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 | ||
20 | #define _GNU_SOURCE | |
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 | |
990570ed | 33 | #include <common/common.h> |
fb3a43a9 DG |
34 | #include <common/utils.h> |
35 | #include <common/compat/poll.h> | |
10a8a223 | 36 | #include <common/kernel-ctl/kernel-ctl.h> |
00e2e675 | 37 | #include <common/sessiond-comm/relayd.h> |
10a8a223 DG |
38 | #include <common/sessiond-comm/sessiond-comm.h> |
39 | #include <common/kernel-consumer/kernel-consumer.h> | |
00e2e675 | 40 | #include <common/relayd/relayd.h> |
10a8a223 DG |
41 | #include <common/ust-consumer/ust-consumer.h> |
42 | ||
43 | #include "consumer.h" | |
3bd1e081 MD |
44 | |
45 | struct lttng_consumer_global_data consumer_data = { | |
3bd1e081 MD |
46 | .stream_count = 0, |
47 | .need_update = 1, | |
48 | .type = LTTNG_CONSUMER_UNKNOWN, | |
49 | }; | |
50 | ||
d8ef542d MD |
51 | enum consumer_channel_action { |
52 | CONSUMER_CHANNEL_ADD, | |
a0cbdd2e | 53 | CONSUMER_CHANNEL_DEL, |
d8ef542d MD |
54 | CONSUMER_CHANNEL_QUIT, |
55 | }; | |
56 | ||
57 | struct consumer_channel_msg { | |
58 | enum consumer_channel_action action; | |
a0cbdd2e MD |
59 | struct lttng_consumer_channel *chan; /* add */ |
60 | uint64_t key; /* del */ | |
d8ef542d MD |
61 | }; |
62 | ||
3bd1e081 MD |
63 | /* |
64 | * Flag to inform the polling thread to quit when all fd hung up. Updated by | |
65 | * the consumer_thread_receive_fds when it notices that all fds has hung up. | |
66 | * Also updated by the signal handler (consumer_should_exit()). Read by the | |
67 | * polling threads. | |
68 | */ | |
a98dae5f | 69 | volatile int consumer_quit; |
3bd1e081 | 70 | |
43c34bc3 | 71 | /* |
43c34bc3 DG |
72 | * Global hash table containing respectively metadata and data streams. The |
73 | * stream element in this ht should only be updated by the metadata poll thread | |
74 | * for the metadata and the data poll thread for the data. | |
75 | */ | |
40dc48e0 DG |
76 | static struct lttng_ht *metadata_ht; |
77 | static struct lttng_ht *data_ht; | |
43c34bc3 | 78 | |
acdb9057 DG |
79 | /* |
80 | * Notify a thread lttng pipe to poll back again. This usually means that some | |
81 | * global state has changed so we just send back the thread in a poll wait | |
82 | * call. | |
83 | */ | |
84 | static void notify_thread_lttng_pipe(struct lttng_pipe *pipe) | |
85 | { | |
86 | struct lttng_consumer_stream *null_stream = NULL; | |
87 | ||
88 | assert(pipe); | |
89 | ||
90 | (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream)); | |
91 | } | |
92 | ||
d8ef542d MD |
93 | static void notify_channel_pipe(struct lttng_consumer_local_data *ctx, |
94 | struct lttng_consumer_channel *chan, | |
a0cbdd2e | 95 | uint64_t key, |
d8ef542d MD |
96 | enum consumer_channel_action action) |
97 | { | |
98 | struct consumer_channel_msg msg; | |
99 | int ret; | |
100 | ||
e56251fc DG |
101 | memset(&msg, 0, sizeof(msg)); |
102 | ||
d8ef542d MD |
103 | msg.action = action; |
104 | msg.chan = chan; | |
f21dae48 | 105 | msg.key = key; |
d8ef542d MD |
106 | do { |
107 | ret = write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg)); | |
108 | } while (ret < 0 && errno == EINTR); | |
109 | } | |
110 | ||
a0cbdd2e MD |
111 | void notify_thread_del_channel(struct lttng_consumer_local_data *ctx, |
112 | uint64_t key) | |
113 | { | |
114 | notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL); | |
115 | } | |
116 | ||
d8ef542d MD |
117 | static int read_channel_pipe(struct lttng_consumer_local_data *ctx, |
118 | struct lttng_consumer_channel **chan, | |
a0cbdd2e | 119 | uint64_t *key, |
d8ef542d MD |
120 | enum consumer_channel_action *action) |
121 | { | |
122 | struct consumer_channel_msg msg; | |
123 | int ret; | |
124 | ||
125 | do { | |
126 | ret = read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg)); | |
127 | } while (ret < 0 && errno == EINTR); | |
128 | if (ret > 0) { | |
129 | *action = msg.action; | |
130 | *chan = msg.chan; | |
a0cbdd2e | 131 | *key = msg.key; |
d8ef542d MD |
132 | } |
133 | return ret; | |
134 | } | |
135 | ||
3bd1e081 MD |
136 | /* |
137 | * Find a stream. The consumer_data.lock must be locked during this | |
138 | * call. | |
139 | */ | |
d88aee68 | 140 | static struct lttng_consumer_stream *find_stream(uint64_t key, |
8389e4f8 | 141 | struct lttng_ht *ht) |
3bd1e081 | 142 | { |
e4421fec | 143 | struct lttng_ht_iter iter; |
d88aee68 | 144 | struct lttng_ht_node_u64 *node; |
e4421fec | 145 | struct lttng_consumer_stream *stream = NULL; |
3bd1e081 | 146 | |
8389e4f8 DG |
147 | assert(ht); |
148 | ||
d88aee68 DG |
149 | /* -1ULL keys are lookup failures */ |
150 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 151 | return NULL; |
7a57cf92 | 152 | } |
e4421fec | 153 | |
6065ceec DG |
154 | rcu_read_lock(); |
155 | ||
d88aee68 DG |
156 | lttng_ht_lookup(ht, &key, &iter); |
157 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
158 | if (node != NULL) { |
159 | stream = caa_container_of(node, struct lttng_consumer_stream, node); | |
3bd1e081 | 160 | } |
e4421fec | 161 | |
6065ceec DG |
162 | rcu_read_unlock(); |
163 | ||
e4421fec | 164 | return stream; |
3bd1e081 MD |
165 | } |
166 | ||
98cf381a | 167 | static void steal_stream_key(uint64_t key, struct lttng_ht *ht) |
7ad0a0cb MD |
168 | { |
169 | struct lttng_consumer_stream *stream; | |
170 | ||
04253271 | 171 | rcu_read_lock(); |
ffe60014 | 172 | stream = find_stream(key, ht); |
04253271 | 173 | if (stream) { |
98cf381a | 174 | stream->key = (uint64_t) -1ULL; |
04253271 MD |
175 | /* |
176 | * We don't want the lookup to match, but we still need | |
177 | * to iterate on this stream when iterating over the hash table. Just | |
178 | * change the node key. | |
179 | */ | |
98cf381a | 180 | stream->node.key = (uint64_t) -1ULL; |
04253271 MD |
181 | } |
182 | rcu_read_unlock(); | |
7ad0a0cb MD |
183 | } |
184 | ||
d56db448 DG |
185 | /* |
186 | * Return a channel object for the given key. | |
187 | * | |
188 | * RCU read side lock MUST be acquired before calling this function and | |
189 | * protects the channel ptr. | |
190 | */ | |
d88aee68 | 191 | struct lttng_consumer_channel *consumer_find_channel(uint64_t key) |
3bd1e081 | 192 | { |
e4421fec | 193 | struct lttng_ht_iter iter; |
d88aee68 | 194 | struct lttng_ht_node_u64 *node; |
e4421fec | 195 | struct lttng_consumer_channel *channel = NULL; |
3bd1e081 | 196 | |
d88aee68 DG |
197 | /* -1ULL keys are lookup failures */ |
198 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 199 | return NULL; |
7a57cf92 | 200 | } |
e4421fec | 201 | |
d88aee68 DG |
202 | lttng_ht_lookup(consumer_data.channel_ht, &key, &iter); |
203 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
204 | if (node != NULL) { |
205 | channel = caa_container_of(node, struct lttng_consumer_channel, node); | |
3bd1e081 | 206 | } |
e4421fec DG |
207 | |
208 | return channel; | |
3bd1e081 MD |
209 | } |
210 | ||
ffe60014 | 211 | static void free_stream_rcu(struct rcu_head *head) |
7ad0a0cb | 212 | { |
d88aee68 DG |
213 | struct lttng_ht_node_u64 *node = |
214 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
ffe60014 DG |
215 | struct lttng_consumer_stream *stream = |
216 | caa_container_of(node, struct lttng_consumer_stream, node); | |
7ad0a0cb | 217 | |
ffe60014 | 218 | free(stream); |
7ad0a0cb MD |
219 | } |
220 | ||
ffe60014 | 221 | static void free_channel_rcu(struct rcu_head *head) |
702b1ea4 | 222 | { |
d88aee68 DG |
223 | struct lttng_ht_node_u64 *node = |
224 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
ffe60014 DG |
225 | struct lttng_consumer_channel *channel = |
226 | caa_container_of(node, struct lttng_consumer_channel, node); | |
702b1ea4 | 227 | |
ffe60014 | 228 | free(channel); |
702b1ea4 MD |
229 | } |
230 | ||
00e2e675 DG |
231 | /* |
232 | * RCU protected relayd socket pair free. | |
233 | */ | |
ffe60014 | 234 | static void free_relayd_rcu(struct rcu_head *head) |
00e2e675 | 235 | { |
d88aee68 DG |
236 | struct lttng_ht_node_u64 *node = |
237 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
00e2e675 DG |
238 | struct consumer_relayd_sock_pair *relayd = |
239 | caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
240 | ||
8994307f DG |
241 | /* |
242 | * Close all sockets. This is done in the call RCU since we don't want the | |
243 | * socket fds to be reassigned thus potentially creating bad state of the | |
244 | * relayd object. | |
245 | * | |
246 | * We do not have to lock the control socket mutex here since at this stage | |
247 | * there is no one referencing to this relayd object. | |
248 | */ | |
249 | (void) relayd_close(&relayd->control_sock); | |
250 | (void) relayd_close(&relayd->data_sock); | |
251 | ||
00e2e675 DG |
252 | free(relayd); |
253 | } | |
254 | ||
255 | /* | |
256 | * Destroy and free relayd socket pair object. | |
257 | * | |
258 | * This function MUST be called with the consumer_data lock acquired. | |
259 | */ | |
d09e1200 | 260 | static void destroy_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
261 | { |
262 | int ret; | |
263 | struct lttng_ht_iter iter; | |
264 | ||
173af62f DG |
265 | if (relayd == NULL) { |
266 | return; | |
267 | } | |
268 | ||
00e2e675 DG |
269 | DBG("Consumer destroy and close relayd socket pair"); |
270 | ||
271 | iter.iter.node = &relayd->node.node; | |
272 | ret = lttng_ht_del(consumer_data.relayd_ht, &iter); | |
173af62f | 273 | if (ret != 0) { |
8994307f | 274 | /* We assume the relayd is being or is destroyed */ |
173af62f DG |
275 | return; |
276 | } | |
00e2e675 | 277 | |
00e2e675 | 278 | /* RCU free() call */ |
ffe60014 DG |
279 | call_rcu(&relayd->node.head, free_relayd_rcu); |
280 | } | |
281 | ||
282 | /* | |
283 | * Remove a channel from the global list protected by a mutex. This function is | |
284 | * also responsible for freeing its data structures. | |
285 | */ | |
286 | void consumer_del_channel(struct lttng_consumer_channel *channel) | |
287 | { | |
288 | int ret; | |
289 | struct lttng_ht_iter iter; | |
f2a444f1 | 290 | struct lttng_consumer_stream *stream, *stmp; |
ffe60014 | 291 | |
d88aee68 | 292 | DBG("Consumer delete channel key %" PRIu64, channel->key); |
ffe60014 DG |
293 | |
294 | pthread_mutex_lock(&consumer_data.lock); | |
72c3879d | 295 | pthread_mutex_lock(&channel->lock); |
ffe60014 DG |
296 | |
297 | switch (consumer_data.type) { | |
298 | case LTTNG_CONSUMER_KERNEL: | |
299 | break; | |
300 | case LTTNG_CONSUMER32_UST: | |
301 | case LTTNG_CONSUMER64_UST: | |
f2a444f1 DG |
302 | /* Delete streams that might have been left in the stream list. */ |
303 | cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head, | |
304 | send_node) { | |
305 | cds_list_del(&stream->send_node); | |
306 | lttng_ustconsumer_del_stream(stream); | |
307 | free(stream); | |
308 | } | |
ffe60014 DG |
309 | lttng_ustconsumer_del_channel(channel); |
310 | break; | |
311 | default: | |
312 | ERR("Unknown consumer_data type"); | |
313 | assert(0); | |
314 | goto end; | |
315 | } | |
316 | ||
317 | rcu_read_lock(); | |
318 | iter.iter.node = &channel->node.node; | |
319 | ret = lttng_ht_del(consumer_data.channel_ht, &iter); | |
320 | assert(!ret); | |
321 | rcu_read_unlock(); | |
322 | ||
323 | call_rcu(&channel->node.head, free_channel_rcu); | |
324 | end: | |
72c3879d | 325 | pthread_mutex_unlock(&channel->lock); |
ffe60014 | 326 | pthread_mutex_unlock(&consumer_data.lock); |
00e2e675 DG |
327 | } |
328 | ||
228b5bf7 DG |
329 | /* |
330 | * Iterate over the relayd hash table and destroy each element. Finally, | |
331 | * destroy the whole hash table. | |
332 | */ | |
333 | static void cleanup_relayd_ht(void) | |
334 | { | |
335 | struct lttng_ht_iter iter; | |
336 | struct consumer_relayd_sock_pair *relayd; | |
337 | ||
338 | rcu_read_lock(); | |
339 | ||
340 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
341 | node.node) { | |
342 | destroy_relayd(relayd); | |
343 | } | |
344 | ||
228b5bf7 | 345 | rcu_read_unlock(); |
36b588ed MD |
346 | |
347 | lttng_ht_destroy(consumer_data.relayd_ht); | |
228b5bf7 DG |
348 | } |
349 | ||
8994307f DG |
350 | /* |
351 | * Update the end point status of all streams having the given network sequence | |
352 | * index (relayd index). | |
353 | * | |
354 | * It's atomically set without having the stream mutex locked which is fine | |
355 | * because we handle the write/read race with a pipe wakeup for each thread. | |
356 | */ | |
98cf381a | 357 | static void update_endpoint_status_by_netidx(uint64_t net_seq_idx, |
8994307f DG |
358 | enum consumer_endpoint_status status) |
359 | { | |
360 | struct lttng_ht_iter iter; | |
361 | struct lttng_consumer_stream *stream; | |
362 | ||
98cf381a | 363 | DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx); |
8994307f DG |
364 | |
365 | rcu_read_lock(); | |
366 | ||
367 | /* Let's begin with metadata */ | |
368 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
369 | if (stream->net_seq_idx == net_seq_idx) { | |
370 | uatomic_set(&stream->endpoint_status, status); | |
371 | DBG("Delete flag set to metadata stream %d", stream->wait_fd); | |
372 | } | |
373 | } | |
374 | ||
375 | /* Follow up by the data streams */ | |
376 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
377 | if (stream->net_seq_idx == net_seq_idx) { | |
378 | uatomic_set(&stream->endpoint_status, status); | |
379 | DBG("Delete flag set to data stream %d", stream->wait_fd); | |
380 | } | |
381 | } | |
382 | rcu_read_unlock(); | |
383 | } | |
384 | ||
385 | /* | |
386 | * Cleanup a relayd object by flagging every associated streams for deletion, | |
387 | * destroying the object meaning removing it from the relayd hash table, | |
388 | * closing the sockets and freeing the memory in a RCU call. | |
389 | * | |
390 | * If a local data context is available, notify the threads that the streams' | |
391 | * state have changed. | |
392 | */ | |
393 | static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd, | |
394 | struct lttng_consumer_local_data *ctx) | |
395 | { | |
98cf381a | 396 | uint64_t netidx; |
8994307f DG |
397 | |
398 | assert(relayd); | |
399 | ||
9617607b DG |
400 | DBG("Cleaning up relayd sockets"); |
401 | ||
8994307f DG |
402 | /* Save the net sequence index before destroying the object */ |
403 | netidx = relayd->net_seq_idx; | |
404 | ||
405 | /* | |
406 | * Delete the relayd from the relayd hash table, close the sockets and free | |
407 | * the object in a RCU call. | |
408 | */ | |
409 | destroy_relayd(relayd); | |
410 | ||
411 | /* Set inactive endpoint to all streams */ | |
412 | update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE); | |
413 | ||
414 | /* | |
415 | * With a local data context, notify the threads that the streams' state | |
416 | * have changed. The write() action on the pipe acts as an "implicit" | |
417 | * memory barrier ordering the updates of the end point status from the | |
418 | * read of this status which happens AFTER receiving this notify. | |
419 | */ | |
420 | if (ctx) { | |
acdb9057 | 421 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
13886d2d | 422 | notify_thread_lttng_pipe(ctx->consumer_metadata_pipe); |
8994307f DG |
423 | } |
424 | } | |
425 | ||
a6ba4fe1 DG |
426 | /* |
