
4069 | } | |
5c786ded | 4070 | |
d07ceecd MD |
4071 | unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos, |
4072 | unsigned long produced_pos, uint64_t nb_packets_per_stream, | |
4073 | uint64_t max_sb_size) | |
5c786ded | 4074 | { |
d07ceecd | 4075 | unsigned long start_pos; |
5c786ded | 4076 | |
d07ceecd MD |
4077 | if (!nb_packets_per_stream) { |
4078 | return consumed_pos; /* Grab everything */ | |
4079 | } | |
4080 | start_pos = produced_pos - offset_align_floor(produced_pos, max_sb_size); | |
4081 | start_pos -= max_sb_size * nb_packets_per_stream; | |
4082 | if ((long) (start_pos - consumed_pos) < 0) { | |
4083 | return consumed_pos; /* Grab everything */ | |
4084 | } | |
4085 | return start_pos; | |
5c786ded | 4086 | } |
a1ae2ea5 | 4087 | |
c0e2451e JG |
4088 | /* Stream lock must be held by the caller. */ |
4089 | static int sample_stream_positions(struct lttng_consumer_stream *stream, | |
4090 | unsigned long *produced, unsigned long *consumed) | |
4091 | { | |
4092 | int ret; | |
4093 | ||
4094 | ASSERT_LOCKED(stream->lock); | |
4095 | ||
4096 | ret = lttng_consumer_sample_snapshot_positions(stream); | |
4097 | if (ret < 0) { | |
4098 | ERR("Failed to sample snapshot positions"); | |
4099 | goto end; | |
4100 | } | |
4101 | ||
4102 | ret = lttng_consumer_get_produced_snapshot(stream, produced); | |
4103 | if (ret < 0) { | |
4104 | ERR("Failed to sample produced position"); | |
4105 | goto end; | |
4106 | } | |
4107 | ||
4108 | ret = lttng_consumer_get_consumed_snapshot(stream, consumed); | |
4109 | if (ret < 0) { | |
4110 | ERR("Failed to sample consumed position"); | |
4111 | goto end; | |
4112 | } | |
4113 | ||
4114 | end: | |
4115 | return ret; | |
4116 | } | |
4117 | ||
b99a8d42 JD |
4118 | /* |
4119 | * Sample the rotate position for all the streams of a channel. If a stream | |
4120 | * is already at the rotate position (produced == consumed), we flag it as | |
4121 | * ready for rotation. The rotation of ready streams occurs after we have | |
4122 | * replied to the session daemon that we have finished sampling the positions. | |
92b7a7f8 | 4123 | * Must be called with RCU read-side lock held to ensure existence of channel. |
b99a8d42 JD |
4124 | * |
4125 | * Returns 0 on success, < 0 on error | |
4126 | */ | |
92b7a7f8 | 4127 | int lttng_consumer_rotate_channel(struct lttng_consumer_channel *channel, |
d2956687 | 4128 | uint64_t key, uint64_t relayd_id, uint32_t metadata, |
b99a8d42 JD |
4129 | struct lttng_consumer_local_data *ctx) |
4130 | { | |
4131 | int ret; | |
b99a8d42 JD |
4132 | struct lttng_consumer_stream *stream; |
4133 | struct lttng_ht_iter iter; | |
4134 | struct lttng_ht *ht = consumer_data.stream_per_chan_id_ht; | |
c35f9726 JG |
4135 | struct lttng_dynamic_array stream_rotation_positions; |
4136 | uint64_t next_chunk_id, stream_count = 0; | |
4137 | enum lttng_trace_chunk_status chunk_status; | |
4138 | const bool is_local_trace = relayd_id == -1ULL; | |
4139 | struct consumer_relayd_sock_pair *relayd = NULL; | |
4140 | bool rotating_to_new_chunk = true; | |
e537f276 JG |
4141 | /* Array of `struct lttng_consumer_stream *` */ |
4142 | struct lttng_dynamic_pointer_array streams_packet_to_open; | |
4143 | size_t stream_idx; | |
b99a8d42 JD |
4144 | |
4145 | DBG("Consumer sample rotate position for channel %" PRIu64, key); | |
4146 | ||
c35f9726 JG |
4147 | lttng_dynamic_array_init(&stream_rotation_positions, |
4148 | sizeof(struct relayd_stream_rotation_position), NULL); | |
e537f276 | 4149 | lttng_dynamic_pointer_array_init(&streams_packet_to_open, NULL); |
c35f9726 | 4150 | |
b99a8d42 JD |
4151 | rcu_read_lock(); |
4152 | ||
b99a8d42 | 4153 | pthread_mutex_lock(&channel->lock); |
c35f9726 JG |
4154 | assert(channel->trace_chunk); |
4155 | chunk_status = lttng_trace_chunk_get_id(channel->trace_chunk, | |
4156 | &next_chunk_id); | |
4157 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { | |
4158 | ret = -1; | |
4159 | goto end_unlock_channel; | |
4160 | } | |
b99a8d42 JD |
4161 | |
4162 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
4163 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
4164 | ht->match_fct, &channel->key, &iter.iter, | |
4165 | stream, node_channel_id.node) { | |
a40a503f | 4166 | unsigned long produced_pos = 0, consumed_pos = 0; |
b99a8d42 JD |
4167 | |
4168 | health_code_update(); | |
4169 | ||
4170 | /* | |
4171 | * Lock stream because we are about to change its state. | |
4172 | */ | |
4173 | pthread_mutex_lock(&stream->lock); | |
4174 | ||
c35f9726 JG |
4175 | if (stream->trace_chunk == stream->chan->trace_chunk) { |
4176 | rotating_to_new_chunk = false; | |
4177 | } | |
4178 | ||
a40a503f | 4179 | /* |
c0e2451e | 4180 | * Do not flush a packet when rotating from a NULL trace |
a9dde553 | 4181 | * chunk. The stream has no means to output data, and the prior |
c0e2451e JG |
4182 | * rotation which rotated to NULL performed that side-effect |
4183 | * already. No new data can be produced when a stream has no | |
4184 | * associated trace chunk (e.g. a stop followed by a rotate). | |
a40a503f | 4185 | */ |
a9dde553 | 4186 | if (stream->trace_chunk) { |
c0e2451e JG |
4187 | bool flush_active; |
4188 | ||
4189 | if (stream->metadata_flag) { | |
4190 | /* | |
4191 | * Don't produce an empty metadata packet, | |
4192 | * simply close the current one. | |
4193 | * | |
4194 | * Metadata is regenerated on every trace chunk | |
4195 | * switch; there is no concern that no data was | |
4196 | * produced. | |
4197 | */ | |
4198 | flush_active = true; | |
4199 | } else { | |
4200 | /* | |
4201 | * Only flush an empty packet if the "packet | |
4202 | * open" could not be performed on transition | |
4203 | * to a new trace chunk and no packets were | |
4204 | * consumed within the chunk's lifetime. | |
4205 | */ | |
4206 | if (stream->opened_packet_in_current_trace_chunk) { | |
4207 | flush_active = true; | |
4208 | } else { | |
4209 | /* | |
4210 | * Stream could have been full at the | |
4211 | * time of rotation, but then have had | |
4212 | * no activity at all. | |
4213 | * | |
4214 | * It is important to flush a packet | |
4215 | * to prevent 0-length files from being | |
4216 | * produced as most viewers choke on | |
4217 | * them. | |
4218 | * | |
4219 | * Unfortunately viewers will not be | |
4220 | * able to know that tracing was active | |
4221 | * for this stream during this trace | |
4222 | * chunk's lifetime. | |
4223 | */ | |
4224 | ret = sample_stream_positions(stream, &produced_pos, &consumed_pos); | |
4225 | if (ret) { | |
4226 | goto end_unlock_stream; | |
4227 | } | |
4228 | ||
4229 | /* | |
