Commit | Line | Data |
---|---|---|
3bd1e081 MD |
1 | /* |
2 | * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca> | |
3 | * Mathieu Desnoyers <mathieu.desnoyers@efficios.com> | |
00e2e675 | 4 | * 2012 - David Goulet <dgoulet@efficios.com> |
3bd1e081 | 5 | * |
d14d33bf AM |
6 | * This program is free software; you can redistribute it and/or modify |
7 | * it under the terms of the GNU General Public License, version 2 only, | |
8 | * as published by the Free Software Foundation. | |
3bd1e081 | 9 | * |
d14d33bf AM |
10 | * This program is distributed in the hope that it will be useful, but WITHOUT |
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
13 | * more details. | |
3bd1e081 | 14 | * |
d14d33bf AM |
15 | * You should have received a copy of the GNU General Public License along |
16 | * with this program; if not, write to the Free Software Foundation, Inc., | |
17 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. | |
3bd1e081 MD |
18 | */ |
19 | ||
6c1c0768 | 20 | #define _LGPL_SOURCE |
3bd1e081 | 21 | #include <assert.h> |
3bd1e081 MD |
22 | #include <poll.h> |
23 | #include <pthread.h> | |
24 | #include <stdlib.h> | |
25 | #include <string.h> | |
26 | #include <sys/mman.h> | |
27 | #include <sys/socket.h> | |
28 | #include <sys/types.h> | |
29 | #include <unistd.h> | |
77c7c900 | 30 | #include <inttypes.h> |
331744e3 | 31 | #include <signal.h> |
3bd1e081 | 32 | |
51a9e1c7 | 33 | #include <bin/lttng-consumerd/health-consumerd.h> |
990570ed | 34 | #include <common/common.h> |
fb3a43a9 DG |
35 | #include <common/utils.h> |
36 | #include <common/compat/poll.h> | |
f263b7fd | 37 | #include <common/compat/endian.h> |
309167d2 | 38 | #include <common/index/index.h> |
10a8a223 | 39 | #include <common/kernel-ctl/kernel-ctl.h> |
00e2e675 | 40 | #include <common/sessiond-comm/relayd.h> |
10a8a223 DG |
41 | #include <common/sessiond-comm/sessiond-comm.h> |
42 | #include <common/kernel-consumer/kernel-consumer.h> | |
00e2e675 | 43 | #include <common/relayd/relayd.h> |
10a8a223 | 44 | #include <common/ust-consumer/ust-consumer.h> |
c8fea79c JR |
45 | #include <common/consumer/consumer-timer.h> |
46 | #include <common/consumer/consumer.h> | |
47 | #include <common/consumer/consumer-stream.h> | |
48 | #include <common/consumer/consumer-testpoint.h> | |
49 | #include <common/align.h> | |
3bd1e081 MD |
50 | |
51 | struct lttng_consumer_global_data consumer_data = { | |
3bd1e081 MD |
52 | .stream_count = 0, |
53 | .need_update = 1, | |
54 | .type = LTTNG_CONSUMER_UNKNOWN, | |
55 | }; | |
56 | ||
d8ef542d MD |
57 | enum consumer_channel_action { |
58 | CONSUMER_CHANNEL_ADD, | |
a0cbdd2e | 59 | CONSUMER_CHANNEL_DEL, |
d8ef542d MD |
60 | CONSUMER_CHANNEL_QUIT, |
61 | }; | |
62 | ||
63 | struct consumer_channel_msg { | |
64 | enum consumer_channel_action action; | |
a0cbdd2e MD |
65 | struct lttng_consumer_channel *chan; /* add */ |
66 | uint64_t key; /* del */ | |
d8ef542d MD |
67 | }; |
68 | ||
3bd1e081 MD |
69 | /* |
70 | * Flag to inform the polling thread to quit when all fd hung up. Updated by | |
71 | * the consumer_thread_receive_fds when it notices that all fds has hung up. | |
72 | * Also updated by the signal handler (consumer_should_exit()). Read by the | |
73 | * polling threads. | |
74 | */ | |
a98dae5f | 75 | volatile int consumer_quit; |
3bd1e081 | 76 | |
43c34bc3 | 77 | /* |
43c34bc3 DG |
78 | * Global hash table containing respectively metadata and data streams. The |
79 | * stream element in this ht should only be updated by the metadata poll thread | |
80 | * for the metadata and the data poll thread for the data. | |
81 | */ | |
40dc48e0 DG |
82 | static struct lttng_ht *metadata_ht; |
83 | static struct lttng_ht *data_ht; | |
43c34bc3 | 84 | |
acdb9057 DG |
85 | /* |
86 | * Notify a thread lttng pipe to poll back again. This usually means that some | |
87 | * global state has changed so we just send back the thread in a poll wait | |
88 | * call. | |
89 | */ | |
90 | static void notify_thread_lttng_pipe(struct lttng_pipe *pipe) | |
91 | { | |
92 | struct lttng_consumer_stream *null_stream = NULL; | |
93 | ||
94 | assert(pipe); | |
95 | ||
96 | (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream)); | |
97 | } | |
98 | ||
5c635c72 MD |
99 | static void notify_health_quit_pipe(int *pipe) |
100 | { | |
6cd525e8 | 101 | ssize_t ret; |
5c635c72 | 102 | |
6cd525e8 MD |
103 | ret = lttng_write(pipe[1], "4", 1); |
104 | if (ret < 1) { | |
5c635c72 MD |
105 | PERROR("write consumer health quit"); |
106 | } | |
107 | } | |
108 | ||
d8ef542d MD |
109 | static void notify_channel_pipe(struct lttng_consumer_local_data *ctx, |
110 | struct lttng_consumer_channel *chan, | |
a0cbdd2e | 111 | uint64_t key, |
d8ef542d MD |
112 | enum consumer_channel_action action) |
113 | { | |
114 | struct consumer_channel_msg msg; | |
6cd525e8 | 115 | ssize_t ret; |
d8ef542d | 116 | |
e56251fc DG |
117 | memset(&msg, 0, sizeof(msg)); |
118 | ||
d8ef542d MD |
119 | msg.action = action; |
120 | msg.chan = chan; | |
f21dae48 | 121 | msg.key = key; |
6cd525e8 MD |
122 | ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg)); |
123 | if (ret < sizeof(msg)) { | |
124 | PERROR("notify_channel_pipe write error"); | |
125 | } | |
d8ef542d MD |
126 | } |
127 | ||
a0cbdd2e MD |
128 | void notify_thread_del_channel(struct lttng_consumer_local_data *ctx, |
129 | uint64_t key) | |
130 | { | |
131 | notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL); | |
132 | } | |
133 | ||
d8ef542d MD |
134 | static int read_channel_pipe(struct lttng_consumer_local_data *ctx, |
135 | struct lttng_consumer_channel **chan, | |
a0cbdd2e | 136 | uint64_t *key, |
d8ef542d MD |
137 | enum consumer_channel_action *action) |
138 | { | |
139 | struct consumer_channel_msg msg; | |
6cd525e8 | 140 | ssize_t ret; |
d8ef542d | 141 | |
6cd525e8 MD |
142 | ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg)); |
143 | if (ret < sizeof(msg)) { | |
144 | ret = -1; | |
145 | goto error; | |
d8ef542d | 146 | } |
6cd525e8 MD |
147 | *action = msg.action; |
148 | *chan = msg.chan; | |
149 | *key = msg.key; | |
150 | error: | |
151 | return (int) ret; | |
d8ef542d MD |
152 | } |
153 | ||
212d67a2 DG |
154 | /* |
155 | * Cleanup the stream list of a channel. Those streams are not yet globally | |
156 | * visible | |
157 | */ | |
158 | static void clean_channel_stream_list(struct lttng_consumer_channel *channel) | |
159 | { | |
160 | struct lttng_consumer_stream *stream, *stmp; | |
161 | ||
162 | assert(channel); | |
163 | ||
164 | /* Delete streams that might have been left in the stream list. */ | |
165 | cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head, | |
166 | send_node) { | |
167 | cds_list_del(&stream->send_node); | |
168 | /* | |
169 | * Once a stream is added to this list, the buffers were created so we | |
170 | * have a guarantee that this call will succeed. Setting the monitor | |
171 | * mode to 0 so we don't lock nor try to delete the stream from the | |
172 | * global hash table. | |
173 | */ | |
174 | stream->monitor = 0; | |
175 | consumer_stream_destroy(stream, NULL); | |
176 | } | |
177 | } | |
178 | ||
3bd1e081 MD |
179 | /* |
180 | * Find a stream. The consumer_data.lock must be locked during this | |
181 | * call. | |
182 | */ | |
d88aee68 | 183 | static struct lttng_consumer_stream *find_stream(uint64_t key, |
8389e4f8 | 184 | struct lttng_ht *ht) |
3bd1e081 | 185 | { |
e4421fec | 186 | struct lttng_ht_iter iter; |
d88aee68 | 187 | struct lttng_ht_node_u64 *node; |
e4421fec | 188 | struct lttng_consumer_stream *stream = NULL; |
3bd1e081 | 189 | |
8389e4f8 DG |
190 | assert(ht); |
191 | ||
d88aee68 DG |
192 | /* -1ULL keys are lookup failures */ |
193 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 194 | return NULL; |
7a57cf92 | 195 | } |
e4421fec | 196 | |
6065ceec DG |
197 | rcu_read_lock(); |
198 | ||
d88aee68 DG |
199 | lttng_ht_lookup(ht, &key, &iter); |
200 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
201 | if (node != NULL) { |
202 | stream = caa_container_of(node, struct lttng_consumer_stream, node); | |
3bd1e081 | 203 | } |
e4421fec | 204 | |
6065ceec DG |
205 | rcu_read_unlock(); |
206 | ||
e4421fec | 207 | return stream; |
3bd1e081 MD |
208 | } |
209 | ||
da009f2c | 210 | static void steal_stream_key(uint64_t key, struct lttng_ht *ht) |
7ad0a0cb MD |
211 | { |
212 | struct lttng_consumer_stream *stream; | |
213 | ||
04253271 | 214 | rcu_read_lock(); |
ffe60014 | 215 | stream = find_stream(key, ht); |
04253271 | 216 | if (stream) { |
da009f2c | 217 | stream->key = (uint64_t) -1ULL; |
04253271 MD |
218 | /* |
219 | * We don't want the lookup to match, but we still need | |
220 | * to iterate on this stream when iterating over the hash table. Just | |
221 | * change the node key. | |
222 | */ | |
da009f2c | 223 | stream->node.key = (uint64_t) -1ULL; |
04253271 MD |
224 | } |
225 | rcu_read_unlock(); | |
7ad0a0cb MD |
226 | } |
227 | ||
d56db448 DG |
228 | /* |
229 | * Return a channel object for the given key. | |
230 | * | |
231 | * RCU read side lock MUST be acquired before calling this function and | |
232 | * protects the channel ptr. | |
233 | */ | |
d88aee68 | 234 | struct lttng_consumer_channel *consumer_find_channel(uint64_t key) |
3bd1e081 | 235 | { |
e4421fec | 236 | struct lttng_ht_iter iter; |
d88aee68 | 237 | struct lttng_ht_node_u64 *node; |
e4421fec | 238 | struct lttng_consumer_channel *channel = NULL; |
3bd1e081 | 239 | |
d88aee68 DG |
240 | /* -1ULL keys are lookup failures */ |
241 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 242 | return NULL; |
7a57cf92 | 243 | } |
e4421fec | 244 | |
d88aee68 DG |
245 | lttng_ht_lookup(consumer_data.channel_ht, &key, &iter); |
246 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
247 | if (node != NULL) { |
248 | channel = caa_container_of(node, struct lttng_consumer_channel, node); | |
3bd1e081 | 249 | } |
e4421fec DG |
250 | |
251 | return channel; | |
3bd1e081 MD |
252 | } |
253 | ||
b5a6470f DG |
254 | /* |
255 | * There is a possibility that the consumer does not have enough time between | |
256 | * the close of the channel on the session daemon and the cleanup in here thus | |
257 | * once we have a channel add with an existing key, we know for sure that this | |
258 | * channel will eventually get cleaned up by all streams being closed. | |
259 | * | |
260 | * This function just nullifies the already existing channel key. | |
261 | */ | |
262 | static void steal_channel_key(uint64_t key) | |
263 | { | |
264 | struct lttng_consumer_channel *channel; | |
265 | ||
266 | rcu_read_lock(); | |
267 | channel = consumer_find_channel(key); | |
268 | if (channel) { | |
269 | channel->key = (uint64_t) -1ULL; | |
270 | /* | |
271 | * We don't want the lookup to match, but we still need to iterate on | |
272 | * this channel when iterating over the hash table. Just change the | |
273 | * node key. | |
274 | */ | |
275 | channel->node.key = (uint64_t) -1ULL; | |
276 | } | |
277 | rcu_read_unlock(); | |
278 | } | |
279 | ||
ffe60014 | 280 | static void free_channel_rcu(struct rcu_head *head) |
702b1ea4 | 281 | { |
d88aee68 DG |
282 | struct lttng_ht_node_u64 *node = |
283 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
ffe60014 DG |
284 | struct lttng_consumer_channel *channel = |
285 | caa_container_of(node, struct lttng_consumer_channel, node); | |
702b1ea4 | 286 | |
b83e03c4 MD |
287 | switch (consumer_data.type) { |
288 | case LTTNG_CONSUMER_KERNEL: | |
289 | break; | |
290 | case LTTNG_CONSUMER32_UST: | |
291 | case LTTNG_CONSUMER64_UST: | |
292 | lttng_ustconsumer_free_channel(channel); | |
293 | break; | |
294 | default: | |
295 | ERR("Unknown consumer_data type"); | |
296 | abort(); | |
297 | } | |
ffe60014 | 298 | free(channel); |
702b1ea4 MD |
299 | } |
300 | ||
00e2e675 DG |
301 | /* |
302 | * RCU protected relayd socket pair free. | |
303 | */ | |
ffe60014 | 304 | static void free_relayd_rcu(struct rcu_head *head) |
00e2e675 | 305 | { |
d88aee68 DG |
306 | struct lttng_ht_node_u64 *node = |
307 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
00e2e675 DG |
308 | struct consumer_relayd_sock_pair *relayd = |
309 | caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
310 | ||
8994307f DG |
311 | /* |
312 | * Close all sockets. This is done in the call RCU since we don't want the | |
313 | * socket fds to be reassigned thus potentially creating bad state of the | |
314 | * relayd object. | |
315 | * | |
316 | * We do not have to lock the control socket mutex here since at this stage | |
317 | * there is no one referencing to this relayd object. | |
318 | */ | |
319 | (void) relayd_close(&relayd->control_sock); | |
320 | (void) relayd_close(&relayd->data_sock); | |
321 | ||
00e2e675 DG |
322 | free(relayd); |
323 | } | |
324 | ||
325 | /* | |
326 | * Destroy and free relayd socket pair object. | |
00e2e675 | 327 | */ |
51230d70 | 328 | void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
329 | { |
330 | int ret; | |
331 | struct lttng_ht_iter iter; | |
332 | ||
173af62f DG |
333 | if (relayd == NULL) { |
334 | return; | |
335 | } | |
336 | ||
00e2e675 DG |
337 | DBG("Consumer destroy and close relayd socket pair"); |
338 | ||
339 | iter.iter.node = &relayd->node.node; | |
340 | ret = lttng_ht_del(consumer_data.relayd_ht, &iter); | |
173af62f | 341 | if (ret != 0) { |
8994307f | 342 | /* We assume the relayd is being or is destroyed */ |
173af62f DG |
343 | return; |
344 | } | |
00e2e675 | 345 | |
00e2e675 | 346 | /* RCU free() call */ |
ffe60014 DG |
347 | call_rcu(&relayd->node.head, free_relayd_rcu); |
348 | } | |
349 | ||
350 | /* | |
351 | * Remove a channel from the global list protected by a mutex. This function is | |
352 | * also responsible for freeing its data structures. | |
353 | */ | |
354 | void consumer_del_channel(struct lttng_consumer_channel *channel) | |
355 | { | |
356 | int ret; | |
357 | struct lttng_ht_iter iter; | |
358 | ||
d88aee68 | 359 | DBG("Consumer delete channel key %" PRIu64, channel->key); |
ffe60014 DG |
360 | |
361 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 362 | pthread_mutex_lock(&channel->lock); |
ffe60014 | 363 | |
212d67a2 DG |
364 | /* Destroy streams that might have been left in the stream list. */ |
365 | clean_channel_stream_list(channel); | |
51e762e5 | 366 | |
d3e2ba59 JD |
367 | if (channel->live_timer_enabled == 1) { |
368 | consumer_timer_live_stop(channel); | |
369 | } | |
370 | ||
ffe60014 DG |
371 | switch (consumer_data.type) { |
372 | case LTTNG_CONSUMER_KERNEL: | |
373 | break; | |
374 | case LTTNG_CONSUMER32_UST: | |
375 | case LTTNG_CONSUMER64_UST: | |
376 | lttng_ustconsumer_del_channel(channel); | |
377 | break; | |
378 | default: | |
379 | ERR("Unknown consumer_data type"); | |
380 | assert(0); | |
381 | goto end; | |
382 | } | |
383 | ||
384 | rcu_read_lock(); | |
385 | iter.iter.node = &channel->node.node; | |
386 | ret = lttng_ht_del(consumer_data.channel_ht, &iter); | |
387 | assert(!ret); | |
388 | rcu_read_unlock(); | |
389 | ||
390 | call_rcu(&channel->node.head, free_channel_rcu); | |
391 | end: | |
a9838785 | 392 | pthread_mutex_unlock(&channel->lock); |
ffe60014 | 393 | pthread_mutex_unlock(&consumer_data.lock); |
00e2e675 DG |
394 | } |
395 | ||
228b5bf7 DG |
396 | /* |
397 | * Iterate over the relayd hash table and destroy each element. Finally, | |
398 | * destroy the whole hash table. | |
399 | */ | |
400 | static void cleanup_relayd_ht(void) | |
401 | { | |
402 | struct lttng_ht_iter iter; | |
403 | struct consumer_relayd_sock_pair *relayd; | |
404 | ||
405 | rcu_read_lock(); | |
406 | ||
407 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
408 | node.node) { | |
51230d70 | 409 | consumer_destroy_relayd(relayd); |
228b5bf7 DG |
410 | } |
411 | ||
228b5bf7 | 412 | rcu_read_unlock(); |
36b588ed MD |
413 | |
414 | lttng_ht_destroy(consumer_data.relayd_ht); | |
228b5bf7 DG |
415 | } |
416 | ||
8994307f DG |
417 | /* |
418 | * Update the end point status of all streams having the given network sequence | |
419 | * index (relayd index). | |
420 | * | |
421 | * It's atomically set without having the stream mutex locked which is fine | |
422 | * because we handle the write/read race with a pipe wakeup for each thread. | |
423 | */ | |
da009f2c | 424 | static void update_endpoint_status_by_netidx(uint64_t net_seq_idx, |
8994307f DG |
425 | enum consumer_endpoint_status status) |
426 | { | |
427 | struct lttng_ht_iter iter; | |
428 | struct lttng_consumer_stream *stream; | |
429 | ||
da009f2c | 430 | DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx); |
8994307f DG |
431 | |
432 | rcu_read_lock(); | |
433 | ||
434 | /* Let's begin with metadata */ | |
435 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
436 | if (stream->net_seq_idx == net_seq_idx) { | |
437 | uatomic_set(&stream->endpoint_status, status); | |
438 | DBG("Delete flag set to metadata stream %d", stream->wait_fd); | |
439 | } | |
440 | } | |
441 | ||
442 | /* Follow up by the data streams */ | |
443 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
444 | if (stream->net_seq_idx == net_seq_idx) { | |
445 | uatomic_set(&stream->endpoint_status, status); | |
446 | DBG("Delete flag set to data stream %d", stream->wait_fd); | |
447 | } | |
448 | } | |
449 | rcu_read_unlock(); | |
450 | } | |
451 | ||
452 | /* | |
453 | * Cleanup a relayd object by flagging every associated streams for deletion, | |
454 | * destroying the object meaning removing it from the relayd hash table, | |
455 | * closing the sockets and freeing the memory in a RCU call. | |
456 | * | |
457 | * If a local data context is available, notify the threads that the streams' | |
458 | * state have changed. | |
459 | */ | |
460 | static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd, | |
461 | struct lttng_consumer_local_data *ctx) | |
462 | { | |
da009f2c | 463 | uint64_t netidx; |
8994307f DG |
464 | |
465 | assert(relayd); | |