427 | * Flag a relayd socket pair for destruction. Destroy it if the refcount | |
428 | * reaches zero. | |
429 | * | |
430 | * RCU read side lock MUST be aquired before calling this function. | |
431 | */ | |
432 | void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) | |
433 | { | |
434 | assert(relayd); | |
435 | ||
436 | /* Set destroy flag for this object */ | |
437 | uatomic_set(&relayd->destroy_flag, 1); | |
438 | ||
439 | /* Destroy the relayd if refcount is 0 */ | |
440 | if (uatomic_read(&relayd->refcount) == 0) { | |
d09e1200 | 441 | destroy_relayd(relayd); |
a6ba4fe1 DG |
442 | } |
443 | } | |
444 | ||
3bd1e081 MD |
445 | /* |
446 | * Remove a stream from the global list protected by a mutex. This | |
447 | * function is also responsible for freeing its data structures. | |
448 | */ | |
e316aad5 DG |
449 | void consumer_del_stream(struct lttng_consumer_stream *stream, |
450 | struct lttng_ht *ht) | |
3bd1e081 MD |
451 | { |
452 | int ret; | |
e4421fec | 453 | struct lttng_ht_iter iter; |
3bd1e081 | 454 | struct lttng_consumer_channel *free_chan = NULL; |
00e2e675 DG |
455 | struct consumer_relayd_sock_pair *relayd; |
456 | ||
457 | assert(stream); | |
3bd1e081 | 458 | |
8994307f DG |
459 | DBG("Consumer del stream %d", stream->wait_fd); |
460 | ||
e316aad5 DG |
461 | if (ht == NULL) { |
462 | /* Means the stream was allocated but not successfully added */ | |
ffe60014 | 463 | goto free_stream_rcu; |
e316aad5 DG |
464 | } |
465 | ||
3bd1e081 | 466 | pthread_mutex_lock(&consumer_data.lock); |
74251bb8 | 467 | pthread_mutex_lock(&stream->lock); |
3bd1e081 MD |
468 | |
469 | switch (consumer_data.type) { | |
470 | case LTTNG_CONSUMER_KERNEL: | |
471 | if (stream->mmap_base != NULL) { | |
472 | ret = munmap(stream->mmap_base, stream->mmap_len); | |
473 | if (ret != 0) { | |
7a57cf92 | 474 | PERROR("munmap"); |
3bd1e081 MD |
475 | } |
476 | } | |
4c95e622 JD |
477 | |
478 | if (stream->wait_fd >= 0) { | |
479 | ret = close(stream->wait_fd); | |
480 | if (ret) { | |
481 | PERROR("close"); | |
482 | } | |
483 | } | |
3bd1e081 | 484 | break; |
7753dea8 MD |
485 | case LTTNG_CONSUMER32_UST: |
486 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
487 | lttng_ustconsumer_del_stream(stream); |
488 | break; | |
489 | default: | |
490 | ERR("Unknown consumer_data type"); | |
491 | assert(0); | |
492 | goto end; | |
493 | } | |
494 | ||
6065ceec | 495 | rcu_read_lock(); |
04253271 | 496 | iter.iter.node = &stream->node.node; |
e316aad5 | 497 | ret = lttng_ht_del(ht, &iter); |
04253271 | 498 | assert(!ret); |
ca22feea | 499 | |
d8ef542d MD |
500 | iter.iter.node = &stream->node_channel_id.node; |
501 | ret = lttng_ht_del(consumer_data.stream_per_chan_id_ht, &iter); | |
502 | assert(!ret); | |
503 | ||
ca22feea DG |
504 | iter.iter.node = &stream->node_session_id.node; |
505 | ret = lttng_ht_del(consumer_data.stream_list_ht, &iter); | |
506 | assert(!ret); | |
6065ceec DG |
507 | rcu_read_unlock(); |
508 | ||
50f8ae69 | 509 | assert(consumer_data.stream_count > 0); |
3bd1e081 | 510 | consumer_data.stream_count--; |
50f8ae69 | 511 | |
3bd1e081 | 512 | if (stream->out_fd >= 0) { |
4c462e79 MD |
513 | ret = close(stream->out_fd); |
514 | if (ret) { | |
515 | PERROR("close"); | |
516 | } | |
3bd1e081 | 517 | } |
00e2e675 DG |
518 | |
519 | /* Check and cleanup relayd */ | |
b0b335c8 | 520 | rcu_read_lock(); |
00e2e675 DG |
521 | relayd = consumer_find_relayd(stream->net_seq_idx); |
522 | if (relayd != NULL) { | |
b0b335c8 MD |
523 | uatomic_dec(&relayd->refcount); |
524 | assert(uatomic_read(&relayd->refcount) >= 0); | |
173af62f | 525 | |
3f8e211f DG |
526 | /* Closing streams requires to lock the control socket. */ |
527 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
173af62f DG |
528 | ret = relayd_send_close_stream(&relayd->control_sock, |
529 | stream->relayd_stream_id, | |
530 | stream->next_net_seq_num - 1); | |
3f8e211f | 531 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
173af62f | 532 | if (ret < 0) { |
a4b92340 DG |
533 | DBG("Unable to close stream on the relayd. Continuing"); |
534 | /* | |
535 | * Continue here. There is nothing we can do for the relayd. | |
536 | * Chances are that the relayd has closed the socket so we just | |
537 | * continue cleaning up. | |
538 | */ | |
173af62f DG |
539 | } |
540 | ||
541 | /* Both conditions are met, we destroy the relayd. */ | |
542 | if (uatomic_read(&relayd->refcount) == 0 && | |
543 | uatomic_read(&relayd->destroy_flag)) { | |
d09e1200 | 544 | destroy_relayd(relayd); |
00e2e675 | 545 | } |
00e2e675 | 546 | } |
b0b335c8 | 547 | rcu_read_unlock(); |
00e2e675 | 548 | |
f2ad556d | 549 | if (!uatomic_sub_return(&stream->chan->refcount, 1) |
ffe60014 | 550 | && !uatomic_read(&stream->chan->nb_init_stream_left)) { |
3bd1e081 | 551 | free_chan = stream->chan; |
00e2e675 DG |
552 | } |
553 | ||
3bd1e081 MD |
554 | end: |
555 | consumer_data.need_update = 1; | |
8994307f | 556 | pthread_mutex_unlock(&stream->lock); |
74251bb8 | 557 | pthread_mutex_unlock(&consumer_data.lock); |
3bd1e081 | 558 | |
c30aaa51 | 559 | if (free_chan) { |
3bd1e081 | 560 | consumer_del_channel(free_chan); |
c30aaa51 | 561 | } |
e316aad5 | 562 | |
ffe60014 DG |
563 | free_stream_rcu: |
564 | call_rcu(&stream->node.head, free_stream_rcu); | |
3bd1e081 MD |
565 | } |
566 | ||
155e9bb8 MD |
567 | /* |
568 | * XXX naming of del vs destroy is all mixed up. | |
569 | */ | |
570 | void consumer_del_stream_for_data(struct lttng_consumer_stream *stream) | |
571 | { | |
572 | consumer_del_stream(stream, data_ht); | |
573 | } | |
574 | ||
575 | void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream) | |
576 | { | |
577 | consumer_del_stream(stream, metadata_ht); | |
578 | } | |
579 | ||
d88aee68 DG |
580 | struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, |
581 | uint64_t stream_key, | |
3bd1e081 | 582 | enum lttng_consumer_stream_state state, |
ffe60014 | 583 | const char *channel_name, |
6df2e2c9 | 584 | uid_t uid, |
00e2e675 | 585 | gid_t gid, |
57a269f2 | 586 | uint64_t relayd_id, |
53632229 | 587 | uint64_t session_id, |
ffe60014 DG |
588 | int cpu, |
589 | int *alloc_ret, | |
590 | enum consumer_channel_type type) | |
3bd1e081 | 591 | { |
ffe60014 | 592 | int ret; |
3bd1e081 | 593 | struct lttng_consumer_stream *stream; |
3bd1e081 | 594 | |
effcf122 | 595 | stream = zmalloc(sizeof(*stream)); |
3bd1e081 | 596 | if (stream == NULL) { |
7a57cf92 | 597 | PERROR("malloc struct lttng_consumer_stream"); |
ffe60014 | 598 | ret = -ENOMEM; |
7a57cf92 | 599 | goto end; |
3bd1e081 | 600 | } |
7a57cf92 | 601 | |
d56db448 DG |
602 | rcu_read_lock(); |
603 | ||
3bd1e081 | 604 | stream->key = stream_key; |
3bd1e081 MD |
605 | stream->out_fd = -1; |
606 | stream->out_fd_offset = 0; | |
607 | stream->state = state; | |
6df2e2c9 MD |
608 | stream->uid = uid; |
609 | stream->gid = gid; | |
ffe60014 | 610 | stream->net_seq_idx = relayd_id; |
53632229 | 611 | stream->session_id = session_id; |
bf9abc51 | 612 | stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE; |
53632229 | 613 | pthread_mutex_init(&stream->lock, NULL); |
58b1f425 | 614 | |
ffe60014 DG |
615 | /* If channel is the metadata, flag this stream as metadata. */ |
616 | if (type == CONSUMER_CHANNEL_TYPE_METADATA) { | |
617 | stream->metadata_flag = 1; | |
618 | /* Metadata is flat out. */ | |
619 | strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name)); | |
58b1f425 | 620 | } else { |
ffe60014 DG |
621 | /* Format stream name to <channel_name>_<cpu_number> */ |
622 | ret = snprintf(stream->name, sizeof(stream->name), "%s_%d", | |
623 | channel_name, cpu); | |
624 | if (ret < 0) { | |
625 | PERROR("snprintf stream name"); | |
626 | goto error; | |
627 | } | |
58b1f425 | 628 | } |
c30aaa51 | 629 | |
ffe60014 | 630 | /* Key is always the wait_fd for streams. */ |
d88aee68 | 631 | lttng_ht_node_init_u64(&stream->node, stream->key); |
ffe60014 | 632 | |
d8ef542d MD |
633 | /* Init node per channel id key */ |
634 | lttng_ht_node_init_u64(&stream->node_channel_id, channel_key); | |
635 | ||
53632229 | 636 | /* Init session id node with the stream session id */ |
d88aee68 | 637 | lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id); |
53632229 | 638 | |
d8ef542d MD |
639 | DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64 " relayd_id %" PRIu64 ", session_id %" PRIu64, |
640 | stream->name, stream->key, channel_key, stream->net_seq_idx, stream->session_id); | |
d56db448 DG |
641 | |
642 | rcu_read_unlock(); | |
3bd1e081 | 643 | return stream; |
c80048c6 MD |
644 | |
645 | error: | |
d56db448 | 646 | rcu_read_unlock(); |
c80048c6 | 647 | free(stream); |
7a57cf92 | 648 | end: |
ffe60014 DG |
649 | if (alloc_ret) { |
650 | *alloc_ret = ret; | |
651 | } | |
c80048c6 | 652 | return NULL; |
3bd1e081 MD |
653 | } |
654 | ||
655 | /* | |
656 | * Add a stream to the global list protected by a mutex. | |
657 | */ | |
155e9bb8 | 658 | int consumer_add_data_stream(struct lttng_consumer_stream *stream) |
3bd1e081 | 659 | { |
155e9bb8 | 660 | struct lttng_ht *ht = data_ht; |
3bd1e081 | 661 | int ret = 0; |
00e2e675 | 662 | struct consumer_relayd_sock_pair *relayd; |
3bd1e081 | 663 | |
e316aad5 | 664 | assert(stream); |
43c34bc3 | 665 | assert(ht); |
c77fc10a | 666 | |
d88aee68 | 667 | DBG3("Adding consumer stream %" PRIu64, stream->key); |
e316aad5 DG |
668 | |
669 | pthread_mutex_lock(&consumer_data.lock); | |
72c3879d | 670 | pthread_mutex_lock(&stream->chan->lock); |
6ca1a6b0 | 671 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 672 | pthread_mutex_lock(&stream->lock); |
b0b335c8 | 673 | rcu_read_lock(); |
e316aad5 | 674 | |
43c34bc3 | 675 | /* Steal stream identifier to avoid having streams with the same key */ |
ffe60014 | 676 | steal_stream_key(stream->key, ht); |
43c34bc3 | 677 | |
d88aee68 | 678 | lttng_ht_add_unique_u64(ht, &stream->node); |
00e2e675 | 679 | |
d8ef542d MD |
680 | lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, |
681 | &stream->node_channel_id); | |
682 | ||
ca22feea DG |
683 | /* |
684 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
685 | * the key since the HT does not use it and we allow to add redundant keys | |
686 | * into this table. | |
687 | */ | |
d88aee68 | 688 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 689 | |
00e2e675 DG |
690 | /* Check and cleanup relayd */ |
691 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
692 | if (relayd != NULL) { | |
b0b335c8 | 693 | uatomic_inc(&relayd->refcount); |
00e2e675 DG |
694 | } |
695 | ||
e316aad5 | 696 | /* |
ffe60014 DG |
697 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
698 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
699 | * causes the count to become 0 also causes a stream to be added. The |
700 | * channel deletion will thus be triggered by the following removal of this | |
701 | * stream. | |
702 | */ | |
ffe60014 | 703 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
704 | /* Increment refcount before decrementing nb_init_stream_left */ |
705 | cmm_smp_wmb(); | |
ffe60014 | 706 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
707 | } |
708 | ||
709 | /* Update consumer data once the node is inserted. */ | |
3bd1e081 MD |
710 | consumer_data.stream_count++; |
711 | consumer_data.need_update = 1; | |
712 | ||
e316aad5 | 713 | rcu_read_unlock(); |
2e818a6a | 714 | pthread_mutex_unlock(&stream->lock); |
6ca1a6b0 | 715 | pthread_mutex_unlock(&stream->chan->timer_lock); |
72c3879d | 716 | pthread_mutex_unlock(&stream->chan->lock); |
3bd1e081 | 717 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 718 | |
3bd1e081 MD |
719 | return ret; |
720 | } | |
721 | ||
155e9bb8 MD |
722 | void consumer_del_data_stream(struct lttng_consumer_stream *stream) |
723 | { | |
724 | consumer_del_stream(stream, data_ht); | |
725 | } | |
726 | ||
00e2e675 | 727 | /* |
3f8e211f DG |
728 | * Add relayd socket to global consumer data hashtable. RCU read side lock MUST |
729 | * be acquired before calling this. | |
00e2e675 | 730 | */ |
d09e1200 | 731 | static int add_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
732 | { |
733 | int ret = 0; | |
d88aee68 | 734 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
735 | struct lttng_ht_iter iter; |
736 | ||
ffe60014 | 737 | assert(relayd); |
00e2e675 | 738 | |
00e2e675 | 739 | lttng_ht_lookup(consumer_data.relayd_ht, |
d88aee68 DG |
740 | &relayd->net_seq_idx, &iter); |
741 | node = lttng_ht_iter_get_node_u64(&iter); | |
00e2e675 | 742 | if (node != NULL) { |
00e2e675 DG |
743 | goto end; |
744 | } | |
d88aee68 | 745 | lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node); |
00e2e675 | 746 | |
00e2e675 DG |
747 | end: |
748 | return ret; | |
749 | } | |
750 | ||
751 | /* | |
752 | * Allocate and return a consumer relayd socket. | |
753 | */ | |
754 | struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( | |
98cf381a | 755 | uint64_t net_seq_idx) |
00e2e675 DG |
756 | { |
757 | struct consumer_relayd_sock_pair *obj = NULL; | |
758 | ||
98cf381a MD |
759 | /* net sequence index of -1 is a failure */ |
760 | if (net_seq_idx == (uint64_t) -1ULL) { | |
00e2e675 DG |
761 | goto error; |
762 | } | |
763 | ||
764 | obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); | |
765 | if (obj == NULL) { | |
766 | PERROR("zmalloc relayd sock"); | |
767 | goto error; | |
768 | } | |
769 | ||
770 | obj->net_seq_idx = net_seq_idx; | |
771 | obj->refcount = 0; | |
173af62f | 772 | obj->destroy_flag = 0; |
f96e4545 MD |
773 | obj->control_sock.sock.fd = -1; |
774 | obj->data_sock.sock.fd = -1; | |
d88aee68 | 775 | lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx); |
00e2e675 DG |
776 | pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); |
777 | ||
778 | error: | |
779 | return obj; | |
780 | } | |
781 | ||
782 | /* | |
783 | * Find a relayd socket pair in the global consumer data. | |
784 | * | |
785 | * Return the object if found else NULL. | |
b0b335c8 MD |
786 | * RCU read-side lock must be held across this call and while using the |
787 | * returned object. | |
00e2e675 | 788 | */ |
d88aee68 | 789 | struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key) |
00e2e675 DG |
790 | { |
791 | struct lttng_ht_iter iter; | |
d88aee68 | 792 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
793 | struct consumer_relayd_sock_pair *relayd = NULL; |
794 | ||
795 | /* Negative keys are lookup failures */ | |
d88aee68 | 796 | if (key == (uint64_t) -1ULL) { |
00e2e675 DG |
797 | goto error; |
798 | } | |
799 | ||
d88aee68 | 800 | lttng_ht_lookup(consumer_data.relayd_ht, &key, |
00e2e675 | 801 | &iter); |
d88aee68 | 802 | node = lttng_ht_iter_get_node_u64(&iter); |
00e2e675 DG |
803 | if (node != NULL) { |
804 | relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
805 | } | |
806 | ||
00e2e675 DG |
807 | error: |
808 | return relayd; | |
809 | } | |
810 | ||
811 | /* | |
812 | * Handle stream for relayd transmission if the stream applies for network | |
813 | * streaming where the net sequence index is set. | |
814 | * | |
815 | * Return destination file descriptor or negative value on error. | |
816 | */ | |
6197aea7 | 817 | static int write_relayd_stream_header(struct lttng_consumer_stream *stream, |
1d4dfdef DG |
818 | size_t data_size, unsigned long padding, |