4230 | * Don't flush an empty packet if data | |
4231 | * was produced; it will be consumed | |
4232 | * before the rotation completes. | |
4233 | */ | |
4234 | flush_active = produced_pos != consumed_pos; | |
4235 | if (!flush_active) { | |
4236 | enum lttng_trace_chunk_status chunk_status; | |
4237 | const char *trace_chunk_name; | |
4238 | uint64_t trace_chunk_id; | |
4239 | ||
4240 | chunk_status = lttng_trace_chunk_get_name( | |
4241 | stream->trace_chunk, | |
4242 | &trace_chunk_name, | |
4243 | NULL); | |
4244 | if (chunk_status == LTTNG_TRACE_CHUNK_STATUS_NONE) { | |
4245 | trace_chunk_name = "none"; | |
4246 | } | |
4247 | ||
4248 | /* | |
4249 | * Consumer trace chunks are | |
4250 | * never anonymous. | |
4251 | */ | |
4252 | chunk_status = lttng_trace_chunk_get_id( | |
4253 | stream->trace_chunk, | |
4254 | &trace_chunk_id); | |
4255 | assert(chunk_status == | |
4256 | LTTNG_TRACE_CHUNK_STATUS_OK); | |
4257 | ||
4258 | DBG("Unable to open packet for stream during trace chunk's lifetime. " | |
4259 | "Flushing an empty packet to prevent an empty file from being created: " | |
4260 | "stream id = %" PRIu64 ", trace chunk name = `%s`, trace chunk id = %" PRIu64, | |
4261 | stream->key, trace_chunk_name, trace_chunk_id); | |
4262 | } | |
4263 | } | |
4264 | } | |
4265 | ||
a9dde553 | 4266 | /* |
c0e2451e JG |
4267 | * Close the current packet before sampling the |
4268 | * ring buffer positions. | |
a9dde553 | 4269 | */ |
c0e2451e | 4270 | ret = consumer_flush_buffer(stream, flush_active); |
a9dde553 MD |
4271 | if (ret < 0) { |
4272 | ERR("Failed to flush stream %" PRIu64 " during channel rotation", | |
4273 | stream->key); | |
4274 | goto end_unlock_stream; | |
4275 | } | |
b99a8d42 JD |
4276 | } |
4277 | ||
a40a503f MD |
4278 | ret = lttng_consumer_take_snapshot(stream); |
4279 | if (ret < 0 && ret != -ENODATA && ret != -EAGAIN) { | |
4280 | ERR("Failed to sample snapshot position during channel rotation"); | |
b99a8d42 JD |
4281 | goto end_unlock_stream; |
4282 | } | |
a40a503f MD |
4283 | if (!ret) { |
4284 | ret = lttng_consumer_get_produced_snapshot(stream, | |
4285 | &produced_pos); | |
4286 | if (ret < 0) { | |
4287 | ERR("Failed to sample produced position during channel rotation"); | |
4288 | goto end_unlock_stream; | |
4289 | } | |
b99a8d42 | 4290 | |
a40a503f MD |
4291 | ret = lttng_consumer_get_consumed_snapshot(stream, |
4292 | &consumed_pos); | |
4293 | if (ret < 0) { | |
4294 | ERR("Failed to sample consumed position during channel rotation"); | |
4295 | goto end_unlock_stream; | |
4296 | } | |
4297 | } | |
4298 | /* | |
4299 | * Align produced position on the start-of-packet boundary of the first | |
4300 | * packet going into the next trace chunk. | |
4301 | */ | |
4302 | produced_pos = ALIGN_FLOOR(produced_pos, stream->max_sb_size); | |
4303 | if (consumed_pos == produced_pos) { | |
f8528c7a MD |
4304 | DBG("Set rotate ready for stream %" PRIu64 " produced = %lu consumed = %lu", |
4305 | stream->key, produced_pos, consumed_pos); | |
b99a8d42 | 4306 | stream->rotate_ready = true; |
f8528c7a MD |
4307 | } else { |
4308 | DBG("Different consumed and produced positions " | |
4309 | "for stream %" PRIu64 " produced = %lu consumed = %lu", | |
4310 | stream->key, produced_pos, consumed_pos); | |
b99a8d42 | 4311 | } |
633d0182 | 4312 | /* |
a40a503f MD |
4313 | * The rotation position is based on the packet_seq_num of the |
4314 | * packet following the last packet that was consumed for this | |
4315 | * stream, incremented by the offset between produced and | |
4316 | * consumed positions. This rotation position is a lower bound | |
4317 | * (inclusive) at which the next trace chunk starts. Since it | |
4318 | * is a lower bound, it is OK if the packet_seq_num does not | |
4319 | * correspond exactly to the same packet identified by the | |
4320 | * consumed_pos, which can happen in overwrite mode. | |
633d0182 | 4321 | */ |
a40a503f MD |
4322 | if (stream->sequence_number_unavailable) { |
4323 | /* | |
4324 | * Rotation should never be performed on a session which | |
4325 | * interacts with a pre-2.8 lttng-modules, which does | |
4326 | * not implement packet sequence number. | |
4327 | */ | |
4328 | ERR("Failure to rotate stream %" PRIu64 ": sequence number unavailable", | |
b99a8d42 | 4329 | stream->key); |
a40a503f | 4330 | ret = -1; |
b99a8d42 JD |
4331 | goto end_unlock_stream; |
4332 | } | |
a40a503f MD |
4333 | stream->rotate_position = stream->last_sequence_number + 1 + |
4334 | ((produced_pos - consumed_pos) / stream->max_sb_size); | |
f8528c7a MD |
4335 | DBG("Set rotation position for stream %" PRIu64 " at position %" PRIu64, |
4336 | stream->key, stream->rotate_position); | |
b99a8d42 | 4337 | |
c35f9726 | 4338 | if (!is_local_trace) { |
633d0182 JG |
4339 | /* |
4340 | * The relay daemon control protocol expects a rotation | |
4341 | * position as "the sequence number of the first packet | |
a40a503f | 4342 | * _after_ the current trace chunk". |
633d0182 | 4343 | */ |
c35f9726 JG |
4344 | const struct relayd_stream_rotation_position position = { |
4345 | .stream_id = stream->relayd_stream_id, | |
a40a503f | 4346 | .rotate_at_seq_num = stream->rotate_position, |
c35f9726 JG |
4347 | }; |
4348 | ||
4349 | ret = lttng_dynamic_array_add_element( | |
4350 | &stream_rotation_positions, | |
4351 | &position); | |
4352 | if (ret) { | |
4353 | ERR("Failed to allocate stream rotation position"); | |
4354 | goto end_unlock_stream; | |
4355 | } | |
4356 | stream_count++; | |
4357 | } | |
fc0c52e9 JG |
4358 | |
4359 | stream->opened_packet_in_current_trace_chunk = false; | |
4360 | ||
4361 | if (rotating_to_new_chunk && !stream->metadata_flag) { | |
4362 | /* | |
4363 | * Attempt to flush an empty packet as close to the | |
4364 | * rotation point as possible. In the event where a | |
4365 | * stream remains inactive after the rotation point, | |
4366 | * this ensures that the new trace chunk has a | |
4367 | * beginning timestamp set at the begining of the | |
4368 | * trace chunk instead of only creating an empty | |
4369 | * packet when the trace chunk is stopped. | |
4370 | * | |
4371 | * This indicates to the viewers that the stream | |
4372 | * was being recorded, but more importantly it | |
4373 | * allows viewers to determine a useable trace | |
4374 | * intersection. | |
4375 | * | |
4376 | * This presents a problem in the case where the | |
4377 | * ring-buffer is completely full. | |
4378 | * | |
4379 | * Consider the following scenario: | |
4380 | * - The consumption of data is slow (slow network, | |
4381 | * for instance), | |