466 | ||
9617607b DG |
467 | DBG("Cleaning up relayd sockets"); |
468 | ||
8994307f DG |
469 | /* Save the net sequence index before destroying the object */ |
470 | netidx = relayd->net_seq_idx; | |
471 | ||
472 | /* | |
473 | * Delete the relayd from the relayd hash table, close the sockets and free | |
474 | * the object in a RCU call. | |
475 | */ | |
51230d70 | 476 | consumer_destroy_relayd(relayd); |
8994307f DG |
477 | |
478 | /* Set inactive endpoint to all streams */ | |
479 | update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE); | |
480 | ||
481 | /* | |
482 | * With a local data context, notify the threads that the streams' state | |
483 | * have changed. The write() action on the pipe acts as an "implicit" | |
484 | * memory barrier ordering the updates of the end point status from the | |
485 | * read of this status which happens AFTER receiving this notify. | |
486 | */ | |
487 | if (ctx) { | |
acdb9057 | 488 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
13886d2d | 489 | notify_thread_lttng_pipe(ctx->consumer_metadata_pipe); |
8994307f DG |
490 | } |
491 | } | |
492 | ||
a6ba4fe1 DG |
493 | /* |
494 | * Flag a relayd socket pair for destruction. Destroy it if the refcount | |
495 | * reaches zero. | |
496 | * | |
497 | * RCU read side lock MUST be aquired before calling this function. | |
498 | */ | |
499 | void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) | |
500 | { | |
501 | assert(relayd); | |
502 | ||
503 | /* Set destroy flag for this object */ | |
504 | uatomic_set(&relayd->destroy_flag, 1); | |
505 | ||
506 | /* Destroy the relayd if refcount is 0 */ | |
507 | if (uatomic_read(&relayd->refcount) == 0) { | |
51230d70 | 508 | consumer_destroy_relayd(relayd); |
a6ba4fe1 DG |
509 | } |
510 | } | |
511 | ||
3bd1e081 | 512 | /* |
1d1a276c DG |
513 | * Completly destroy stream from every visiable data structure and the given |
514 | * hash table if one. | |
515 | * | |
516 | * One this call returns, the stream object is not longer usable nor visible. | |
3bd1e081 | 517 | */ |
e316aad5 DG |
518 | void consumer_del_stream(struct lttng_consumer_stream *stream, |
519 | struct lttng_ht *ht) | |
3bd1e081 | 520 | { |
1d1a276c | 521 | consumer_stream_destroy(stream, ht); |
3bd1e081 MD |
522 | } |
523 | ||
5ab66908 MD |
524 | /* |
525 | * XXX naming of del vs destroy is all mixed up. | |
526 | */ | |
527 | void consumer_del_stream_for_data(struct lttng_consumer_stream *stream) | |
528 | { | |
529 | consumer_stream_destroy(stream, data_ht); | |
530 | } | |
531 | ||
532 | void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream) | |
533 | { | |
534 | consumer_stream_destroy(stream, metadata_ht); | |
535 | } | |
536 | ||
d88aee68 DG |
537 | struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, |
538 | uint64_t stream_key, | |
3bd1e081 | 539 | enum lttng_consumer_stream_state state, |
ffe60014 | 540 | const char *channel_name, |
6df2e2c9 | 541 | uid_t uid, |
00e2e675 | 542 | gid_t gid, |
57a269f2 | 543 | uint64_t relayd_id, |
53632229 | 544 | uint64_t session_id, |
ffe60014 DG |
545 | int cpu, |
546 | int *alloc_ret, | |
4891ece8 DG |
547 | enum consumer_channel_type type, |
548 | unsigned int monitor) | |
3bd1e081 | 549 | { |
ffe60014 | 550 | int ret; |
3bd1e081 | 551 | struct lttng_consumer_stream *stream; |
3bd1e081 | 552 | |
effcf122 | 553 | stream = zmalloc(sizeof(*stream)); |
3bd1e081 | 554 | if (stream == NULL) { |
7a57cf92 | 555 | PERROR("malloc struct lttng_consumer_stream"); |
ffe60014 | 556 | ret = -ENOMEM; |
7a57cf92 | 557 | goto end; |
3bd1e081 | 558 | } |
7a57cf92 | 559 | |
d56db448 DG |
560 | rcu_read_lock(); |
561 | ||
3bd1e081 | 562 | stream->key = stream_key; |
3bd1e081 MD |
563 | stream->out_fd = -1; |
564 | stream->out_fd_offset = 0; | |
e5d1a9b3 | 565 | stream->output_written = 0; |
3bd1e081 | 566 | stream->state = state; |
6df2e2c9 MD |
567 | stream->uid = uid; |
568 | stream->gid = gid; | |
ffe60014 | 569 | stream->net_seq_idx = relayd_id; |
53632229 | 570 | stream->session_id = session_id; |
4891ece8 | 571 | stream->monitor = monitor; |
774d490c | 572 | stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE; |
309167d2 | 573 | stream->index_fd = -1; |
53632229 | 574 | pthread_mutex_init(&stream->lock, NULL); |
c585821b | 575 | pthread_mutex_init(&stream->metadata_timer_lock, NULL); |
58b1f425 | 576 | |
ffe60014 DG |
577 | /* If channel is the metadata, flag this stream as metadata. */ |
578 | if (type == CONSUMER_CHANNEL_TYPE_METADATA) { | |
579 | stream->metadata_flag = 1; | |
580 | /* Metadata is flat out. */ | |
581 | strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name)); | |
94d49140 JD |
582 | /* Live rendez-vous point. */ |
583 | pthread_cond_init(&stream->metadata_rdv, NULL); | |
584 | pthread_mutex_init(&stream->metadata_rdv_lock, NULL); | |
58b1f425 | 585 | } else { |
ffe60014 DG |
586 | /* Format stream name to <channel_name>_<cpu_number> */ |
587 | ret = snprintf(stream->name, sizeof(stream->name), "%s_%d", | |
588 | channel_name, cpu); | |
589 | if (ret < 0) { | |
590 | PERROR("snprintf stream name"); | |
591 | goto error; | |
592 | } | |
58b1f425 | 593 | } |
c30aaa51 | 594 | |
ffe60014 | 595 | /* Key is always the wait_fd for streams. */ |
d88aee68 | 596 | lttng_ht_node_init_u64(&stream->node, stream->key); |
ffe60014 | 597 | |
d8ef542d MD |
598 | /* Init node per channel id key */ |
599 | lttng_ht_node_init_u64(&stream->node_channel_id, channel_key); | |
600 | ||
53632229 | 601 | /* Init session id node with the stream session id */ |
d88aee68 | 602 | lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id); |
53632229 | 603 | |
07b86b52 JD |
604 | DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64 |
605 | " relayd_id %" PRIu64 ", session_id %" PRIu64, | |
606 | stream->name, stream->key, channel_key, | |
607 | stream->net_seq_idx, stream->session_id); | |
d56db448 DG |
608 | |
609 | rcu_read_unlock(); | |
3bd1e081 | 610 | return stream; |
c80048c6 MD |
611 | |
612 | error: | |
d56db448 | 613 | rcu_read_unlock(); |
c80048c6 | 614 | free(stream); |
7a57cf92 | 615 | end: |
ffe60014 DG |
616 | if (alloc_ret) { |
617 | *alloc_ret = ret; | |
618 | } | |
c80048c6 | 619 | return NULL; |
3bd1e081 MD |
620 | } |
621 | ||
622 | /* | |
623 | * Add a stream to the global list protected by a mutex. | |
624 | */ | |
5ab66908 | 625 | int consumer_add_data_stream(struct lttng_consumer_stream *stream) |
3bd1e081 | 626 | { |
5ab66908 | 627 | struct lttng_ht *ht = data_ht; |
3bd1e081 MD |
628 | int ret = 0; |
629 | ||
e316aad5 | 630 | assert(stream); |
43c34bc3 | 631 | assert(ht); |
c77fc10a | 632 | |
d88aee68 | 633 | DBG3("Adding consumer stream %" PRIu64, stream->key); |
e316aad5 DG |
634 | |
635 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 636 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 637 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 638 | pthread_mutex_lock(&stream->lock); |
b0b335c8 | 639 | rcu_read_lock(); |
e316aad5 | 640 | |
43c34bc3 | 641 | /* Steal stream identifier to avoid having streams with the same key */ |
ffe60014 | 642 | steal_stream_key(stream->key, ht); |
43c34bc3 | 643 | |
d88aee68 | 644 | lttng_ht_add_unique_u64(ht, &stream->node); |
00e2e675 | 645 | |
d8ef542d MD |
646 | lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, |
647 | &stream->node_channel_id); | |
648 | ||
ca22feea DG |
649 | /* |
650 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
651 | * the key since the HT does not use it and we allow to add redundant keys | |
652 | * into this table. | |
653 | */ | |
d88aee68 | 654 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 655 | |
e316aad5 | 656 | /* |
ffe60014 DG |
657 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
658 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
659 | * causes the count to become 0 also causes a stream to be added. The |
660 | * channel deletion will thus be triggered by the following removal of this | |
661 | * stream. | |
662 | */ | |
ffe60014 | 663 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
664 | /* Increment refcount before decrementing nb_init_stream_left */ |
665 | cmm_smp_wmb(); | |
ffe60014 | 666 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
667 | } |
668 | ||
669 | /* Update consumer data once the node is inserted. */ | |
3bd1e081 MD |
670 | consumer_data.stream_count++; |
671 | consumer_data.need_update = 1; | |
672 | ||
e316aad5 | 673 | rcu_read_unlock(); |
2e818a6a | 674 | pthread_mutex_unlock(&stream->lock); |
ec6ea7d0 | 675 | pthread_mutex_unlock(&stream->chan->timer_lock); |
a9838785 | 676 | pthread_mutex_unlock(&stream->chan->lock); |
3bd1e081 | 677 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 678 | |
3bd1e081 MD |
679 | return ret; |
680 | } | |
681 | ||
5ab66908 MD |
682 | void consumer_del_data_stream(struct lttng_consumer_stream *stream) |
683 | { | |
684 | consumer_del_stream(stream, data_ht); | |
685 | } | |
686 | ||
00e2e675 | 687 | /* |
3f8e211f DG |
688 | * Add relayd socket to global consumer data hashtable. RCU read side lock MUST |
689 | * be acquired before calling this. | |
00e2e675 | 690 | */ |
d09e1200 | 691 | static int add_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
692 | { |
693 | int ret = 0; | |
d88aee68 | 694 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
695 | struct lttng_ht_iter iter; |
696 | ||
ffe60014 | 697 | assert(relayd); |
00e2e675 | 698 | |
00e2e675 | 699 | lttng_ht_lookup(consumer_data.relayd_ht, |
d88aee68 DG |
700 | &relayd->net_seq_idx, &iter); |
701 | node = lttng_ht_iter_get_node_u64(&iter); | |
00e2e675 | 702 | if (node != NULL) { |
00e2e675 DG |
703 | goto end; |
704 | } | |
d88aee68 | 705 | lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node); |
00e2e675 | 706 | |
00e2e675 DG |
707 | end: |
708 | return ret; | |
709 | } | |
710 | ||
711 | /* | |
712 | * Allocate and return a consumer relayd socket. | |
713 | */ | |
027a694f | 714 | static struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( |
da009f2c | 715 | uint64_t net_seq_idx) |
00e2e675 DG |
716 | { |
717 | struct consumer_relayd_sock_pair *obj = NULL; | |
718 | ||
da009f2c MD |
719 | /* net sequence index of -1 is a failure */ |
720 | if (net_seq_idx == (uint64_t) -1ULL) { | |
00e2e675 DG |
721 | goto error; |
722 | } | |
723 | ||
724 | obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); | |
725 | if (obj == NULL) { | |
726 | PERROR("zmalloc relayd sock"); | |
727 | goto error; | |
728 | } | |
729 | ||
730 | obj->net_seq_idx = net_seq_idx; | |
731 | obj->refcount = 0; | |
173af62f | 732 | obj->destroy_flag = 0; |
f96e4545 MD |
733 | obj->control_sock.sock.fd = -1; |
734 | obj->data_sock.sock.fd = -1; | |
d88aee68 | 735 | lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx); |
00e2e675 DG |
736 | pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); |
737 | ||
738 | error: | |
739 | return obj; | |
740 | } | |
741 | ||
742 | /* | |
743 | * Find a relayd socket pair in the global consumer data. | |
744 | * | |
745 | * Return the object if found else NULL. | |
b0b335c8 MD |
746 | * RCU read-side lock must be held across this call and while using the |
747 | * returned object. | |
00e2e675 | 748 | */ |
d88aee68 | 749 | struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key) |
00e2e675 DG |
750 | { |
751 | struct lttng_ht_iter iter; | |
d88aee68 | 752 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
753 | struct consumer_relayd_sock_pair *relayd = NULL; |
754 | ||
755 | /* Negative keys are lookup failures */ | |
d88aee68 | 756 | if (key == (uint64_t) -1ULL) { |
00e2e675 DG |
757 | goto error; |
758 | } | |
759 | ||
d88aee68 | 760 | lttng_ht_lookup(consumer_data.relayd_ht, &key, |
00e2e675 | 761 | &iter); |
d88aee68 | 762 | node = lttng_ht_iter_get_node_u64(&iter); |
00e2e675 DG |
763 | if (node != NULL) { |
764 | relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
765 | } | |
766 | ||
00e2e675 DG |
767 | error: |
768 | return relayd; | |
769 | } | |
770 | ||
10a50311 JD |
771 | /* |
772 | * Find a relayd and send the stream | |
773 | * | |
774 | * Returns 0 on success, < 0 on error | |
775 | */ | |
776 | int consumer_send_relayd_stream(struct lttng_consumer_stream *stream, | |
777 | char *path) | |
778 | { | |
779 | int ret = 0; | |
780 | struct consumer_relayd_sock_pair *relayd; | |
781 | ||
782 | assert(stream); | |
783 | assert(stream->net_seq_idx != -1ULL); | |
784 | assert(path); | |
785 | ||
786 | /* The stream is not metadata. Get relayd reference if exists. */ | |
787 | rcu_read_lock(); | |
788 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
789 | if (relayd != NULL) { | |
790 | /* Add stream on the relayd */ | |
791 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
792 | ret = relayd_add_stream(&relayd->control_sock, stream->name, | |
793 | path, &stream->relayd_stream_id, | |
794 | stream->chan->tracefile_size, stream->chan->tracefile_count); | |
795 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
796 | if (ret < 0) { | |
797 | goto end; | |
798 | } | |
1c20f0e2 | 799 | |
10a50311 | 800 | uatomic_inc(&relayd->refcount); |
d01178b6 | 801 | stream->sent_to_relayd = 1; |
10a50311 JD |
802 | } else { |
803 | ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.", | |
804 | stream->key, stream->net_seq_idx); | |
805 | ret = -1; | |
806 | goto end; | |
807 | } | |
808 | ||
809 | DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64, | |
810 | stream->name, stream->key, stream->net_seq_idx); | |
811 | ||
812 | end: | |
813 | rcu_read_unlock(); | |
814 | return ret; | |
815 | } | |
816 | ||
a4baae1b JD |
817 | /* |
818 | * Find a relayd and send the streams sent message | |
819 | * | |
820 | * Returns 0 on success, < 0 on error | |
821 | */ | |
822 | int consumer_send_relayd_streams_sent(uint64_t net_seq_idx) | |
823 | { | |
824 | int ret = 0; | |
825 | struct consumer_relayd_sock_pair *relayd; | |
826 | ||
827 | assert(net_seq_idx != -1ULL); | |
828 | ||
829 | /* The stream is not metadata. Get relayd reference if exists. */ | |
830 | rcu_read_lock(); | |
831 | relayd = consumer_find_relayd(net_seq_idx); | |
832 | if (relayd != NULL) { | |
833 | /* Add stream on the relayd */ | |
834 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
835 | ret = relayd_streams_sent(&relayd->control_sock); | |
836 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
837 | if (ret < 0) { | |
838 | goto end; | |
839 | } | |
840 | } else { | |
841 | ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.", | |
842 | net_seq_idx); | |
843 | ret = -1; | |
844 | goto end; | |
845 | } | |
846 | ||
847 | ret = 0; | |
848 | DBG("All streams sent relayd id %" PRIu64, net_seq_idx); | |
849 | ||
850 | end: | |
851 | rcu_read_unlock(); | |
852 | return ret; | |
853 | } | |
854 | ||
10a50311 JD |
855 | /* |
856 | * Find a relayd and close the stream | |
857 | */ | |
858 | void close_relayd_stream(struct lttng_consumer_stream *stream) | |
859 | { | |
860 | struct consumer_relayd_sock_pair *relayd; | |
861 | ||
862 | /* The stream is not metadata. Get relayd reference if exists. */ | |
863 | rcu_read_lock(); | |
864 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
865 | if (relayd) { | |
866 | consumer_stream_relayd_close(stream, relayd); | |
867 | } | |
868 | rcu_read_unlock(); | |
869 | } | |
870 | ||
00e2e675 DG |
871 | /* |
872 | * Handle stream for relayd transmission if the stream applies for network | |
873 | * streaming where the net sequence index is set. | |
874 | * | |
875 | * Return destination file descriptor or negative value on error. | |
876 | */ | |
6197aea7 | 877 | static int write_relayd_stream_header(struct lttng_consumer_stream *stream, |
1d4dfdef DG |
878 | size_t data_size, unsigned long padding, |
879 | struct consumer_relayd_sock_pair *relayd) | |
00e2e675 DG |
880 | { |
881 | int outfd = -1, ret; | |
00e2e675 DG |
882 | struct lttcomm_relayd_data_hdr data_hdr; |
883 | ||
884 | /* Safety net */ | |
885 | assert(stream); | |
6197aea7 | 886 | assert(relayd); |
00e2e675 DG |
887 | |
888 | /* Reset data header */ | |
889 | memset(&data_hdr, 0, sizeof(data_hdr)); | |
890 | ||
00e2e675 DG |
891 | if (stream->metadata_flag) { |
892 | /* Caller MUST acquire the relayd control socket lock */ | |
893 | ret = relayd_send_metadata(&relayd->control_sock, data_size); | |
894 | if (ret < 0) { | |
895 | goto error; | |
896 | } | |
897 | ||
898 | /* Metadata are always sent on the control socket. */ | |
6151a90f | 899 | outfd = relayd->control_sock.sock.fd; |
00e2e675 DG |
900 | } else { |
901 | /* Set header with stream information */ | |
902 | data_hdr.stream_id = htobe64(stream->relayd_stream_id); | |
903 | data_hdr.data_size = htobe32(data_size); | |
1d4dfdef | 904 | data_hdr.padding_size = htobe32(padding); |
39df6d9f DG |
905 | /* |
906 | * Note that net_seq_num below is assigned with the *current* value of | |
907 | * next_net_seq_num and only after that the next_net_seq_num will be | |
908 | * increment. This is why when issuing a command on the relayd using | |
909 | * this next value, 1 should always be substracted in order to compare | |
910 | * the last seen sequence number on the relayd side to the last sent. | |
911 | */ | |
3604f373 | 912 | data_hdr.net_seq_num = htobe64(stream->next_net_seq_num); |
00e2e675 DG |
913 | /* Other fields are zeroed previously */ |
914 | ||
915 | ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, | |
916 | sizeof(data_hdr)); | |
917 | if (ret < 0) { | |
918 | goto error; | |
919 | } | |
920 | ||
3604f373 DG |
921 | ++stream->next_net_seq_num; |
922 | ||
00e2e675 | 923 | /* Set to go on data socket */ |
6151a90f | 924 | outfd = relayd->data_sock.sock.fd; |
00e2e675 DG |
925 | } |
926 | ||
927 | error: | |
928 | return outfd; | |
929 | } | |
930 | ||
3bd1e081 | 931 | /* |
ffe60014 DG |
932 | * Allocate and return a new lttng_consumer_channel object using the given key |
933 | * to initialize the hash table node. | |