819 | struct consumer_relayd_sock_pair *relayd) | |
00e2e675 DG |
820 | { |
821 | int outfd = -1, ret; | |
00e2e675 DG |
822 | struct lttcomm_relayd_data_hdr data_hdr; |
823 | ||
824 | /* Safety net */ | |
825 | assert(stream); | |
6197aea7 | 826 | assert(relayd); |
00e2e675 DG |
827 | |
828 | /* Reset data header */ | |
829 | memset(&data_hdr, 0, sizeof(data_hdr)); | |
830 | ||
00e2e675 DG |
831 | if (stream->metadata_flag) { |
832 | /* Caller MUST acquire the relayd control socket lock */ | |
833 | ret = relayd_send_metadata(&relayd->control_sock, data_size); | |
834 | if (ret < 0) { | |
835 | goto error; | |
836 | } | |
837 | ||
838 | /* Metadata are always sent on the control socket. */ | |
6151a90f | 839 | outfd = relayd->control_sock.sock.fd; |
00e2e675 DG |
840 | } else { |
841 | /* Set header with stream information */ | |
842 | data_hdr.stream_id = htobe64(stream->relayd_stream_id); | |
843 | data_hdr.data_size = htobe32(data_size); | |
1d4dfdef | 844 | data_hdr.padding_size = htobe32(padding); |
39df6d9f DG |
845 | /* |
846 | * Note that net_seq_num below is assigned with the *current* value of | |
847 | * next_net_seq_num and only after that the next_net_seq_num will be | |
848 | * increment. This is why when issuing a command on the relayd using | |
849 | * this next value, 1 should always be substracted in order to compare | |
850 | * the last seen sequence number on the relayd side to the last sent. | |
851 | */ | |
3604f373 | 852 | data_hdr.net_seq_num = htobe64(stream->next_net_seq_num); |
00e2e675 DG |
853 | /* Other fields are zeroed previously */ |
854 | ||
855 | ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, | |
856 | sizeof(data_hdr)); | |
857 | if (ret < 0) { | |
858 | goto error; | |
859 | } | |
860 | ||
3604f373 DG |
861 | ++stream->next_net_seq_num; |
862 | ||
00e2e675 | 863 | /* Set to go on data socket */ |
6151a90f | 864 | outfd = relayd->data_sock.sock.fd; |
00e2e675 DG |
865 | } |
866 | ||
867 | error: | |
868 | return outfd; | |
869 | } | |
870 | ||
3bd1e081 | 871 | /* |
ffe60014 DG |
872 | * Allocate and return a new lttng_consumer_channel object using the given key |
873 | * to initialize the hash table node. | |
874 | * | |
875 | * On error, return NULL. | |
3bd1e081 | 876 | */ |
886224ff | 877 | struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, |
ffe60014 DG |
878 | uint64_t session_id, |
879 | const char *pathname, | |
880 | const char *name, | |
881 | uid_t uid, | |
882 | gid_t gid, | |
57a269f2 | 883 | uint64_t relayd_id, |
1624d5b7 JD |
884 | enum lttng_event_output output, |
885 | uint64_t tracefile_size, | |
cb7c6909 JD |
886 | uint64_t tracefile_count, |
887 | uint64_t session_id_per_pid) | |
3bd1e081 MD |
888 | { |
889 | struct lttng_consumer_channel *channel; | |
3bd1e081 | 890 | |
276b26d1 | 891 | channel = zmalloc(sizeof(*channel)); |
3bd1e081 | 892 | if (channel == NULL) { |
7a57cf92 | 893 | PERROR("malloc struct lttng_consumer_channel"); |
3bd1e081 MD |
894 | goto end; |
895 | } | |
ffe60014 DG |
896 | |
897 | channel->key = key; | |
3bd1e081 | 898 | channel->refcount = 0; |
ffe60014 | 899 | channel->session_id = session_id; |
cb7c6909 | 900 | channel->session_id_per_pid = session_id_per_pid; |
ffe60014 DG |
901 | channel->uid = uid; |
902 | channel->gid = gid; | |
903 | channel->relayd_id = relayd_id; | |
904 | channel->output = output; | |
1624d5b7 JD |
905 | channel->tracefile_size = tracefile_size; |
906 | channel->tracefile_count = tracefile_count; | |
72c3879d | 907 | pthread_mutex_init(&channel->lock, NULL); |
6ca1a6b0 | 908 | pthread_mutex_init(&channel->timer_lock, NULL); |
ffe60014 DG |
909 | |
910 | strncpy(channel->pathname, pathname, sizeof(channel->pathname)); | |
911 | channel->pathname[sizeof(channel->pathname) - 1] = '\0'; | |
912 | ||
913 | strncpy(channel->name, name, sizeof(channel->name)); | |
914 | channel->name[sizeof(channel->name) - 1] = '\0'; | |
915 | ||
d88aee68 | 916 | lttng_ht_node_init_u64(&channel->node, channel->key); |
d8ef542d MD |
917 | |
918 | channel->wait_fd = -1; | |
919 | ||
ffe60014 DG |
920 | CDS_INIT_LIST_HEAD(&channel->streams.head); |
921 | ||
d88aee68 | 922 | DBG("Allocated channel (key %" PRIu64 ")", channel->key) |
3bd1e081 | 923 | |
3bd1e081 MD |
924 | end: |
925 | return channel; | |
926 | } | |
927 | ||
928 | /* | |
929 | * Add a channel to the global list protected by a mutex. | |
821fffb2 DG |
930 | * |
931 | * On success 0 is returned else a negative value. | |
3bd1e081 | 932 | */ |
d8ef542d MD |
933 | int consumer_add_channel(struct lttng_consumer_channel *channel, |
934 | struct lttng_consumer_local_data *ctx) | |
3bd1e081 | 935 | { |
ffe60014 | 936 | int ret = 0; |
d88aee68 | 937 | struct lttng_ht_node_u64 *node; |
c77fc10a DG |
938 | struct lttng_ht_iter iter; |
939 | ||
3bd1e081 | 940 | pthread_mutex_lock(&consumer_data.lock); |
72c3879d | 941 | pthread_mutex_lock(&channel->lock); |
6ca1a6b0 | 942 | pthread_mutex_lock(&channel->timer_lock); |
6065ceec | 943 | rcu_read_lock(); |
c77fc10a | 944 | |
7972aab2 | 945 | lttng_ht_lookup(consumer_data.channel_ht, &channel->key, &iter); |
d88aee68 | 946 | node = lttng_ht_iter_get_node_u64(&iter); |
c77fc10a DG |
947 | if (node != NULL) { |
948 | /* Channel already exist. Ignore the insertion */ | |
d88aee68 DG |
949 | ERR("Consumer add channel key %" PRIu64 " already exists!", |
950 | channel->key); | |
821fffb2 | 951 | ret = -EEXIST; |
c77fc10a DG |
952 | goto end; |
953 | } | |
954 | ||
d88aee68 | 955 | lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node); |
c77fc10a DG |
956 | |
957 | end: | |
6065ceec | 958 | rcu_read_unlock(); |
6ca1a6b0 | 959 | pthread_mutex_unlock(&channel->timer_lock); |
72c3879d | 960 | pthread_mutex_unlock(&channel->lock); |
3bd1e081 | 961 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 962 | |
d8ef542d MD |
963 | if (!ret && channel->wait_fd != -1 && |
964 | channel->metadata_stream == NULL) { | |
a0cbdd2e | 965 | notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD); |
d8ef542d | 966 | } |
ffe60014 | 967 | return ret; |
3bd1e081 MD |
968 | } |
969 | ||
970 | /* | |
971 | * Allocate the pollfd structure and the local view of the out fds to avoid | |
972 | * doing a lookup in the linked list and concurrency issues when writing is | |
973 | * needed. Called with consumer_data.lock held. | |
974 | * | |
975 | * Returns the number of fds in the structures. | |
976 | */ | |
ffe60014 DG |
977 | static int update_poll_array(struct lttng_consumer_local_data *ctx, |
978 | struct pollfd **pollfd, struct lttng_consumer_stream **local_stream, | |
979 | struct lttng_ht *ht) | |
3bd1e081 | 980 | { |
3bd1e081 | 981 | int i = 0; |
e4421fec DG |
982 | struct lttng_ht_iter iter; |
983 | struct lttng_consumer_stream *stream; | |
3bd1e081 | 984 | |
ffe60014 DG |
985 | assert(ctx); |
986 | assert(ht); | |
987 | assert(pollfd); | |
988 | assert(local_stream); | |
989 | ||
3bd1e081 | 990 | DBG("Updating poll fd array"); |
481d6c57 | 991 | rcu_read_lock(); |
43c34bc3 | 992 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
8994307f DG |
993 | /* |
994 | * Only active streams with an active end point can be added to the | |
995 | * poll set and local stream storage of the thread. | |
996 | * | |
997 | * There is a potential race here for endpoint_status to be updated | |
998 | * just after the check. However, this is OK since the stream(s) will | |
999 | * be deleted once the thread is notified that the end point state has | |
1000 | * changed where this function will be called back again. | |
1001 | */ | |
1002 | if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM || | |
79d4ffb7 | 1003 | stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { |
3bd1e081 MD |
1004 | continue; |
1005 | } | |
7972aab2 DG |
1006 | /* |
1007 | * This clobbers way too much the debug output. Uncomment that if you | |
1008 | * need it for debugging purposes. | |
1009 | * | |
1010 | * DBG("Active FD %d", stream->wait_fd); | |
1011 | */ | |
e4421fec | 1012 | (*pollfd)[i].fd = stream->wait_fd; |
3bd1e081 | 1013 | (*pollfd)[i].events = POLLIN | POLLPRI; |
e4421fec | 1014 | local_stream[i] = stream; |
3bd1e081 MD |
1015 | i++; |
1016 | } | |
481d6c57 | 1017 | rcu_read_unlock(); |
3bd1e081 MD |
1018 | |
1019 | /* | |
50f8ae69 | 1020 | * Insert the consumer_data_pipe at the end of the array and don't |
3bd1e081 MD |
1021 | * increment i so nb_fd is the number of real FD. |
1022 | */ | |
acdb9057 | 1023 | (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe); |
509bb1cf | 1024 | (*pollfd)[i].events = POLLIN | POLLPRI; |
3bd1e081 MD |
1025 | return i; |
1026 | } | |
1027 | ||
1028 | /* | |
1029 | * Poll on the should_quit pipe and the command socket return -1 on error and | |
1030 | * should exit, 0 if data is available on the command socket | |
1031 | */ | |
1032 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) | |
1033 | { | |
1034 | int num_rdy; | |
1035 | ||
88f2b785 | 1036 | restart: |
3bd1e081 MD |
1037 | num_rdy = poll(consumer_sockpoll, 2, -1); |
1038 | if (num_rdy == -1) { | |
88f2b785 MD |
1039 | /* |
1040 | * Restart interrupted system call. | |
1041 | */ | |
1042 | if (errno == EINTR) { | |
1043 | goto restart; | |
1044 | } | |
7a57cf92 | 1045 | PERROR("Poll error"); |
3bd1e081 MD |
1046 | goto exit; |
1047 | } | |
509bb1cf | 1048 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
3bd1e081 MD |
1049 | DBG("consumer_should_quit wake up"); |
1050 | goto exit; | |
1051 | } | |
1052 | return 0; | |
1053 | ||
1054 | exit: | |
1055 | return -1; | |
1056 | } | |
1057 | ||
1058 | /* | |
1059 | * Set the error socket. | |
1060 | */ | |
ffe60014 DG |
1061 | void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx, |
1062 | int sock) | |
3bd1e081 MD |
1063 | { |
1064 | ctx->consumer_error_socket = sock; | |
1065 | } | |
1066 | ||
1067 | /* | |
1068 | * Set the command socket path. | |
1069 | */ | |
3bd1e081 MD |
1070 | void lttng_consumer_set_command_sock_path( |
1071 | struct lttng_consumer_local_data *ctx, char *sock) | |
1072 | { | |
1073 | ctx->consumer_command_sock_path = sock; | |
1074 | } | |
1075 | ||
1076 | /* | |
1077 | * Send return code to the session daemon. | |
1078 | * If the socket is not defined, we return 0, it is not a fatal error | |
1079 | */ | |
ffe60014 | 1080 | int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd) |
3bd1e081 MD |
1081 | { |
1082 | if (ctx->consumer_error_socket > 0) { | |
1083 | return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd, | |
1084 | sizeof(enum lttcomm_sessiond_command)); | |
1085 | } | |
1086 | ||
1087 | return 0; | |
1088 | } | |
1089 | ||
1090 | /* | |
228b5bf7 DG |
1091 | * Close all the tracefiles and stream fds and MUST be called when all |
1092 | * instances are destroyed i.e. when all threads were joined and are ended. | |
3bd1e081 MD |
1093 | */ |
1094 | void lttng_consumer_cleanup(void) | |
1095 | { | |
e4421fec | 1096 | struct lttng_ht_iter iter; |
ffe60014 | 1097 | struct lttng_consumer_channel *channel; |
6065ceec DG |
1098 | |
1099 | rcu_read_lock(); | |
3bd1e081 | 1100 | |
ffe60014 DG |
1101 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel, |
1102 | node.node) { | |
702b1ea4 | 1103 | consumer_del_channel(channel); |
3bd1e081 | 1104 | } |
6065ceec DG |
1105 | |
1106 | rcu_read_unlock(); | |
d6ce1df2 | 1107 | |
d6ce1df2 | 1108 | lttng_ht_destroy(consumer_data.channel_ht); |
228b5bf7 DG |
1109 | |
1110 | cleanup_relayd_ht(); | |
1111 | ||
d8ef542d MD |
1112 | lttng_ht_destroy(consumer_data.stream_per_chan_id_ht); |
1113 | ||
228b5bf7 DG |
1114 | /* |
1115 | * This HT contains streams that are freed by either the metadata thread or | |
1116 | * the data thread so we do *nothing* on the hash table and simply destroy | |
1117 | * it. | |
1118 | */ | |
1119 | lttng_ht_destroy(consumer_data.stream_list_ht); | |
3bd1e081 MD |
1120 | } |
1121 | ||
1122 | /* | |
1123 | * Called from signal handler. | |
1124 | */ | |
1125 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) | |
1126 | { | |
1127 | int ret; | |
1128 | consumer_quit = 1; | |
6f94560a MD |
1129 | do { |
1130 | ret = write(ctx->consumer_should_quit[1], "4", 1); | |
1131 | } while (ret < 0 && errno == EINTR); | |
4cec016f | 1132 | if (ret < 0 || ret != 1) { |
7a57cf92 | 1133 | PERROR("write consumer quit"); |
3bd1e081 | 1134 | } |
ab1027f4 DG |
1135 | |
1136 | DBG("Consumer flag that it should quit"); | |
3bd1e081 MD |
1137 | } |
1138 | ||
00e2e675 DG |
1139 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
1140 | off_t orig_offset) | |
3bd1e081 MD |
1141 | { |
1142 | int outfd = stream->out_fd; | |
1143 | ||
1144 | /* | |
1145 | * This does a blocking write-and-wait on any page that belongs to the | |
1146 | * subbuffer prior to the one we just wrote. | |
1147 | * Don't care about error values, as these are just hints and ways to | |
1148 | * limit the amount of page cache used. | |
1149 | */ | |
ffe60014 | 1150 | if (orig_offset < stream->max_sb_size) { |
3bd1e081 MD |
1151 | return; |
1152 | } | |
ffe60014 DG |
1153 | lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size, |
1154 | stream->max_sb_size, | |
3bd1e081 MD |
1155 | SYNC_FILE_RANGE_WAIT_BEFORE |
1156 | | SYNC_FILE_RANGE_WRITE | |
1157 | | SYNC_FILE_RANGE_WAIT_AFTER); | |
1158 | /* | |
1159 | * Give hints to the kernel about how we access the file: | |
1160 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after | |
1161 | * we write it. | |
1162 | * | |
1163 | * We need to call fadvise again after the file grows because the | |
1164 | * kernel does not seem to apply fadvise to non-existing parts of the | |
1165 | * file. | |
1166 | * | |
1167 | * Call fadvise _after_ having waited for the page writeback to | |
1168 | * complete because the dirty page writeback semantic is not well | |
1169 | * defined. So it can be expected to lead to lower throughput in | |
1170 | * streaming. | |
1171 | */ | |
ffe60014 DG |
1172 | posix_fadvise(outfd, orig_offset - stream->max_sb_size, |
1173 | stream->max_sb_size, POSIX_FADV_DONTNEED); | |
3bd1e081 MD |
1174 | } |
1175 | ||
1176 | /* | |
1177 | * Initialise the necessary environnement : | |
1178 | * - create a new context | |
1179 | * - create the poll_pipe | |
1180 | * - create the should_quit pipe (for signal handler) | |
1181 | * - create the thread pipe (for splice) | |
1182 | * | |
1183 | * Takes a function pointer as argument, this function is called when data is | |
1184 | * available on a buffer. This function is responsible to do the | |
1185 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the | |
1186 | * buffer configuration and then kernctl_put_next_subbuf at the end. | |
1187 | * | |
1188 | * Returns a pointer to the new context or NULL on error. | |
1189 | */ | |
1190 | struct lttng_consumer_local_data *lttng_consumer_create( | |
1191 | enum lttng_consumer_type type, | |
4078b776 | 1192 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
d41f73b7 | 1193 | struct lttng_consumer_local_data *ctx), |
3bd1e081 MD |
1194 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
1195 | int (*recv_stream)(struct lttng_consumer_stream *stream), | |
bcfa3ff9 | 1196 | int (*update_stream)(uint64_t stream_key, uint32_t state)) |
3bd1e081 | 1197 | { |
d8ef542d | 1198 | int ret; |