4382 | * - The ring buffer is full, | |
4383 | * - A rotation is initiated, | |
4384 | * - The flush below does nothing (no space left to | |
4385 | * open a new packet), | |
4386 | * - The other streams rotate very soon, and new | |
4387 | * data is produced in the new chunk, | |
4388 | * - This stream completes its rotation long after the | |
4389 | * rotation was initiated | |
4390 | * - The session is stopped before any event can be | |
4391 | * produced in this stream's buffers. | |
4392 | * | |
4393 | * The resulting trace chunk will have a single packet | |
4394 | * temporaly at the end of the trace chunk for this | |
4395 | * stream making the stream intersection more narrow | |
4396 | * than it should be. | |
4397 | * | |
4398 | * To work-around this, an empty flush is performed | |
4399 | * after the first consumption of a packet during a | |
4400 | * rotation if open_packet fails. The idea is that | |
4401 | * consuming a packet frees enough space to switch | |
4402 | * packets in this scenario and allows the tracer to | |
4403 | * "stamp" the beginning of the new trace chunk at the | |
4404 | * earliest possible point. | |
e537f276 JG |
4405 | * |
4406 | * The packet open is performed after the channel | |
4407 | * rotation to ensure that no attempt to open a packet | |
4408 | * is performed in a stream that has no active trace | |
4409 | * chunk. | |
fc0c52e9 | 4410 | */ |
e537f276 JG |
4411 | ret = lttng_dynamic_pointer_array_add_pointer( |
4412 | &streams_packet_to_open, stream); | |
4413 | if (ret) { | |
4414 | PERROR("Failed to add a stream pointer to array of streams in which to open a packet"); | |
fc0c52e9 JG |
4415 | ret = -1; |
4416 | goto end_unlock_stream; | |
fc0c52e9 JG |
4417 | } |
4418 | } | |
4419 | ||
b99a8d42 JD |
4420 | pthread_mutex_unlock(&stream->lock); |
4421 | } | |
c35f9726 | 4422 | stream = NULL; |
b99a8d42 | 4423 | |
e537f276 JG |
4424 | if (!is_local_trace) { |
4425 | relayd = consumer_find_relayd(relayd_id); | |
4426 | if (!relayd) { | |
4427 | ERR("Failed to find relayd %" PRIu64, relayd_id); | |
4428 | ret = -1; | |
4429 | goto end_unlock_channel; | |
4430 | } | |
c35f9726 | 4431 | |
e537f276 JG |
4432 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
4433 | ret = relayd_rotate_streams(&relayd->control_sock, stream_count, | |
4434 | rotating_to_new_chunk ? &next_chunk_id : NULL, | |
4435 | (const struct relayd_stream_rotation_position *) | |
4436 | stream_rotation_positions.buffer | |
4437 | .data); | |
4438 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
4439 | if (ret < 0) { | |
4440 | ERR("Relayd rotate stream failed. Cleaning up relayd %" PRIu64, | |
4441 | relayd->net_seq_idx); | |
4442 | lttng_consumer_cleanup_relayd(relayd); | |
4443 | goto end_unlock_channel; | |
4444 | } | |
c35f9726 JG |
4445 | } |
4446 | ||
e537f276 JG |
4447 | for (stream_idx = 0; |
4448 | stream_idx < lttng_dynamic_pointer_array_get_count( | |
4449 | &streams_packet_to_open); | |
4450 | stream_idx++) { | |
4451 | enum open_packet_status status; | |
4452 | ||
4453 | stream = lttng_dynamic_pointer_array_get_pointer( | |
4454 | &streams_packet_to_open, stream_idx); | |
4455 | ||
4456 | pthread_mutex_lock(&stream->lock); | |
4457 | status = open_packet(stream); | |
4458 | pthread_mutex_unlock(&stream->lock); | |
4459 | switch (status) { | |
4460 | case OPEN_PACKET_STATUS_OPENED: | |
4461 | DBG("Opened a packet after a rotation: stream id = %" PRIu64 | |
4462 | ", channel name = %s, session id = %" PRIu64, | |
4463 | stream->key, stream->chan->name, | |
4464 | stream->chan->session_id); | |
4465 | break; | |
4466 | case OPEN_PACKET_STATUS_NO_SPACE: | |
4467 | /* | |
4468 | * Can't open a packet as there is no space left | |
4469 | * in the buffer. A new packet will be opened | |
4470 | * once one has been consumed. | |
4471 | */ | |
4472 | DBG("No space left to open a packet after a rotation: stream id = %" PRIu64 | |
4473 | ", channel name = %s, session id = %" PRIu64, | |
4474 | stream->key, stream->chan->name, | |
4475 | stream->chan->session_id); | |
4476 | break; | |
4477 | case OPEN_PACKET_STATUS_ERROR: | |
4478 | /* Logged by callee. */ | |
4479 | ret = -1; | |
5e285896 | 4480 | goto end_unlock_channel; |
e537f276 JG |
4481 | default: |
4482 | abort(); | |
4483 | } | |
c35f9726 JG |
4484 | } |
4485 | ||
e537f276 | 4486 | pthread_mutex_unlock(&channel->lock); |
b99a8d42 JD |
4487 | ret = 0; |
4488 | goto end; | |
4489 | ||
4490 | end_unlock_stream: | |
4491 | pthread_mutex_unlock(&stream->lock); | |
c35f9726 | 4492 | end_unlock_channel: |
b99a8d42 JD |
4493 | pthread_mutex_unlock(&channel->lock); |
4494 | end: | |
4495 | rcu_read_unlock(); | |
c35f9726 | 4496 | lttng_dynamic_array_reset(&stream_rotation_positions); |
e537f276 | 4497 | lttng_dynamic_pointer_array_reset(&streams_packet_to_open); |
b99a8d42 JD |
4498 | return ret; |
4499 | } | |
4500 | ||
5f3aff8b MD |
4501 | static |
4502 | int consumer_clear_buffer(struct lttng_consumer_stream *stream) | |
4503 | { | |
4504 | int ret = 0; | |
4505 | unsigned long consumed_pos_before, consumed_pos_after; | |
4506 | ||
4507 | ret = lttng_consumer_sample_snapshot_positions(stream); | |
4508 | if (ret < 0) { | |
4509 | ERR("Taking snapshot positions"); | |
4510 | goto end; | |
4511 | } | |
4512 | ||
4513 | ret = lttng_consumer_get_consumed_snapshot(stream, &consumed_pos_before); | |
4514 | if (ret < 0) { | |
4515 | ERR("Consumed snapshot position"); | |
4516 | goto end; | |
4517 | } | |
4518 | ||
4519 | switch (consumer_data.type) { | |
4520 | case LTTNG_CONSUMER_KERNEL: | |
4521 | ret = kernctl_buffer_clear(stream->wait_fd); | |
4522 | if (ret < 0) { | |
95c4203e | 4523 | ERR("Failed to clear kernel stream (ret = %d)", ret); |
5f3aff8b MD |
4524 | goto end; |
4525 | } | |
4526 | break; | |
4527 | case LTTNG_CONSUMER32_UST: | |
4528 | case LTTNG_CONSUMER64_UST: | |
4529 | lttng_ustconsumer_clear_buffer(stream); | |
4530 | break; | |
4531 | default: | |
4532 | ERR("Unknown consumer_data type"); | |
4533 | abort(); | |
4534 | } | |
4535 | ||
4536 | ret = lttng_consumer_sample_snapshot_positions(stream); | |
4537 | if (ret < 0) { | |
4538 | ERR("Taking snapshot positions"); | |
4539 | goto end; | |
4540 | } | |
4541 | ret = lttng_consumer_get_consumed_snapshot(stream, &consumed_pos_after); | |
4542 | if (ret < 0) { | |
4543 | ERR("Consumed snapshot position"); | |
4544 | goto end; | |
4545 | } | |
4546 | DBG("clear: before: %lu after: %lu", consumed_pos_before, consumed_pos_after); | |
4547 | end: | |
4548 | return ret; | |
4549 | } | |
4550 | ||
4551 | static | |