934 | * | |
935 | * On error, return NULL. | |
3bd1e081 | 936 | */ |
886224ff | 937 | struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, |
ffe60014 DG |
938 | uint64_t session_id, |
939 | const char *pathname, | |
940 | const char *name, | |
941 | uid_t uid, | |
942 | gid_t gid, | |
57a269f2 | 943 | uint64_t relayd_id, |
1624d5b7 JD |
944 | enum lttng_event_output output, |
945 | uint64_t tracefile_size, | |
2bba9e53 | 946 | uint64_t tracefile_count, |
1950109e | 947 | uint64_t session_id_per_pid, |
ecc48a90 | 948 | unsigned int monitor, |
d7ba1388 | 949 | unsigned int live_timer_interval, |
3d071855 | 950 | const char *root_shm_path, |
d7ba1388 | 951 | const char *shm_path) |
3bd1e081 MD |
952 | { |
953 | struct lttng_consumer_channel *channel; | |
3bd1e081 | 954 | |
276b26d1 | 955 | channel = zmalloc(sizeof(*channel)); |
3bd1e081 | 956 | if (channel == NULL) { |
7a57cf92 | 957 | PERROR("malloc struct lttng_consumer_channel"); |
3bd1e081 MD |
958 | goto end; |
959 | } | |
ffe60014 DG |
960 | |
961 | channel->key = key; | |
3bd1e081 | 962 | channel->refcount = 0; |
ffe60014 | 963 | channel->session_id = session_id; |
1950109e | 964 | channel->session_id_per_pid = session_id_per_pid; |
ffe60014 DG |
965 | channel->uid = uid; |
966 | channel->gid = gid; | |
967 | channel->relayd_id = relayd_id; | |
1624d5b7 JD |
968 | channel->tracefile_size = tracefile_size; |
969 | channel->tracefile_count = tracefile_count; | |
2bba9e53 | 970 | channel->monitor = monitor; |
ecc48a90 | 971 | channel->live_timer_interval = live_timer_interval; |
a9838785 | 972 | pthread_mutex_init(&channel->lock, NULL); |
ec6ea7d0 | 973 | pthread_mutex_init(&channel->timer_lock, NULL); |
ffe60014 | 974 | |
0c759fc9 DG |
975 | switch (output) { |
976 | case LTTNG_EVENT_SPLICE: | |
977 | channel->output = CONSUMER_CHANNEL_SPLICE; | |
978 | break; | |
979 | case LTTNG_EVENT_MMAP: | |
980 | channel->output = CONSUMER_CHANNEL_MMAP; | |
981 | break; | |
982 | default: | |
983 | assert(0); | |
984 | free(channel); | |
985 | channel = NULL; | |
986 | goto end; | |
987 | } | |
988 | ||
07b86b52 JD |
989 | /* |
990 | * In monitor mode, the streams associated with the channel will be put in | |
991 | * a special list ONLY owned by this channel. So, the refcount is set to 1 | |
992 | * here meaning that the channel itself has streams that are referenced. | |
993 | * | |
994 | * On a channel deletion, once the channel is no longer visible, the | |
995 | * refcount is decremented and checked for a zero value to delete it. With | |
996 | * streams in no monitor mode, it will now be safe to destroy the channel. | |
997 | */ | |
998 | if (!channel->monitor) { | |
999 | channel->refcount = 1; | |
1000 | } | |
1001 | ||
ffe60014 DG |
1002 | strncpy(channel->pathname, pathname, sizeof(channel->pathname)); |
1003 | channel->pathname[sizeof(channel->pathname) - 1] = '\0'; | |
1004 | ||
1005 | strncpy(channel->name, name, sizeof(channel->name)); | |
1006 | channel->name[sizeof(channel->name) - 1] = '\0'; | |
1007 | ||
3d071855 MD |
1008 | if (root_shm_path) { |
1009 | strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path)); | |
1010 | channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0'; | |
1011 | } | |
d7ba1388 MD |
1012 | if (shm_path) { |
1013 | strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path)); | |
1014 | channel->shm_path[sizeof(channel->shm_path) - 1] = '\0'; | |
1015 | } | |
1016 | ||
d88aee68 | 1017 | lttng_ht_node_init_u64(&channel->node, channel->key); |
d8ef542d MD |
1018 | |
1019 | channel->wait_fd = -1; | |
1020 | ||
ffe60014 DG |
1021 | CDS_INIT_LIST_HEAD(&channel->streams.head); |
1022 | ||
d88aee68 | 1023 | DBG("Allocated channel (key %" PRIu64 ")", channel->key) |
3bd1e081 | 1024 | |
3bd1e081 MD |
1025 | end: |
1026 | return channel; | |
1027 | } | |
1028 | ||
1029 | /* | |
1030 | * Add a channel to the global list protected by a mutex. | |
821fffb2 | 1031 | * |
b5a6470f | 1032 | * Always return 0 indicating success. |
3bd1e081 | 1033 | */ |
d8ef542d MD |
1034 | int consumer_add_channel(struct lttng_consumer_channel *channel, |
1035 | struct lttng_consumer_local_data *ctx) | |
3bd1e081 | 1036 | { |
3bd1e081 | 1037 | pthread_mutex_lock(&consumer_data.lock); |
a9838785 | 1038 | pthread_mutex_lock(&channel->lock); |
ec6ea7d0 | 1039 | pthread_mutex_lock(&channel->timer_lock); |
c77fc10a | 1040 | |
b5a6470f DG |
1041 | /* |
1042 | * This gives us a guarantee that the channel we are about to add to the | |
1043 | * channel hash table will be unique. See this function comment on the why | |
1044 | * we need to steel the channel key at this stage. | |
1045 | */ | |
1046 | steal_channel_key(channel->key); | |
c77fc10a | 1047 | |
b5a6470f | 1048 | rcu_read_lock(); |
d88aee68 | 1049 | lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node); |
6065ceec | 1050 | rcu_read_unlock(); |
b5a6470f | 1051 | |
ec6ea7d0 | 1052 | pthread_mutex_unlock(&channel->timer_lock); |
a9838785 | 1053 | pthread_mutex_unlock(&channel->lock); |
3bd1e081 | 1054 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 1055 | |
b5a6470f | 1056 | if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) { |
a0cbdd2e | 1057 | notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD); |
d8ef542d | 1058 | } |
b5a6470f DG |
1059 | |
1060 | return 0; | |
3bd1e081 MD |
1061 | } |
1062 | ||
1063 | /* | |
1064 | * Allocate the pollfd structure and the local view of the out fds to avoid | |
1065 | * doing a lookup in the linked list and concurrency issues when writing is | |
1066 | * needed. Called with consumer_data.lock held. | |
1067 | * | |
1068 | * Returns the number of fds in the structures. | |
1069 | */ | |
ffe60014 DG |
1070 | static int update_poll_array(struct lttng_consumer_local_data *ctx, |
1071 | struct pollfd **pollfd, struct lttng_consumer_stream **local_stream, | |
1072 | struct lttng_ht *ht) | |
3bd1e081 | 1073 | { |
3bd1e081 | 1074 | int i = 0; |
e4421fec DG |
1075 | struct lttng_ht_iter iter; |
1076 | struct lttng_consumer_stream *stream; | |
3bd1e081 | 1077 | |
ffe60014 DG |
1078 | assert(ctx); |
1079 | assert(ht); | |
1080 | assert(pollfd); | |
1081 | assert(local_stream); | |
1082 | ||
3bd1e081 | 1083 | DBG("Updating poll fd array"); |
481d6c57 | 1084 | rcu_read_lock(); |
43c34bc3 | 1085 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
8994307f DG |
1086 | /* |
1087 | * Only active streams with an active end point can be added to the | |
1088 | * poll set and local stream storage of the thread. | |
1089 | * | |
1090 | * There is a potential race here for endpoint_status to be updated | |
1091 | * just after the check. However, this is OK since the stream(s) will | |
1092 | * be deleted once the thread is notified that the end point state has | |
1093 | * changed where this function will be called back again. | |
1094 | */ | |
1095 | if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM || | |
79d4ffb7 | 1096 | stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { |
3bd1e081 MD |
1097 | continue; |
1098 | } | |
7972aab2 DG |
1099 | /* |
1100 | * This clobbers way too much the debug output. Uncomment that if you | |
1101 | * need it for debugging purposes. | |
1102 | * | |
1103 | * DBG("Active FD %d", stream->wait_fd); | |
1104 | */ | |
e4421fec | 1105 | (*pollfd)[i].fd = stream->wait_fd; |
3bd1e081 | 1106 | (*pollfd)[i].events = POLLIN | POLLPRI; |
e4421fec | 1107 | local_stream[i] = stream; |
3bd1e081 MD |
1108 | i++; |
1109 | } | |
481d6c57 | 1110 | rcu_read_unlock(); |
3bd1e081 MD |
1111 | |
1112 | /* | |
50f8ae69 | 1113 | * Insert the consumer_data_pipe at the end of the array and don't |
3bd1e081 MD |
1114 | * increment i so nb_fd is the number of real FD. |
1115 | */ | |
acdb9057 | 1116 | (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe); |
509bb1cf | 1117 | (*pollfd)[i].events = POLLIN | POLLPRI; |
02b3d176 DG |
1118 | |
1119 | (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe); | |
1120 | (*pollfd)[i + 1].events = POLLIN | POLLPRI; | |
3bd1e081 MD |
1121 | return i; |
1122 | } | |
1123 | ||
1124 | /* | |
84382d49 MD |
1125 | * Poll on the should_quit pipe and the command socket return -1 on |
1126 | * error, 1 if should exit, 0 if data is available on the command socket | |
3bd1e081 MD |
1127 | */ |
1128 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) | |
1129 | { | |
1130 | int num_rdy; | |
1131 | ||
88f2b785 | 1132 | restart: |
3bd1e081 MD |
1133 | num_rdy = poll(consumer_sockpoll, 2, -1); |
1134 | if (num_rdy == -1) { | |
88f2b785 MD |
1135 | /* |
1136 | * Restart interrupted system call. | |
1137 | */ | |
1138 | if (errno == EINTR) { | |
1139 | goto restart; | |
1140 | } | |
7a57cf92 | 1141 | PERROR("Poll error"); |
84382d49 | 1142 | return -1; |
3bd1e081 | 1143 | } |
509bb1cf | 1144 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
3bd1e081 | 1145 | DBG("consumer_should_quit wake up"); |
84382d49 | 1146 | return 1; |
3bd1e081 MD |
1147 | } |
1148 | return 0; | |
3bd1e081 MD |
1149 | } |
1150 | ||
1151 | /* | |
1152 | * Set the error socket. | |
1153 | */ | |
ffe60014 DG |
1154 | void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx, |
1155 | int sock) | |
3bd1e081 MD |
1156 | { |
1157 | ctx->consumer_error_socket = sock; | |
1158 | } | |
1159 | ||
1160 | /* | |
1161 | * Set the command socket path. | |
1162 | */ | |
3bd1e081 MD |
1163 | void lttng_consumer_set_command_sock_path( |
1164 | struct lttng_consumer_local_data *ctx, char *sock) | |
1165 | { | |
1166 | ctx->consumer_command_sock_path = sock; | |
1167 | } | |
1168 | ||
1169 | /* | |
1170 | * Send return code to the session daemon. | |
1171 | * If the socket is not defined, we return 0, it is not a fatal error | |
1172 | */ | |
ffe60014 | 1173 | int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd) |
3bd1e081 MD |
1174 | { |
1175 | if (ctx->consumer_error_socket > 0) { | |
1176 | return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd, | |
1177 | sizeof(enum lttcomm_sessiond_command)); | |
1178 | } | |
1179 | ||
1180 | return 0; | |
1181 | } | |
1182 | ||
1183 | /* | |
228b5bf7 DG |
1184 | * Close all the tracefiles and stream fds and MUST be called when all |
1185 | * instances are destroyed i.e. when all threads were joined and are ended. | |
3bd1e081 MD |
1186 | */ |
1187 | void lttng_consumer_cleanup(void) | |
1188 | { | |
e4421fec | 1189 | struct lttng_ht_iter iter; |
ffe60014 | 1190 | struct lttng_consumer_channel *channel; |
6065ceec DG |
1191 | |
1192 | rcu_read_lock(); | |
3bd1e081 | 1193 | |
ffe60014 DG |
1194 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel, |
1195 | node.node) { | |
702b1ea4 | 1196 | consumer_del_channel(channel); |
3bd1e081 | 1197 | } |
6065ceec DG |
1198 | |
1199 | rcu_read_unlock(); | |
d6ce1df2 | 1200 | |
d6ce1df2 | 1201 | lttng_ht_destroy(consumer_data.channel_ht); |
228b5bf7 DG |
1202 | |
1203 | cleanup_relayd_ht(); | |
1204 | ||
d8ef542d MD |
1205 | lttng_ht_destroy(consumer_data.stream_per_chan_id_ht); |
1206 | ||
228b5bf7 DG |
1207 | /* |
1208 | * This HT contains streams that are freed by either the metadata thread or | |
1209 | * the data thread so we do *nothing* on the hash table and simply destroy | |
1210 | * it. | |
1211 | */ | |
1212 | lttng_ht_destroy(consumer_data.stream_list_ht); | |
3bd1e081 MD |
1213 | } |
1214 | ||
1215 | /* | |
1216 | * Called from signal handler. | |
1217 | */ | |
1218 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) | |
1219 | { | |
6cd525e8 MD |
1220 | ssize_t ret; |
1221 | ||
3bd1e081 | 1222 | consumer_quit = 1; |
6cd525e8 MD |
1223 | ret = lttng_write(ctx->consumer_should_quit[1], "4", 1); |
1224 | if (ret < 1) { | |
7a57cf92 | 1225 | PERROR("write consumer quit"); |
3bd1e081 | 1226 | } |
ab1027f4 DG |
1227 | |
1228 | DBG("Consumer flag that it should quit"); | |
3bd1e081 MD |
1229 | } |
1230 | ||
00e2e675 DG |
1231 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
1232 | off_t orig_offset) | |
3bd1e081 MD |
1233 | { |
1234 | int outfd = stream->out_fd; | |
1235 | ||
1236 | /* | |
1237 | * This does a blocking write-and-wait on any page that belongs to the | |
1238 | * subbuffer prior to the one we just wrote. | |
1239 | * Don't care about error values, as these are just hints and ways to | |
1240 | * limit the amount of page cache used. | |
1241 | */ | |
ffe60014 | 1242 | if (orig_offset < stream->max_sb_size) { |
3bd1e081 MD |
1243 | return; |
1244 | } | |
ffe60014 DG |
1245 | lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size, |
1246 | stream->max_sb_size, | |
3bd1e081 MD |
1247 | SYNC_FILE_RANGE_WAIT_BEFORE |
1248 | | SYNC_FILE_RANGE_WRITE | |
1249 | | SYNC_FILE_RANGE_WAIT_AFTER); | |
1250 | /* | |
1251 | * Give hints to the kernel about how we access the file: | |
1252 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after | |
1253 | * we write it. | |
1254 | * | |
1255 | * We need to call fadvise again after the file grows because the | |
1256 | * kernel does not seem to apply fadvise to non-existing parts of the | |
1257 | * file. | |
1258 | * | |
1259 | * Call fadvise _after_ having waited for the page writeback to | |
1260 | * complete because the dirty page writeback semantic is not well | |
1261 | * defined. So it can be expected to lead to lower throughput in | |
1262 | * streaming. | |
1263 | */ | |
ffe60014 DG |
1264 | posix_fadvise(outfd, orig_offset - stream->max_sb_size, |
1265 | stream->max_sb_size, POSIX_FADV_DONTNEED); | |
3bd1e081 MD |
1266 | } |
1267 | ||
1268 | /* | |
1269 | * Initialise the necessary environnement : | |
1270 | * - create a new context | |
1271 | * - create the poll_pipe | |
1272 | * - create the should_quit pipe (for signal handler) | |
1273 | * - create the thread pipe (for splice) | |
1274 | * | |
1275 | * Takes a function pointer as argument, this function is called when data is | |
1276 | * available on a buffer. This function is responsible to do the | |
1277 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the | |
1278 | * buffer configuration and then kernctl_put_next_subbuf at the end. | |
1279 | * | |
1280 | * Returns a pointer to the new context or NULL on error. | |
1281 | */ | |
1282 | struct lttng_consumer_local_data *lttng_consumer_create( | |
1283 | enum lttng_consumer_type type, | |
4078b776 | 1284 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
d41f73b7 | 1285 | struct lttng_consumer_local_data *ctx), |
3bd1e081 MD |
1286 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
1287 | int (*recv_stream)(struct lttng_consumer_stream *stream), | |
30319bcb | 1288 | int (*update_stream)(uint64_t stream_key, uint32_t state)) |
3bd1e081 | 1289 | { |
d8ef542d | 1290 | int ret; |
3bd1e081 MD |
1291 | struct lttng_consumer_local_data *ctx; |
1292 | ||
1293 | assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || | |
1294 | consumer_data.type == type); | |
1295 | consumer_data.type = type; | |
1296 | ||
effcf122 | 1297 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
3bd1e081 | 1298 | if (ctx == NULL) { |
7a57cf92 | 1299 | PERROR("allocating context"); |
3bd1e081 MD |
1300 | goto error; |
1301 | } | |
1302 | ||
1303 | ctx->consumer_error_socket = -1; | |
331744e3 | 1304 | ctx->consumer_metadata_socket = -1; |
75d83e50 | 1305 | pthread_mutex_init(&ctx->metadata_socket_lock, NULL); |
3bd1e081 MD |
1306 | /* assign the callbacks */ |
1307 | ctx->on_buffer_ready = buffer_ready; | |
1308 | ctx->on_recv_channel = recv_channel; | |
1309 | ctx->on_recv_stream = recv_stream; | |
1310 | ctx->on_update_stream = update_stream; | |
1311 | ||
acdb9057 DG |
1312 | ctx->consumer_data_pipe = lttng_pipe_open(0); |
1313 | if (!ctx->consumer_data_pipe) { | |
3bd1e081 MD |
1314 | goto error_poll_pipe; |
1315 | } | |
1316 | ||
02b3d176 DG |
1317 | ctx->consumer_wakeup_pipe = lttng_pipe_open(0); |
1318 | if (!ctx->consumer_wakeup_pipe) { | |
1319 | goto error_wakeup_pipe; | |
1320 | } | |
1321 | ||
3bd1e081 MD |
1322 | ret = pipe(ctx->consumer_should_quit); |
1323 | if (ret < 0) { | |
7a57cf92 | 1324 | PERROR("Error creating recv pipe"); |
3bd1e081 MD |
1325 | goto error_quit_pipe; |
1326 | } | |
1327 | ||
d8ef542d MD |
1328 | ret = pipe(ctx->consumer_channel_pipe); |
1329 | if (ret < 0) { | |
1330 | PERROR("Error creating channel pipe"); | |
1331 | goto error_channel_pipe; | |
1332 | } | |
1333 | ||
13886d2d DG |
1334 | ctx->consumer_metadata_pipe = lttng_pipe_open(0); |
1335 | if (!ctx->consumer_metadata_pipe) { | |
fb3a43a9 DG |
1336 | goto error_metadata_pipe; |
1337 | } | |
3bd1e081 | 1338 | |
fb3a43a9 | 1339 | return ctx; |
3bd1e081 | 1340 | |
fb3a43a9 | 1341 | error_metadata_pipe: |
d8ef542d MD |
1342 | utils_close_pipe(ctx->consumer_channel_pipe); |
1343 | error_channel_pipe: | |
d8ef542d | 1344 | utils_close_pipe(ctx->consumer_should_quit); |
3bd1e081 | 1345 | error_quit_pipe: |
02b3d176 DG |
1346 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
1347 | error_wakeup_pipe: | |
acdb9057 | 1348 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
3bd1e081 MD |
1349 | error_poll_pipe: |
1350 | free(ctx); | |
1351 | error: | |
1352 | return NULL; | |
1353 | } | |
1354 | ||
282dadbc MD |
1355 | /* |
1356 | * Iterate over all streams of the hashtable and free them properly. | |
1357 | */ | |
1358 | static void destroy_data_stream_ht(struct lttng_ht *ht) | |
1359 | { | |
1360 | struct lttng_ht_iter iter; | |
1361 | struct lttng_consumer_stream *stream; | |
1362 | ||
1363 | if (ht == NULL) { | |
1364 | return; | |
1365 | } | |
1366 | ||
1367 | rcu_read_lock(); | |
1368 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1369 | /* | |
1370 | * Ignore return value since we are currently cleaning up so any error | |
1371 | * can't be handled. | |
1372 | */ | |
1373 | (void) consumer_del_stream(stream, ht); | |
1374 | } | |
1375 | rcu_read_unlock(); | |
1376 | ||
1377 | lttng_ht_destroy(ht); | |
1378 | } | |
1379 | ||
1380 | /* | |
1381 | * Iterate over all streams of the metadata hashtable and free them | |