3bd1e081 MD |
1199 | struct lttng_consumer_local_data *ctx; |
1200 | ||
1201 | assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || | |
1202 | consumer_data.type == type); | |
1203 | consumer_data.type = type; | |
1204 | ||
effcf122 | 1205 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
3bd1e081 | 1206 | if (ctx == NULL) { |
7a57cf92 | 1207 | PERROR("allocating context"); |
3bd1e081 MD |
1208 | goto error; |
1209 | } | |
1210 | ||
1211 | ctx->consumer_error_socket = -1; | |
331744e3 | 1212 | ctx->consumer_metadata_socket = -1; |
3bd1e081 MD |
1213 | /* assign the callbacks */ |
1214 | ctx->on_buffer_ready = buffer_ready; | |
1215 | ctx->on_recv_channel = recv_channel; | |
1216 | ctx->on_recv_stream = recv_stream; | |
1217 | ctx->on_update_stream = update_stream; | |
1218 | ||
acdb9057 DG |
1219 | ctx->consumer_data_pipe = lttng_pipe_open(0); |
1220 | if (!ctx->consumer_data_pipe) { | |
3bd1e081 MD |
1221 | goto error_poll_pipe; |
1222 | } | |
1223 | ||
1224 | ret = pipe(ctx->consumer_should_quit); | |
1225 | if (ret < 0) { | |
7a57cf92 | 1226 | PERROR("Error creating recv pipe"); |
3bd1e081 MD |
1227 | goto error_quit_pipe; |
1228 | } | |
1229 | ||
1230 | ret = pipe(ctx->consumer_thread_pipe); | |
1231 | if (ret < 0) { | |
7a57cf92 | 1232 | PERROR("Error creating thread pipe"); |
3bd1e081 MD |
1233 | goto error_thread_pipe; |
1234 | } | |
1235 | ||
d8ef542d MD |
1236 | ret = pipe(ctx->consumer_channel_pipe); |
1237 | if (ret < 0) { | |
1238 | PERROR("Error creating channel pipe"); | |
1239 | goto error_channel_pipe; | |
1240 | } | |
1241 | ||
13886d2d DG |
1242 | ctx->consumer_metadata_pipe = lttng_pipe_open(0); |
1243 | if (!ctx->consumer_metadata_pipe) { | |
fb3a43a9 DG |
1244 | goto error_metadata_pipe; |
1245 | } | |
3bd1e081 | 1246 | |
fb3a43a9 DG |
1247 | ret = utils_create_pipe(ctx->consumer_splice_metadata_pipe); |
1248 | if (ret < 0) { | |
1249 | goto error_splice_pipe; | |
1250 | } | |
1251 | ||
1252 | return ctx; | |
3bd1e081 | 1253 | |
fb3a43a9 | 1254 | error_splice_pipe: |
13886d2d | 1255 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
fb3a43a9 | 1256 | error_metadata_pipe: |
d8ef542d MD |
1257 | utils_close_pipe(ctx->consumer_channel_pipe); |
1258 | error_channel_pipe: | |
fb3a43a9 | 1259 | utils_close_pipe(ctx->consumer_thread_pipe); |
3bd1e081 | 1260 | error_thread_pipe: |
d8ef542d | 1261 | utils_close_pipe(ctx->consumer_should_quit); |
3bd1e081 | 1262 | error_quit_pipe: |
acdb9057 | 1263 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
3bd1e081 MD |
1264 | error_poll_pipe: |
1265 | free(ctx); | |
1266 | error: | |
1267 | return NULL; | |
1268 | } | |
1269 | ||
1270 | /* | |
1271 | * Close all fds associated with the instance and free the context. | |
1272 | */ | |
1273 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) | |
1274 | { | |
4c462e79 MD |
1275 | int ret; |
1276 | ||
ab1027f4 DG |
1277 | DBG("Consumer destroying it. Closing everything."); |
1278 | ||
4c462e79 MD |
1279 | ret = close(ctx->consumer_error_socket); |
1280 | if (ret) { | |
1281 | PERROR("close"); | |
1282 | } | |
331744e3 JD |
1283 | ret = close(ctx->consumer_metadata_socket); |
1284 | if (ret) { | |
1285 | PERROR("close"); | |
1286 | } | |
d8ef542d MD |
1287 | utils_close_pipe(ctx->consumer_thread_pipe); |
1288 | utils_close_pipe(ctx->consumer_channel_pipe); | |
acdb9057 | 1289 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
13886d2d | 1290 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
d8ef542d | 1291 | utils_close_pipe(ctx->consumer_should_quit); |
fb3a43a9 DG |
1292 | utils_close_pipe(ctx->consumer_splice_metadata_pipe); |
1293 | ||
3bd1e081 MD |
1294 | unlink(ctx->consumer_command_sock_path); |
1295 | free(ctx); | |
1296 | } | |
1297 | ||
6197aea7 DG |
1298 | /* |
1299 | * Write the metadata stream id on the specified file descriptor. | |
1300 | */ | |
1301 | static int write_relayd_metadata_id(int fd, | |
1302 | struct lttng_consumer_stream *stream, | |
ffe60014 | 1303 | struct consumer_relayd_sock_pair *relayd, unsigned long padding) |
6197aea7 DG |
1304 | { |
1305 | int ret; | |
1d4dfdef | 1306 | struct lttcomm_relayd_metadata_payload hdr; |
6197aea7 | 1307 | |
1d4dfdef DG |
1308 | hdr.stream_id = htobe64(stream->relayd_stream_id); |
1309 | hdr.padding_size = htobe32(padding); | |
6197aea7 | 1310 | do { |
1d4dfdef | 1311 | ret = write(fd, (void *) &hdr, sizeof(hdr)); |
6197aea7 | 1312 | } while (ret < 0 && errno == EINTR); |
4cec016f | 1313 | if (ret < 0 || ret != sizeof(hdr)) { |
d7b75ec8 DG |
1314 | /* |
1315 | * This error means that the fd's end is closed so ignore the perror | |
1316 | * not to clubber the error output since this can happen in a normal | |
1317 | * code path. | |
1318 | */ | |
1319 | if (errno != EPIPE) { | |
1320 | PERROR("write metadata stream id"); | |
1321 | } | |
1322 | DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno); | |
534d2592 DG |
1323 | /* |
1324 | * Set ret to a negative value because if ret != sizeof(hdr), we don't | |
1325 | * handle writting the missing part so report that as an error and | |
1326 | * don't lie to the caller. | |
1327 | */ | |
1328 | ret = -1; | |
6197aea7 DG |
1329 | goto end; |
1330 | } | |
1d4dfdef DG |
1331 | DBG("Metadata stream id %" PRIu64 " with padding %lu written before data", |
1332 | stream->relayd_stream_id, padding); | |
6197aea7 DG |
1333 | |
1334 | end: | |
1335 | return ret; | |
1336 | } | |
1337 | ||
3bd1e081 | 1338 | /* |
09e26845 DG |
1339 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
1340 | * core function for writing trace buffers to either the local filesystem or | |
1341 | * the network. | |
1342 | * | |
79d4ffb7 DG |
1343 | * It must be called with the stream lock held. |
1344 | * | |
09e26845 | 1345 | * Careful review MUST be put if any changes occur! |
3bd1e081 MD |
1346 | * |
1347 | * Returns the number of bytes written | |
1348 | */ | |
4078b776 | 1349 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
3bd1e081 | 1350 | struct lttng_consumer_local_data *ctx, |
1d4dfdef DG |
1351 | struct lttng_consumer_stream *stream, unsigned long len, |
1352 | unsigned long padding) | |
3bd1e081 | 1353 | { |
f02e1e8a | 1354 | unsigned long mmap_offset; |
ffe60014 | 1355 | void *mmap_base; |
f02e1e8a DG |
1356 | ssize_t ret = 0, written = 0; |
1357 | off_t orig_offset = stream->out_fd_offset; | |
1358 | /* Default is on the disk */ | |
1359 | int outfd = stream->out_fd; | |
f02e1e8a | 1360 | struct consumer_relayd_sock_pair *relayd = NULL; |
8994307f | 1361 | unsigned int relayd_hang_up = 0; |
f02e1e8a DG |
1362 | |
1363 | /* RCU lock for the relayd pointer */ | |
1364 | rcu_read_lock(); | |
1365 | ||
1366 | /* Flag that the current stream if set for network streaming. */ | |
98cf381a | 1367 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1368 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1369 | if (relayd == NULL) { | |
da3d1b70 | 1370 | ret = -EPIPE; |
f02e1e8a DG |
1371 | goto end; |
1372 | } | |
1373 | } | |
1374 | ||
1375 | /* get the offset inside the fd to mmap */ | |
3bd1e081 MD |
1376 | switch (consumer_data.type) { |
1377 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1378 | mmap_base = stream->mmap_base; |
f02e1e8a | 1379 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); |
da3d1b70 MD |
1380 | if (ret != 0) { |
1381 | PERROR("tracer ctl get_mmap_read_offset"); | |
1382 | written = -errno; | |
1383 | goto end; | |
1384 | } | |
f02e1e8a | 1385 | break; |
7753dea8 MD |
1386 | case LTTNG_CONSUMER32_UST: |
1387 | case LTTNG_CONSUMER64_UST: | |
ffe60014 DG |
1388 | mmap_base = lttng_ustctl_get_mmap_base(stream); |
1389 | if (!mmap_base) { | |
1390 | ERR("read mmap get mmap base for stream %s", stream->name); | |
da3d1b70 | 1391 | written = -EPERM; |
ffe60014 DG |
1392 | goto end; |
1393 | } | |
1394 | ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset); | |
da3d1b70 MD |
1395 | if (ret != 0) { |
1396 | PERROR("tracer ctl get_mmap_read_offset"); | |
1397 | written = ret; | |
1398 | goto end; | |
1399 | } | |
f02e1e8a | 1400 | break; |
3bd1e081 MD |
1401 | default: |
1402 | ERR("Unknown consumer_data type"); | |
1403 | assert(0); | |
1404 | } | |
b9182dd9 | 1405 | |
f02e1e8a DG |
1406 | /* Handle stream on the relayd if the output is on the network */ |
1407 | if (relayd) { | |
1408 | unsigned long netlen = len; | |
1409 | ||
1410 | /* | |
1411 | * Lock the control socket for the complete duration of the function | |
1412 | * since from this point on we will use the socket. | |
1413 | */ | |
1414 | if (stream->metadata_flag) { | |
1415 | /* Metadata requires the control socket. */ | |
1416 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1d4dfdef | 1417 | netlen += sizeof(struct lttcomm_relayd_metadata_payload); |
f02e1e8a DG |
1418 | } |
1419 | ||
1d4dfdef | 1420 | ret = write_relayd_stream_header(stream, netlen, padding, relayd); |
f02e1e8a DG |
1421 | if (ret >= 0) { |
1422 | /* Use the returned socket. */ | |
1423 | outfd = ret; | |
1424 | ||
1425 | /* Write metadata stream id before payload */ | |
1426 | if (stream->metadata_flag) { | |
1d4dfdef | 1427 | ret = write_relayd_metadata_id(outfd, stream, relayd, padding); |
f02e1e8a | 1428 | if (ret < 0) { |
f02e1e8a | 1429 | written = ret; |
8994307f DG |
1430 | /* Socket operation failed. We consider the relayd dead */ |
1431 | if (ret == -EPIPE || ret == -EINVAL) { | |
1432 | relayd_hang_up = 1; | |
1433 | goto write_error; | |
1434 | } | |
f02e1e8a DG |
1435 | goto end; |
1436 | } | |
f02e1e8a | 1437 | } |
8994307f DG |
1438 | } else { |
1439 | /* Socket operation failed. We consider the relayd dead */ | |
1440 | if (ret == -EPIPE || ret == -EINVAL) { | |
1441 | relayd_hang_up = 1; | |
1442 | goto write_error; | |
1443 | } | |
1444 | /* Else, use the default set before which is the filesystem. */ | |
f02e1e8a | 1445 | } |
1d4dfdef DG |
1446 | } else { |
1447 | /* No streaming, we have to set the len with the full padding */ | |
1448 | len += padding; | |
1624d5b7 JD |
1449 | |
1450 | /* | |
1451 | * Check if we need to change the tracefile before writing the packet. | |
1452 | */ | |
1453 | if (stream->chan->tracefile_size > 0 && | |
1454 | (stream->tracefile_size_current + len) > | |
1455 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1456 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1457 | stream->name, stream->chan->tracefile_size, | |
1458 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
1459 | stream->out_fd, &(stream->tracefile_count_current)); | |
1624d5b7 JD |
1460 | if (ret < 0) { |
1461 | ERR("Rotating output file"); | |
1462 | goto end; | |
1463 | } | |
fe4477ee | 1464 | outfd = stream->out_fd = ret; |
a6976990 DG |
1465 | /* Reset current size because we just perform a rotation. */ |
1466 | stream->tracefile_size_current = 0; | |
e32c867f JD |
1467 | stream->out_fd_offset = 0; |
1468 | orig_offset = 0; | |
1624d5b7 JD |
1469 | } |
1470 | stream->tracefile_size_current += len; | |
f02e1e8a DG |
1471 | } |
1472 | ||
1473 | while (len > 0) { | |
1474 | do { | |
ffe60014 | 1475 | ret = write(outfd, mmap_base + mmap_offset, len); |
f02e1e8a | 1476 | } while (ret < 0 && errno == EINTR); |
1d4dfdef | 1477 | DBG("Consumer mmap write() ret %zd (len %lu)", ret, len); |
f02e1e8a | 1478 | if (ret < 0) { |
c5c45efa DG |
1479 | /* |
1480 | * This is possible if the fd is closed on the other side (outfd) | |
1481 | * or any write problem. It can be verbose a bit for a normal | |
1482 | * execution if for instance the relayd is stopped abruptly. This | |
1483 | * can happen so set this to a DBG statement. | |
1484 | */ | |
1485 | DBG("Error in file write mmap"); | |
f02e1e8a | 1486 | if (written == 0) { |
da3d1b70 | 1487 | written = -errno; |
f02e1e8a | 1488 | } |
8994307f DG |
1489 | /* Socket operation failed. We consider the relayd dead */ |
1490 | if (errno == EPIPE || errno == EINVAL) { | |
1491 | relayd_hang_up = 1; | |
1492 | goto write_error; | |
1493 | } | |
f02e1e8a DG |
1494 | goto end; |
1495 | } else if (ret > len) { | |
77c7c900 | 1496 | PERROR("Error in file write (ret %zd > len %lu)", ret, len); |
f02e1e8a DG |
1497 | written += ret; |
1498 | goto end; | |
1499 | } else { | |
1500 | len -= ret; | |
1501 | mmap_offset += ret; | |
1502 | } | |
f02e1e8a DG |
1503 | |
1504 | /* This call is useless on a socket so better save a syscall. */ | |
1505 | if (!relayd) { | |
1506 | /* This won't block, but will start writeout asynchronously */ | |
1507 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret, | |
1508 | SYNC_FILE_RANGE_WRITE); | |
1509 | stream->out_fd_offset += ret; | |
1510 | } | |
1511 | written += ret; | |
1512 | } | |
1513 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1514 | ||
8994307f DG |
1515 | write_error: |
1516 | /* | |
1517 | * This is a special case that the relayd has closed its socket. Let's | |
1518 | * cleanup the relayd object and all associated streams. | |
1519 | */ | |
1520 | if (relayd && relayd_hang_up) { | |
1521 | cleanup_relayd(relayd, ctx); | |
1522 | } | |
1523 | ||
f02e1e8a DG |
1524 | end: |
1525 | /* Unlock only if ctrl socket used */ | |
1526 | if (relayd && stream->metadata_flag) { | |
1527 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1528 | } | |
1529 | ||
1530 | rcu_read_unlock(); | |
1531 | return written; | |
3bd1e081 MD |
1532 | } |
1533 | ||
1534 | /* | |
1535 | * Splice the data from the ring buffer to the tracefile. | |
1536 | * | |
79d4ffb7 DG |
1537 | * It must be called with the stream lock held. |
1538 | * | |
3bd1e081 MD |
1539 | * Returns the number of bytes spliced. |
1540 | */ | |
4078b776 | 1541 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
3bd1e081 | 1542 | struct lttng_consumer_local_data *ctx, |
1d4dfdef DG |
1543 | struct lttng_consumer_stream *stream, unsigned long len, |
1544 | unsigned long padding) | |
3bd1e081 | 1545 | { |
f02e1e8a DG |
1546 | ssize_t ret = 0, written = 0, ret_splice = 0; |
1547 | loff_t offset = 0; | |
1548 | off_t orig_offset = stream->out_fd_offset; | |
1549 | int fd = stream->wait_fd; | |
1550 | /* Default is on the disk */ | |
1551 | int outfd = stream->out_fd; | |
f02e1e8a | 1552 | struct consumer_relayd_sock_pair *relayd = NULL; |
fb3a43a9 | 1553 | int *splice_pipe; |
8994307f | 1554 | unsigned int relayd_hang_up = 0; |
f02e1e8a | 1555 | |
3bd1e081 MD |
1556 | switch (consumer_data.type) { |
1557 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a | 1558 | break; |
7753dea8 MD |
1559 | case LTTNG_CONSUMER32_UST: |
1560 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a | 1561 | /* Not supported for user space tracing */ |
3bd1e081 MD |
1562 | return -ENOSYS; |
1563 | default: | |
1564 | ERR("Unknown consumer_data type"); | |
1565 | assert(0); | |
3bd1e081 MD |
1566 | } |
1567 | ||
f02e1e8a DG |
1568 | /* RCU lock for the relayd pointer */ |
1569 | rcu_read_lock(); | |
1570 | ||
1571 | /* Flag that the current stream if set for network streaming. */ | |
98cf381a | 1572 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1573 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1574 | if (relayd == NULL) { | |
da3d1b70 | 1575 | ret = -EPIPE; |
f02e1e8a DG |
1576 | goto end; |
1577 | } | |
1578 | } | |
1579 | ||
fb3a43a9 DG |
1580 | /* |