4552 | int consumer_clear_stream(struct lttng_consumer_stream *stream) | |
4553 | { | |
4554 | int ret; | |
4555 | ||
4556 | ret = consumer_flush_buffer(stream, 1); | |
4557 | if (ret < 0) { | |
4558 | ERR("Failed to flush stream %" PRIu64 " during channel clear", | |
4559 | stream->key); | |
4560 | ret = LTTCOMM_CONSUMERD_FATAL; | |
4561 | goto error; | |
4562 | } | |
4563 | ||
4564 | ret = consumer_clear_buffer(stream); | |
4565 | if (ret < 0) { | |
4566 | ERR("Failed to clear stream %" PRIu64 " during channel clear", | |
4567 | stream->key); | |
4568 | ret = LTTCOMM_CONSUMERD_FATAL; | |
4569 | goto error; | |
4570 | } | |
4571 | ||
4572 | ret = LTTCOMM_CONSUMERD_SUCCESS; | |
4573 | error: | |
4574 | return ret; | |
4575 | } | |
4576 | ||
4577 | static | |
4578 | int consumer_clear_unmonitored_channel(struct lttng_consumer_channel *channel) | |
4579 | { | |
4580 | int ret; | |
4581 | struct lttng_consumer_stream *stream; | |
4582 | ||
4583 | rcu_read_lock(); | |
4584 | pthread_mutex_lock(&channel->lock); | |
4585 | cds_list_for_each_entry(stream, &channel->streams.head, send_node) { | |
4586 | health_code_update(); | |
4587 | pthread_mutex_lock(&stream->lock); | |
4588 | ret = consumer_clear_stream(stream); | |
4589 | if (ret) { | |
4590 | goto error_unlock; | |
4591 | } | |
4592 | pthread_mutex_unlock(&stream->lock); | |
4593 | } | |
4594 | pthread_mutex_unlock(&channel->lock); | |
4595 | rcu_read_unlock(); | |
4596 | return 0; | |
4597 | ||
4598 | error_unlock: | |
4599 | pthread_mutex_unlock(&stream->lock); | |
4600 | pthread_mutex_unlock(&channel->lock); | |
4601 | rcu_read_unlock(); | |
5f3aff8b MD |
4602 | return ret; |
4603 | } | |
4604 | ||
02d02e31 JD |
4605 | /* |
4606 | * Check if a stream is ready to be rotated after extracting it. | |
4607 | * | |
4608 | * Return 1 if it is ready for rotation, 0 if it is not, a negative value on | |
4609 | * error. Stream lock must be held. | |
4610 | */ | |
4611 | int lttng_consumer_stream_is_rotate_ready(struct lttng_consumer_stream *stream) | |
4612 | { | |
f8528c7a MD |
4613 | DBG("Check is rotate ready for stream %" PRIu64 |
4614 | " ready %u rotate_position %" PRIu64 | |
4615 | " last_sequence_number %" PRIu64, | |
4616 | stream->key, stream->rotate_ready, | |
4617 | stream->rotate_position, stream->last_sequence_number); | |
02d02e31 | 4618 | if (stream->rotate_ready) { |
a40a503f | 4619 | return 1; |
02d02e31 JD |
4620 | } |
4621 | ||
4622 | /* | |
a40a503f MD |
4623 | * If packet seq num is unavailable, it means we are interacting |
4624 | * with a pre-2.8 lttng-modules which does not implement the | |
4625 | * sequence number. Rotation should never be used by sessiond in this | |
4626 | * scenario. | |
02d02e31 | 4627 | */ |
a40a503f MD |
4628 | if (stream->sequence_number_unavailable) { |
4629 | ERR("Internal error: rotation used on stream %" PRIu64 | |
4630 | " with unavailable sequence number", | |
4631 | stream->key); | |
4632 | return -1; | |
02d02e31 JD |
4633 | } |
4634 | ||
a40a503f MD |
4635 | if (stream->rotate_position == -1ULL || |
4636 | stream->last_sequence_number == -1ULL) { | |
4637 | return 0; | |
02d02e31 JD |
4638 | } |
4639 | ||
a40a503f MD |
4640 | /* |
4641 | * Rotate position not reached yet. The stream rotate position is | |
4642 | * the position of the next packet belonging to the next trace chunk, | |
4643 | * but consumerd considers rotation ready when reaching the last | |
4644 | * packet of the current chunk, hence the "rotate_position - 1". | |
4645 | */ | |
f8528c7a MD |
4646 | |
4647 | DBG("Check is rotate ready for stream %" PRIu64 | |
4648 | " last_sequence_number %" PRIu64 | |
4649 | " rotate_position %" PRIu64, | |
4650 | stream->key, stream->last_sequence_number, | |
4651 | stream->rotate_position); | |
a40a503f MD |
4652 | if (stream->last_sequence_number >= stream->rotate_position - 1) { |
4653 | return 1; | |
02d02e31 | 4654 | } |
02d02e31 | 4655 | |
a40a503f | 4656 | return 0; |
02d02e31 JD |
4657 | } |
4658 | ||
d73bf3d7 JD |
4659 | /* |
4660 | * Reset the state for a stream after a rotation occurred. | |
4661 | */ | |
4662 | void lttng_consumer_reset_stream_rotate_state(struct lttng_consumer_stream *stream) | |
4663 | { | |
f8528c7a MD |
4664 | DBG("lttng_consumer_reset_stream_rotate_state for stream %" PRIu64, |
4665 | stream->key); | |
a40a503f | 4666 | stream->rotate_position = -1ULL; |
d73bf3d7 JD |
4667 | stream->rotate_ready = false; |
4668 | } | |
4669 | ||
4670 | /* | |
4671 | * Perform the rotation a local stream file. | |
4672 | */ | |
d2956687 | 4673 | static |
d73bf3d7 JD |
4674 | int rotate_local_stream(struct lttng_consumer_local_data *ctx, |
4675 | struct lttng_consumer_stream *stream) | |
4676 | { | |
d2956687 | 4677 | int ret = 0; |
d73bf3d7 | 4678 | |
d2956687 | 4679 | DBG("Rotate local stream: stream key %" PRIu64 ", channel key %" PRIu64, |
d73bf3d7 | 4680 | stream->key, |
d2956687 | 4681 | stream->chan->key); |
d73bf3d7 | 4682 | stream->tracefile_size_current = 0; |
d2956687 | 4683 | stream->tracefile_count_current = 0; |
d73bf3d7 | 4684 | |
d2956687 JG |
4685 | if (stream->out_fd >= 0) { |
4686 | ret = close(stream->out_fd); | |
4687 | if (ret) { | |
4688 | PERROR("Failed to close stream out_fd of channel \"%s\"", | |
4689 | stream->chan->name); | |
4690 | } | |
4691 | stream->out_fd = -1; | |
4692 | } | |
d73bf3d7 | 4693 | |
d2956687 | 4694 | if (stream->index_file) { |
d73bf3d7 | 4695 | lttng_index_file_put(stream->index_file); |
d2956687 | 4696 | stream->index_file = NULL; |
d73bf3d7 JD |
4697 | } |
4698 | ||
d2956687 JG |
4699 | if (!stream->trace_chunk) { |
4700 | goto end; | |
4701 | } | |
d73bf3d7 | 4702 | |
d2956687 | 4703 | ret = consumer_stream_create_output_files(stream, true); |
d73bf3d7 JD |
4704 | end: |
4705 | return ret; | |
d73bf3d7 JD |
4706 | } |
4707 | ||
d73bf3d7 JD |
4708 | /* |
4709 | * Performs the stream rotation for the rotate session feature if needed. | |
d2956687 | 4710 | * It must be called with the channel and stream locks held. |
d73bf3d7 JD |
4711 | * |
4712 | * Return 0 on success, a negative number of error. | |
4713 | */ | |
4714 | int lttng_consumer_rotate_stream(struct lttng_consumer_local_data *ctx, | |
d2956687 | 4715 | struct lttng_consumer_stream *stream) |
d73bf3d7 JD |
4716 | { |
4717 | int ret; | |
4718 | ||
4719 | DBG("Consumer rotate stream %" PRIu64, stream->key); | |
4720 | ||
d2956687 JG |
4721 | /* |
4722 | * Update the stream's 'current' chunk to the session's (channel) | |
4723 | * now-current chunk. | |
4724 | */ | |
4725 | lttng_trace_chunk_put(stream->trace_chunk); | |
4726 | if (stream->chan->trace_chunk == stream->trace_chunk) { | |