1382 | * properly. | |
1383 | */ | |
1384 | static void destroy_metadata_stream_ht(struct lttng_ht *ht) | |
1385 | { | |
1386 | struct lttng_ht_iter iter; | |
1387 | struct lttng_consumer_stream *stream; | |
1388 | ||
1389 | if (ht == NULL) { | |
1390 | return; | |
1391 | } | |
1392 | ||
1393 | rcu_read_lock(); | |
1394 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1395 | /* | |
1396 | * Ignore return value since we are currently cleaning up so any error | |
1397 | * can't be handled. | |
1398 | */ | |
1399 | (void) consumer_del_metadata_stream(stream, ht); | |
1400 | } | |
1401 | rcu_read_unlock(); | |
1402 | ||
1403 | lttng_ht_destroy(ht); | |
1404 | } | |
1405 | ||
3bd1e081 MD |
1406 | /* |
1407 | * Close all fds associated with the instance and free the context. | |
1408 | */ | |
1409 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) | |
1410 | { | |
4c462e79 MD |
1411 | int ret; |
1412 | ||
ab1027f4 DG |
1413 | DBG("Consumer destroying it. Closing everything."); |
1414 | ||
4f2e75b9 DG |
1415 | if (!ctx) { |
1416 | return; | |
1417 | } | |
1418 | ||
282dadbc MD |
1419 | destroy_data_stream_ht(data_ht); |
1420 | destroy_metadata_stream_ht(metadata_ht); | |
1421 | ||
4c462e79 MD |
1422 | ret = close(ctx->consumer_error_socket); |
1423 | if (ret) { | |
1424 | PERROR("close"); | |
1425 | } | |
331744e3 JD |
1426 | ret = close(ctx->consumer_metadata_socket); |
1427 | if (ret) { | |
1428 | PERROR("close"); | |
1429 | } | |
d8ef542d | 1430 | utils_close_pipe(ctx->consumer_channel_pipe); |
acdb9057 | 1431 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
13886d2d | 1432 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
02b3d176 | 1433 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
d8ef542d | 1434 | utils_close_pipe(ctx->consumer_should_quit); |
fb3a43a9 | 1435 | |
3bd1e081 MD |
1436 | unlink(ctx->consumer_command_sock_path); |
1437 | free(ctx); | |
1438 | } | |
1439 | ||
6197aea7 DG |
1440 | /* |
1441 | * Write the metadata stream id on the specified file descriptor. | |
1442 | */ | |
1443 | static int write_relayd_metadata_id(int fd, | |
1444 | struct lttng_consumer_stream *stream, | |
ffe60014 | 1445 | struct consumer_relayd_sock_pair *relayd, unsigned long padding) |
6197aea7 | 1446 | { |
6cd525e8 | 1447 | ssize_t ret; |
1d4dfdef | 1448 | struct lttcomm_relayd_metadata_payload hdr; |
6197aea7 | 1449 | |
1d4dfdef DG |
1450 | hdr.stream_id = htobe64(stream->relayd_stream_id); |
1451 | hdr.padding_size = htobe32(padding); | |
6cd525e8 MD |
1452 | ret = lttng_write(fd, (void *) &hdr, sizeof(hdr)); |
1453 | if (ret < sizeof(hdr)) { | |
d7b75ec8 | 1454 | /* |
6f04ed72 | 1455 | * This error means that the fd's end is closed so ignore the PERROR |
d7b75ec8 DG |
1456 | * not to clubber the error output since this can happen in a normal |
1457 | * code path. | |
1458 | */ | |
1459 | if (errno != EPIPE) { | |
1460 | PERROR("write metadata stream id"); | |
1461 | } | |
1462 | DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno); | |
534d2592 DG |
1463 | /* |
1464 | * Set ret to a negative value because if ret != sizeof(hdr), we don't | |
1465 | * handle writting the missing part so report that as an error and | |
1466 | * don't lie to the caller. | |
1467 | */ | |
1468 | ret = -1; | |
6197aea7 DG |
1469 | goto end; |
1470 | } | |
1d4dfdef DG |
1471 | DBG("Metadata stream id %" PRIu64 " with padding %lu written before data", |
1472 | stream->relayd_stream_id, padding); | |
6197aea7 DG |
1473 | |
1474 | end: | |
6cd525e8 | 1475 | return (int) ret; |
6197aea7 DG |
1476 | } |
1477 | ||
3bd1e081 | 1478 | /* |
09e26845 DG |
1479 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
1480 | * core function for writing trace buffers to either the local filesystem or | |
1481 | * the network. | |
1482 | * | |
79d4ffb7 DG |
1483 | * It must be called with the stream lock held. |
1484 | * | |
09e26845 | 1485 | * Careful review MUST be put if any changes occur! |
3bd1e081 MD |
1486 | * |
1487 | * Returns the number of bytes written | |
1488 | */ | |
4078b776 | 1489 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
3bd1e081 | 1490 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1491 | struct lttng_consumer_stream *stream, unsigned long len, |
309167d2 | 1492 | unsigned long padding, |
50adc264 | 1493 | struct ctf_packet_index *index) |
3bd1e081 | 1494 | { |
f02e1e8a | 1495 | unsigned long mmap_offset; |
ffe60014 | 1496 | void *mmap_base; |
994ab360 | 1497 | ssize_t ret = 0; |
f02e1e8a DG |
1498 | off_t orig_offset = stream->out_fd_offset; |
1499 | /* Default is on the disk */ | |
1500 | int outfd = stream->out_fd; | |
f02e1e8a | 1501 | struct consumer_relayd_sock_pair *relayd = NULL; |
8994307f | 1502 | unsigned int relayd_hang_up = 0; |
f02e1e8a DG |
1503 | |
1504 | /* RCU lock for the relayd pointer */ | |
1505 | rcu_read_lock(); | |
1506 | ||
1507 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1508 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1509 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1510 | if (relayd == NULL) { | |
56591bac | 1511 | ret = -EPIPE; |
f02e1e8a DG |
1512 | goto end; |
1513 | } | |
1514 | } | |
1515 | ||
1516 | /* get the offset inside the fd to mmap */ | |
3bd1e081 MD |
1517 | switch (consumer_data.type) { |
1518 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1519 | mmap_base = stream->mmap_base; |
f02e1e8a | 1520 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); |
994ab360 DG |
1521 | if (ret < 0) { |
1522 | ret = -errno; | |
56591bac | 1523 | PERROR("tracer ctl get_mmap_read_offset"); |
56591bac MD |
1524 | goto end; |
1525 | } | |
f02e1e8a | 1526 | break; |
7753dea8 MD |
1527 | case LTTNG_CONSUMER32_UST: |
1528 | case LTTNG_CONSUMER64_UST: | |
ffe60014 DG |
1529 | mmap_base = lttng_ustctl_get_mmap_base(stream); |
1530 | if (!mmap_base) { | |
1531 | ERR("read mmap get mmap base for stream %s", stream->name); | |
994ab360 | 1532 | ret = -EPERM; |
ffe60014 DG |
1533 | goto end; |
1534 | } | |
1535 | ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset); | |
56591bac MD |
1536 | if (ret != 0) { |
1537 | PERROR("tracer ctl get_mmap_read_offset"); | |
994ab360 | 1538 | ret = -EINVAL; |
56591bac MD |
1539 | goto end; |
1540 | } | |
f02e1e8a | 1541 | break; |
3bd1e081 MD |
1542 | default: |
1543 | ERR("Unknown consumer_data type"); | |
1544 | assert(0); | |
1545 | } | |
b9182dd9 | 1546 | |
f02e1e8a DG |
1547 | /* Handle stream on the relayd if the output is on the network */ |
1548 | if (relayd) { | |
1549 | unsigned long netlen = len; | |
1550 | ||
1551 | /* | |
1552 | * Lock the control socket for the complete duration of the function | |
1553 | * since from this point on we will use the socket. | |
1554 | */ | |
1555 | if (stream->metadata_flag) { | |
1556 | /* Metadata requires the control socket. */ | |
1557 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1d4dfdef | 1558 | netlen += sizeof(struct lttcomm_relayd_metadata_payload); |
f02e1e8a DG |
1559 | } |
1560 | ||
1d4dfdef | 1561 | ret = write_relayd_stream_header(stream, netlen, padding, relayd); |
994ab360 DG |
1562 | if (ret < 0) { |
1563 | relayd_hang_up = 1; | |
1564 | goto write_error; | |
1565 | } | |
1566 | /* Use the returned socket. */ | |
1567 | outfd = ret; | |
f02e1e8a | 1568 | |
994ab360 DG |
1569 | /* Write metadata stream id before payload */ |
1570 | if (stream->metadata_flag) { | |
1571 | ret = write_relayd_metadata_id(outfd, stream, relayd, padding); | |
1572 | if (ret < 0) { | |
8994307f DG |
1573 | relayd_hang_up = 1; |
1574 | goto write_error; | |
1575 | } | |
f02e1e8a | 1576 | } |
1d4dfdef DG |
1577 | } else { |
1578 | /* No streaming, we have to set the len with the full padding */ | |
1579 | len += padding; | |
1624d5b7 JD |
1580 | |
1581 | /* | |
1582 | * Check if we need to change the tracefile before writing the packet. | |
1583 | */ | |
1584 | if (stream->chan->tracefile_size > 0 && | |
1585 | (stream->tracefile_size_current + len) > | |
1586 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1587 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1588 | stream->name, stream->chan->tracefile_size, | |
1589 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
309167d2 JD |
1590 | stream->out_fd, &(stream->tracefile_count_current), |
1591 | &stream->out_fd); | |
1624d5b7 JD |
1592 | if (ret < 0) { |
1593 | ERR("Rotating output file"); | |
1594 | goto end; | |
1595 | } | |
309167d2 JD |
1596 | outfd = stream->out_fd; |
1597 | ||
1598 | if (stream->index_fd >= 0) { | |
1599 | ret = index_create_file(stream->chan->pathname, | |
1600 | stream->name, stream->uid, stream->gid, | |
1601 | stream->chan->tracefile_size, | |
1602 | stream->tracefile_count_current); | |
1603 | if (ret < 0) { | |
1604 | goto end; | |
1605 | } | |
1606 | stream->index_fd = ret; | |
1607 | } | |
1608 | ||
a6976990 DG |
1609 | /* Reset current size because we just perform a rotation. */ |
1610 | stream->tracefile_size_current = 0; | |
a1ae300f JD |
1611 | stream->out_fd_offset = 0; |
1612 | orig_offset = 0; | |
1624d5b7 JD |
1613 | } |
1614 | stream->tracefile_size_current += len; | |
309167d2 JD |
1615 | if (index) { |
1616 | index->offset = htobe64(stream->out_fd_offset); | |
1617 | } | |
f02e1e8a DG |
1618 | } |
1619 | ||
d02b8372 DG |
1620 | /* |
1621 | * This call guarantee that len or less is returned. It's impossible to | |
1622 | * receive a ret value that is bigger than len. | |
1623 | */ | |
1624 | ret = lttng_write(outfd, mmap_base + mmap_offset, len); | |
1625 | DBG("Consumer mmap write() ret %zd (len %lu)", ret, len); | |
1626 | if (ret < 0 || ((size_t) ret != len)) { | |
1627 | /* | |
1628 | * Report error to caller if nothing was written else at least send the | |
1629 | * amount written. | |
1630 | */ | |
1631 | if (ret < 0) { | |
994ab360 | 1632 | ret = -errno; |
f02e1e8a | 1633 | } |
994ab360 | 1634 | relayd_hang_up = 1; |
f02e1e8a | 1635 | |
d02b8372 | 1636 | /* Socket operation failed. We consider the relayd dead */ |
994ab360 | 1637 | if (errno == EPIPE || errno == EINVAL || errno == EBADF) { |
d02b8372 DG |
1638 | /* |
1639 | * This is possible if the fd is closed on the other side | |
1640 | * (outfd) or any write problem. It can be verbose a bit for a | |
1641 | * normal execution if for instance the relayd is stopped | |
1642 | * abruptly. This can happen so set this to a DBG statement. | |
1643 | */ | |
1644 | DBG("Consumer mmap write detected relayd hang up"); | |
994ab360 DG |
1645 | } else { |
1646 | /* Unhandled error, print it and stop function right now. */ | |
1647 | PERROR("Error in write mmap (ret %zd != len %lu)", ret, len); | |
f02e1e8a | 1648 | } |
994ab360 | 1649 | goto write_error; |
d02b8372 DG |
1650 | } |
1651 | stream->output_written += ret; | |
d02b8372 DG |
1652 | |
1653 | /* This call is useless on a socket so better save a syscall. */ | |
1654 | if (!relayd) { | |
1655 | /* This won't block, but will start writeout asynchronously */ | |
1656 | lttng_sync_file_range(outfd, stream->out_fd_offset, len, | |
1657 | SYNC_FILE_RANGE_WRITE); | |
1658 | stream->out_fd_offset += len; | |
f02e1e8a DG |
1659 | } |
1660 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1661 | ||
8994307f DG |
1662 | write_error: |
1663 | /* | |
1664 | * This is a special case that the relayd has closed its socket. Let's | |
1665 | * cleanup the relayd object and all associated streams. | |
1666 | */ | |
1667 | if (relayd && relayd_hang_up) { | |
1668 | cleanup_relayd(relayd, ctx); | |
1669 | } | |
1670 | ||
f02e1e8a DG |
1671 | end: |
1672 | /* Unlock only if ctrl socket used */ | |
1673 | if (relayd && stream->metadata_flag) { | |
1674 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1675 | } | |
1676 | ||
1677 | rcu_read_unlock(); | |
994ab360 | 1678 | return ret; |
3bd1e081 MD |
1679 | } |
1680 | ||
1681 | /* | |
1682 | * Splice the data from the ring buffer to the tracefile. | |
1683 | * | |
79d4ffb7 DG |
1684 | * It must be called with the stream lock held. |
1685 | * | |
3bd1e081 MD |
1686 | * Returns the number of bytes spliced. |
1687 | */ | |
4078b776 | 1688 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
3bd1e081 | 1689 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1690 | struct lttng_consumer_stream *stream, unsigned long len, |
309167d2 | 1691 | unsigned long padding, |
50adc264 | 1692 | struct ctf_packet_index *index) |
3bd1e081 | 1693 | { |
f02e1e8a DG |
1694 | ssize_t ret = 0, written = 0, ret_splice = 0; |
1695 | loff_t offset = 0; | |
1696 | off_t orig_offset = stream->out_fd_offset; | |
1697 | int fd = stream->wait_fd; | |
1698 | /* Default is on the disk */ | |
1699 | int outfd = stream->out_fd; | |
f02e1e8a | 1700 | struct consumer_relayd_sock_pair *relayd = NULL; |
fb3a43a9 | 1701 | int *splice_pipe; |
8994307f | 1702 | unsigned int relayd_hang_up = 0; |
f02e1e8a | 1703 | |
3bd1e081 MD |
1704 | switch (consumer_data.type) { |
1705 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a | 1706 | break; |
7753dea8 MD |
1707 | case LTTNG_CONSUMER32_UST: |
1708 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a | 1709 | /* Not supported for user space tracing */ |
3bd1e081 MD |
1710 | return -ENOSYS; |
1711 | default: | |
1712 | ERR("Unknown consumer_data type"); | |
1713 | assert(0); | |
3bd1e081 MD |
1714 | } |
1715 | ||
f02e1e8a DG |
1716 | /* RCU lock for the relayd pointer */ |
1717 | rcu_read_lock(); | |
1718 | ||
1719 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1720 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1721 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1722 | if (relayd == NULL) { | |
ad0b0d23 | 1723 | written = -ret; |
f02e1e8a DG |
1724 | goto end; |
1725 | } | |
1726 | } | |
a2361a61 | 1727 | splice_pipe = stream->splice_pipe; |
fb3a43a9 | 1728 | |
f02e1e8a | 1729 | /* Write metadata stream id before payload */ |
1d4dfdef | 1730 | if (relayd) { |
ad0b0d23 | 1731 | unsigned long total_len = len; |
f02e1e8a | 1732 | |
1d4dfdef DG |
1733 | if (stream->metadata_flag) { |
1734 | /* | |
1735 | * Lock the control socket for the complete duration of the function | |
1736 | * since from this point on we will use the socket. | |
1737 | */ | |
1738 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1739 | ||
1740 | ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd, | |
1741 | padding); | |
1742 | if (ret < 0) { | |
1743 | written = ret; | |
ad0b0d23 DG |
1744 | relayd_hang_up = 1; |
1745 | goto write_error; | |
1d4dfdef DG |
1746 | } |
1747 | ||
1748 | total_len += sizeof(struct lttcomm_relayd_metadata_payload); | |
1749 | } | |
1750 | ||
1751 | ret = write_relayd_stream_header(stream, total_len, padding, relayd); | |
ad0b0d23 DG |
1752 | if (ret < 0) { |
1753 | written = ret; | |
1754 | relayd_hang_up = 1; | |
1755 | goto write_error; | |
f02e1e8a | 1756 | } |
ad0b0d23 DG |
1757 | /* Use the returned socket. */ |
1758 | outfd = ret; | |
1d4dfdef DG |
1759 | } else { |
1760 | /* No streaming, we have to set the len with the full padding */ | |
1761 | len += padding; | |
1624d5b7 JD |
1762 | |
1763 | /* | |
1764 | * Check if we need to change the tracefile before writing the packet. | |
1765 | */ | |
1766 | if (stream->chan->tracefile_size > 0 && | |
1767 | (stream->tracefile_size_current + len) > | |
1768 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1769 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1770 | stream->name, stream->chan->tracefile_size, | |
1771 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
309167d2 JD |
1772 | stream->out_fd, &(stream->tracefile_count_current), |
1773 | &stream->out_fd); | |
1624d5b7 | 1774 | if (ret < 0) { |
ad0b0d23 | 1775 | written = ret; |
1624d5b7 JD |
1776 | ERR("Rotating output file"); |
1777 | goto end; | |
1778 | } | |
309167d2 JD |
1779 | outfd = stream->out_fd; |
1780 | ||
1781 | if (stream->index_fd >= 0) { | |
1782 | ret = index_create_file(stream->chan->pathname, | |
1783 | stream->name, stream->uid, stream->gid, | |
1784 | stream->chan->tracefile_size, | |
1785 | stream->tracefile_count_current); | |
1786 | if (ret < 0) { | |
ad0b0d23 | 1787 | written = ret; |
309167d2 JD |
1788 | goto end; |
1789 | } | |
1790 | stream->index_fd = ret; | |
1791 | } | |
1792 | ||
a6976990 DG |
1793 | /* Reset current size because we just perform a rotation. */ |
1794 | stream->tracefile_size_current = 0; | |
a1ae300f JD |
1795 | stream->out_fd_offset = 0; |
1796 | orig_offset = 0; | |
1624d5b7 JD |
1797 | } |
1798 | stream->tracefile_size_current += len; | |
309167d2 | 1799 | index->offset = htobe64(stream->out_fd_offset); |
f02e1e8a DG |
1800 | } |
1801 | ||
1802 | while (len > 0) { | |
1d4dfdef DG |
1803 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)", |
1804 | (unsigned long)offset, len, fd, splice_pipe[1]); | |
fb3a43a9 | 1805 | ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len, |
f02e1e8a DG |
1806 | SPLICE_F_MOVE | SPLICE_F_MORE); |
1807 | DBG("splice chan to pipe, ret %zd", ret_splice); | |
1808 | if (ret_splice < 0) { | |
d02b8372 | 1809 | ret = errno; |
ad0b0d23 | 1810 | written = -ret; |
d02b8372 | 1811 | PERROR("Error in relay splice"); |
f02e1e8a DG |
1812 | goto splice_error; |
1813 | } | |
1814 | ||
1815 | /* Handle stream on the relayd if the output is on the network */ | |
ad0b0d23 DG |
1816 | if (relayd && stream->metadata_flag) { |
1817 | size_t metadata_payload_size = | |
1818 | sizeof(struct lttcomm_relayd_metadata_payload); | |
1819 | ||
1820 | /* Update counter to fit the spliced data */ | |
1821 | ret_splice += metadata_payload_size; | |
1822 | len += metadata_payload_size; | |
1823 | /* | |
1824 | * We do this so the return value can match the len passed as | |
1825 | * argument to this function. | |
1826 | */ | |
1827 | written -= metadata_payload_size; | |
f02e1e8a DG |
1828 | } |
1829 | ||
1830 | /* Splice data out */ | |
fb3a43a9 | 1831 | ret_splice = splice(splice_pipe[0], NULL, outfd, NULL, |