1581 | * Choose right pipe for splice. Metadata and trace data are handled by | |
1582 | * different threads hence the use of two pipes in order not to race or | |
1583 | * corrupt the written data. | |
1584 | */ | |
1585 | if (stream->metadata_flag) { | |
1586 | splice_pipe = ctx->consumer_splice_metadata_pipe; | |
1587 | } else { | |
1588 | splice_pipe = ctx->consumer_thread_pipe; | |
1589 | } | |
1590 | ||
f02e1e8a | 1591 | /* Write metadata stream id before payload */ |
1d4dfdef DG |
1592 | if (relayd) { |
1593 | int total_len = len; | |
f02e1e8a | 1594 | |
1d4dfdef DG |
1595 | if (stream->metadata_flag) { |
1596 | /* | |
1597 | * Lock the control socket for the complete duration of the function | |
1598 | * since from this point on we will use the socket. | |
1599 | */ | |
1600 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1601 | ||
1602 | ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd, | |
1603 | padding); | |
1604 | if (ret < 0) { | |
1605 | written = ret; | |
8994307f DG |
1606 | /* Socket operation failed. We consider the relayd dead */ |
1607 | if (ret == -EBADF) { | |
1608 | WARN("Remote relayd disconnected. Stopping"); | |
1609 | relayd_hang_up = 1; | |
1610 | goto write_error; | |
1611 | } | |
1d4dfdef DG |
1612 | goto end; |
1613 | } | |
1614 | ||
1615 | total_len += sizeof(struct lttcomm_relayd_metadata_payload); | |
1616 | } | |
1617 | ||
1618 | ret = write_relayd_stream_header(stream, total_len, padding, relayd); | |
1619 | if (ret >= 0) { | |
1620 | /* Use the returned socket. */ | |
1621 | outfd = ret; | |
1622 | } else { | |
8994307f DG |
1623 | /* Socket operation failed. We consider the relayd dead */ |
1624 | if (ret == -EBADF) { | |
1625 | WARN("Remote relayd disconnected. Stopping"); | |
1626 | relayd_hang_up = 1; | |
1627 | goto write_error; | |
1628 | } | |
f02e1e8a DG |
1629 | goto end; |
1630 | } | |
1d4dfdef DG |
1631 | } else { |
1632 | /* No streaming, we have to set the len with the full padding */ | |
1633 | len += padding; | |
1624d5b7 JD |
1634 | |
1635 | /* | |
1636 | * Check if we need to change the tracefile before writing the packet. | |
1637 | */ | |
1638 | if (stream->chan->tracefile_size > 0 && | |
1639 | (stream->tracefile_size_current + len) > | |
1640 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1641 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1642 | stream->name, stream->chan->tracefile_size, | |
1643 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
1644 | stream->out_fd, &(stream->tracefile_count_current)); | |
1624d5b7 JD |
1645 | if (ret < 0) { |
1646 | ERR("Rotating output file"); | |
1647 | goto end; | |
1648 | } | |
fe4477ee | 1649 | outfd = stream->out_fd = ret; |
a6976990 DG |
1650 | /* Reset current size because we just perform a rotation. */ |
1651 | stream->tracefile_size_current = 0; | |
e32c867f JD |
1652 | stream->out_fd_offset = 0; |
1653 | orig_offset = 0; | |
1624d5b7 JD |
1654 | } |
1655 | stream->tracefile_size_current += len; | |
f02e1e8a DG |
1656 | } |
1657 | ||
1658 | while (len > 0) { | |
1d4dfdef DG |
1659 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)", |
1660 | (unsigned long)offset, len, fd, splice_pipe[1]); | |
fb3a43a9 | 1661 | ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len, |
f02e1e8a DG |
1662 | SPLICE_F_MOVE | SPLICE_F_MORE); |
1663 | DBG("splice chan to pipe, ret %zd", ret_splice); | |
1664 | if (ret_splice < 0) { | |
1665 | PERROR("Error in relay splice"); | |
1666 | if (written == 0) { | |
1667 | written = ret_splice; | |
1668 | } | |
1669 | ret = errno; | |
1670 | goto splice_error; | |
1671 | } | |
1672 | ||
1673 | /* Handle stream on the relayd if the output is on the network */ | |
1674 | if (relayd) { | |
1675 | if (stream->metadata_flag) { | |
1d4dfdef DG |
1676 | size_t metadata_payload_size = |
1677 | sizeof(struct lttcomm_relayd_metadata_payload); | |
1678 | ||
f02e1e8a | 1679 | /* Update counter to fit the spliced data */ |
1d4dfdef DG |
1680 | ret_splice += metadata_payload_size; |
1681 | len += metadata_payload_size; | |
f02e1e8a DG |
1682 | /* |
1683 | * We do this so the return value can match the len passed as | |
1684 | * argument to this function. | |
1685 | */ | |
1d4dfdef | 1686 | written -= metadata_payload_size; |
f02e1e8a DG |
1687 | } |
1688 | } | |
1689 | ||
1690 | /* Splice data out */ | |
fb3a43a9 | 1691 | ret_splice = splice(splice_pipe[0], NULL, outfd, NULL, |
f02e1e8a | 1692 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); |
1d4dfdef | 1693 | DBG("Consumer splice pipe to file, ret %zd", ret_splice); |
f02e1e8a DG |
1694 | if (ret_splice < 0) { |
1695 | PERROR("Error in file splice"); | |
1696 | if (written == 0) { | |
1697 | written = ret_splice; | |
1698 | } | |
8994307f | 1699 | /* Socket operation failed. We consider the relayd dead */ |
00c8752b | 1700 | if (errno == EBADF || errno == EPIPE) { |
8994307f DG |
1701 | WARN("Remote relayd disconnected. Stopping"); |
1702 | relayd_hang_up = 1; | |
1703 | goto write_error; | |
1704 | } | |
f02e1e8a DG |
1705 | ret = errno; |
1706 | goto splice_error; | |
1707 | } else if (ret_splice > len) { | |
1708 | errno = EINVAL; | |
1709 | PERROR("Wrote more data than requested %zd (len: %lu)", | |
1710 | ret_splice, len); | |
1711 | written += ret_splice; | |
1712 | ret = errno; | |
1713 | goto splice_error; | |
1714 | } | |
1715 | len -= ret_splice; | |
1716 | ||
1717 | /* This call is useless on a socket so better save a syscall. */ | |
1718 | if (!relayd) { | |
1719 | /* This won't block, but will start writeout asynchronously */ | |
1720 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, | |
1721 | SYNC_FILE_RANGE_WRITE); | |
1722 | stream->out_fd_offset += ret_splice; | |
1723 | } | |
1724 | written += ret_splice; | |
1725 | } | |
1726 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1727 | ||
1728 | ret = ret_splice; | |
1729 | ||
1730 | goto end; | |
1731 | ||
8994307f DG |
1732 | write_error: |
1733 | /* | |
1734 | * This is a special case that the relayd has closed its socket. Let's | |
1735 | * cleanup the relayd object and all associated streams. | |
1736 | */ | |
1737 | if (relayd && relayd_hang_up) { | |
1738 | cleanup_relayd(relayd, ctx); | |
1739 | /* Skip splice error so the consumer does not fail */ | |
1740 | goto end; | |
1741 | } | |
1742 | ||
f02e1e8a DG |
1743 | splice_error: |
1744 | /* send the appropriate error description to sessiond */ | |
1745 | switch (ret) { | |
f02e1e8a | 1746 | case EINVAL: |
f73fabfd | 1747 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); |
f02e1e8a DG |
1748 | break; |
1749 | case ENOMEM: | |
f73fabfd | 1750 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); |
f02e1e8a DG |
1751 | break; |
1752 | case ESPIPE: | |
f73fabfd | 1753 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); |
f02e1e8a DG |
1754 | break; |
1755 | } | |
1756 | ||
1757 | end: | |
1758 | if (relayd && stream->metadata_flag) { | |
1759 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1760 | } | |
1761 | ||
1762 | rcu_read_unlock(); | |
1763 | return written; | |
3bd1e081 MD |
1764 | } |
1765 | ||
1766 | /* | |
1767 | * Take a snapshot for a specific fd | |
1768 | * | |
1769 | * Returns 0 on success, < 0 on error | |
1770 | */ | |
ffe60014 | 1771 | int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) |
3bd1e081 MD |
1772 | { |
1773 | switch (consumer_data.type) { | |
1774 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1775 | return lttng_kconsumer_take_snapshot(stream); |
7753dea8 MD |
1776 | case LTTNG_CONSUMER32_UST: |
1777 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 1778 | return lttng_ustconsumer_take_snapshot(stream); |
3bd1e081 MD |
1779 | default: |
1780 | ERR("Unknown consumer_data type"); | |
1781 | assert(0); | |
1782 | return -ENOSYS; | |
1783 | } | |
3bd1e081 MD |
1784 | } |
1785 | ||
1786 | /* | |
1787 | * Get the produced position | |
1788 | * | |
1789 | * Returns 0 on success, < 0 on error | |
1790 | */ | |
ffe60014 | 1791 | int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, |
3bd1e081 MD |
1792 | unsigned long *pos) |
1793 | { | |
1794 | switch (consumer_data.type) { | |
1795 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1796 | return lttng_kconsumer_get_produced_snapshot(stream, pos); |
7753dea8 MD |
1797 | case LTTNG_CONSUMER32_UST: |
1798 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 1799 | return lttng_ustconsumer_get_produced_snapshot(stream, pos); |
3bd1e081 MD |
1800 | default: |
1801 | ERR("Unknown consumer_data type"); | |
1802 | assert(0); | |
1803 | return -ENOSYS; | |
1804 | } | |
1805 | } | |
1806 | ||
1807 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, | |
1808 | int sock, struct pollfd *consumer_sockpoll) | |
1809 | { | |
1810 | switch (consumer_data.type) { | |
1811 | case LTTNG_CONSUMER_KERNEL: | |
1812 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
7753dea8 MD |
1813 | case LTTNG_CONSUMER32_UST: |
1814 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1815 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
1816 | default: | |
1817 | ERR("Unknown consumer_data type"); | |
1818 | assert(0); | |
1819 | return -ENOSYS; | |
1820 | } | |
1821 | } | |
1822 | ||
43c34bc3 DG |
1823 | /* |
1824 | * Iterate over all streams of the hashtable and free them properly. | |
1825 | * | |
1826 | * WARNING: *MUST* be used with data stream only. | |
1827 | */ | |
1828 | static void destroy_data_stream_ht(struct lttng_ht *ht) | |
1829 | { | |
43c34bc3 DG |
1830 | struct lttng_ht_iter iter; |
1831 | struct lttng_consumer_stream *stream; | |
1832 | ||
1833 | if (ht == NULL) { | |
1834 | return; | |
1835 | } | |
1836 | ||
1837 | rcu_read_lock(); | |
1838 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
5c540210 DG |
1839 | /* |
1840 | * Ignore return value since we are currently cleaning up so any error | |
1841 | * can't be handled. | |
1842 | */ | |
1843 | (void) consumer_del_stream(stream, ht); | |
43c34bc3 DG |
1844 | } |
1845 | rcu_read_unlock(); | |
1846 | ||
1847 | lttng_ht_destroy(ht); | |
1848 | } | |
1849 | ||
fb3a43a9 | 1850 | /* |
f724d81e | 1851 | * Iterate over all streams of the hashtable and free them properly. |
e316aad5 DG |
1852 | * |
1853 | * XXX: Should not be only for metadata stream or else use an other name. | |
fb3a43a9 DG |
1854 | */ |
1855 | static void destroy_stream_ht(struct lttng_ht *ht) | |
1856 | { | |
fb3a43a9 DG |
1857 | struct lttng_ht_iter iter; |
1858 | struct lttng_consumer_stream *stream; | |
1859 | ||
1860 | if (ht == NULL) { | |
1861 | return; | |
1862 | } | |
1863 | ||
d09e1200 | 1864 | rcu_read_lock(); |
58b1f425 | 1865 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
5c540210 DG |
1866 | /* |
1867 | * Ignore return value since we are currently cleaning up so any error | |
1868 | * can't be handled. | |
1869 | */ | |
1870 | (void) consumer_del_metadata_stream(stream, ht); | |
fb3a43a9 | 1871 | } |
d09e1200 | 1872 | rcu_read_unlock(); |
fb3a43a9 DG |
1873 | |
1874 | lttng_ht_destroy(ht); | |
1875 | } | |
1876 | ||
d88aee68 DG |
1877 | void lttng_consumer_close_metadata(void) |
1878 | { | |
1879 | switch (consumer_data.type) { | |
1880 | case LTTNG_CONSUMER_KERNEL: | |
1881 | /* | |
1882 | * The Kernel consumer has a different metadata scheme so we don't | |
1883 | * close anything because the stream will be closed by the session | |
1884 | * daemon. | |
1885 | */ | |
1886 | break; | |
1887 | case LTTNG_CONSUMER32_UST: | |
1888 | case LTTNG_CONSUMER64_UST: | |
1889 | /* | |
1890 | * Close all metadata streams. The metadata hash table is passed and | |
1891 | * this call iterates over it by closing all wakeup fd. This is safe | |
1892 | * because at this point we are sure that the metadata producer is | |
1893 | * either dead or blocked. | |
1894 | */ | |
1895 | lttng_ustconsumer_close_metadata(metadata_ht); | |
1896 | break; | |
1897 | default: | |
1898 | ERR("Unknown consumer_data type"); | |
1899 | assert(0); | |
1900 | } | |
1901 | } | |
1902 | ||
fb3a43a9 DG |
1903 | /* |
1904 | * Clean up a metadata stream and free its memory. | |
1905 | */ | |
e316aad5 DG |
1906 | void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, |
1907 | struct lttng_ht *ht) | |
fb3a43a9 DG |
1908 | { |
1909 | int ret; | |
e316aad5 DG |
1910 | struct lttng_ht_iter iter; |
1911 | struct lttng_consumer_channel *free_chan = NULL; | |
fb3a43a9 DG |
1912 | struct consumer_relayd_sock_pair *relayd; |
1913 | ||
1914 | assert(stream); | |
1915 | /* | |
1916 | * This call should NEVER receive regular stream. It must always be | |
1917 | * metadata stream and this is crucial for data structure synchronization. | |
1918 | */ | |
1919 | assert(stream->metadata_flag); | |
1920 | ||
e316aad5 DG |
1921 | DBG3("Consumer delete metadata stream %d", stream->wait_fd); |
1922 | ||
1923 | if (ht == NULL) { | |
1924 | /* Means the stream was allocated but not successfully added */ | |
ffe60014 | 1925 | goto free_stream_rcu; |
e316aad5 DG |
1926 | } |
1927 | ||
74251bb8 | 1928 | pthread_mutex_lock(&consumer_data.lock); |
72c3879d | 1929 | pthread_mutex_lock(&stream->chan->lock); |
8994307f DG |
1930 | pthread_mutex_lock(&stream->lock); |
1931 | ||
fb3a43a9 DG |
1932 | switch (consumer_data.type) { |
1933 | case LTTNG_CONSUMER_KERNEL: | |
1934 | if (stream->mmap_base != NULL) { | |
1935 | ret = munmap(stream->mmap_base, stream->mmap_len); | |
1936 | if (ret != 0) { | |
1937 | PERROR("munmap metadata stream"); | |
1938 | } | |
1939 | } | |
4c95e622 JD |
1940 | |
1941 | if (stream->wait_fd >= 0) { | |
1942 | ret = close(stream->wait_fd); | |
1943 | if (ret < 0) { | |
1944 | PERROR("close kernel metadata wait_fd"); | |
1945 | } | |
1946 | } | |
fb3a43a9 DG |
1947 | break; |
1948 | case LTTNG_CONSUMER32_UST: | |
1949 | case LTTNG_CONSUMER64_UST: | |
1950 | lttng_ustconsumer_del_stream(stream); | |
1951 | break; | |
1952 | default: | |
1953 | ERR("Unknown consumer_data type"); | |
1954 | assert(0); | |
e316aad5 | 1955 | goto end; |
fb3a43a9 | 1956 | } |
fb3a43a9 | 1957 | |
c869f647 | 1958 | rcu_read_lock(); |
58b1f425 | 1959 | iter.iter.node = &stream->node.node; |
c869f647 DG |
1960 | ret = lttng_ht_del(ht, &iter); |
1961 | assert(!ret); | |
ca22feea | 1962 | |
d8ef542d MD |
1963 | iter.iter.node = &stream->node_channel_id.node; |
1964 | ret = lttng_ht_del(consumer_data.stream_per_chan_id_ht, &iter); | |
1965 | assert(!ret); | |
1966 | ||
ca22feea DG |
1967 | iter.iter.node = &stream->node_session_id.node; |
1968 | ret = lttng_ht_del(consumer_data.stream_list_ht, &iter); | |
1969 | assert(!ret); | |
c869f647 DG |
1970 | rcu_read_unlock(); |
1971 | ||
fb3a43a9 DG |
1972 | if (stream->out_fd >= 0) { |
1973 | ret = close(stream->out_fd); | |
1974 | if (ret) { | |
1975 | PERROR("close"); | |
1976 | } | |
1977 | } | |
1978 | ||
fb3a43a9 DG |
1979 | /* Check and cleanup relayd */ |
1980 | rcu_read_lock(); | |
1981 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
1982 | if (relayd != NULL) { | |
1983 | uatomic_dec(&relayd->refcount); | |
1984 | assert(uatomic_read(&relayd->refcount) >= 0); | |
1985 | ||
1986 | /* Closing streams requires to lock the control socket. */ | |
1987 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1988 | ret = relayd_send_close_stream(&relayd->control_sock, | |