4727 | /* | |
4728 | * A channel can be rotated and not have a "next" chunk | |
4729 | * to transition to. In that case, the channel's "current chunk" | |
4730 | * has not been closed yet, but it has not been updated to | |
4731 | * a "next" trace chunk either. Hence, the stream, like its | |
4732 | * parent channel, becomes part of no chunk and can't output | |
4733 | * anything until a new trace chunk is created. | |
4734 | */ | |
4735 | stream->trace_chunk = NULL; | |
4736 | } else if (stream->chan->trace_chunk && | |
4737 | !lttng_trace_chunk_get(stream->chan->trace_chunk)) { | |
4738 | ERR("Failed to acquire a reference to channel's trace chunk during stream rotation"); | |
4739 | ret = -1; | |
4740 | goto error; | |
4741 | } else { | |
4742 | /* | |
4743 | * Update the stream's trace chunk to its parent channel's | |
4744 | * current trace chunk. | |
4745 | */ | |
4746 | stream->trace_chunk = stream->chan->trace_chunk; | |
4747 | } | |
4748 | ||
c35f9726 | 4749 | if (stream->net_seq_idx == (uint64_t) -1ULL) { |
d73bf3d7 | 4750 | ret = rotate_local_stream(ctx, stream); |
c35f9726 JG |
4751 | if (ret < 0) { |
4752 | ERR("Failed to rotate stream, ret = %i", ret); | |
4753 | goto error; | |
4754 | } | |
d73bf3d7 JD |
4755 | } |
4756 | ||
d2956687 JG |
4757 | if (stream->metadata_flag && stream->trace_chunk) { |
4758 | /* | |
4759 | * If the stream has transitioned to a new trace | |
4760 | * chunk, the metadata should be re-dumped to the | |
4761 | * newest chunk. | |
4762 | * | |
4763 | * However, it is possible for a stream to transition to | |
4764 | * a "no-chunk" state. This can happen if a rotation | |
4765 | * occurs on an inactive session. In such cases, the metadata | |
4766 | * regeneration will happen when the next trace chunk is | |
4767 | * created. | |
4768 | */ | |
4769 | ret = consumer_metadata_stream_dump(stream); | |
4770 | if (ret) { | |
4771 | goto error; | |
d73bf3d7 JD |
4772 | } |
4773 | } | |
4774 | lttng_consumer_reset_stream_rotate_state(stream); | |
4775 | ||
4776 | ret = 0; | |
4777 | ||
4778 | error: | |
4779 | return ret; | |
4780 | } | |
4781 | ||
b99a8d42 JD |
4782 | /* |
4783 | * Rotate all the ready streams now. | |
4784 | * | |
4785 | * This is especially important for low throughput streams that have already | |
4786 | * been consumed, we cannot wait for their next packet to perform the | |
4787 | * rotation. | |
92b7a7f8 MD |
4788 | * Need to be called with RCU read-side lock held to ensure existence of |
4789 | * channel. | |
b99a8d42 JD |
4790 | * |
4791 | * Returns 0 on success, < 0 on error | |
4792 | */ | |
92b7a7f8 MD |
4793 | int lttng_consumer_rotate_ready_streams(struct lttng_consumer_channel *channel, |
4794 | uint64_t key, struct lttng_consumer_local_data *ctx) | |
b99a8d42 JD |
4795 | { |
4796 | int ret; | |
b99a8d42 JD |
4797 | struct lttng_consumer_stream *stream; |
4798 | struct lttng_ht_iter iter; | |
4799 | struct lttng_ht *ht = consumer_data.stream_per_chan_id_ht; | |
4800 | ||
4801 | rcu_read_lock(); | |
4802 | ||
4803 | DBG("Consumer rotate ready streams in channel %" PRIu64, key); | |
4804 | ||
b99a8d42 JD |
4805 | cds_lfht_for_each_entry_duplicate(ht->ht, |
4806 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
4807 | ht->match_fct, &channel->key, &iter.iter, | |
4808 | stream, node_channel_id.node) { | |
4809 | health_code_update(); | |
4810 | ||
d2956687 | 4811 | pthread_mutex_lock(&stream->chan->lock); |
b99a8d42 JD |
4812 | pthread_mutex_lock(&stream->lock); |
4813 | ||
4814 | if (!stream->rotate_ready) { | |
4815 | pthread_mutex_unlock(&stream->lock); | |
d2956687 | 4816 | pthread_mutex_unlock(&stream->chan->lock); |
b99a8d42 JD |
4817 | continue; |
4818 | } | |
4819 | DBG("Consumer rotate ready stream %" PRIu64, stream->key); | |
4820 | ||
d2956687 | 4821 | ret = lttng_consumer_rotate_stream(ctx, stream); |
b99a8d42 | 4822 | pthread_mutex_unlock(&stream->lock); |
d2956687 | 4823 | pthread_mutex_unlock(&stream->chan->lock); |
b99a8d42 JD |
4824 | if (ret) { |
4825 | goto end; | |
4826 | } | |
4827 | } | |
4828 | ||
4829 | ret = 0; | |
4830 | ||
4831 | end: | |
4832 | rcu_read_unlock(); | |
4833 | return ret; | |
4834 | } | |
4835 | ||
d2956687 JG |
4836 | enum lttcomm_return_code lttng_consumer_init_command( |
4837 | struct lttng_consumer_local_data *ctx, | |
4838 | const lttng_uuid sessiond_uuid) | |
00fb02ac | 4839 | { |
d2956687 | 4840 | enum lttcomm_return_code ret; |
c70636a7 | 4841 | char uuid_str[LTTNG_UUID_STR_LEN]; |
00fb02ac | 4842 | |
d2956687 JG |
4843 | if (ctx->sessiond_uuid.is_set) { |
4844 | ret = LTTCOMM_CONSUMERD_ALREADY_SET; | |
00fb02ac JD |
4845 | goto end; |
4846 | } | |
4847 | ||
d2956687 JG |
4848 | ctx->sessiond_uuid.is_set = true; |
4849 | memcpy(ctx->sessiond_uuid.value, sessiond_uuid, sizeof(lttng_uuid)); | |
4850 | ret = LTTCOMM_CONSUMERD_SUCCESS; | |
4851 | lttng_uuid_to_str(sessiond_uuid, uuid_str); | |
4852 | DBG("Received session daemon UUID: %s", uuid_str); | |
00fb02ac JD |
4853 | end: |
4854 | return ret; | |
4855 | } | |
4856 | ||
d2956687 JG |
4857 | enum lttcomm_return_code lttng_consumer_create_trace_chunk( |
4858 | const uint64_t *relayd_id, uint64_t session_id, | |
4859 | uint64_t chunk_id, | |
4860 | time_t chunk_creation_timestamp, | |
4861 | const char *chunk_override_name, | |
4862 | const struct lttng_credentials *credentials, | |
4863 | struct lttng_directory_handle *chunk_directory_handle) | |
00fb02ac JD |
4864 | { |
4865 | int ret; | |
d2956687 | 4866 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
7ea24db3 | 4867 | struct lttng_trace_chunk *created_chunk = NULL, *published_chunk = NULL; |
d2956687 JG |
4868 | enum lttng_trace_chunk_status chunk_status; |
4869 | char relayd_id_buffer[MAX_INT_DEC_LEN(*relayd_id)]; | |
4870 | char creation_timestamp_buffer[ISO8601_STR_LEN]; | |
4871 | const char *relayd_id_str = "(none)"; | |
4872 | const char *creation_timestamp_str; | |
4873 | struct lttng_ht_iter iter; | |
4874 | struct lttng_consumer_channel *channel; | |
92816cc3 | 4875 | |
d2956687 JG |
4876 | if (relayd_id) { |
4877 | /* Only used for logging purposes. */ | |
4878 | ret = snprintf(relayd_id_buffer, sizeof(relayd_id_buffer), | |
4879 | "%" PRIu64, *relayd_id); | |
4880 | if (ret > 0 && ret < sizeof(relayd_id_buffer)) { | |
4881 | relayd_id_str = relayd_id_buffer; | |
4882 | } else { | |
4883 | relayd_id_str = "(formatting error)"; | |
4884 | } | |
4885 | } | |
4886 | ||
4887 | /* Local protocol error. */ | |
4888 | assert(chunk_creation_timestamp); | |
4889 | ret = time_to_iso8601_str(chunk_creation_timestamp, | |
4890 | creation_timestamp_buffer, | |