f02e1e8a | 1832 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); |
a2361a61 JD |
1833 | DBG("Consumer splice pipe to file (out_fd: %d), ret %zd", |
1834 | outfd, ret_splice); | |
f02e1e8a | 1835 | if (ret_splice < 0) { |
d02b8372 | 1836 | ret = errno; |
ad0b0d23 DG |
1837 | written = -ret; |
1838 | relayd_hang_up = 1; | |
1839 | goto write_error; | |
f02e1e8a | 1840 | } else if (ret_splice > len) { |
d02b8372 DG |
1841 | /* |
1842 | * We don't expect this code path to be executed but you never know | |
1843 | * so this is an extra protection agains a buggy splice(). | |
1844 | */ | |
f02e1e8a | 1845 | ret = errno; |
ad0b0d23 | 1846 | written += ret_splice; |
d02b8372 DG |
1847 | PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice, |
1848 | len); | |
f02e1e8a | 1849 | goto splice_error; |
d02b8372 DG |
1850 | } else { |
1851 | /* All good, update current len and continue. */ | |
1852 | len -= ret_splice; | |
f02e1e8a | 1853 | } |
f02e1e8a DG |
1854 | |
1855 | /* This call is useless on a socket so better save a syscall. */ | |
1856 | if (!relayd) { | |
1857 | /* This won't block, but will start writeout asynchronously */ | |
1858 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, | |
1859 | SYNC_FILE_RANGE_WRITE); | |
1860 | stream->out_fd_offset += ret_splice; | |
1861 | } | |
e5d1a9b3 | 1862 | stream->output_written += ret_splice; |
f02e1e8a DG |
1863 | written += ret_splice; |
1864 | } | |
1865 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
f02e1e8a DG |
1866 | goto end; |
1867 | ||
8994307f DG |
1868 | write_error: |
1869 | /* | |
1870 | * This is a special case that the relayd has closed its socket. Let's | |
1871 | * cleanup the relayd object and all associated streams. | |
1872 | */ | |
1873 | if (relayd && relayd_hang_up) { | |
1874 | cleanup_relayd(relayd, ctx); | |
1875 | /* Skip splice error so the consumer does not fail */ | |
1876 | goto end; | |
1877 | } | |
1878 | ||
f02e1e8a DG |
1879 | splice_error: |
1880 | /* send the appropriate error description to sessiond */ | |
1881 | switch (ret) { | |
f02e1e8a | 1882 | case EINVAL: |
f73fabfd | 1883 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); |
f02e1e8a DG |
1884 | break; |
1885 | case ENOMEM: | |
f73fabfd | 1886 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); |
f02e1e8a DG |
1887 | break; |
1888 | case ESPIPE: | |
f73fabfd | 1889 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); |
f02e1e8a DG |
1890 | break; |
1891 | } | |
1892 | ||
1893 | end: | |
1894 | if (relayd && stream->metadata_flag) { | |
1895 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1896 | } | |
1897 | ||
1898 | rcu_read_unlock(); | |
1899 | return written; | |
3bd1e081 MD |
1900 | } |
1901 | ||
1902 | /* | |
1903 | * Take a snapshot for a specific fd | |
1904 | * | |
1905 | * Returns 0 on success, < 0 on error | |
1906 | */ | |
ffe60014 | 1907 | int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) |
3bd1e081 MD |
1908 | { |
1909 | switch (consumer_data.type) { | |
1910 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1911 | return lttng_kconsumer_take_snapshot(stream); |
7753dea8 MD |
1912 | case LTTNG_CONSUMER32_UST: |
1913 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 1914 | return lttng_ustconsumer_take_snapshot(stream); |
3bd1e081 MD |
1915 | default: |
1916 | ERR("Unknown consumer_data type"); | |
1917 | assert(0); | |
1918 | return -ENOSYS; | |
1919 | } | |
3bd1e081 MD |
1920 | } |
1921 | ||
1922 | /* | |
1923 | * Get the produced position | |
1924 | * | |
1925 | * Returns 0 on success, < 0 on error | |
1926 | */ | |
ffe60014 | 1927 | int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, |
3bd1e081 MD |
1928 | unsigned long *pos) |
1929 | { | |
1930 | switch (consumer_data.type) { | |
1931 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1932 | return lttng_kconsumer_get_produced_snapshot(stream, pos); |
7753dea8 MD |
1933 | case LTTNG_CONSUMER32_UST: |
1934 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 1935 | return lttng_ustconsumer_get_produced_snapshot(stream, pos); |
3bd1e081 MD |
1936 | default: |
1937 | ERR("Unknown consumer_data type"); | |
1938 | assert(0); | |
1939 | return -ENOSYS; | |
1940 | } | |
1941 | } | |
1942 | ||
1943 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, | |
1944 | int sock, struct pollfd *consumer_sockpoll) | |
1945 | { | |
1946 | switch (consumer_data.type) { | |
1947 | case LTTNG_CONSUMER_KERNEL: | |
1948 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
7753dea8 MD |
1949 | case LTTNG_CONSUMER32_UST: |
1950 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1951 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
1952 | default: | |
1953 | ERR("Unknown consumer_data type"); | |
1954 | assert(0); | |
1955 | return -ENOSYS; | |
1956 | } | |
1957 | } | |
1958 | ||
6d574024 | 1959 | void lttng_consumer_close_all_metadata(void) |
d88aee68 DG |
1960 | { |
1961 | switch (consumer_data.type) { | |
1962 | case LTTNG_CONSUMER_KERNEL: | |
1963 | /* | |
1964 | * The Kernel consumer has a different metadata scheme so we don't | |
1965 | * close anything because the stream will be closed by the session | |
1966 | * daemon. | |
1967 | */ | |
1968 | break; | |
1969 | case LTTNG_CONSUMER32_UST: | |
1970 | case LTTNG_CONSUMER64_UST: | |
1971 | /* | |
1972 | * Close all metadata streams. The metadata hash table is passed and | |
1973 | * this call iterates over it by closing all wakeup fd. This is safe | |
1974 | * because at this point we are sure that the metadata producer is | |
1975 | * either dead or blocked. | |
1976 | */ | |
6d574024 | 1977 | lttng_ustconsumer_close_all_metadata(metadata_ht); |
d88aee68 DG |
1978 | break; |
1979 | default: | |
1980 | ERR("Unknown consumer_data type"); | |
1981 | assert(0); | |
1982 | } | |
1983 | } | |
1984 | ||
fb3a43a9 DG |
1985 | /* |
1986 | * Clean up a metadata stream and free its memory. | |
1987 | */ | |
e316aad5 DG |
1988 | void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, |
1989 | struct lttng_ht *ht) | |
fb3a43a9 | 1990 | { |
e316aad5 | 1991 | struct lttng_consumer_channel *free_chan = NULL; |
fb3a43a9 DG |
1992 | |
1993 | assert(stream); | |
1994 | /* | |
1995 | * This call should NEVER receive regular stream. It must always be | |
1996 | * metadata stream and this is crucial for data structure synchronization. | |
1997 | */ | |
1998 | assert(stream->metadata_flag); | |
1999 | ||
e316aad5 DG |
2000 | DBG3("Consumer delete metadata stream %d", stream->wait_fd); |
2001 | ||
74251bb8 | 2002 | pthread_mutex_lock(&consumer_data.lock); |
a9838785 | 2003 | pthread_mutex_lock(&stream->chan->lock); |
8994307f DG |
2004 | pthread_mutex_lock(&stream->lock); |
2005 | ||
6d574024 DG |
2006 | /* Remove any reference to that stream. */ |
2007 | consumer_stream_delete(stream, ht); | |
ca22feea | 2008 | |
6d574024 DG |
2009 | /* Close down everything including the relayd if one. */ |
2010 | consumer_stream_close(stream); | |
2011 | /* Destroy tracer buffers of the stream. */ | |
2012 | consumer_stream_destroy_buffers(stream); | |
fb3a43a9 DG |
2013 | |
2014 | /* Atomically decrement channel refcount since other threads can use it. */ | |
f2ad556d | 2015 | if (!uatomic_sub_return(&stream->chan->refcount, 1) |
ffe60014 | 2016 | && !uatomic_read(&stream->chan->nb_init_stream_left)) { |
c30aaa51 | 2017 | /* Go for channel deletion! */ |
e316aad5 | 2018 | free_chan = stream->chan; |
fb3a43a9 DG |
2019 | } |
2020 | ||
73811ecc DG |
2021 | /* |
2022 | * Nullify the stream reference so it is not used after deletion. The | |
6d574024 DG |
2023 | * channel lock MUST be acquired before being able to check for a NULL |
2024 | * pointer value. | |
73811ecc DG |
2025 | */ |
2026 | stream->chan->metadata_stream = NULL; | |
2027 | ||
8994307f | 2028 | pthread_mutex_unlock(&stream->lock); |
a9838785 | 2029 | pthread_mutex_unlock(&stream->chan->lock); |
74251bb8 | 2030 | pthread_mutex_unlock(&consumer_data.lock); |
e316aad5 DG |
2031 | |
2032 | if (free_chan) { | |
2033 | consumer_del_channel(free_chan); | |
2034 | } | |
2035 | ||
6d574024 | 2036 | consumer_stream_free(stream); |
fb3a43a9 DG |
2037 | } |
2038 | ||
2039 | /* | |
2040 | * Action done with the metadata stream when adding it to the consumer internal | |
2041 | * data structures to handle it. | |
2042 | */ | |
5ab66908 | 2043 | int consumer_add_metadata_stream(struct lttng_consumer_stream *stream) |
fb3a43a9 | 2044 | { |
5ab66908 | 2045 | struct lttng_ht *ht = metadata_ht; |
e316aad5 | 2046 | int ret = 0; |
76082088 | 2047 | struct lttng_ht_iter iter; |
d88aee68 | 2048 | struct lttng_ht_node_u64 *node; |
fb3a43a9 | 2049 | |
e316aad5 DG |
2050 | assert(stream); |
2051 | assert(ht); | |
2052 | ||
d88aee68 | 2053 | DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key); |
e316aad5 DG |
2054 | |
2055 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 2056 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 2057 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 2058 | pthread_mutex_lock(&stream->lock); |
e316aad5 | 2059 | |
e316aad5 DG |
2060 | /* |
2061 | * From here, refcounts are updated so be _careful_ when returning an error | |
2062 | * after this point. | |
2063 | */ | |
2064 | ||
fb3a43a9 | 2065 | rcu_read_lock(); |
76082088 DG |
2066 | |
2067 | /* | |
2068 | * Lookup the stream just to make sure it does not exist in our internal | |
2069 | * state. This should NEVER happen. | |
2070 | */ | |
d88aee68 DG |
2071 | lttng_ht_lookup(ht, &stream->key, &iter); |
2072 | node = lttng_ht_iter_get_node_u64(&iter); | |
76082088 DG |
2073 | assert(!node); |
2074 | ||
e316aad5 | 2075 | /* |
ffe60014 DG |
2076 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
2077 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
2078 | * causes the count to become 0 also causes a stream to be added. The |
2079 | * channel deletion will thus be triggered by the following removal of this | |
2080 | * stream. | |
2081 | */ | |
ffe60014 | 2082 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
2083 | /* Increment refcount before decrementing nb_init_stream_left */ |
2084 | cmm_smp_wmb(); | |
ffe60014 | 2085 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
2086 | } |
2087 | ||
d88aee68 | 2088 | lttng_ht_add_unique_u64(ht, &stream->node); |
ca22feea | 2089 | |
d8ef542d MD |
2090 | lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht, |
2091 | &stream->node_channel_id); | |
2092 | ||
ca22feea DG |
2093 | /* |
2094 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
2095 | * the key since the HT does not use it and we allow to add redundant keys | |
2096 | * into this table. | |
2097 | */ | |
d88aee68 | 2098 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 2099 | |
fb3a43a9 | 2100 | rcu_read_unlock(); |
e316aad5 | 2101 | |
2e818a6a | 2102 | pthread_mutex_unlock(&stream->lock); |
a9838785 | 2103 | pthread_mutex_unlock(&stream->chan->lock); |
ec6ea7d0 | 2104 | pthread_mutex_unlock(&stream->chan->timer_lock); |
e316aad5 DG |
2105 | pthread_mutex_unlock(&consumer_data.lock); |
2106 | return ret; | |
fb3a43a9 DG |
2107 | } |
2108 | ||
8994307f DG |
2109 | /* |
2110 | * Delete data stream that are flagged for deletion (endpoint_status). | |
2111 | */ | |
2112 | static void validate_endpoint_status_data_stream(void) | |
2113 | { | |
2114 | struct lttng_ht_iter iter; | |
2115 | struct lttng_consumer_stream *stream; | |
2116 | ||
2117 | DBG("Consumer delete flagged data stream"); | |
2118 | ||
2119 | rcu_read_lock(); | |
2120 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
2121 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2122 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2123 | continue; |
2124 | } | |
2125 | /* Delete it right now */ | |
2126 | consumer_del_stream(stream, data_ht); | |
2127 | } | |
2128 | rcu_read_unlock(); | |
2129 | } | |
2130 | ||
2131 | /* | |
2132 | * Delete metadata stream that are flagged for deletion (endpoint_status). | |
2133 | */ | |
2134 | static void validate_endpoint_status_metadata_stream( | |
2135 | struct lttng_poll_event *pollset) | |
2136 | { | |
2137 | struct lttng_ht_iter iter; | |
2138 | struct lttng_consumer_stream *stream; | |
2139 | ||
2140 | DBG("Consumer delete flagged metadata stream"); | |
2141 | ||
2142 | assert(pollset); | |
2143 | ||
2144 | rcu_read_lock(); | |
2145 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
2146 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2147 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2148 | continue; |
2149 | } | |
2150 | /* | |
2151 | * Remove from pollset so the metadata thread can continue without | |
2152 | * blocking on a deleted stream. | |
2153 | */ | |
2154 | lttng_poll_del(pollset, stream->wait_fd); | |
2155 | ||
2156 | /* Delete it right now */ | |
2157 | consumer_del_metadata_stream(stream, metadata_ht); | |
2158 | } | |
2159 | rcu_read_unlock(); | |
2160 | } | |
2161 | ||
fb3a43a9 DG |
2162 | /* |
2163 | * Thread polls on metadata file descriptor and write them on disk or on the | |
2164 | * network. | |
2165 | */ | |
7d980def | 2166 | void *consumer_thread_metadata_poll(void *data) |
fb3a43a9 | 2167 | { |
1fc79fb4 | 2168 | int ret, i, pollfd, err = -1; |
fb3a43a9 | 2169 | uint32_t revents, nb_fd; |
e316aad5 | 2170 | struct lttng_consumer_stream *stream = NULL; |
fb3a43a9 | 2171 | struct lttng_ht_iter iter; |
d88aee68 | 2172 | struct lttng_ht_node_u64 *node; |
fb3a43a9 DG |
2173 | struct lttng_poll_event events; |
2174 | struct lttng_consumer_local_data *ctx = data; | |
2175 | ssize_t len; | |
2176 | ||
2177 | rcu_register_thread(); | |
2178 | ||
1fc79fb4 MD |
2179 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA); |
2180 | ||
2d57de81 MD |
2181 | if (testpoint(consumerd_thread_metadata)) { |
2182 | goto error_testpoint; | |
2183 | } | |
2184 | ||
9ce5646a MD |
2185 | health_code_update(); |
2186 | ||
fb3a43a9 DG |
2187 | DBG("Thread metadata poll started"); |
2188 | ||
fb3a43a9 DG |
2189 | /* Size is set to 1 for the consumer_metadata pipe */ |
2190 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2191 | if (ret < 0) { | |
2192 | ERR("Poll set creation failed"); | |
d8ef542d | 2193 | goto end_poll; |
fb3a43a9 DG |
2194 | } |
2195 | ||
13886d2d DG |
2196 | ret = lttng_poll_add(&events, |
2197 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN); | |
fb3a43a9 DG |
2198 | if (ret < 0) { |
2199 | goto end; | |
2200 | } | |
2201 | ||
2202 | /* Main loop */ | |
2203 | DBG("Metadata main loop started"); | |
2204 | ||
2205 | while (1) { | |
fb3a43a9 | 2206 | restart: |
7fa2082e | 2207 | health_code_update(); |
9ce5646a | 2208 | health_poll_entry(); |
7fa2082e | 2209 | DBG("Metadata poll wait"); |
fb3a43a9 | 2210 | ret = lttng_poll_wait(&events, -1); |
7fa2082e MD |
2211 | DBG("Metadata poll return from wait with %d fd(s)", |
2212 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2213 | health_poll_exit(); |
fb3a43a9 DG |
2214 | DBG("Metadata event catched in thread"); |
2215 | if (ret < 0) { | |
2216 | if (errno == EINTR) { | |
e316aad5 | 2217 | ERR("Poll EINTR catched"); |
fb3a43a9 DG |
2218 | goto restart; |
2219 | } | |
d9607cd7 MD |
2220 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2221 | err = 0; /* All is OK */ | |
2222 | } | |
2223 | goto end; | |
fb3a43a9 DG |
2224 | } |
2225 | ||
0d9c5d77 DG |
2226 | nb_fd = ret; |
2227 | ||
e316aad5 | 2228 | /* From here, the event is a metadata wait fd */ |
fb3a43a9 | 2229 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2230 | health_code_update(); |
2231 | ||
fb3a43a9 DG |
2232 | revents = LTTNG_POLL_GETEV(&events, i); |
2233 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2234 | ||
fd20dac9 MD |
2235 | if (!revents) { |
2236 | /* No activity for this FD (poll implementation). */ | |
2237 | continue; | |
2238 | } | |
2239 | ||
13886d2d | 2240 | if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) { |
03e43155 | 2241 | if (revents & LPOLLIN) { |
13886d2d DG |
2242 | ssize_t pipe_len; |
2243 | ||
2244 | pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe, | |
2245 | &stream, sizeof(stream)); | |
6cd525e8 | 2246 | if (pipe_len < sizeof(stream)) { |
03e43155 MD |
2247 | if (pipe_len < 0) { |
2248 | PERROR("read metadata stream"); | |
2249 | } | |
fb3a43a9 | 2250 | /* |
03e43155 MD |
2251 | * Remove the pipe from the poll set and continue the loop |
2252 | * since their might be data to consume. | |
fb3a43a9 | 2253 | */ |
03e43155 MD |
2254 | lttng_poll_del(&events, |
2255 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2256 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
fb3a43a9 DG |
2257 | continue; |
2258 | } | |
2259 | ||
8994307f DG |
2260 | /* A NULL stream means that the state has changed. */ |
2261 | if (stream == NULL) { | |
2262 | /* Check for deleted streams. */ | |
2263 | validate_endpoint_status_metadata_stream(&events); | |
3714380f | 2264 | goto restart; |
8994307f DG |
2265 | } |
2266 | ||
fb3a43a9 DG |
2267 | DBG("Adding metadata stream %d to poll set", |
2268 | stream->wait_fd); | |
2269 | ||
fb3a43a9 DG |
2270 | /* Add metadata stream to the global poll events list */ |
2271 | lttng_poll_add(&events, stream->wait_fd, | |
6d574024 | 2272 | LPOLLIN | LPOLLPRI | LPOLLHUP); |
03e43155 MD |
2273 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
2274 | DBG("Metadata thread pipe hung up"); | |
2275 | /* | |
2276 | * Remove the pipe from the poll set and continue the loop | |
2277 | * since their might be data to consume. | |
2278 | */ | |
2279 | lttng_poll_del(&events, | |
2280 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2281 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
2282 | continue; | |
2283 | } else { | |
2284 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
2285 | goto end; | |
fb3a43a9 DG |
2286 | } |
2287 | ||
e316aad5 | 2288 | /* Handle other stream */ |
fb3a43a9 DG |
2289 | continue; |
2290 | } | |
2291 | ||
d09e1200 | 2292 | rcu_read_lock(); |
d88aee68 DG |
2293 | { |