1989 | stream->relayd_stream_id, stream->next_net_seq_num - 1); | |
1990 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1991 | if (ret < 0) { | |
1992 | DBG("Unable to close stream on the relayd. Continuing"); | |
1993 | /* | |
1994 | * Continue here. There is nothing we can do for the relayd. | |
1995 | * Chances are that the relayd has closed the socket so we just | |
1996 | * continue cleaning up. | |
1997 | */ | |
1998 | } | |
1999 | ||
2000 | /* Both conditions are met, we destroy the relayd. */ | |
2001 | if (uatomic_read(&relayd->refcount) == 0 && | |
2002 | uatomic_read(&relayd->destroy_flag)) { | |
d09e1200 | 2003 | destroy_relayd(relayd); |
fb3a43a9 DG |
2004 | } |
2005 | } | |
2006 | rcu_read_unlock(); | |
2007 | ||
2008 | /* Atomically decrement channel refcount since other threads can use it. */ | |
f2ad556d | 2009 | if (!uatomic_sub_return(&stream->chan->refcount, 1) |
ffe60014 | 2010 | && !uatomic_read(&stream->chan->nb_init_stream_left)) { |
c30aaa51 | 2011 | /* Go for channel deletion! */ |
e316aad5 | 2012 | free_chan = stream->chan; |
fb3a43a9 DG |
2013 | } |
2014 | ||
e316aad5 | 2015 | end: |
73811ecc DG |
2016 | /* |
2017 | * Nullify the stream reference so it is not used after deletion. The | |
f0e1c785 MD |
2018 | * channel lock MUST be acquired before being able to check for |
2019 | * a NULL pointer value. | |
73811ecc DG |
2020 | */ |
2021 | stream->chan->metadata_stream = NULL; | |
2022 | ||
8994307f | 2023 | pthread_mutex_unlock(&stream->lock); |
72c3879d | 2024 | pthread_mutex_unlock(&stream->chan->lock); |
74251bb8 | 2025 | pthread_mutex_unlock(&consumer_data.lock); |
e316aad5 DG |
2026 | |
2027 | if (free_chan) { | |
2028 | consumer_del_channel(free_chan); | |
2029 | } | |
2030 | ||
ffe60014 DG |
2031 | free_stream_rcu: |
2032 | call_rcu(&stream->node.head, free_stream_rcu); | |
fb3a43a9 DG |
2033 | } |
2034 | ||
2035 | /* | |
2036 | * Action done with the metadata stream when adding it to the consumer internal | |
2037 | * data structures to handle it. | |
2038 | */ | |
155e9bb8 | 2039 | int consumer_add_metadata_stream(struct lttng_consumer_stream *stream) |
fb3a43a9 | 2040 | { |
155e9bb8 | 2041 | struct lttng_ht *ht = metadata_ht; |
e316aad5 | 2042 | int ret = 0; |
fb3a43a9 | 2043 | struct consumer_relayd_sock_pair *relayd; |
76082088 | 2044 | struct lttng_ht_iter iter; |
d88aee68 | 2045 | struct lttng_ht_node_u64 *node; |
fb3a43a9 | 2046 | |
e316aad5 DG |
2047 | assert(stream); |
2048 | assert(ht); | |
2049 | ||
d88aee68 | 2050 | DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key); |
e316aad5 DG |
2051 | |
2052 | pthread_mutex_lock(&consumer_data.lock); | |
72c3879d | 2053 | pthread_mutex_lock(&stream->chan->lock); |
6ca1a6b0 | 2054 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 2055 | pthread_mutex_lock(&stream->lock); |
e316aad5 | 2056 | |
e316aad5 DG |
2057 | /* |
2058 | * From here, refcounts are updated so be _careful_ when returning an error | |
2059 | * after this point. | |
2060 | */ | |
2061 | ||
fb3a43a9 | 2062 | rcu_read_lock(); |
76082088 DG |
2063 | |
2064 | /* | |
2065 | * Lookup the stream just to make sure it does not exist in our internal | |
2066 | * state. This should NEVER happen. | |
2067 | */ | |
d88aee68 DG |
2068 | lttng_ht_lookup(ht, &stream->key, &iter); |
2069 | node = lttng_ht_iter_get_node_u64(&iter); | |
76082088 DG |
2070 | assert(!node); |
2071 | ||
e316aad5 | 2072 | /* Find relayd and, if one is found, increment refcount. */ |
fb3a43a9 DG |
2073 | relayd = consumer_find_relayd(stream->net_seq_idx); |
2074 | if (relayd != NULL) { | |
2075 | uatomic_inc(&relayd->refcount); | |
2076 | } | |
e316aad5 | 2077 | |
e316aad5 | 2078 | /* |
ffe60014 DG |
2079 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
2080 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
2081 | * causes the count to become 0 also causes a stream to be added. The |
2082 | * channel deletion will thus be triggered by the following removal of this | |
2083 | * stream. | |
2084 | */ | |
ffe60014 | 2085 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
2086 | /* Increment refcount before decrementing nb_init_stream_left */ |
2087 | cmm_smp_wmb(); | |
ffe60014 | 2088 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
2089 | } |
2090 | ||
d88aee68 | 2091 | lttng_ht_add_unique_u64(ht, &stream->node); |
ca22feea | 2092 | |
d8ef542d MD |
2093 | lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht, |
2094 | &stream->node_channel_id); | |
2095 | ||
ca22feea DG |
2096 | /* |
2097 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
2098 | * the key since the HT does not use it and we allow to add redundant keys | |
2099 | * into this table. | |
2100 | */ | |
d88aee68 | 2101 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 2102 | |
fb3a43a9 | 2103 | rcu_read_unlock(); |
e316aad5 | 2104 | |
2e818a6a | 2105 | pthread_mutex_unlock(&stream->lock); |
72c3879d | 2106 | pthread_mutex_unlock(&stream->chan->lock); |
6ca1a6b0 | 2107 | pthread_mutex_unlock(&stream->chan->timer_lock); |
e316aad5 DG |
2108 | pthread_mutex_unlock(&consumer_data.lock); |
2109 | return ret; | |
fb3a43a9 DG |
2110 | } |
2111 | ||
8994307f DG |
2112 | /* |
2113 | * Delete data stream that are flagged for deletion (endpoint_status). | |
2114 | */ | |
2115 | static void validate_endpoint_status_data_stream(void) | |
2116 | { | |
2117 | struct lttng_ht_iter iter; | |
2118 | struct lttng_consumer_stream *stream; | |
2119 | ||
2120 | DBG("Consumer delete flagged data stream"); | |
2121 | ||
2122 | rcu_read_lock(); | |
2123 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
2124 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2125 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2126 | continue; |
2127 | } | |
2128 | /* Delete it right now */ | |
2129 | consumer_del_stream(stream, data_ht); | |
2130 | } | |
2131 | rcu_read_unlock(); | |
2132 | } | |
2133 | ||
2134 | /* | |
2135 | * Delete metadata stream that are flagged for deletion (endpoint_status). | |
2136 | */ | |
2137 | static void validate_endpoint_status_metadata_stream( | |
2138 | struct lttng_poll_event *pollset) | |
2139 | { | |
2140 | struct lttng_ht_iter iter; | |
2141 | struct lttng_consumer_stream *stream; | |
2142 | ||
2143 | DBG("Consumer delete flagged metadata stream"); | |
2144 | ||
2145 | assert(pollset); | |
2146 | ||
2147 | rcu_read_lock(); | |
2148 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
2149 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2150 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2151 | continue; |
2152 | } | |
2153 | /* | |
2154 | * Remove from pollset so the metadata thread can continue without | |
2155 | * blocking on a deleted stream. | |
2156 | */ | |
2157 | lttng_poll_del(pollset, stream->wait_fd); | |
2158 | ||
2159 | /* Delete it right now */ | |
2160 | consumer_del_metadata_stream(stream, metadata_ht); | |
2161 | } | |
2162 | rcu_read_unlock(); | |
2163 | } | |
2164 | ||
fb3a43a9 DG |
2165 | /* |
2166 | * Thread polls on metadata file descriptor and write them on disk or on the | |
2167 | * network. | |
2168 | */ | |
7d980def | 2169 | void *consumer_thread_metadata_poll(void *data) |
fb3a43a9 DG |
2170 | { |
2171 | int ret, i, pollfd; | |
2172 | uint32_t revents, nb_fd; | |
e316aad5 | 2173 | struct lttng_consumer_stream *stream = NULL; |
fb3a43a9 | 2174 | struct lttng_ht_iter iter; |
d88aee68 | 2175 | struct lttng_ht_node_u64 *node; |
fb3a43a9 DG |
2176 | struct lttng_poll_event events; |
2177 | struct lttng_consumer_local_data *ctx = data; | |
2178 | ssize_t len; | |
2179 | ||
2180 | rcu_register_thread(); | |
2181 | ||
d88aee68 | 2182 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
04bb2b64 DG |
2183 | if (!metadata_ht) { |
2184 | /* ENOMEM at this point. Better to bail out. */ | |
d8ef542d | 2185 | goto end_ht; |
04bb2b64 DG |
2186 | } |
2187 | ||
fb3a43a9 DG |
2188 | DBG("Thread metadata poll started"); |
2189 | ||
fb3a43a9 DG |
2190 | /* Size is set to 1 for the consumer_metadata pipe */ |
2191 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2192 | if (ret < 0) { | |
2193 | ERR("Poll set creation failed"); | |
d8ef542d | 2194 | goto end_poll; |
fb3a43a9 DG |
2195 | } |
2196 | ||
13886d2d DG |
2197 | ret = lttng_poll_add(&events, |
2198 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN); | |
fb3a43a9 DG |
2199 | if (ret < 0) { |
2200 | goto end; | |
2201 | } | |
2202 | ||
2203 | /* Main loop */ | |
2204 | DBG("Metadata main loop started"); | |
2205 | ||
2206 | while (1) { | |
fb3a43a9 | 2207 | /* Only the metadata pipe is set */ |
d21b0d71 | 2208 | if (LTTNG_POLL_GETNB(&events) == 0 && consumer_quit == 1) { |
fb3a43a9 DG |
2209 | goto end; |
2210 | } | |
2211 | ||
2212 | restart: | |
d21b0d71 | 2213 | DBG("Metadata poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events)); |
fb3a43a9 DG |
2214 | ret = lttng_poll_wait(&events, -1); |
2215 | DBG("Metadata event catched in thread"); | |
2216 | if (ret < 0) { | |
2217 | if (errno == EINTR) { | |
e316aad5 | 2218 | ERR("Poll EINTR catched"); |
fb3a43a9 DG |
2219 | goto restart; |
2220 | } | |
2221 | goto error; | |
2222 | } | |
2223 | ||
0d9c5d77 DG |
2224 | nb_fd = ret; |
2225 | ||
e316aad5 | 2226 | /* From here, the event is a metadata wait fd */ |
fb3a43a9 DG |
2227 | for (i = 0; i < nb_fd; i++) { |
2228 | revents = LTTNG_POLL_GETEV(&events, i); | |
2229 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2230 | ||
e316aad5 DG |
2231 | /* Just don't waste time if no returned events for the fd */ |
2232 | if (!revents) { | |
2233 | continue; | |
2234 | } | |
2235 | ||
13886d2d | 2236 | if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) { |
4adabd61 | 2237 | if (revents & (LPOLLERR | LPOLLHUP )) { |
fb3a43a9 DG |
2238 | DBG("Metadata thread pipe hung up"); |
2239 | /* | |
2240 | * Remove the pipe from the poll set and continue the loop | |
2241 | * since their might be data to consume. | |
2242 | */ | |
13886d2d DG |
2243 | lttng_poll_del(&events, |
2244 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2245 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
fb3a43a9 DG |
2246 | continue; |
2247 | } else if (revents & LPOLLIN) { | |
13886d2d DG |
2248 | ssize_t pipe_len; |
2249 | ||
2250 | pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe, | |
2251 | &stream, sizeof(stream)); | |
2252 | if (pipe_len < 0) { | |
c5c46de1 | 2253 | ERR("read metadata stream, ret: %zd", pipe_len); |
fb3a43a9 | 2254 | /* |
13886d2d | 2255 | * Continue here to handle the rest of the streams. |
fb3a43a9 DG |
2256 | */ |
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, | |
2272 | LPOLLIN | LPOLLPRI); | |
fb3a43a9 DG |
2273 | } |
2274 | ||
e316aad5 | 2275 | /* Handle other stream */ |
fb3a43a9 DG |
2276 | continue; |
2277 | } | |
2278 | ||
d09e1200 | 2279 | rcu_read_lock(); |
d88aee68 DG |
2280 | { |
2281 | uint64_t tmp_id = (uint64_t) pollfd; | |
2282 | ||
2283 | lttng_ht_lookup(metadata_ht, &tmp_id, &iter); | |
2284 | } | |
2285 | node = lttng_ht_iter_get_node_u64(&iter); | |
e316aad5 | 2286 | assert(node); |
fb3a43a9 DG |
2287 | |
2288 | stream = caa_container_of(node, struct lttng_consumer_stream, | |
58b1f425 | 2289 | node); |
fb3a43a9 | 2290 | |
e316aad5 | 2291 | /* Check for error event */ |
4adabd61 | 2292 | if (revents & (LPOLLERR | LPOLLHUP)) { |
e316aad5 | 2293 | DBG("Metadata fd %d is hup|err.", pollfd); |
fb3a43a9 DG |
2294 | if (!stream->hangup_flush_done |
2295 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2296 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2297 | DBG("Attempting to flush and consume the UST buffers"); | |
2298 | lttng_ustconsumer_on_stream_hangup(stream); | |
2299 | ||
2300 | /* We just flushed the stream now read it. */ | |
4bb94b75 DG |
2301 | do { |
2302 | len = ctx->on_buffer_ready(stream, ctx); | |
2303 | /* | |
2304 | * We don't check the return value here since if we get | |
2305 | * a negative len, it means an error occured thus we | |
2306 | * simply remove it from the poll set and free the | |
2307 | * stream. | |
2308 | */ | |
2309 | } while (len > 0); | |
fb3a43a9 DG |
2310 | } |
2311 | ||
fb3a43a9 | 2312 | lttng_poll_del(&events, stream->wait_fd); |
e316aad5 DG |
2313 | /* |
2314 | * This call update the channel states, closes file descriptors | |
2315 | * and securely free the stream. | |
2316 | */ | |
2317 | consumer_del_metadata_stream(stream, metadata_ht); | |
2318 | } else if (revents & (LPOLLIN | LPOLLPRI)) { | |
2319 | /* Get the data out of the metadata file descriptor */ | |
2320 | DBG("Metadata available on fd %d", pollfd); | |
2321 | assert(stream->wait_fd == pollfd); | |
2322 | ||
2323 | len = ctx->on_buffer_ready(stream, ctx); | |
2324 | /* It's ok to have an unavailable sub-buffer */ | |
b64403e3 | 2325 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2326 | /* Clean up stream from consumer and free it. */ |
2327 | lttng_poll_del(&events, stream->wait_fd); | |
2328 | consumer_del_metadata_stream(stream, metadata_ht); | |
e316aad5 DG |
2329 | } else if (len > 0) { |
2330 | stream->data_read = 1; | |
2331 | } | |
fb3a43a9 | 2332 | } |
e316aad5 DG |
2333 | |
2334 | /* Release RCU lock for the stream looked up */ | |
d09e1200 | 2335 | rcu_read_unlock(); |
fb3a43a9 DG |
2336 | } |
2337 | } | |
2338 | ||
2339 | error: | |
2340 | end: | |
2341 | DBG("Metadata poll thread exiting"); | |
fb3a43a9 | 2342 | |
d8ef542d MD |
2343 | lttng_poll_clean(&events); |
2344 | end_poll: | |
04bb2b64 | 2345 | destroy_stream_ht(metadata_ht); |
d8ef542d | 2346 | end_ht: |
fb3a43a9 DG |
2347 | rcu_unregister_thread(); |
2348 | return NULL; | |
2349 | } | |
2350 | ||
3bd1e081 | 2351 | /* |
e4421fec | 2352 | * This thread polls the fds in the set to consume the data and write |
3bd1e081 MD |
2353 | * it to tracefile if necessary. |
2354 | */ | |
7d980def | 2355 | void *consumer_thread_data_poll(void *data) |
3bd1e081 MD |
2356 | { |
2357 | int num_rdy, num_hup, high_prio, ret, i; | |
2358 | struct pollfd *pollfd = NULL; | |
2359 | /* local view of the streams */ | |
c869f647 | 2360 | struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; |
3bd1e081 MD |
2361 | /* local view of consumer_data.fds_count */ |
2362 | int nb_fd = 0; | |
3bd1e081 | 2363 | struct lttng_consumer_local_data *ctx = data; |
00e2e675 | 2364 | ssize_t len; |
3bd1e081 | 2365 | |
e7b994a3 DG |
2366 | rcu_register_thread(); |
2367 | ||
d88aee68 | 2368 | data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
43c34bc3 | 2369 | if (data_ht == NULL) { |
04bb2b64 | 2370 | /* ENOMEM at this point. Better to bail out. */ |
43c34bc3 DG |
2371 | goto end; |
2372 | } | |
2373 | ||
e1eb1e0e MD |
2374 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream *)); |
2375 | if (local_stream == NULL) { | |
2376 | PERROR("local_stream malloc"); | |
2377 | goto end; | |
2378 | } | |
3bd1e081 MD |
2379 | |
2380 | while (1) { | |
2381 | high_prio = 0; | |
2382 | num_hup = 0; | |
2383 | ||
2384 | /* | |
e4421fec | 2385 | * the fds set has been updated, we need to update our |
3bd1e081 MD |
2386 | * local array as well |
2387 | */ | |
2388 | pthread_mutex_lock(&consumer_data.lock); | |
2389 | if (consumer_data.need_update) { | |
0e428499 DG |
2390 | free(pollfd); |