4891 | sizeof(creation_timestamp_buffer)); | |
4892 | creation_timestamp_str = !ret ? creation_timestamp_buffer : | |
4893 | "(formatting error)"; | |
4894 | ||
4895 | DBG("Consumer create trace chunk command: relay_id = %s" | |
4896 | ", session_id = %" PRIu64 ", chunk_id = %" PRIu64 | |
4897 | ", chunk_override_name = %s" | |
4898 | ", chunk_creation_timestamp = %s", | |
4899 | relayd_id_str, session_id, chunk_id, | |
4900 | chunk_override_name ? : "(none)", | |
4901 | creation_timestamp_str); | |
92816cc3 JG |
4902 | |
4903 | /* | |
d2956687 JG |
4904 | * The trace chunk registry, as used by the consumer daemon, implicitly |
4905 | * owns the trace chunks. This is only needed in the consumer since | |
4906 | * the consumer has no notion of a session beyond session IDs being | |
4907 | * used to identify other objects. | |
4908 | * | |
4909 | * The lttng_trace_chunk_registry_publish() call below provides a | |
4910 | * reference which is not released; it implicitly becomes the session | |
4911 | * daemon's reference to the chunk in the consumer daemon. | |
4912 | * | |
4913 | * The lifetime of trace chunks in the consumer daemon is managed by | |
4914 | * the session daemon through the LTTNG_CONSUMER_CREATE_TRACE_CHUNK | |
4915 | * and LTTNG_CONSUMER_DESTROY_TRACE_CHUNK commands. | |
92816cc3 | 4916 | */ |
d2956687 | 4917 | created_chunk = lttng_trace_chunk_create(chunk_id, |
a7ceb342 | 4918 | chunk_creation_timestamp, NULL); |
d2956687 JG |
4919 | if (!created_chunk) { |
4920 | ERR("Failed to create trace chunk"); | |
4921 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; | |
7ea24db3 | 4922 | goto error; |
d2956687 | 4923 | } |
92816cc3 | 4924 | |
d2956687 JG |
4925 | if (chunk_override_name) { |
4926 | chunk_status = lttng_trace_chunk_override_name(created_chunk, | |
4927 | chunk_override_name); | |
4928 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { | |
4929 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; | |
7ea24db3 | 4930 | goto error; |
92816cc3 JG |
4931 | } |
4932 | } | |
4933 | ||
d2956687 JG |
4934 | if (chunk_directory_handle) { |
4935 | chunk_status = lttng_trace_chunk_set_credentials(created_chunk, | |
4936 | credentials); | |
4937 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { | |
4938 | ERR("Failed to set trace chunk credentials"); | |
4939 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; | |
7ea24db3 | 4940 | goto error; |
d2956687 JG |
4941 | } |
4942 | /* | |
4943 | * The consumer daemon has no ownership of the chunk output | |
4944 | * directory. | |
4945 | */ | |
4946 | chunk_status = lttng_trace_chunk_set_as_user(created_chunk, | |
4947 | chunk_directory_handle); | |
cbf53d23 | 4948 | chunk_directory_handle = NULL; |
d2956687 JG |
4949 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { |
4950 | ERR("Failed to set trace chunk's directory handle"); | |
4951 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; | |
7ea24db3 | 4952 | goto error; |
92816cc3 JG |
4953 | } |
4954 | } | |
4955 | ||
d2956687 JG |
4956 | published_chunk = lttng_trace_chunk_registry_publish_chunk( |
4957 | consumer_data.chunk_registry, session_id, | |
4958 | created_chunk); | |
4959 | lttng_trace_chunk_put(created_chunk); | |
4960 | created_chunk = NULL; | |
4961 | if (!published_chunk) { | |
4962 | ERR("Failed to publish trace chunk"); | |
4963 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; | |
7ea24db3 | 4964 | goto error; |
d88744a4 JD |
4965 | } |
4966 | ||
d2956687 JG |
4967 | rcu_read_lock(); |
4968 | cds_lfht_for_each_entry_duplicate(consumer_data.channels_by_session_id_ht->ht, | |
4969 | consumer_data.channels_by_session_id_ht->hash_fct( | |
4970 | &session_id, lttng_ht_seed), | |
4971 | consumer_data.channels_by_session_id_ht->match_fct, | |
4972 | &session_id, &iter.iter, channel, | |
4973 | channels_by_session_id_ht_node.node) { | |
4974 | ret = lttng_consumer_channel_set_trace_chunk(channel, | |
4975 | published_chunk); | |
4976 | if (ret) { | |
4977 | /* | |
4978 | * Roll-back the creation of this chunk. | |
4979 | * | |
4980 | * This is important since the session daemon will | |
4981 | * assume that the creation of this chunk failed and | |
4982 | * will never ask for it to be closed, resulting | |
4983 | * in a leak and an inconsistent state for some | |
4984 | * channels. | |
4985 | */ | |
4986 | enum lttcomm_return_code close_ret; | |
ecd1a12f | 4987 | char path[LTTNG_PATH_MAX]; |
d2956687 JG |
4988 | |
4989 | DBG("Failed to set new trace chunk on existing channels, rolling back"); | |
4990 | close_ret = lttng_consumer_close_trace_chunk(relayd_id, | |
4991 | session_id, chunk_id, | |
ecd1a12f MD |
4992 | chunk_creation_timestamp, NULL, |
4993 | path); | |
d2956687 JG |
4994 | if (close_ret != LTTCOMM_CONSUMERD_SUCCESS) { |
4995 | ERR("Failed to roll-back the creation of new chunk: session_id = %" PRIu64 ", chunk_id = %" PRIu64, | |
4996 | session_id, chunk_id); | |
4997 | } | |
a1ae2ea5 | 4998 | |
d2956687 JG |
4999 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; |
5000 | break; | |
5001 | } | |
a1ae2ea5 JD |
5002 | } |
5003 | ||
e5add6d0 JG |
5004 | if (relayd_id) { |
5005 | struct consumer_relayd_sock_pair *relayd; | |
5006 | ||
5007 | relayd = consumer_find_relayd(*relayd_id); | |
5008 | if (relayd) { | |
5009 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
5010 | ret = relayd_create_trace_chunk( | |
5011 | &relayd->control_sock, published_chunk); | |
5012 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
5013 | } else { | |
5014 | ERR("Failed to find relay daemon socket: relayd_id = %" PRIu64, *relayd_id); | |
5015 | } | |
5016 | ||
5017 | if (!relayd || ret) { | |
5018 | enum lttcomm_return_code close_ret; | |
ecd1a12f | 5019 | char path[LTTNG_PATH_MAX]; |
e5add6d0 JG |
5020 | |
5021 | close_ret = lttng_consumer_close_trace_chunk(relayd_id, | |
5022 | session_id, | |
5023 | chunk_id, | |
bbc4768c | 5024 | chunk_creation_timestamp, |
ecd1a12f | 5025 | NULL, path); |
e5add6d0 JG |
5026 | if (close_ret != LTTCOMM_CONSUMERD_SUCCESS) { |
5027 | ERR("Failed to roll-back the creation of new chunk: session_id = %" PRIu64 ", chunk_id = %" PRIu64, | |
5028 | session_id, | |
5029 | chunk_id); | |
5030 | } | |
5031 | ||
5032 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; | |
7ea24db3 | 5033 | goto error_unlock; |
e5add6d0 JG |
5034 | } |
5035 | } | |
7ea24db3 | 5036 | error_unlock: |
e5add6d0 | 5037 | rcu_read_unlock(); |
7ea24db3 | 5038 | error: |
d2956687 JG |
5039 | /* Release the reference returned by the "publish" operation. */ |
5040 | lttng_trace_chunk_put(published_chunk); | |