2294 | uint64_t tmp_id = (uint64_t) pollfd; | |
2295 | ||
2296 | lttng_ht_lookup(metadata_ht, &tmp_id, &iter); | |
2297 | } | |
2298 | node = lttng_ht_iter_get_node_u64(&iter); | |
e316aad5 | 2299 | assert(node); |
fb3a43a9 DG |
2300 | |
2301 | stream = caa_container_of(node, struct lttng_consumer_stream, | |
58b1f425 | 2302 | node); |
fb3a43a9 | 2303 | |
03e43155 MD |
2304 | if (revents & (LPOLLIN | LPOLLPRI)) { |
2305 | /* Get the data out of the metadata file descriptor */ | |
2306 | DBG("Metadata available on fd %d", pollfd); | |
2307 | assert(stream->wait_fd == pollfd); | |
2308 | ||
2309 | do { | |
2310 | health_code_update(); | |
2311 | ||
2312 | len = ctx->on_buffer_ready(stream, ctx); | |
2313 | /* | |
2314 | * We don't check the return value here since if we get | |
2315 | * a negative len, it means an error occured thus we | |
2316 | * simply remove it from the poll set and free the | |
2317 | * stream. | |
2318 | */ | |
2319 | } while (len > 0); | |
2320 | ||
2321 | /* It's ok to have an unavailable sub-buffer */ | |
2322 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { | |
2323 | /* Clean up stream from consumer and free it. */ | |
2324 | lttng_poll_del(&events, stream->wait_fd); | |
2325 | consumer_del_metadata_stream(stream, metadata_ht); | |
2326 | } | |
2327 | } else if (revents & (LPOLLERR | LPOLLHUP)) { | |
e316aad5 | 2328 | DBG("Metadata fd %d is hup|err.", pollfd); |
fb3a43a9 DG |
2329 | if (!stream->hangup_flush_done |
2330 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2331 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2332 | DBG("Attempting to flush and consume the UST buffers"); | |
2333 | lttng_ustconsumer_on_stream_hangup(stream); | |
2334 | ||
2335 | /* We just flushed the stream now read it. */ | |
4bb94b75 | 2336 | do { |
9ce5646a MD |
2337 | health_code_update(); |
2338 | ||
4bb94b75 DG |
2339 | len = ctx->on_buffer_ready(stream, ctx); |
2340 | /* | |
2341 | * We don't check the return value here since if we get | |
2342 | * a negative len, it means an error occured thus we | |
2343 | * simply remove it from the poll set and free the | |
2344 | * stream. | |
2345 | */ | |
2346 | } while (len > 0); | |
fb3a43a9 DG |
2347 | } |
2348 | ||
fb3a43a9 | 2349 | lttng_poll_del(&events, stream->wait_fd); |
e316aad5 DG |
2350 | /* |
2351 | * This call update the channel states, closes file descriptors | |
2352 | * and securely free the stream. | |
2353 | */ | |
2354 | consumer_del_metadata_stream(stream, metadata_ht); | |
03e43155 MD |
2355 | } else { |
2356 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
6f2f1a70 | 2357 | rcu_read_unlock(); |
03e43155 | 2358 | goto end; |
fb3a43a9 | 2359 | } |
e316aad5 | 2360 | /* Release RCU lock for the stream looked up */ |
d09e1200 | 2361 | rcu_read_unlock(); |
fb3a43a9 DG |
2362 | } |
2363 | } | |
2364 | ||
1fc79fb4 MD |
2365 | /* All is OK */ |
2366 | err = 0; | |
fb3a43a9 DG |
2367 | end: |
2368 | DBG("Metadata poll thread exiting"); | |
fb3a43a9 | 2369 | |
d8ef542d MD |
2370 | lttng_poll_clean(&events); |
2371 | end_poll: | |
2d57de81 | 2372 | error_testpoint: |
1fc79fb4 MD |
2373 | if (err) { |
2374 | health_error(); | |
2375 | ERR("Health error occurred in %s", __func__); | |
2376 | } | |
2377 | health_unregister(health_consumerd); | |
fb3a43a9 DG |
2378 | rcu_unregister_thread(); |
2379 | return NULL; | |
2380 | } | |
2381 | ||
3bd1e081 | 2382 | /* |
e4421fec | 2383 | * This thread polls the fds in the set to consume the data and write |
3bd1e081 MD |
2384 | * it to tracefile if necessary. |
2385 | */ | |
7d980def | 2386 | void *consumer_thread_data_poll(void *data) |
3bd1e081 | 2387 | { |
1fc79fb4 | 2388 | int num_rdy, num_hup, high_prio, ret, i, err = -1; |
3bd1e081 MD |
2389 | struct pollfd *pollfd = NULL; |
2390 | /* local view of the streams */ | |
c869f647 | 2391 | struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; |
3bd1e081 MD |
2392 | /* local view of consumer_data.fds_count */ |
2393 | int nb_fd = 0; | |
3bd1e081 | 2394 | struct lttng_consumer_local_data *ctx = data; |
00e2e675 | 2395 | ssize_t len; |
3bd1e081 | 2396 | |
e7b994a3 DG |
2397 | rcu_register_thread(); |
2398 | ||
1fc79fb4 MD |
2399 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA); |
2400 | ||
2d57de81 MD |
2401 | if (testpoint(consumerd_thread_data)) { |
2402 | goto error_testpoint; | |
2403 | } | |
2404 | ||
9ce5646a MD |
2405 | health_code_update(); |
2406 | ||
4df6c8cb MD |
2407 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream *)); |
2408 | if (local_stream == NULL) { | |
2409 | PERROR("local_stream malloc"); | |
2410 | goto end; | |
2411 | } | |
3bd1e081 MD |
2412 | |
2413 | while (1) { | |
9ce5646a MD |
2414 | health_code_update(); |
2415 | ||
3bd1e081 MD |
2416 | high_prio = 0; |
2417 | num_hup = 0; | |
2418 | ||
2419 | /* | |
e4421fec | 2420 | * the fds set has been updated, we need to update our |
3bd1e081 MD |
2421 | * local array as well |
2422 | */ | |
2423 | pthread_mutex_lock(&consumer_data.lock); | |
2424 | if (consumer_data.need_update) { | |
0e428499 DG |
2425 | free(pollfd); |
2426 | pollfd = NULL; | |
2427 | ||
2428 | free(local_stream); | |
2429 | local_stream = NULL; | |
3bd1e081 | 2430 | |
02b3d176 DG |
2431 | /* |
2432 | * Allocate for all fds +1 for the consumer_data_pipe and +1 for | |
2433 | * wake up pipe. | |
2434 | */ | |
2435 | pollfd = zmalloc((consumer_data.stream_count + 2) * sizeof(struct pollfd)); | |
3bd1e081 | 2436 | if (pollfd == NULL) { |
7a57cf92 | 2437 | PERROR("pollfd malloc"); |
3bd1e081 MD |
2438 | pthread_mutex_unlock(&consumer_data.lock); |
2439 | goto end; | |
2440 | } | |
2441 | ||
02b3d176 | 2442 | local_stream = zmalloc((consumer_data.stream_count + 2) * |
747f8642 | 2443 | sizeof(struct lttng_consumer_stream *)); |
3bd1e081 | 2444 | if (local_stream == NULL) { |
7a57cf92 | 2445 | PERROR("local_stream malloc"); |
3bd1e081 MD |
2446 | pthread_mutex_unlock(&consumer_data.lock); |
2447 | goto end; | |
2448 | } | |
ffe60014 | 2449 | ret = update_poll_array(ctx, &pollfd, local_stream, |
43c34bc3 | 2450 | data_ht); |
3bd1e081 MD |
2451 | if (ret < 0) { |
2452 | ERR("Error in allocating pollfd or local_outfds"); | |
f73fabfd | 2453 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2454 | pthread_mutex_unlock(&consumer_data.lock); |
2455 | goto end; | |
2456 | } | |
2457 | nb_fd = ret; | |
2458 | consumer_data.need_update = 0; | |
2459 | } | |
2460 | pthread_mutex_unlock(&consumer_data.lock); | |
2461 | ||
4078b776 MD |
2462 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
2463 | if (nb_fd == 0 && consumer_quit == 1) { | |
1fc79fb4 | 2464 | err = 0; /* All is OK */ |
4078b776 MD |
2465 | goto end; |
2466 | } | |
3bd1e081 | 2467 | /* poll on the array of fds */ |
88f2b785 | 2468 | restart: |
02b3d176 | 2469 | DBG("polling on %d fd", nb_fd + 2); |
9ce5646a | 2470 | health_poll_entry(); |
02b3d176 | 2471 | num_rdy = poll(pollfd, nb_fd + 2, -1); |
9ce5646a | 2472 | health_poll_exit(); |
3bd1e081 MD |
2473 | DBG("poll num_rdy : %d", num_rdy); |
2474 | if (num_rdy == -1) { | |
88f2b785 MD |
2475 | /* |
2476 | * Restart interrupted system call. | |
2477 | */ | |
2478 | if (errno == EINTR) { | |
2479 | goto restart; | |
2480 | } | |
7a57cf92 | 2481 | PERROR("Poll error"); |
f73fabfd | 2482 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2483 | goto end; |
2484 | } else if (num_rdy == 0) { | |
2485 | DBG("Polling thread timed out"); | |
2486 | goto end; | |
2487 | } | |
2488 | ||
3bd1e081 | 2489 | /* |
50f8ae69 | 2490 | * If the consumer_data_pipe triggered poll go directly to the |
00e2e675 DG |
2491 | * beginning of the loop to update the array. We want to prioritize |
2492 | * array update over low-priority reads. | |
3bd1e081 | 2493 | */ |
509bb1cf | 2494 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
ab30f567 | 2495 | ssize_t pipe_readlen; |
04fdd819 | 2496 | |
50f8ae69 | 2497 | DBG("consumer_data_pipe wake up"); |
acdb9057 DG |
2498 | pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe, |
2499 | &new_stream, sizeof(new_stream)); | |
6cd525e8 MD |
2500 | if (pipe_readlen < sizeof(new_stream)) { |
2501 | PERROR("Consumer data pipe"); | |
23f5f35d DG |
2502 | /* Continue so we can at least handle the current stream(s). */ |
2503 | continue; | |
2504 | } | |
c869f647 DG |
2505 | |
2506 | /* | |
2507 | * If the stream is NULL, just ignore it. It's also possible that | |
2508 | * the sessiond poll thread changed the consumer_quit state and is | |
2509 | * waking us up to test it. | |
2510 | */ | |
2511 | if (new_stream == NULL) { | |
8994307f | 2512 | validate_endpoint_status_data_stream(); |
c869f647 DG |
2513 | continue; |
2514 | } | |
2515 | ||
c869f647 | 2516 | /* Continue to update the local streams and handle prio ones */ |
3bd1e081 MD |
2517 | continue; |
2518 | } | |
2519 | ||
02b3d176 DG |
2520 | /* Handle wakeup pipe. */ |
2521 | if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) { | |
2522 | char dummy; | |
2523 | ssize_t pipe_readlen; | |
2524 | ||
2525 | pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy, | |
2526 | sizeof(dummy)); | |
2527 | if (pipe_readlen < 0) { | |
2528 | PERROR("Consumer data wakeup pipe"); | |
2529 | } | |
2530 | /* We've been awakened to handle stream(s). */ | |
2531 | ctx->has_wakeup = 0; | |
2532 | } | |
2533 | ||
3bd1e081 MD |
2534 | /* Take care of high priority channels first. */ |
2535 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2536 | health_code_update(); |
2537 | ||
9617607b DG |
2538 | if (local_stream[i] == NULL) { |
2539 | continue; | |
2540 | } | |
fb3a43a9 | 2541 | if (pollfd[i].revents & POLLPRI) { |
d41f73b7 MD |
2542 | DBG("Urgent read on fd %d", pollfd[i].fd); |
2543 | high_prio = 1; | |
4078b776 | 2544 | len = ctx->on_buffer_ready(local_stream[i], ctx); |
d41f73b7 | 2545 | /* it's ok to have an unavailable sub-buffer */ |
b64403e3 | 2546 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2547 | /* Clean the stream and free it. */ |
2548 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2549 | local_stream[i] = NULL; |
4078b776 MD |
2550 | } else if (len > 0) { |
2551 | local_stream[i]->data_read = 1; | |
d41f73b7 | 2552 | } |
3bd1e081 MD |
2553 | } |
2554 | } | |
2555 | ||
4078b776 MD |
2556 | /* |
2557 | * If we read high prio channel in this loop, try again | |
2558 | * for more high prio data. | |
2559 | */ | |
2560 | if (high_prio) { | |
3bd1e081 MD |
2561 | continue; |
2562 | } | |
2563 | ||
2564 | /* Take care of low priority channels. */ | |
4078b776 | 2565 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2566 | health_code_update(); |
2567 | ||
9617607b DG |
2568 | if (local_stream[i] == NULL) { |
2569 | continue; | |
2570 | } | |
4078b776 | 2571 | if ((pollfd[i].revents & POLLIN) || |
02b3d176 DG |
2572 | local_stream[i]->hangup_flush_done || |
2573 | local_stream[i]->has_data) { | |
4078b776 MD |
2574 | DBG("Normal read on fd %d", pollfd[i].fd); |
2575 | len = ctx->on_buffer_ready(local_stream[i], ctx); | |
2576 | /* it's ok to have an unavailable sub-buffer */ | |
b64403e3 | 2577 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2578 | /* Clean the stream and free it. */ |
2579 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2580 | local_stream[i] = NULL; |
4078b776 MD |
2581 | } else if (len > 0) { |
2582 | local_stream[i]->data_read = 1; | |
2583 | } | |
2584 | } | |
2585 | } | |
2586 | ||
2587 | /* Handle hangup and errors */ | |
2588 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2589 | health_code_update(); |
2590 | ||
9617607b DG |
2591 | if (local_stream[i] == NULL) { |
2592 | continue; | |
2593 | } | |
4078b776 MD |
2594 | if (!local_stream[i]->hangup_flush_done |
2595 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) | |
2596 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2597 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2598 | DBG("fd %d is hup|err|nval. Attempting flush and read.", | |
9617607b | 2599 | pollfd[i].fd); |
4078b776 MD |
2600 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); |
2601 | /* Attempt read again, for the data we just flushed. */ | |
2602 | local_stream[i]->data_read = 1; | |
2603 | } | |
2604 | /* | |
2605 | * If the poll flag is HUP/ERR/NVAL and we have | |
2606 | * read no data in this pass, we can remove the | |
2607 | * stream from its hash table. | |
2608 | */ | |
2609 | if ((pollfd[i].revents & POLLHUP)) { | |
2610 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); | |
2611 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2612 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2613 | local_stream[i] = NULL; |
4078b776 MD |
2614 | num_hup++; |
2615 | } | |
2616 | } else if (pollfd[i].revents & POLLERR) { | |
2617 | ERR("Error returned in polling fd %d.", pollfd[i].fd); | |
2618 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2619 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2620 | local_stream[i] = NULL; |
4078b776 MD |
2621 | num_hup++; |
2622 | } | |
2623 | } else if (pollfd[i].revents & POLLNVAL) { | |
2624 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); | |
2625 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2626 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2627 | local_stream[i] = NULL; |
4078b776 | 2628 | num_hup++; |
3bd1e081 MD |
2629 | } |
2630 | } | |
9617607b DG |
2631 | if (local_stream[i] != NULL) { |
2632 | local_stream[i]->data_read = 0; | |
2633 | } | |
3bd1e081 MD |
2634 | } |
2635 | } | |
1fc79fb4 MD |
2636 | /* All is OK */ |
2637 | err = 0; | |
3bd1e081 MD |
2638 | end: |
2639 | DBG("polling thread exiting"); | |
0e428499 DG |
2640 | free(pollfd); |
2641 | free(local_stream); | |
fb3a43a9 DG |
2642 | |
2643 | /* | |
2644 | * Close the write side of the pipe so epoll_wait() in | |
7d980def DG |
2645 | * consumer_thread_metadata_poll can catch it. The thread is monitoring the |
2646 | * read side of the pipe. If we close them both, epoll_wait strangely does | |
2647 | * not return and could create a endless wait period if the pipe is the | |
2648 | * only tracked fd in the poll set. The thread will take care of closing | |
2649 | * the read side. | |
fb3a43a9 | 2650 | */ |
13886d2d | 2651 | (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe); |
fb3a43a9 | 2652 | |
2d57de81 | 2653 | error_testpoint: |
1fc79fb4 MD |
2654 | if (err) { |
2655 | health_error(); | |
2656 | ERR("Health error occurred in %s", __func__); | |
2657 | } | |
2658 | health_unregister(health_consumerd); | |
2659 | ||
e7b994a3 | 2660 | rcu_unregister_thread(); |
3bd1e081 MD |
2661 | return NULL; |
2662 | } | |
2663 | ||
d8ef542d MD |
2664 | /* |
2665 | * Close wake-up end of each stream belonging to the channel. This will | |
2666 | * allow the poll() on the stream read-side to detect when the | |
2667 | * write-side (application) finally closes them. | |
2668 | */ | |
2669 | static | |
2670 | void consumer_close_channel_streams(struct lttng_consumer_channel *channel) | |
2671 | { | |
2672 | struct lttng_ht *ht; | |
2673 | struct lttng_consumer_stream *stream; | |
2674 | struct lttng_ht_iter iter; | |
2675 | ||
2676 | ht = consumer_data.stream_per_chan_id_ht; | |
2677 | ||
2678 | rcu_read_lock(); | |
2679 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
2680 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
2681 | ht->match_fct, &channel->key, | |
2682 | &iter.iter, stream, node_channel_id.node) { | |
f2ad556d MD |
2683 | /* |
2684 | * Protect against teardown with mutex. | |
2685 | */ | |
2686 | pthread_mutex_lock(&stream->lock); | |
2687 | if (cds_lfht_is_node_deleted(&stream->node.node)) { | |
2688 | goto next; | |
2689 | } | |
d8ef542d MD |
2690 | switch (consumer_data.type) { |
2691 | case LTTNG_CONSUMER_KERNEL: | |
2692 | break; | |
2693 | case LTTNG_CONSUMER32_UST: | |
2694 | case LTTNG_CONSUMER64_UST: | |
b4a650f3 DG |
2695 | if (stream->metadata_flag) { |
2696 | /* Safe and protected by the stream lock. */ | |
2697 | lttng_ustconsumer_close_metadata(stream->chan); | |
2698 | } else { | |
2699 | /* | |
2700 | * Note: a mutex is taken internally within | |
2701 | * liblttng-ust-ctl to protect timer wakeup_fd | |
2702 | * use from concurrent close. | |
2703 | */ | |
2704 | lttng_ustconsumer_close_stream_wakeup(stream); | |
2705 | } | |
d8ef542d MD |
2706 | break; |
2707 | default: | |
2708 | ERR("Unknown consumer_data type"); | |
2709 | assert(0); | |
2710 | } | |
f2ad556d MD |
2711 | next: |
2712 | pthread_mutex_unlock(&stream->lock); | |
d8ef542d MD |
2713 | } |
2714 | rcu_read_unlock(); | |
2715 | } | |
2716 | ||
2717 | static void destroy_channel_ht(struct lttng_ht *ht) | |
2718 | { | |
2719 | struct lttng_ht_iter iter; | |
2720 | struct lttng_consumer_channel *channel; | |
2721 | int ret; | |
2722 | ||
2723 | if (ht == NULL) { | |
2724 | return; | |
2725 | } | |
2726 | ||
2727 | rcu_read_lock(); | |
2728 | cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) { | |
2729 | ret = lttng_ht_del(ht, &iter); | |
2730 | assert(ret != 0); | |
2731 | } | |
2732 | rcu_read_unlock(); | |
2733 | ||
2734 | lttng_ht_destroy(ht); | |
2735 | } | |
2736 | ||
2737 | /* | |
2738 | * This thread polls the channel fds to detect when they are being | |
2739 | * closed. It closes all related streams if the channel is detected as | |
2740 | * closed. It is currently only used as a shim layer for UST because the | |
2741 | * consumerd needs to keep the per-stream wakeup end of pipes open for | |
2742 | * periodical flush. | |
2743 | */ | |
2744 | void *consumer_thread_channel_poll(void *data) | |
2745 | { | |
1fc79fb4 | 2746 | int ret, i, pollfd, err = -1; |
d8ef542d MD |
2747 | uint32_t revents, nb_fd; |
2748 | struct lttng_consumer_channel *chan = NULL; | |
2749 | struct lttng_ht_iter iter; | |
2750 | struct lttng_ht_node_u64 *node; | |
2751 | struct lttng_poll_event events; | |
2752 | struct lttng_consumer_local_data *ctx = data; | |
2753 | struct lttng_ht *channel_ht; | |
2754 | ||
2755 | rcu_register_thread(); | |