2391 | pollfd = NULL; | |
2392 | ||
2393 | free(local_stream); | |
2394 | local_stream = NULL; | |
3bd1e081 | 2395 | |
50f8ae69 | 2396 | /* allocate for all fds + 1 for the consumer_data_pipe */ |
effcf122 | 2397 | pollfd = zmalloc((consumer_data.stream_count + 1) * sizeof(struct pollfd)); |
3bd1e081 | 2398 | if (pollfd == NULL) { |
7a57cf92 | 2399 | PERROR("pollfd malloc"); |
3bd1e081 MD |
2400 | pthread_mutex_unlock(&consumer_data.lock); |
2401 | goto end; | |
2402 | } | |
2403 | ||
50f8ae69 | 2404 | /* allocate for all fds + 1 for the consumer_data_pipe */ |
effcf122 | 2405 | local_stream = zmalloc((consumer_data.stream_count + 1) * |
747f8642 | 2406 | sizeof(struct lttng_consumer_stream *)); |
3bd1e081 | 2407 | if (local_stream == NULL) { |
7a57cf92 | 2408 | PERROR("local_stream malloc"); |
3bd1e081 MD |
2409 | pthread_mutex_unlock(&consumer_data.lock); |
2410 | goto end; | |
2411 | } | |
ffe60014 | 2412 | ret = update_poll_array(ctx, &pollfd, local_stream, |
43c34bc3 | 2413 | data_ht); |
3bd1e081 MD |
2414 | if (ret < 0) { |
2415 | ERR("Error in allocating pollfd or local_outfds"); | |
f73fabfd | 2416 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2417 | pthread_mutex_unlock(&consumer_data.lock); |
2418 | goto end; | |
2419 | } | |
2420 | nb_fd = ret; | |
2421 | consumer_data.need_update = 0; | |
2422 | } | |
2423 | pthread_mutex_unlock(&consumer_data.lock); | |
2424 | ||
4078b776 MD |
2425 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
2426 | if (nb_fd == 0 && consumer_quit == 1) { | |
2427 | goto end; | |
2428 | } | |
3bd1e081 | 2429 | /* poll on the array of fds */ |
88f2b785 | 2430 | restart: |
3bd1e081 | 2431 | DBG("polling on %d fd", nb_fd + 1); |
cb365c03 | 2432 | num_rdy = poll(pollfd, nb_fd + 1, -1); |
3bd1e081 MD |
2433 | DBG("poll num_rdy : %d", num_rdy); |
2434 | if (num_rdy == -1) { | |
88f2b785 MD |
2435 | /* |
2436 | * Restart interrupted system call. | |
2437 | */ | |
2438 | if (errno == EINTR) { | |
2439 | goto restart; | |
2440 | } | |
7a57cf92 | 2441 | PERROR("Poll error"); |
f73fabfd | 2442 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2443 | goto end; |
2444 | } else if (num_rdy == 0) { | |
2445 | DBG("Polling thread timed out"); | |
2446 | goto end; | |
2447 | } | |
2448 | ||
3bd1e081 | 2449 | /* |
50f8ae69 | 2450 | * If the consumer_data_pipe triggered poll go directly to the |
00e2e675 DG |
2451 | * beginning of the loop to update the array. We want to prioritize |
2452 | * array update over low-priority reads. | |
3bd1e081 | 2453 | */ |
509bb1cf | 2454 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
ab30f567 | 2455 | ssize_t pipe_readlen; |
04fdd819 | 2456 | |
50f8ae69 | 2457 | DBG("consumer_data_pipe wake up"); |
acdb9057 DG |
2458 | pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe, |
2459 | &new_stream, sizeof(new_stream)); | |
23f5f35d | 2460 | if (pipe_readlen < 0) { |
c5c46de1 | 2461 | ERR("Consumer data pipe ret %zd", pipe_readlen); |
23f5f35d DG |
2462 | /* Continue so we can at least handle the current stream(s). */ |
2463 | continue; | |
2464 | } | |
c869f647 DG |
2465 | |
2466 | /* | |
2467 | * If the stream is NULL, just ignore it. It's also possible that | |
2468 | * the sessiond poll thread changed the consumer_quit state and is | |
2469 | * waking us up to test it. | |
2470 | */ | |
2471 | if (new_stream == NULL) { | |
8994307f | 2472 | validate_endpoint_status_data_stream(); |
c869f647 DG |
2473 | continue; |
2474 | } | |
2475 | ||
c869f647 | 2476 | /* Continue to update the local streams and handle prio ones */ |
3bd1e081 MD |
2477 | continue; |
2478 | } | |
2479 | ||
2480 | /* Take care of high priority channels first. */ | |
2481 | for (i = 0; i < nb_fd; i++) { | |
9617607b DG |
2482 | if (local_stream[i] == NULL) { |
2483 | continue; | |
2484 | } | |
fb3a43a9 | 2485 | if (pollfd[i].revents & POLLPRI) { |
d41f73b7 MD |
2486 | DBG("Urgent read on fd %d", pollfd[i].fd); |
2487 | high_prio = 1; | |
4078b776 | 2488 | len = ctx->on_buffer_ready(local_stream[i], ctx); |
d41f73b7 | 2489 | /* it's ok to have an unavailable sub-buffer */ |
b64403e3 | 2490 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2491 | /* Clean the stream and free it. */ |
2492 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2493 | local_stream[i] = NULL; |
4078b776 MD |
2494 | } else if (len > 0) { |
2495 | local_stream[i]->data_read = 1; | |
d41f73b7 | 2496 | } |
3bd1e081 MD |
2497 | } |
2498 | } | |
2499 | ||
4078b776 MD |
2500 | /* |
2501 | * If we read high prio channel in this loop, try again | |
2502 | * for more high prio data. | |
2503 | */ | |
2504 | if (high_prio) { | |
3bd1e081 MD |
2505 | continue; |
2506 | } | |
2507 | ||
2508 | /* Take care of low priority channels. */ | |
4078b776 | 2509 | for (i = 0; i < nb_fd; i++) { |
9617607b DG |
2510 | if (local_stream[i] == NULL) { |
2511 | continue; | |
2512 | } | |
4078b776 MD |
2513 | if ((pollfd[i].revents & POLLIN) || |
2514 | local_stream[i]->hangup_flush_done) { | |
4078b776 MD |
2515 | DBG("Normal read on fd %d", pollfd[i].fd); |
2516 | len = ctx->on_buffer_ready(local_stream[i], ctx); | |
2517 | /* it's ok to have an unavailable sub-buffer */ | |
b64403e3 | 2518 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2519 | /* Clean the stream and free it. */ |
2520 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2521 | local_stream[i] = NULL; |
4078b776 MD |
2522 | } else if (len > 0) { |
2523 | local_stream[i]->data_read = 1; | |
2524 | } | |
2525 | } | |
2526 | } | |
2527 | ||
2528 | /* Handle hangup and errors */ | |
2529 | for (i = 0; i < nb_fd; i++) { | |
9617607b DG |
2530 | if (local_stream[i] == NULL) { |
2531 | continue; | |
2532 | } | |
4078b776 MD |
2533 | if (!local_stream[i]->hangup_flush_done |
2534 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) | |
2535 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2536 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2537 | DBG("fd %d is hup|err|nval. Attempting flush and read.", | |
9617607b | 2538 | pollfd[i].fd); |
4078b776 MD |
2539 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); |
2540 | /* Attempt read again, for the data we just flushed. */ | |
2541 | local_stream[i]->data_read = 1; | |
2542 | } | |
2543 | /* | |
2544 | * If the poll flag is HUP/ERR/NVAL and we have | |
2545 | * read no data in this pass, we can remove the | |
2546 | * stream from its hash table. | |
2547 | */ | |
2548 | if ((pollfd[i].revents & POLLHUP)) { | |
2549 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); | |
2550 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2551 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2552 | local_stream[i] = NULL; |
4078b776 MD |
2553 | num_hup++; |
2554 | } | |
2555 | } else if (pollfd[i].revents & POLLERR) { | |
2556 | ERR("Error returned in polling fd %d.", pollfd[i].fd); | |
2557 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2558 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2559 | local_stream[i] = NULL; |
4078b776 MD |
2560 | num_hup++; |
2561 | } | |
2562 | } else if (pollfd[i].revents & POLLNVAL) { | |
2563 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); | |
2564 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2565 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2566 | local_stream[i] = NULL; |
4078b776 | 2567 | num_hup++; |
3bd1e081 MD |
2568 | } |
2569 | } | |
9617607b DG |
2570 | if (local_stream[i] != NULL) { |
2571 | local_stream[i]->data_read = 0; | |
2572 | } | |
3bd1e081 MD |
2573 | } |
2574 | } | |
2575 | end: | |
2576 | DBG("polling thread exiting"); | |
0e428499 DG |
2577 | free(pollfd); |
2578 | free(local_stream); | |
fb3a43a9 DG |
2579 | |
2580 | /* | |
2581 | * Close the write side of the pipe so epoll_wait() in | |
7d980def DG |
2582 | * consumer_thread_metadata_poll can catch it. The thread is monitoring the |
2583 | * read side of the pipe. If we close them both, epoll_wait strangely does | |
2584 | * not return and could create a endless wait period if the pipe is the | |
2585 | * only tracked fd in the poll set. The thread will take care of closing | |
2586 | * the read side. | |
fb3a43a9 | 2587 | */ |
13886d2d | 2588 | (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe); |
fb3a43a9 | 2589 | |
04bb2b64 | 2590 | destroy_data_stream_ht(data_ht); |
43c34bc3 | 2591 | |
e7b994a3 | 2592 | rcu_unregister_thread(); |
3bd1e081 MD |
2593 | return NULL; |
2594 | } | |
2595 | ||
d8ef542d MD |
2596 | /* |
2597 | * Close wake-up end of each stream belonging to the channel. This will | |
2598 | * allow the poll() on the stream read-side to detect when the | |
2599 | * write-side (application) finally closes them. | |
2600 | */ | |
2601 | static | |
2602 | void consumer_close_channel_streams(struct lttng_consumer_channel *channel) | |
2603 | { | |
2604 | struct lttng_ht *ht; | |
2605 | struct lttng_consumer_stream *stream; | |
2606 | struct lttng_ht_iter iter; | |
2607 | ||
2608 | ht = consumer_data.stream_per_chan_id_ht; | |
2609 | ||
2610 | rcu_read_lock(); | |
2611 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
2612 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
2613 | ht->match_fct, &channel->key, | |
2614 | &iter.iter, stream, node_channel_id.node) { | |
f2ad556d MD |
2615 | /* |
2616 | * Protect against teardown with mutex. | |
2617 | */ | |
2618 | pthread_mutex_lock(&stream->lock); | |
2619 | if (cds_lfht_is_node_deleted(&stream->node.node)) { | |
2620 | goto next; | |
2621 | } | |
d8ef542d MD |
2622 | switch (consumer_data.type) { |
2623 | case LTTNG_CONSUMER_KERNEL: | |
2624 | break; | |
2625 | case LTTNG_CONSUMER32_UST: | |
2626 | case LTTNG_CONSUMER64_UST: | |
2627 | /* | |
2628 | * Note: a mutex is taken internally within | |
2629 | * liblttng-ust-ctl to protect timer wakeup_fd | |
2630 | * use from concurrent close. | |
2631 | */ | |
2632 | lttng_ustconsumer_close_stream_wakeup(stream); | |
2633 | break; | |
2634 | default: | |
2635 | ERR("Unknown consumer_data type"); | |
2636 | assert(0); | |
2637 | } | |
f2ad556d MD |
2638 | next: |
2639 | pthread_mutex_unlock(&stream->lock); | |
d8ef542d MD |
2640 | } |
2641 | rcu_read_unlock(); | |
2642 | } | |
2643 | ||
2644 | static void destroy_channel_ht(struct lttng_ht *ht) | |
2645 | { | |
2646 | struct lttng_ht_iter iter; | |
2647 | struct lttng_consumer_channel *channel; | |
2648 | int ret; | |
2649 | ||
2650 | if (ht == NULL) { | |
2651 | return; | |
2652 | } | |
2653 | ||
2654 | rcu_read_lock(); | |
2655 | cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) { | |
2656 | ret = lttng_ht_del(ht, &iter); | |
2657 | assert(ret != 0); | |
2658 | } | |
2659 | rcu_read_unlock(); | |
2660 | ||
2661 | lttng_ht_destroy(ht); | |
2662 | } | |
2663 | ||
2664 | /* | |
2665 | * This thread polls the channel fds to detect when they are being | |
2666 | * closed. It closes all related streams if the channel is detected as | |
2667 | * closed. It is currently only used as a shim layer for UST because the | |
2668 | * consumerd needs to keep the per-stream wakeup end of pipes open for | |
2669 | * periodical flush. | |
2670 | */ | |
2671 | void *consumer_thread_channel_poll(void *data) | |
2672 | { | |
2673 | int ret, i, pollfd; | |
2674 | uint32_t revents, nb_fd; | |
2675 | struct lttng_consumer_channel *chan = NULL; | |
2676 | struct lttng_ht_iter iter; | |
2677 | struct lttng_ht_node_u64 *node; | |
2678 | struct lttng_poll_event events; | |
2679 | struct lttng_consumer_local_data *ctx = data; | |
2680 | struct lttng_ht *channel_ht; | |
2681 | ||
2682 | rcu_register_thread(); | |
2683 | ||
2684 | channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
2685 | if (!channel_ht) { | |
2686 | /* ENOMEM at this point. Better to bail out. */ | |
2687 | goto end_ht; | |
2688 | } | |
2689 | ||
2690 | DBG("Thread channel poll started"); | |
2691 | ||
2692 | /* Size is set to 1 for the consumer_channel pipe */ | |
2693 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2694 | if (ret < 0) { | |
2695 | ERR("Poll set creation failed"); | |
2696 | goto end_poll; | |
2697 | } | |
2698 | ||
2699 | ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN); | |
2700 | if (ret < 0) { | |
2701 | goto end; | |
2702 | } | |
2703 | ||
2704 | /* Main loop */ | |
2705 | DBG("Channel main loop started"); | |
2706 | ||
2707 | while (1) { | |
2708 | /* Only the channel pipe is set */ | |
2709 | if (LTTNG_POLL_GETNB(&events) == 0 && consumer_quit == 1) { | |
2710 | goto end; | |
2711 | } | |
2712 | ||
2713 | restart: | |
2714 | DBG("Channel poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events)); | |
2715 | ret = lttng_poll_wait(&events, -1); | |
2716 | DBG("Channel event catched in thread"); | |
2717 | if (ret < 0) { | |
2718 | if (errno == EINTR) { | |
2719 | ERR("Poll EINTR catched"); | |
2720 | goto restart; | |
2721 | } | |
2722 | goto end; | |
2723 | } | |
2724 | ||
2725 | nb_fd = ret; | |
2726 | ||
2727 | /* From here, the event is a channel wait fd */ | |
2728 | for (i = 0; i < nb_fd; i++) { | |
2729 | revents = LTTNG_POLL_GETEV(&events, i); | |
2730 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2731 | ||
2732 | /* Just don't waste time if no returned events for the fd */ | |
2733 | if (!revents) { | |
2734 | continue; | |
2735 | } | |
2736 | if (pollfd == ctx->consumer_channel_pipe[0]) { | |
2737 | if (revents & (LPOLLERR | LPOLLHUP)) { | |
2738 | DBG("Channel thread pipe hung up"); | |
2739 | /* | |
2740 | * Remove the pipe from the poll set and continue the loop | |
2741 | * since their might be data to consume. | |
2742 | */ | |
2743 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
2744 | continue; | |
2745 | } else if (revents & LPOLLIN) { | |
2746 | enum consumer_channel_action action; | |
a0cbdd2e | 2747 | uint64_t key; |
d8ef542d | 2748 | |
a0cbdd2e | 2749 | ret = read_channel_pipe(ctx, &chan, &key, &action); |
d8ef542d MD |
2750 | if (ret <= 0) { |
2751 | ERR("Error reading channel pipe"); | |
2752 | continue; | |
2753 | } | |
2754 | ||
2755 | switch (action) { | |
2756 | case CONSUMER_CHANNEL_ADD: | |
2757 | DBG("Adding channel %d to poll set", | |
2758 | chan->wait_fd); | |
2759 | ||
2760 | lttng_ht_node_init_u64(&chan->wait_fd_node, | |
2761 | chan->wait_fd); | |
c7260a81 | 2762 | rcu_read_lock(); |
d8ef542d MD |
2763 | lttng_ht_add_unique_u64(channel_ht, |
2764 | &chan->wait_fd_node); | |
c7260a81 | 2765 | rcu_read_unlock(); |
d8ef542d MD |
2766 | /* Add channel to the global poll events list */ |
2767 | lttng_poll_add(&events, chan->wait_fd, | |
2768 | LPOLLIN | LPOLLPRI); | |
2769 | break; | |
a0cbdd2e MD |
2770 | case CONSUMER_CHANNEL_DEL: |
2771 | { | |
f2a444f1 DG |
2772 | struct lttng_consumer_stream *stream, *stmp; |
2773 | ||
c7260a81 | 2774 | rcu_read_lock(); |
a0cbdd2e MD |
2775 | chan = consumer_find_channel(key); |
2776 | if (!chan) { | |
c7260a81 | 2777 | rcu_read_unlock(); |
a0cbdd2e MD |
2778 | ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key); |
2779 | break; | |
2780 | } | |
2781 | lttng_poll_del(&events, chan->wait_fd); | |
f623cc0b | 2782 | iter.iter.node = &chan->wait_fd_node.node; |
a0cbdd2e MD |
2783 | ret = lttng_ht_del(channel_ht, &iter); |
2784 | assert(ret == 0); | |
2785 | consumer_close_channel_streams(chan); | |
2786 | ||
f2a444f1 DG |