9bb5f1f8 | 5041 | lttng_trace_chunk_put(created_chunk); |
d2956687 | 5042 | return ret_code; |
a1ae2ea5 JD |
5043 | } |
5044 | ||
d2956687 JG |
5045 | enum lttcomm_return_code lttng_consumer_close_trace_chunk( |
5046 | const uint64_t *relayd_id, uint64_t session_id, | |
bbc4768c | 5047 | uint64_t chunk_id, time_t chunk_close_timestamp, |
ecd1a12f MD |
5048 | const enum lttng_trace_chunk_command_type *close_command, |
5049 | char *path) | |
a1ae2ea5 | 5050 | { |
d2956687 JG |
5051 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
5052 | struct lttng_trace_chunk *chunk; | |
5053 | char relayd_id_buffer[MAX_INT_DEC_LEN(*relayd_id)]; | |
5054 | const char *relayd_id_str = "(none)"; | |
bbc4768c | 5055 | const char *close_command_name = "none"; |
d2956687 JG |
5056 | struct lttng_ht_iter iter; |
5057 | struct lttng_consumer_channel *channel; | |
5058 | enum lttng_trace_chunk_status chunk_status; | |
a1ae2ea5 | 5059 | |
d2956687 JG |
5060 | if (relayd_id) { |
5061 | int ret; | |
5062 | ||
5063 | /* Only used for logging purposes. */ | |
5064 | ret = snprintf(relayd_id_buffer, sizeof(relayd_id_buffer), | |
5065 | "%" PRIu64, *relayd_id); | |
5066 | if (ret > 0 && ret < sizeof(relayd_id_buffer)) { | |
5067 | relayd_id_str = relayd_id_buffer; | |
5068 | } else { | |
5069 | relayd_id_str = "(formatting error)"; | |
5070 | } | |
bbc4768c JG |
5071 | } |
5072 | if (close_command) { | |
5073 | close_command_name = lttng_trace_chunk_command_type_get_name( | |
5074 | *close_command); | |
5075 | } | |
d2956687 JG |
5076 | |
5077 | DBG("Consumer close trace chunk command: relayd_id = %s" | |
bbc4768c JG |
5078 | ", session_id = %" PRIu64 ", chunk_id = %" PRIu64 |
5079 | ", close command = %s", | |
5080 | relayd_id_str, session_id, chunk_id, | |
5081 | close_command_name); | |
5082 | ||
d2956687 | 5083 | chunk = lttng_trace_chunk_registry_find_chunk( |
bbc4768c JG |
5084 | consumer_data.chunk_registry, session_id, chunk_id); |
5085 | if (!chunk) { | |
d2956687 JG |
5086 | ERR("Failed to find chunk: session_id = %" PRIu64 |
5087 | ", chunk_id = %" PRIu64, | |
5088 | session_id, chunk_id); | |
5089 | ret_code = LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK; | |
a1ae2ea5 JD |
5090 | goto end; |
5091 | } | |
5092 | ||
d2956687 JG |
5093 | chunk_status = lttng_trace_chunk_set_close_timestamp(chunk, |
5094 | chunk_close_timestamp); | |
5095 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { | |
5096 | ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED; | |
5097 | goto end; | |
45f1d9a1 | 5098 | } |
bbc4768c JG |
5099 | |
5100 | if (close_command) { | |
5101 | chunk_status = lttng_trace_chunk_set_close_command( | |
5102 | chunk, *close_command); | |
5103 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { | |
5104 | ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED; | |
5105 | goto end; | |
5106 | } | |
5107 | } | |
a1ae2ea5 | 5108 | |
d2956687 JG |
5109 | /* |
5110 | * chunk is now invalid to access as we no longer hold a reference to | |
5111 | * it; it is only kept around to compare it (by address) to the | |
5112 | * current chunk found in the session's channels. | |
5113 | */ | |
5114 | rcu_read_lock(); | |
5115 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, | |
5116 | channel, node.node) { | |
5117 | int ret; | |
a1ae2ea5 | 5118 | |
d2956687 JG |
5119 | /* |
5120 | * Only change the channel's chunk to NULL if it still | |
5121 | * references the chunk being closed. The channel may | |
5122 | * reference a newer channel in the case of a session | |
5123 | * rotation. When a session rotation occurs, the "next" | |
5124 | * chunk is created before the "current" chunk is closed. | |
5125 | */ | |
5126 | if (channel->trace_chunk != chunk) { | |
5127 | continue; | |
5128 | } | |
5129 | ret = lttng_consumer_channel_set_trace_chunk(channel, NULL); | |
5130 | if (ret) { | |
5131 | /* | |
5132 | * Attempt to close the chunk on as many channels as | |
5133 | * possible. | |
5134 | */ | |
5135 | ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED; | |
5136 | } | |
a1ae2ea5 | 5137 | } |
bbc4768c JG |
5138 | |
5139 | if (relayd_id) { | |
5140 | int ret; | |
5141 | struct consumer_relayd_sock_pair *relayd; | |
5142 | ||
5143 | relayd = consumer_find_relayd(*relayd_id); | |
5144 | if (relayd) { | |
5145 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
5146 | ret = relayd_close_trace_chunk( | |
ecd1a12f MD |
5147 | &relayd->control_sock, chunk, |
5148 | path); | |
bbc4768c JG |
5149 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
5150 | } else { | |
5151 | ERR("Failed to find relay daemon socket: relayd_id = %" PRIu64, | |
5152 | *relayd_id); | |
5153 | } | |
5154 | ||
5155 | if (!relayd || ret) { | |
5156 | ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED; | |
5157 | goto error_unlock; | |
5158 | } | |
5159 | } | |
5160 | error_unlock: | |
d2956687 JG |
5161 | rcu_read_unlock(); |
5162 | end: | |
bbc4768c JG |
5163 | /* |
5164 | * Release the reference returned by the "find" operation and | |
5165 | * the session daemon's implicit reference to the chunk. | |
5166 | */ | |
5167 | lttng_trace_chunk_put(chunk); | |
5168 | lttng_trace_chunk_put(chunk); | |
5169 | ||
d2956687 | 5170 | return ret_code; |
a1ae2ea5 | 5171 | } |
3654ed19 | 5172 | |
d2956687 JG |
5173 | enum lttcomm_return_code lttng_consumer_trace_chunk_exists( |
5174 | const uint64_t *relayd_id, uint64_t session_id, | |
5175 | uint64_t chunk_id) | |
3654ed19 | 5176 | { |
c35f9726 | 5177 | int ret; |
d2956687 | 5178 | enum lttcomm_return_code ret_code; |
d2956687 JG |
5179 | char relayd_id_buffer[MAX_INT_DEC_LEN(*relayd_id)]; |
5180 | const char *relayd_id_str = "(none)"; | |
c35f9726 JG |
5181 | const bool is_local_trace = !relayd_id; |
5182 | struct consumer_relayd_sock_pair *relayd = NULL; | |
6b584c2e | 5183 | bool chunk_exists_local, chunk_exists_remote; |
d2956687 JG |
5184 | |
5185 | if (relayd_id) { | |
5186 | int ret; | |
5187 | ||
5188 | /* Only used for logging purposes. */ | |
5189 | ret = snprintf(relayd_id_buffer, sizeof(relayd_id_buffer), | |
5190 | "%" PRIu64, *relayd_id); | |
5191 | if (ret > 0 && ret < sizeof(relayd_id_buffer)) { | |
5192 | relayd_id_str = relayd_id_buffer; | |
5193 | } else { | |
5194 | relayd_id_str = "(formatting error)"; | |
5195 | } | |
5196 | } | |
5197 | ||
5198 | DBG("Consumer trace chunk exists command: relayd_id = %s" | |
d2956687 | 5199 | ", chunk_id = %" PRIu64, relayd_id_str, |
c35f9726 | 5200 | chunk_id); |
6b584c2e | 5201 | ret = lttng_trace_chunk_registry_chunk_exists( |
d2956687 | 5202 | consumer_data.chunk_registry, session_id, |