2756 | ||
1fc79fb4 MD |
2757 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL); |
2758 | ||
2d57de81 MD |
2759 | if (testpoint(consumerd_thread_channel)) { |
2760 | goto error_testpoint; | |
2761 | } | |
2762 | ||
9ce5646a MD |
2763 | health_code_update(); |
2764 | ||
d8ef542d MD |
2765 | channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
2766 | if (!channel_ht) { | |
2767 | /* ENOMEM at this point. Better to bail out. */ | |
2768 | goto end_ht; | |
2769 | } | |
2770 | ||
2771 | DBG("Thread channel poll started"); | |
2772 | ||
2773 | /* Size is set to 1 for the consumer_channel pipe */ | |
2774 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2775 | if (ret < 0) { | |
2776 | ERR("Poll set creation failed"); | |
2777 | goto end_poll; | |
2778 | } | |
2779 | ||
2780 | ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN); | |
2781 | if (ret < 0) { | |
2782 | goto end; | |
2783 | } | |
2784 | ||
2785 | /* Main loop */ | |
2786 | DBG("Channel main loop started"); | |
2787 | ||
2788 | while (1) { | |
d8ef542d | 2789 | restart: |
7fa2082e MD |
2790 | health_code_update(); |
2791 | DBG("Channel poll wait"); | |
9ce5646a | 2792 | health_poll_entry(); |
d8ef542d | 2793 | ret = lttng_poll_wait(&events, -1); |
7fa2082e MD |
2794 | DBG("Channel poll return from wait with %d fd(s)", |
2795 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2796 | health_poll_exit(); |
d8ef542d MD |
2797 | DBG("Channel event catched in thread"); |
2798 | if (ret < 0) { | |
2799 | if (errno == EINTR) { | |
2800 | ERR("Poll EINTR catched"); | |
2801 | goto restart; | |
2802 | } | |
d9607cd7 MD |
2803 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2804 | err = 0; /* All is OK */ | |
2805 | } | |
d8ef542d MD |
2806 | goto end; |
2807 | } | |
2808 | ||
2809 | nb_fd = ret; | |
2810 | ||
2811 | /* From here, the event is a channel wait fd */ | |
2812 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2813 | health_code_update(); |
2814 | ||
d8ef542d MD |
2815 | revents = LTTNG_POLL_GETEV(&events, i); |
2816 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2817 | ||
d8ef542d | 2818 | if (!revents) { |
fd20dac9 | 2819 | /* No activity for this FD (poll implementation). */ |
d8ef542d MD |
2820 | continue; |
2821 | } | |
fd20dac9 | 2822 | |
d8ef542d | 2823 | if (pollfd == ctx->consumer_channel_pipe[0]) { |
03e43155 | 2824 | if (revents & LPOLLIN) { |
d8ef542d | 2825 | enum consumer_channel_action action; |
a0cbdd2e | 2826 | uint64_t key; |
d8ef542d | 2827 | |
a0cbdd2e | 2828 | ret = read_channel_pipe(ctx, &chan, &key, &action); |
d8ef542d | 2829 | if (ret <= 0) { |
03e43155 MD |
2830 | if (ret < 0) { |
2831 | ERR("Error reading channel pipe"); | |
2832 | } | |
2833 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
d8ef542d MD |
2834 | continue; |
2835 | } | |
2836 | ||
2837 | switch (action) { | |
2838 | case CONSUMER_CHANNEL_ADD: | |
2839 | DBG("Adding channel %d to poll set", | |
2840 | chan->wait_fd); | |
2841 | ||
2842 | lttng_ht_node_init_u64(&chan->wait_fd_node, | |
2843 | chan->wait_fd); | |
c7260a81 | 2844 | rcu_read_lock(); |
d8ef542d MD |
2845 | lttng_ht_add_unique_u64(channel_ht, |
2846 | &chan->wait_fd_node); | |
c7260a81 | 2847 | rcu_read_unlock(); |
d8ef542d MD |
2848 | /* Add channel to the global poll events list */ |
2849 | lttng_poll_add(&events, chan->wait_fd, | |
03e43155 | 2850 | LPOLLERR | LPOLLHUP); |
d8ef542d | 2851 | break; |
a0cbdd2e MD |
2852 | case CONSUMER_CHANNEL_DEL: |
2853 | { | |
b4a650f3 DG |
2854 | /* |
2855 | * This command should never be called if the channel | |
2856 | * has streams monitored by either the data or metadata | |
2857 | * thread. The consumer only notify this thread with a | |
2858 | * channel del. command if it receives a destroy | |
2859 | * channel command from the session daemon that send it | |
2860 | * if a command prior to the GET_CHANNEL failed. | |
2861 | */ | |
2862 | ||
c7260a81 | 2863 | rcu_read_lock(); |
a0cbdd2e MD |
2864 | chan = consumer_find_channel(key); |
2865 | if (!chan) { | |
c7260a81 | 2866 | rcu_read_unlock(); |
a0cbdd2e MD |
2867 | ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key); |
2868 | break; | |
2869 | } | |
2870 | lttng_poll_del(&events, chan->wait_fd); | |
f623cc0b | 2871 | iter.iter.node = &chan->wait_fd_node.node; |
a0cbdd2e MD |
2872 | ret = lttng_ht_del(channel_ht, &iter); |
2873 | assert(ret == 0); | |
a0cbdd2e | 2874 | |
f2a444f1 DG |
2875 | switch (consumer_data.type) { |
2876 | case LTTNG_CONSUMER_KERNEL: | |
2877 | break; | |
2878 | case LTTNG_CONSUMER32_UST: | |
2879 | case LTTNG_CONSUMER64_UST: | |
212d67a2 DG |
2880 | health_code_update(); |
2881 | /* Destroy streams that might have been left in the stream list. */ | |
2882 | clean_channel_stream_list(chan); | |
f2a444f1 DG |
2883 | break; |
2884 | default: | |
2885 | ERR("Unknown consumer_data type"); | |
2886 | assert(0); | |
2887 | } | |
2888 | ||
a0cbdd2e MD |
2889 | /* |
2890 | * Release our own refcount. Force channel deletion even if | |
2891 | * streams were not initialized. | |
2892 | */ | |
2893 | if (!uatomic_sub_return(&chan->refcount, 1)) { | |
2894 | consumer_del_channel(chan); | |
2895 | } | |
c7260a81 | 2896 | rcu_read_unlock(); |
a0cbdd2e MD |
2897 | goto restart; |
2898 | } | |
d8ef542d MD |
2899 | case CONSUMER_CHANNEL_QUIT: |
2900 | /* | |
2901 | * Remove the pipe from the poll set and continue the loop | |
2902 | * since their might be data to consume. | |
2903 | */ | |
2904 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
2905 | continue; | |
2906 | default: | |
2907 | ERR("Unknown action"); | |
2908 | break; | |
2909 | } | |
03e43155 MD |
2910 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
2911 | DBG("Channel thread pipe hung up"); | |
2912 | /* | |
2913 | * Remove the pipe from the poll set and continue the loop | |
2914 | * since their might be data to consume. | |
2915 | */ | |
2916 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
2917 | continue; | |
2918 | } else { | |
2919 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
2920 | goto end; | |
d8ef542d MD |
2921 | } |
2922 | ||
2923 | /* Handle other stream */ | |
2924 | continue; | |
2925 | } | |
2926 | ||
2927 | rcu_read_lock(); | |
2928 | { | |
2929 | uint64_t tmp_id = (uint64_t) pollfd; | |
2930 | ||
2931 | lttng_ht_lookup(channel_ht, &tmp_id, &iter); | |
2932 | } | |
2933 | node = lttng_ht_iter_get_node_u64(&iter); | |
2934 | assert(node); | |
2935 | ||
2936 | chan = caa_container_of(node, struct lttng_consumer_channel, | |
2937 | wait_fd_node); | |
2938 | ||
2939 | /* Check for error event */ | |
2940 | if (revents & (LPOLLERR | LPOLLHUP)) { | |
2941 | DBG("Channel fd %d is hup|err.", pollfd); | |
2942 | ||
2943 | lttng_poll_del(&events, chan->wait_fd); | |
2944 | ret = lttng_ht_del(channel_ht, &iter); | |
2945 | assert(ret == 0); | |
b4a650f3 DG |
2946 | |
2947 | /* | |
2948 | * This will close the wait fd for each stream associated to | |
2949 | * this channel AND monitored by the data/metadata thread thus | |
2950 | * will be clean by the right thread. | |
2951 | */ | |
d8ef542d | 2952 | consumer_close_channel_streams(chan); |
f2ad556d MD |
2953 | |
2954 | /* Release our own refcount */ | |
2955 | if (!uatomic_sub_return(&chan->refcount, 1) | |
2956 | && !uatomic_read(&chan->nb_init_stream_left)) { | |
2957 | consumer_del_channel(chan); | |
2958 | } | |
03e43155 MD |
2959 | } else { |
2960 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
2961 | rcu_read_unlock(); | |
2962 | goto end; | |
d8ef542d MD |
2963 | } |
2964 | ||
2965 | /* Release RCU lock for the channel looked up */ | |
2966 | rcu_read_unlock(); | |
2967 | } | |
2968 | } | |
2969 | ||
1fc79fb4 MD |
2970 | /* All is OK */ |
2971 | err = 0; | |
d8ef542d MD |
2972 | end: |
2973 | lttng_poll_clean(&events); | |
2974 | end_poll: | |
2975 | destroy_channel_ht(channel_ht); | |
2976 | end_ht: | |
2d57de81 | 2977 | error_testpoint: |
d8ef542d | 2978 | DBG("Channel poll thread exiting"); |
1fc79fb4 MD |
2979 | if (err) { |
2980 | health_error(); | |
2981 | ERR("Health error occurred in %s", __func__); | |
2982 | } | |
2983 | health_unregister(health_consumerd); | |
d8ef542d MD |
2984 | rcu_unregister_thread(); |
2985 | return NULL; | |
2986 | } | |
2987 | ||
331744e3 JD |
2988 | static int set_metadata_socket(struct lttng_consumer_local_data *ctx, |
2989 | struct pollfd *sockpoll, int client_socket) | |
2990 | { | |
2991 | int ret; | |
2992 | ||
2993 | assert(ctx); | |
2994 | assert(sockpoll); | |
2995 | ||
84382d49 MD |
2996 | ret = lttng_consumer_poll_socket(sockpoll); |
2997 | if (ret) { | |
331744e3 JD |
2998 | goto error; |
2999 | } | |
3000 | DBG("Metadata connection on client_socket"); | |
3001 | ||
3002 | /* Blocking call, waiting for transmission */ | |
3003 | ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket); | |
3004 | if (ctx->consumer_metadata_socket < 0) { | |
3005 | WARN("On accept metadata"); | |
3006 | ret = -1; | |
3007 | goto error; | |
3008 | } | |
3009 | ret = 0; | |
3010 | ||
3011 | error: | |
3012 | return ret; | |
3013 | } | |
3014 | ||
3bd1e081 MD |
3015 | /* |
3016 | * This thread listens on the consumerd socket and receives the file | |
3017 | * descriptors from the session daemon. | |
3018 | */ | |
7d980def | 3019 | void *consumer_thread_sessiond_poll(void *data) |
3bd1e081 | 3020 | { |
1fc79fb4 | 3021 | int sock = -1, client_socket, ret, err = -1; |
3bd1e081 MD |
3022 | /* |
3023 | * structure to poll for incoming data on communication socket avoids | |
3024 | * making blocking sockets. | |
3025 | */ | |
3026 | struct pollfd consumer_sockpoll[2]; | |
3027 | struct lttng_consumer_local_data *ctx = data; | |
3028 | ||
e7b994a3 DG |
3029 | rcu_register_thread(); |
3030 | ||
1fc79fb4 MD |
3031 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND); |
3032 | ||
2d57de81 MD |
3033 | if (testpoint(consumerd_thread_sessiond)) { |
3034 | goto error_testpoint; | |
3035 | } | |
3036 | ||
9ce5646a MD |
3037 | health_code_update(); |
3038 | ||
3bd1e081 MD |
3039 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
3040 | unlink(ctx->consumer_command_sock_path); | |
3041 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); | |
3042 | if (client_socket < 0) { | |
3043 | ERR("Cannot create command socket"); | |
3044 | goto end; | |
3045 | } | |
3046 | ||
3047 | ret = lttcomm_listen_unix_sock(client_socket); | |
3048 | if (ret < 0) { | |
3049 | goto end; | |
3050 | } | |
3051 | ||
32258573 | 3052 | DBG("Sending ready command to lttng-sessiond"); |
f73fabfd | 3053 | ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); |
3bd1e081 MD |
3054 | /* return < 0 on error, but == 0 is not fatal */ |
3055 | if (ret < 0) { | |
32258573 | 3056 | ERR("Error sending ready command to lttng-sessiond"); |
3bd1e081 MD |
3057 | goto end; |
3058 | } | |
3059 | ||
3bd1e081 MD |
3060 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ |
3061 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; | |
3062 | consumer_sockpoll[0].events = POLLIN | POLLPRI; | |
3063 | consumer_sockpoll[1].fd = client_socket; | |
3064 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3065 | ||
84382d49 MD |
3066 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3067 | if (ret) { | |
3068 | if (ret > 0) { | |
3069 | /* should exit */ | |
3070 | err = 0; | |
3071 | } | |
3bd1e081 MD |
3072 | goto end; |
3073 | } | |
3074 | DBG("Connection on client_socket"); | |
3075 | ||
3076 | /* Blocking call, waiting for transmission */ | |
3077 | sock = lttcomm_accept_unix_sock(client_socket); | |
534d2592 | 3078 | if (sock < 0) { |
3bd1e081 MD |
3079 | WARN("On accept"); |
3080 | goto end; | |
3081 | } | |
3bd1e081 | 3082 | |
331744e3 JD |
3083 | /* |
3084 | * Setup metadata socket which is the second socket connection on the | |
3085 | * command unix socket. | |
3086 | */ | |
3087 | ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket); | |
84382d49 MD |
3088 | if (ret) { |
3089 | if (ret > 0) { | |
3090 | /* should exit */ | |
3091 | err = 0; | |
3092 | } | |
331744e3 JD |
3093 | goto end; |
3094 | } | |
3095 | ||
d96f09c6 DG |
3096 | /* This socket is not useful anymore. */ |
3097 | ret = close(client_socket); | |
3098 | if (ret < 0) { | |
3099 | PERROR("close client_socket"); | |
3100 | } | |
3101 | client_socket = -1; | |
3102 | ||
3bd1e081 MD |
3103 | /* update the polling structure to poll on the established socket */ |
3104 | consumer_sockpoll[1].fd = sock; | |
3105 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3106 | ||
3107 | while (1) { | |
9ce5646a MD |
3108 | health_code_update(); |
3109 | ||
3110 | health_poll_entry(); | |
3111 | ret = lttng_consumer_poll_socket(consumer_sockpoll); | |
3112 | health_poll_exit(); | |
84382d49 MD |
3113 | if (ret) { |
3114 | if (ret > 0) { | |
3115 | /* should exit */ | |
3116 | err = 0; | |
3117 | } | |
3bd1e081 MD |
3118 | goto end; |
3119 | } | |
3120 | DBG("Incoming command on sock"); | |
3121 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
4cbc1a04 DG |
3122 | if (ret <= 0) { |
3123 | /* | |
3124 | * This could simply be a session daemon quitting. Don't output | |
3125 | * ERR() here. | |
3126 | */ | |
3127 | DBG("Communication interrupted on command socket"); | |
41ba6035 | 3128 | err = 0; |
3bd1e081 MD |
3129 | goto end; |
3130 | } | |
3131 | if (consumer_quit) { | |
3132 | DBG("consumer_thread_receive_fds received quit from signal"); | |
1fc79fb4 | 3133 | err = 0; /* All is OK */ |
3bd1e081 MD |
3134 | goto end; |
3135 | } | |
ffe60014 | 3136 | DBG("received command on sock"); |
3bd1e081 | 3137 | } |
1fc79fb4 MD |
3138 | /* All is OK */ |
3139 | err = 0; | |
3140 | ||
3bd1e081 | 3141 | end: |
ffe60014 | 3142 | DBG("Consumer thread sessiond poll exiting"); |
3bd1e081 | 3143 | |
d88aee68 DG |
3144 | /* |
3145 | * Close metadata streams since the producer is the session daemon which | |
3146 | * just died. | |
3147 | * | |
3148 | * NOTE: for now, this only applies to the UST tracer. | |
3149 | */ | |
6d574024 | 3150 | lttng_consumer_close_all_metadata(); |
d88aee68 | 3151 | |
3bd1e081 MD |
3152 | /* |
3153 | * when all fds have hung up, the polling thread | |
3154 | * can exit cleanly | |
3155 | */ | |
3156 | consumer_quit = 1; | |
3157 | ||
04fdd819 | 3158 | /* |
c869f647 | 3159 | * Notify the data poll thread to poll back again and test the |
8994307f | 3160 | * consumer_quit state that we just set so to quit gracefully. |
04fdd819 | 3161 | */ |
acdb9057 | 3162 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
c869f647 | 3163 | |
a0cbdd2e | 3164 | notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT); |
d8ef542d | 3165 | |
5c635c72 MD |
3166 | notify_health_quit_pipe(health_quit_pipe); |
3167 | ||
d96f09c6 DG |
3168 | /* Cleaning up possibly open sockets. */ |
3169 | if (sock >= 0) { | |
3170 | ret = close(sock); | |
3171 | if (ret < 0) { | |
3172 | PERROR("close sock sessiond poll"); | |
3173 | } | |
3174 | } | |
3175 | if (client_socket >= 0) { | |
38476d24 | 3176 | ret = close(client_socket); |
d96f09c6 DG |
3177 | if (ret < 0) { |
3178 | PERROR("close client_socket sessiond poll"); | |
3179 | } | |
3180 | } | |
3181 | ||
2d57de81 | 3182 | error_testpoint: |
1fc79fb4 MD |
3183 | if (err) { |
3184 | health_error(); | |
3185 | ERR("Health error occurred in %s", __func__); | |
3186 | } | |
3187 | health_unregister(health_consumerd); | |
3188 | ||
e7b994a3 | 3189 | rcu_unregister_thread(); |
3bd1e081 MD |
3190 | return NULL; |
3191 | } | |
d41f73b7 | 3192 | |
4078b776 | 3193 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
d41f73b7 MD |
3194 | struct lttng_consumer_local_data *ctx) |
3195 | { | |
74251bb8 DG |
3196 | ssize_t ret; |
3197 | ||
3198 | pthread_mutex_lock(&stream->lock); | |
94d49140 JD |
3199 | if (stream->metadata_flag) { |
3200 | pthread_mutex_lock(&stream->metadata_rdv_lock); | |
3201 | } | |
74251bb8 | 3202 | |
d41f73b7 MD |
3203 | switch (consumer_data.type) { |
3204 | case LTTNG_CONSUMER_KERNEL: | |
74251bb8 DG |
3205 | ret = lttng_kconsumer_read_subbuffer(stream, ctx); |
3206 | break; | |
7753dea8 MD |
3207 | case LTTNG_CONSUMER32_UST: |
3208 | case LTTNG_CONSUMER64_UST: | |
74251bb8 DG |
3209 | ret = lttng_ustconsumer_read_subbuffer(stream, ctx); |
3210 | break; | |
d41f73b7 MD |
3211 | default: |
3212 | ERR("Unknown consumer_data type"); | |
3213 | assert(0); | |
74251bb8 DG |
3214 | ret = -ENOSYS; |
3215 | break; | |
d41f73b7 | 3216 | } |
74251bb8 | 3217 | |
94d49140 JD |
3218 | if (stream->metadata_flag) { |
3219 | pthread_cond_broadcast(&stream->metadata_rdv); | |
3220 | pthread_mutex_unlock(&stream->metadata_rdv_lock); | |
3221 | } | |
74251bb8 DG |
3222 | pthread_mutex_unlock(&stream->lock); |
3223 | return ret; | |
d41f73b7 MD |
3224 | } |
3225 | ||
3226 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
3227 | { | |
3228 | switch (consumer_data.type) { | |
3229 | case LTTNG_CONSUMER_KERNEL: | |
3230 | return lttng_kconsumer_on_recv_stream(stream); | |
7753dea8 MD |
3231 | case LTTNG_CONSUMER32_UST: |
3232 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
3233 | return lttng_ustconsumer_on_recv_stream(stream); |
3234 | default: | |
3235 | ERR("Unknown consumer_data type"); | |
3236 | assert(0); | |
3237 | return -ENOSYS; | |
3238 | } | |
3239 | } | |
e4421fec DG |
3240 | |
3241 | /* | |
3242 | * Allocate and set consumer data hash tables. | |
3243 | */ | |
282dadbc | 3244 | int lttng_consumer_init(void) |
e4421fec | 3245 | { |
d88aee68 | 3246 | consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3247 | if (!consumer_data.channel_ht) { |
3248 | goto error; | |
3249 | } | |
3250 | ||
d88aee68 | 3251 | consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3252 | if (!consumer_data.relayd_ht) { |
3253 | goto error; | |
3254 | } | |
3255 | ||
d88aee68 | 3256 | consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3257 | if (!consumer_data.stream_list_ht) { |
3258 | goto error; | |
3259 | } | |
3260 | ||
d8ef542d | 3261 | consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3262 | if (!consumer_data.stream_per_chan_id_ht) { |
3263 | goto error; | |
3264 | } | |
3265 | ||
3266 | data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3267 | if (!data_ht) { | |
3268 | goto error; | |
3269 | } | |
3270 | ||
3271 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3272 | if (!metadata_ht) { | |
3273 | goto error; | |
3274 | } | |
3275 | ||
3276 | return 0; | |
3277 | ||
3278 | error: | |
3279 | return -1; | |
e4421fec | 3280 | } |
7735ef9e DG |
3281 | |
3282 | /* | |
3283 | * Process the ADD_RELAYD command receive by a consumer. | |
3284 | * | |
3285 | * This will create a relayd socket pair and add it to the relayd hash table. | |
3286 | * The caller MUST acquire a RCU read side lock before calling it. | |
3287 | */ | |
da009f2c | 3288 | int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type, |
7735ef9e | 3289 | struct lttng_consumer_local_data *ctx, int sock, |
6151a90f | 3290 | struct pollfd *consumer_sockpoll, |
d3e2ba59 JD |
3291 | struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id, |
3292 | uint64_t relayd_session_id) | |
7735ef9e | 3293 | { |
cd2b09ed | 3294 | int fd = -1, ret = -1, relayd_created = 0; |
0c759fc9 | 3295 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
d4298c99 | 3296 | struct consumer_relayd_sock_pair *relayd = NULL; |
7735ef9e | 3297 | |
6151a90f JD |
3298 | assert(ctx); |
3299 | assert(relayd_sock); | |
3300 | ||
da009f2c | 3301 | DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx); |
7735ef9e DG |
3302 | |
3303 | /* Get relayd reference if exists. */ | |
3304 | relayd = consumer_find_relayd(net_seq_idx); | |
3305 | if (relayd == NULL) { | |
da009f2c | 3306 | assert(sock_type == LTTNG_STREAM_CONTROL); |
7735ef9e DG |
3307 | /* Not found. Allocate one. */ |
3308 | relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); | |
3309 | if (relayd == NULL) { | |
0d08d75e | 3310 | ret = -ENOMEM; |
618a6a28 MD |
3311 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
3312 | goto error; | |
0d08d75e | 3313 | } else { |
30319bcb | 3314 | relayd->sessiond_session_id = sessiond_id; |
0d08d75e | 3315 | relayd_created = 1; |
7735ef9e | 3316 | } |
0d08d75e DG |
3317 | |
3318 | /* | |
3319 | * This code path MUST continue to the consumer send status message to | |
3320 | * we can notify the session daemon and continue our work without | |
3321 | * killing everything. | |
3322 | */ | |
da009f2c MD |
3323 | } else { |
3324 | /* | |
3325 | * relayd key should never be found for control socket. | |
3326 | */ | |
3327 | assert(sock_type != LTTNG_STREAM_CONTROL); | |
0d08d75e DG |
3328 | } |
3329 | ||
3330 | /* First send a status message before receiving the fds. */ | |
0c759fc9 | 3331 | ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS); |
618a6a28 | 3332 | if (ret < 0) { |
0d08d75e | 3333 | /* Somehow, the session daemon is not responding anymore. */ |
618a6a28 MD |
3334 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); |
3335 | goto error_nosignal; | |
7735ef9e DG |
3336 | } |
3337 | ||
3338 | /* Poll on consumer socket. */ | |
84382d49 MD |
3339 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3340 | if (ret) { | |
3341 | /* Needing to exit in the middle of a command: error. */ | |
0d08d75e | 3342 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
7735ef9e | 3343 | ret = -EINTR; |
618a6a28 | 3344 | goto error_nosignal; |
7735ef9e DG |
3345 | } |
3346 | ||
3347 | /* Get relayd socket from session daemon */ | |
3348 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); | |
3349 | if (ret != sizeof(fd)) { | |
7735ef9e | 3350 | ret = -1; |
4028eeb9 | 3351 | fd = -1; /* Just in case it gets set with an invalid value. */ |
0d08d75e DG |
3352 | |
3353 | /* | |
3354 | * Failing to receive FDs might indicate a major problem such as | |
3355 | * reaching a fd limit during the receive where the kernel returns a | |
3356 | * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we | |
3357 | * don't take any chances and stop everything. | |
3358 | * | |
3359 | * XXX: Feature request #558 will fix that and avoid this possible | |
3360 | * issue when reaching the fd limit. | |
3361 | */ | |
3362 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); | |
618a6a28 | 3363 | ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD; |
f50f23d9 DG |
3364 | goto error; |
3365 | } | |
3366 | ||
7735ef9e DG |
3367 | /* Copy socket information and received FD */ |
3368 | switch (sock_type) { | |
3369 | case LTTNG_STREAM_CONTROL: | |
3370 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3371 | lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock); |
3372 | ret = lttcomm_create_sock(&relayd->control_sock.sock); | |
4028eeb9 | 3373 | /* Handle create_sock error. */ |
f66c074c | 3374 | if (ret < 0) { |
618a6a28 | 3375 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3376 | goto error; |
f66c074c | 3377 | } |
da009f2c MD |
3378 | /* |
3379 | * Close the socket created internally by | |
3380 | * lttcomm_create_sock, so we can replace it by the one | |
3381 | * received from sessiond. | |
3382 | */ | |
3383 | if (close(relayd->control_sock.sock.fd)) { | |
3384 | PERROR("close"); | |
3385 | } | |
7735ef9e DG |
3386 | |
3387 | /* Assign new file descriptor */ | |
6151a90f | 3388 | relayd->control_sock.sock.fd = fd; |
4b29f1ce | 3389 | fd = -1; /* For error path */ |
6151a90f JD |
3390 | /* Assign version values. */ |
3391 | relayd->control_sock.major = relayd_sock->major; | |
3392 | relayd->control_sock.minor = relayd_sock->minor; | |
c5b6f4f0 | 3393 | |
d3e2ba59 | 3394 | relayd->relayd_session_id = relayd_session_id; |
c5b6f4f0 | 3395 | |
7735ef9e DG |
3396 | break; |
3397 | case LTTNG_STREAM_DATA: | |
3398 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3399 | lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock); |
3400 | ret = lttcomm_create_sock(&relayd->data_sock.sock); | |
4028eeb9 | 3401 | /* Handle create_sock error. */ |
f66c074c | 3402 | if (ret < 0) { |
618a6a28 | 3403 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3404 | goto error; |
f66c074c | 3405 | } |
da009f2c MD |
3406 | /* |
3407 | * Close the socket created internally by | |
3408 | * lttcomm_create_sock, so we can replace it by the one | |
3409 | * received from sessiond. | |
3410 | */ | |
3411 | if (close(relayd->data_sock.sock.fd)) { | |
3412 | PERROR("close"); | |
3413 | } | |
7735ef9e DG |
3414 | |
3415 | /* Assign new file descriptor */ | |
6151a90f | 3416 | relayd->data_sock.sock.fd = fd; |
4b29f1ce | 3417 | fd = -1; /* for eventual error paths */ |
6151a90f JD |
3418 | /* Assign version values. */ |
3419 | relayd->data_sock.major = relayd_sock->major; | |
3420 | relayd->data_sock.minor = relayd_sock->minor; | |
7735ef9e DG |
3421 | break; |
3422 | default: | |
3423 | ERR("Unknown relayd socket type (%d)", sock_type); | |
59e71485 | 3424 | ret = -1; |
618a6a28 | 3425 | ret_code = LTTCOMM_CONSUMERD_FATAL; |
7735ef9e DG |
3426 | goto error; |
3427 | } | |
3428 | ||
d88aee68 | 3429 | DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)", |
7735ef9e DG |
3430 | sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", |
3431 | relayd->net_seq_idx, fd); | |
3432 | ||
618a6a28 MD |
3433 | /* We successfully added the socket. Send status back. */ |
3434 | ret = consumer_send_status_msg(sock, ret_code); | |
3435 | if (ret < 0) { | |
3436 | /* Somehow, the session daemon is not responding anymore. */ | |
3437 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3438 | goto error_nosignal; | |
3439 | } | |
3440 | ||
7735ef9e DG |
3441 | /* |
3442 | * Add relayd socket pair to consumer data hashtable. If object already | |
3443 | * exists or on error, the function gracefully returns. | |
3444 | */ | |
d09e1200 | 3445 | add_relayd(relayd); |
7735ef9e DG |
3446 | |
3447 | /* All good! */ | |
4028eeb9 | 3448 | return 0; |
7735ef9e DG |
3449 | |
3450 | error: | |
618a6a28 MD |
3451 | if (consumer_send_status_msg(sock, ret_code) < 0) { |
3452 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3453 | } | |
3454 | ||
3455 | error_nosignal: | |
4028eeb9 DG |
3456 | /* Close received socket if valid. */ |
3457 | if (fd >= 0) { | |
3458 | if (close(fd)) { | |
3459 | PERROR("close received socket"); | |
3460 | } | |
3461 | } | |
cd2b09ed DG |
3462 | |
3463 | if (relayd_created) { | |
cd2b09ed DG |
3464 | free(relayd); |
3465 | } | |
3466 | ||
7735ef9e DG |
3467 | return ret; |
3468 | } | |
ca22feea | 3469 | |
4e9a4686 DG |
3470 | /* |
3471 | * Try to lock the stream mutex. | |
3472 | * | |
3473 | * On success, 1 is returned else 0 indicating that the mutex is NOT lock. | |
3474 | */ | |
3475 | static int stream_try_lock(struct lttng_consumer_stream *stream) | |
3476 | { | |
3477 | int ret; | |
3478 | ||
3479 | assert(stream); | |
3480 | ||
3481 | /* | |
3482 | * Try to lock the stream mutex. On failure, we know that the stream is | |
3483 | * being used else where hence there is data still being extracted. | |
3484 | */ | |
3485 | ret = pthread_mutex_trylock(&stream->lock); | |
3486 | if (ret) { | |
3487 | /* For both EBUSY and EINVAL error, the mutex is NOT locked. */ | |
3488 | ret = 0; | |
3489 | goto end; | |
3490 | } | |
3491 | ||
3492 | ret = 1; | |
3493 | ||
3494 | end: | |
3495 | return ret; | |
3496 | } | |
3497 | ||
f7079f67 DG |
3498 | /* |
3499 | * Search for a relayd associated to the session id and return the reference. | |
3500 | * | |
3501 | * A rcu read side lock MUST be acquire before calling this function and locked | |
3502 | * until the relayd object is no longer necessary. | |
3503 | */ | |
3504 | static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id) | |
3505 | { | |
3506 | struct lttng_ht_iter iter; | |
f7079f67 | 3507 | struct consumer_relayd_sock_pair *relayd = NULL; |
f7079f67 DG |
3508 | |
3509 | /* Iterate over all relayd since they are indexed by net_seq_idx. */ | |
3510 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
3511 | node.node) { | |
18261bd1 DG |
3512 | /* |
3513 | * Check by sessiond id which is unique here where the relayd session | |
3514 | * id might not be when having multiple relayd. | |
3515 | */ | |
3516 | if (relayd->sessiond_session_id == id) { | |
f7079f67 | 3517 | /* Found the relayd. There can be only one per id. */ |
18261bd1 | 3518 | goto found; |
f7079f67 DG |
3519 | } |
3520 | } | |
3521 | ||
18261bd1 DG |
3522 | return NULL; |
3523 | ||
3524 | found: | |
f7079f67 DG |
3525 | return relayd; |
3526 | } | |
3527 | ||
ca22feea DG |
3528 | /* |
3529 | * Check if for a given session id there is still data needed to be extract | |
3530 | * from the buffers. | |
3531 | * | |
6d805429 | 3532 | * Return 1 if data is pending or else 0 meaning ready to be read. |
ca22feea | 3533 | */ |
6d805429 | 3534 | int consumer_data_pending(uint64_t id) |
ca22feea DG |
3535 | { |
3536 | int ret; | |
3537 | struct lttng_ht_iter iter; | |
3538 | struct lttng_ht *ht; | |
3539 | struct lttng_consumer_stream *stream; | |
f7079f67 | 3540 | struct consumer_relayd_sock_pair *relayd = NULL; |
6d805429 | 3541 | int (*data_pending)(struct lttng_consumer_stream *); |
ca22feea | 3542 | |
6d805429 | 3543 | DBG("Consumer data pending command on session id %" PRIu64, id); |
ca22feea | 3544 | |
6f6eda74 | 3545 | rcu_read_lock(); |
ca22feea DG |
3546 | pthread_mutex_lock(&consumer_data.lock); |
3547 | ||
3548 | switch (consumer_data.type) { | |
3549 | case LTTNG_CONSUMER_KERNEL: | |
6d805429 | 3550 | data_pending = lttng_kconsumer_data_pending; |
ca22feea DG |
3551 | break; |
3552 | case LTTNG_CONSUMER32_UST: | |
3553 | case LTTNG_CONSUMER64_UST: | |
6d805429 | 3554 | data_pending = lttng_ustconsumer_data_pending; |
ca22feea DG |
3555 | break; |
3556 | default: | |
3557 | ERR("Unknown consumer data type"); | |
3558 | assert(0); | |
3559 | } | |
3560 | ||
3561 | /* Ease our life a bit */ | |
3562 | ht = consumer_data.stream_list_ht; | |
3563 | ||
f7079f67 DG |
3564 | relayd = find_relayd_by_session_id(id); |
3565 | if (relayd) { | |
3566 | /* Send init command for data pending. */ | |
3567 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3568 | ret = relayd_begin_data_pending(&relayd->control_sock, | |
3569 | relayd->relayd_session_id); | |
3570 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3571 | if (ret < 0) { | |
3572 | /* Communication error thus the relayd so no data pending. */ | |
3573 | goto data_not_pending; | |
3574 | } | |
3575 | } | |
3576 | ||
c8f59ee5 | 3577 | cds_lfht_for_each_entry_duplicate(ht->ht, |
d88aee68 DG |
3578 | ht->hash_fct(&id, lttng_ht_seed), |
3579 | ht->match_fct, &id, | |
ca22feea | 3580 | &iter.iter, stream, node_session_id.node) { |
4e9a4686 DG |
3581 | /* If this call fails, the stream is being used hence data pending. */ |
3582 | ret = stream_try_lock(stream); | |
3583 | if (!ret) { | |
f7079f67 | 3584 | goto data_pending; |
ca22feea | 3585 | } |
ca22feea | 3586 | |
4e9a4686 DG |
3587 | /* |
3588 | * A removed node from the hash table indicates that the stream has | |
3589 | * been deleted thus having a guarantee that the buffers are closed | |
3590 | * on the consumer side. However, data can still be transmitted | |
3591 | * over the network so don't skip the relayd check. | |
3592 | */ | |
3593 | ret = cds_lfht_is_node_deleted(&stream->node.node); | |
3594 | if (!ret) { | |
3595 | /* Check the stream if there is data in the buffers. */ | |
6d805429 DG |
3596 | ret = data_pending(stream); |
3597 | if (ret == 1) { | |
4e9a4686 | 3598 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3599 | goto data_pending; |
4e9a4686 DG |
3600 | } |
3601 | } | |
3602 | ||
3603 | /* Relayd check */ | |
f7079f67 | 3604 | if (relayd) { |
c8f59ee5 DG |
3605 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
3606 | if (stream->metadata_flag) { | |
ad7051c0 DG |
3607 | ret = relayd_quiescent_control(&relayd->control_sock, |
3608 | stream->relayd_stream_id); | |
c8f59ee5 | 3609 | } else { |
6d805429 | 3610 | ret = relayd_data_pending(&relayd->control_sock, |
39df6d9f DG |
3611 | stream->relayd_stream_id, |
3612 | stream->next_net_seq_num - 1); | |
c8f59ee5 DG |
3613 | } |
3614 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
6d805429 | 3615 | if (ret == 1) { |
4e9a4686 | 3616 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3617 | goto data_pending; |
c8f59ee5 DG |
3618 | } |
3619 | } | |
4e9a4686 | 3620 | pthread_mutex_unlock(&stream->lock); |
c8f59ee5 | 3621 | } |
ca22feea | 3622 | |
f7079f67 DG |
3623 | if (relayd) { |
3624 | unsigned int is_data_inflight = 0; | |
3625 | ||
3626 | /* Send init command for data pending. */ | |
3627 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3628 | ret = relayd_end_data_pending(&relayd->control_sock, | |
3629 | relayd->relayd_session_id, &is_data_inflight); | |
3630 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
bdd88757 | 3631 | if (ret < 0) { |
f7079f67 DG |
3632 | goto data_not_pending; |
3633 | } | |
bdd88757 DG |
3634 | if (is_data_inflight) { |
3635 | goto data_pending; | |
3636 | } | |
f7079f67 DG |
3637 | } |
3638 | ||
ca22feea | 3639 | /* |
f7079f67 DG |
3640 | * Finding _no_ node in the hash table and no inflight data means that the |
3641 | * stream(s) have been removed thus data is guaranteed to be available for | |
3642 | * analysis from the trace files. | |
ca22feea DG |
3643 | */ |
3644 | ||
f7079f67 | 3645 | data_not_pending: |
ca22feea DG |
3646 | /* Data is available to be read by a viewer. */ |
3647 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3648 | rcu_read_unlock(); |
6d805429 | 3649 | return 0; |
ca22feea | 3650 | |
f7079f67 | 3651 | data_pending: |
ca22feea DG |
3652 | /* Data is still being extracted from buffers. */ |
3653 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3654 | rcu_read_unlock(); |
6d805429 | 3655 | return 1; |
ca22feea | 3656 | } |
f50f23d9 DG |
3657 | |
3658 | /* | |
3659 | * Send a ret code status message to the sessiond daemon. | |
3660 | * | |
3661 | * Return the sendmsg() return value. | |
3662 | */ | |
3663 | int consumer_send_status_msg(int sock, int ret_code) | |
3664 | { | |
3665 | struct lttcomm_consumer_status_msg msg; | |
3666 | ||
53efb85a | 3667 | memset(&msg, 0, sizeof(msg)); |
f50f23d9 DG |
3668 | msg.ret_code = ret_code; |
3669 | ||
3670 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3671 | } | |
ffe60014 DG |
3672 | |
3673 | /* | |
3674 | * Send a channel status message to the sessiond daemon. | |
3675 | * | |
3676 | * Return the sendmsg() return value. | |
3677 | */ | |
3678 | int consumer_send_status_channel(int sock, | |
3679 | struct lttng_consumer_channel *channel) | |
3680 | { | |
3681 | struct lttcomm_consumer_status_channel msg; | |
3682 | ||
3683 | assert(sock >= 0); | |
3684 | ||
53efb85a | 3685 | memset(&msg, 0, sizeof(msg)); |
ffe60014 | 3686 | if (!channel) { |
0c759fc9 | 3687 | msg.ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL; |
ffe60014 | 3688 | } else { |
0c759fc9 | 3689 | msg.ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
ffe60014 DG |
3690 | msg.key = channel->key; |
3691 | msg.stream_count = channel->streams.count; | |
3692 | } | |
3693 | ||
3694 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3695 | } | |
5c786ded | 3696 | |
d07ceecd MD |
3697 | unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos, |
3698 | unsigned long produced_pos, uint64_t nb_packets_per_stream, | |
3699 | uint64_t max_sb_size) | |
5c786ded | 3700 | { |
d07ceecd | 3701 | unsigned long start_pos; |
5c786ded | 3702 | |
d07ceecd MD |
3703 | if (!nb_packets_per_stream) { |
3704 | return consumed_pos; /* Grab everything */ | |
3705 | } | |
3706 | start_pos = produced_pos - offset_align_floor(produced_pos, max_sb_size); | |
3707 | start_pos -= max_sb_size * nb_packets_per_stream; | |
3708 | if ((long) (start_pos - consumed_pos) < 0) { | |
3709 | return consumed_pos; /* Grab everything */ | |
3710 | } | |
3711 | return start_pos; | |
5c786ded | 3712 | } |