6b584c2e JG |
5203 | chunk_id, &chunk_exists_local); |
5204 | if (ret) { | |
5205 | /* Internal error. */ | |
5206 | ERR("Failed to query the existence of a trace chunk"); | |
5207 | ret_code = LTTCOMM_CONSUMERD_FATAL; | |
13e3b280 | 5208 | goto end; |
6b584c2e JG |
5209 | } |
5210 | DBG("Trace chunk %s locally", | |
5211 | chunk_exists_local ? "exists" : "does not exist"); | |
5212 | if (chunk_exists_local) { | |
c35f9726 | 5213 | ret_code = LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_LOCAL; |
c35f9726 JG |
5214 | goto end; |
5215 | } else if (is_local_trace) { | |
5216 | ret_code = LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK; | |
5217 | goto end; | |
5218 | } | |
5219 | ||
5220 | rcu_read_lock(); | |
5221 | relayd = consumer_find_relayd(*relayd_id); | |
5222 | if (!relayd) { | |
5223 | ERR("Failed to find relayd %" PRIu64, *relayd_id); | |
5224 | ret_code = LTTCOMM_CONSUMERD_INVALID_PARAMETERS; | |
5225 | goto end_rcu_unlock; | |
5226 | } | |
5227 | DBG("Looking up existence of trace chunk on relay daemon"); | |
5228 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
5229 | ret = relayd_trace_chunk_exists(&relayd->control_sock, chunk_id, | |
5230 | &chunk_exists_remote); | |
5231 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
5232 | if (ret < 0) { | |
5233 | ERR("Failed to look-up the existence of trace chunk on relay daemon"); | |
5234 | ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL; | |
5235 | goto end_rcu_unlock; | |
5236 | } | |
5237 | ||
5238 | ret_code = chunk_exists_remote ? | |
5239 | LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_REMOTE : | |
d2956687 | 5240 | LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK; |
c35f9726 JG |
5241 | DBG("Trace chunk %s on relay daemon", |
5242 | chunk_exists_remote ? "exists" : "does not exist"); | |
d2956687 | 5243 | |
c35f9726 JG |
5244 | end_rcu_unlock: |
5245 | rcu_read_unlock(); | |
5246 | end: | |
d2956687 | 5247 | return ret_code; |
3654ed19 | 5248 | } |
5f3aff8b MD |
5249 | |
5250 | static | |
5251 | int consumer_clear_monitored_channel(struct lttng_consumer_channel *channel) | |
5252 | { | |
5253 | struct lttng_ht *ht; | |
5254 | struct lttng_consumer_stream *stream; | |
5255 | struct lttng_ht_iter iter; | |
5256 | int ret; | |
5257 | ||
5258 | ht = consumer_data.stream_per_chan_id_ht; | |
5259 | ||
5260 | rcu_read_lock(); | |
5261 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
5262 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
5263 | ht->match_fct, &channel->key, | |
5264 | &iter.iter, stream, node_channel_id.node) { | |
5265 | /* | |
5266 | * Protect against teardown with mutex. | |
5267 | */ | |
5268 | pthread_mutex_lock(&stream->lock); | |
5269 | if (cds_lfht_is_node_deleted(&stream->node.node)) { | |
5270 | goto next; | |
5271 | } | |
5272 | ret = consumer_clear_stream(stream); | |
5273 | if (ret) { | |
5274 | goto error_unlock; | |
5275 | } | |
5276 | next: | |
5277 | pthread_mutex_unlock(&stream->lock); | |
5278 | } | |
5279 | rcu_read_unlock(); | |
5280 | return LTTCOMM_CONSUMERD_SUCCESS; | |
5281 | ||
5282 | error_unlock: | |
5283 | pthread_mutex_unlock(&stream->lock); | |
5284 | rcu_read_unlock(); | |
5285 | return ret; | |
5286 | } | |
5287 | ||
5288 | int lttng_consumer_clear_channel(struct lttng_consumer_channel *channel) | |
5289 | { | |
5290 | int ret; | |
5291 | ||
5292 | DBG("Consumer clear channel %" PRIu64, channel->key); | |
5293 | ||
5294 | if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) { | |
5295 | /* | |
5296 | * Nothing to do for the metadata channel/stream. | |
5297 | * Snapshot mechanism already take care of the metadata | |
5298 | * handling/generation, and monitored channels only need to | |
5299 | * have their data stream cleared.. | |
5300 | */ | |
5301 | ret = LTTCOMM_CONSUMERD_SUCCESS; | |
5302 | goto end; | |
5303 | } | |
5304 | ||
5305 | if (!channel->monitor) { | |
5306 | ret = consumer_clear_unmonitored_channel(channel); | |
5307 | } else { | |
5308 | ret = consumer_clear_monitored_channel(channel); | |
5309 | } | |
5310 | end: | |
5311 | return ret; | |
5312 | } | |
1223361a JG |
5313 | |
5314 | enum lttcomm_return_code lttng_consumer_open_channel_packets( | |
5315 | struct lttng_consumer_channel *channel) | |
5316 | { | |
5317 | struct lttng_consumer_stream *stream; | |
5318 | enum lttcomm_return_code ret = LTTCOMM_CONSUMERD_SUCCESS; | |
5319 | ||
5320 | if (channel->metadata_stream) { | |
5321 | ERR("Open channel packets command attempted on a metadata channel"); | |
5322 | ret = LTTCOMM_CONSUMERD_INVALID_PARAMETERS; | |
5323 | goto end; | |
5324 | } | |
5325 | ||
5326 | rcu_read_lock(); | |
5327 | cds_list_for_each_entry(stream, &channel->streams.head, send_node) { | |
5328 | enum open_packet_status status; | |
5329 | ||
5330 | pthread_mutex_lock(&stream->lock); | |
5331 | if (cds_lfht_is_node_deleted(&stream->node.node)) { | |
5332 | goto next; | |
5333 | } | |
5334 | ||
5335 | status = open_packet(stream); | |
5336 | switch (status) { | |
5337 | case OPEN_PACKET_STATUS_OPENED: | |
5338 | DBG("Opened a packet in \"open channel packets\" command: stream id = %" PRIu64 | |
5339 | ", channel name = %s, session id = %" PRIu64, | |
5340 | stream->key, stream->chan->name, | |
5341 | stream->chan->session_id); | |
5342 | stream->opened_packet_in_current_trace_chunk = true; | |
5343 | break; | |
5344 | case OPEN_PACKET_STATUS_NO_SPACE: | |
5345 | DBG("No space left to open a packet in \"open channel packets\" command: stream id = %" PRIu64 | |
5346 | ", channel name = %s, session id = %" PRIu64, | |
5347 | stream->key, stream->chan->name, | |
5348 | stream->chan->session_id); | |
5349 | break; | |
5350 | case OPEN_PACKET_STATUS_ERROR: | |
5351 | /* | |
5352 | * Only unexpected internal errors can lead to this | |
5353 | * failing. Report an unknown error. | |
5354 | */ | |
5355 | ERR("Failed to flush empty buffer in \"open channel packets\" command: stream id = %" PRIu64 | |
5356 | ", channel id = %" PRIu64 | |
5357 | ", channel name = %s" | |
5358 | ", session id = %" PRIu64, | |
5359 | stream->key, channel->key, | |
5360 | channel->name, channel->session_id); | |
5361 | ret = LTTCOMM_CONSUMERD_UNKNOWN_ERROR; | |
5362 | goto error_unlock; | |
5363 | default: | |
5364 | abort(); | |
5365 | } | |
5366 | ||
5367 | next: | |
5368 | pthread_mutex_unlock(&stream->lock); | |
5369 | } | |
5370 | ||
5371 | end_rcu_unlock: | |
5372 | rcu_read_unlock(); | |
5373 | end: | |
5374 | return ret; | |
5375 | ||
5376 | error_unlock: | |
5377 | pthread_mutex_unlock(&stream->lock); | |
5378 | goto end_rcu_unlock; | |
5379 | } |