2787 | switch (consumer_data.type) { |
2788 | case LTTNG_CONSUMER_KERNEL: | |
2789 | break; | |
2790 | case LTTNG_CONSUMER32_UST: | |
2791 | case LTTNG_CONSUMER64_UST: | |
2792 | /* Delete streams that might have been left in the stream list. */ | |
2793 | cds_list_for_each_entry_safe(stream, stmp, &chan->streams.head, | |
2794 | send_node) { | |
2795 | cds_list_del(&stream->send_node); | |
2796 | lttng_ustconsumer_del_stream(stream); | |
2797 | uatomic_sub(&stream->chan->refcount, 1); | |
2798 | assert(&chan->refcount); | |
2799 | free(stream); | |
2800 | } | |
2801 | break; | |
2802 | default: | |
2803 | ERR("Unknown consumer_data type"); | |
2804 | assert(0); | |
2805 | } | |
2806 | ||
a0cbdd2e MD |
2807 | /* |
2808 | * Release our own refcount. Force channel deletion even if | |
2809 | * streams were not initialized. | |
2810 | */ | |
2811 | if (!uatomic_sub_return(&chan->refcount, 1)) { | |
2812 | consumer_del_channel(chan); | |
2813 | } | |
c7260a81 | 2814 | rcu_read_unlock(); |
a0cbdd2e MD |
2815 | goto restart; |
2816 | } | |
d8ef542d MD |
2817 | case CONSUMER_CHANNEL_QUIT: |
2818 | /* | |
2819 | * Remove the pipe from the poll set and continue the loop | |
2820 | * since their might be data to consume. | |
2821 | */ | |
2822 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
2823 | continue; | |
2824 | default: | |
2825 | ERR("Unknown action"); | |
2826 | break; | |
2827 | } | |
2828 | } | |
2829 | ||
2830 | /* Handle other stream */ | |
2831 | continue; | |
2832 | } | |
2833 | ||
2834 | rcu_read_lock(); | |
2835 | { | |
2836 | uint64_t tmp_id = (uint64_t) pollfd; | |
2837 | ||
2838 | lttng_ht_lookup(channel_ht, &tmp_id, &iter); | |
2839 | } | |
2840 | node = lttng_ht_iter_get_node_u64(&iter); | |
2841 | assert(node); | |
2842 | ||
2843 | chan = caa_container_of(node, struct lttng_consumer_channel, | |
2844 | wait_fd_node); | |
2845 | ||
2846 | /* Check for error event */ | |
2847 | if (revents & (LPOLLERR | LPOLLHUP)) { | |
2848 | DBG("Channel fd %d is hup|err.", pollfd); | |
2849 | ||
2850 | lttng_poll_del(&events, chan->wait_fd); | |
2851 | ret = lttng_ht_del(channel_ht, &iter); | |
2852 | assert(ret == 0); | |
f2a444f1 | 2853 | assert(cds_list_empty(&chan->streams.head)); |
d8ef542d | 2854 | consumer_close_channel_streams(chan); |
f2ad556d MD |
2855 | |
2856 | /* Release our own refcount */ | |
2857 | if (!uatomic_sub_return(&chan->refcount, 1) | |
2858 | && !uatomic_read(&chan->nb_init_stream_left)) { | |
2859 | consumer_del_channel(chan); | |
2860 | } | |
d8ef542d MD |
2861 | } |
2862 | ||
2863 | /* Release RCU lock for the channel looked up */ | |
2864 | rcu_read_unlock(); | |
2865 | } | |
2866 | } | |
2867 | ||
2868 | end: | |
2869 | lttng_poll_clean(&events); | |
2870 | end_poll: | |
2871 | destroy_channel_ht(channel_ht); | |
2872 | end_ht: | |
2873 | DBG("Channel poll thread exiting"); | |
2874 | rcu_unregister_thread(); | |
2875 | return NULL; | |
2876 | } | |
2877 | ||
331744e3 JD |
2878 | static int set_metadata_socket(struct lttng_consumer_local_data *ctx, |
2879 | struct pollfd *sockpoll, int client_socket) | |
2880 | { | |
2881 | int ret; | |
2882 | ||
2883 | assert(ctx); | |
2884 | assert(sockpoll); | |
2885 | ||
2886 | if (lttng_consumer_poll_socket(sockpoll) < 0) { | |
2887 | ret = -1; | |
2888 | goto error; | |
2889 | } | |
2890 | DBG("Metadata connection on client_socket"); | |
2891 | ||
2892 | /* Blocking call, waiting for transmission */ | |
2893 | ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket); | |
2894 | if (ctx->consumer_metadata_socket < 0) { | |
2895 | WARN("On accept metadata"); | |
2896 | ret = -1; | |
2897 | goto error; | |
2898 | } | |
2899 | ret = 0; | |
2900 | ||
2901 | error: | |
2902 | return ret; | |
2903 | } | |
2904 | ||
3bd1e081 MD |
2905 | /* |
2906 | * This thread listens on the consumerd socket and receives the file | |
2907 | * descriptors from the session daemon. | |
2908 | */ | |
7d980def | 2909 | void *consumer_thread_sessiond_poll(void *data) |
3bd1e081 | 2910 | { |
d96f09c6 | 2911 | int sock = -1, client_socket, ret; |
3bd1e081 MD |
2912 | /* |
2913 | * structure to poll for incoming data on communication socket avoids | |
2914 | * making blocking sockets. | |
2915 | */ | |
2916 | struct pollfd consumer_sockpoll[2]; | |
2917 | struct lttng_consumer_local_data *ctx = data; | |
2918 | ||
e7b994a3 DG |
2919 | rcu_register_thread(); |
2920 | ||
3bd1e081 MD |
2921 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
2922 | unlink(ctx->consumer_command_sock_path); | |
2923 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); | |
2924 | if (client_socket < 0) { | |
2925 | ERR("Cannot create command socket"); | |
2926 | goto end; | |
2927 | } | |
2928 | ||
2929 | ret = lttcomm_listen_unix_sock(client_socket); | |
2930 | if (ret < 0) { | |
2931 | goto end; | |
2932 | } | |
2933 | ||
32258573 | 2934 | DBG("Sending ready command to lttng-sessiond"); |
f73fabfd | 2935 | ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); |
3bd1e081 MD |
2936 | /* return < 0 on error, but == 0 is not fatal */ |
2937 | if (ret < 0) { | |
32258573 | 2938 | ERR("Error sending ready command to lttng-sessiond"); |
3bd1e081 MD |
2939 | goto end; |
2940 | } | |
2941 | ||
3bd1e081 MD |
2942 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ |
2943 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; | |
2944 | consumer_sockpoll[0].events = POLLIN | POLLPRI; | |
2945 | consumer_sockpoll[1].fd = client_socket; | |
2946 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
2947 | ||
2948 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
2949 | goto end; | |
2950 | } | |
2951 | DBG("Connection on client_socket"); | |
2952 | ||
2953 | /* Blocking call, waiting for transmission */ | |
2954 | sock = lttcomm_accept_unix_sock(client_socket); | |
534d2592 | 2955 | if (sock < 0) { |
3bd1e081 MD |
2956 | WARN("On accept"); |
2957 | goto end; | |
2958 | } | |
3bd1e081 | 2959 | |
331744e3 JD |
2960 | /* |
2961 | * Setup metadata socket which is the second socket connection on the | |
2962 | * command unix socket. | |
2963 | */ | |
2964 | ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket); | |
2965 | if (ret < 0) { | |
2966 | goto end; | |
2967 | } | |
2968 | ||
d96f09c6 DG |
2969 | /* This socket is not useful anymore. */ |
2970 | ret = close(client_socket); | |
2971 | if (ret < 0) { | |
2972 | PERROR("close client_socket"); | |
2973 | } | |
2974 | client_socket = -1; | |
2975 | ||
3bd1e081 MD |
2976 | /* update the polling structure to poll on the established socket */ |
2977 | consumer_sockpoll[1].fd = sock; | |
2978 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
2979 | ||
2980 | while (1) { | |
2981 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
2982 | goto end; | |
2983 | } | |
2984 | DBG("Incoming command on sock"); | |
2985 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
2986 | if (ret == -ENOENT) { | |
2987 | DBG("Received STOP command"); | |
2988 | goto end; | |
2989 | } | |
4cbc1a04 DG |
2990 | if (ret <= 0) { |
2991 | /* | |
2992 | * This could simply be a session daemon quitting. Don't output | |
2993 | * ERR() here. | |
2994 | */ | |
2995 | DBG("Communication interrupted on command socket"); | |
3bd1e081 MD |
2996 | goto end; |
2997 | } | |
2998 | if (consumer_quit) { | |
2999 | DBG("consumer_thread_receive_fds received quit from signal"); | |
3000 | goto end; | |
3001 | } | |
ffe60014 | 3002 | DBG("received command on sock"); |
3bd1e081 MD |
3003 | } |
3004 | end: | |
ffe60014 | 3005 | DBG("Consumer thread sessiond poll exiting"); |
3bd1e081 | 3006 | |
d88aee68 DG |
3007 | /* |
3008 | * Close metadata streams since the producer is the session daemon which | |
3009 | * just died. | |
3010 | * | |
3011 | * NOTE: for now, this only applies to the UST tracer. | |
3012 | */ | |
3013 | lttng_consumer_close_metadata(); | |
3014 | ||
3bd1e081 MD |
3015 | /* |
3016 | * when all fds have hung up, the polling thread | |
3017 | * can exit cleanly | |
3018 | */ | |
3019 | consumer_quit = 1; | |
3020 | ||
04fdd819 | 3021 | /* |
c869f647 | 3022 | * Notify the data poll thread to poll back again and test the |
8994307f | 3023 | * consumer_quit state that we just set so to quit gracefully. |
04fdd819 | 3024 | */ |
acdb9057 | 3025 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
c869f647 | 3026 | |
a0cbdd2e | 3027 | notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT); |
d8ef542d | 3028 | |
d96f09c6 DG |
3029 | /* Cleaning up possibly open sockets. */ |
3030 | if (sock >= 0) { | |
3031 | ret = close(sock); | |
3032 | if (ret < 0) { | |
3033 | PERROR("close sock sessiond poll"); | |
3034 | } | |
3035 | } | |
3036 | if (client_socket >= 0) { | |
38476d24 | 3037 | ret = close(client_socket); |
d96f09c6 DG |
3038 | if (ret < 0) { |
3039 | PERROR("close client_socket sessiond poll"); | |
3040 | } | |
3041 | } | |
3042 | ||
e7b994a3 | 3043 | rcu_unregister_thread(); |
3bd1e081 MD |
3044 | return NULL; |
3045 | } | |
d41f73b7 | 3046 | |
4078b776 | 3047 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
d41f73b7 MD |
3048 | struct lttng_consumer_local_data *ctx) |
3049 | { | |
74251bb8 DG |
3050 | ssize_t ret; |
3051 | ||
3052 | pthread_mutex_lock(&stream->lock); | |
3053 | ||
d41f73b7 MD |
3054 | switch (consumer_data.type) { |
3055 | case LTTNG_CONSUMER_KERNEL: | |
74251bb8 DG |
3056 | ret = lttng_kconsumer_read_subbuffer(stream, ctx); |
3057 | break; | |
7753dea8 MD |
3058 | case LTTNG_CONSUMER32_UST: |
3059 | case LTTNG_CONSUMER64_UST: | |
74251bb8 DG |
3060 | ret = lttng_ustconsumer_read_subbuffer(stream, ctx); |
3061 | break; | |
d41f73b7 MD |
3062 | default: |
3063 | ERR("Unknown consumer_data type"); | |
3064 | assert(0); | |
74251bb8 DG |
3065 | ret = -ENOSYS; |
3066 | break; | |
d41f73b7 | 3067 | } |
74251bb8 DG |
3068 | |
3069 | pthread_mutex_unlock(&stream->lock); | |
3070 | return ret; | |
d41f73b7 MD |
3071 | } |
3072 | ||
3073 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
3074 | { | |
3075 | switch (consumer_data.type) { | |
3076 | case LTTNG_CONSUMER_KERNEL: | |
3077 | return lttng_kconsumer_on_recv_stream(stream); | |
7753dea8 MD |
3078 | case LTTNG_CONSUMER32_UST: |
3079 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
3080 | return lttng_ustconsumer_on_recv_stream(stream); |
3081 | default: | |
3082 | ERR("Unknown consumer_data type"); | |
3083 | assert(0); | |
3084 | return -ENOSYS; | |
3085 | } | |
3086 | } | |
e4421fec DG |
3087 | |
3088 | /* | |
3089 | * Allocate and set consumer data hash tables. | |
3090 | */ | |
3091 | void lttng_consumer_init(void) | |
3092 | { | |
d88aee68 DG |
3093 | consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
3094 | consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3095 | consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
d8ef542d | 3096 | consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
e4421fec | 3097 | } |
7735ef9e DG |
3098 | |
3099 | /* | |
3100 | * Process the ADD_RELAYD command receive by a consumer. | |
3101 | * | |
3102 | * This will create a relayd socket pair and add it to the relayd hash table. | |
3103 | * The caller MUST acquire a RCU read side lock before calling it. | |
3104 | */ | |
98cf381a | 3105 | int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type, |
7735ef9e | 3106 | struct lttng_consumer_local_data *ctx, int sock, |
6151a90f | 3107 | struct pollfd *consumer_sockpoll, |
bcfa3ff9 | 3108 | struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id) |
7735ef9e | 3109 | { |
cd2b09ed | 3110 | int fd = -1, ret = -1, relayd_created = 0; |
f50f23d9 | 3111 | enum lttng_error_code ret_code = LTTNG_OK; |
d4298c99 | 3112 | struct consumer_relayd_sock_pair *relayd = NULL; |
7735ef9e | 3113 | |
6151a90f JD |
3114 | assert(ctx); |
3115 | assert(relayd_sock); | |
3116 | ||
98cf381a | 3117 | DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx); |
7735ef9e DG |
3118 | |
3119 | /* Get relayd reference if exists. */ | |
3120 | relayd = consumer_find_relayd(net_seq_idx); | |
3121 | if (relayd == NULL) { | |
98cf381a | 3122 | assert(sock_type == LTTNG_STREAM_CONTROL); |
7735ef9e DG |
3123 | /* Not found. Allocate one. */ |
3124 | relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); | |
3125 | if (relayd == NULL) { | |
0d08d75e | 3126 | ret = -ENOMEM; |
e124929f MD |
3127 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
3128 | goto error; | |
0d08d75e | 3129 | } else { |
bcfa3ff9 | 3130 | relayd->sessiond_session_id = sessiond_id; |
0d08d75e | 3131 | relayd_created = 1; |
7735ef9e | 3132 | } |
0d08d75e DG |
3133 | |
3134 | /* | |
3135 | * This code path MUST continue to the consumer send status message to | |
3136 | * we can notify the session daemon and continue our work without | |
3137 | * killing everything. | |
3138 | */ | |
98cf381a MD |
3139 | } else { |
3140 | /* | |
3141 | * relayd key should never be found for control socket. | |
3142 | */ | |
3143 | assert(sock_type != LTTNG_STREAM_CONTROL); | |
0d08d75e DG |
3144 | } |
3145 | ||
3146 | /* First send a status message before receiving the fds. */ | |
e124929f MD |
3147 | ret = consumer_send_status_msg(sock, LTTNG_OK); |
3148 | if (ret < 0) { | |
0d08d75e | 3149 | /* Somehow, the session daemon is not responding anymore. */ |
e124929f MD |
3150 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); |
3151 | goto error_nosignal; | |
7735ef9e DG |
3152 | } |
3153 | ||
3154 | /* Poll on consumer socket. */ | |
3155 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
0d08d75e | 3156 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
7735ef9e | 3157 | ret = -EINTR; |
e124929f | 3158 | goto error_nosignal; |
7735ef9e DG |
3159 | } |
3160 | ||
3161 | /* Get relayd socket from session daemon */ | |
3162 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); | |
3163 | if (ret != sizeof(fd)) { | |
7735ef9e | 3164 | ret = -1; |
4028eeb9 | 3165 | fd = -1; /* Just in case it gets set with an invalid value. */ |
0d08d75e DG |
3166 | |
3167 | /* | |
3168 | * Failing to receive FDs might indicate a major problem such as | |
3169 | * reaching a fd limit during the receive where the kernel returns a | |
3170 | * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we | |
3171 | * don't take any chances and stop everything. | |
3172 | * | |
3173 | * XXX: Feature request #558 will fix that and avoid this possible | |
3174 | * issue when reaching the fd limit. | |
3175 | */ | |
3176 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); | |
e124929f | 3177 | ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD; |
f50f23d9 DG |
3178 | goto error; |
3179 | } | |
3180 | ||
7735ef9e DG |
3181 | /* Copy socket information and received FD */ |
3182 | switch (sock_type) { | |
3183 | case LTTNG_STREAM_CONTROL: | |
3184 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3185 | lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock); |
3186 | ret = lttcomm_create_sock(&relayd->control_sock.sock); | |
4028eeb9 | 3187 | /* Handle create_sock error. */ |
f66c074c | 3188 | if (ret < 0) { |
e124929f | 3189 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3190 | goto error; |
f66c074c | 3191 | } |
98cf381a MD |
3192 | /* |
3193 | * Close the socket created internally by | |
3194 | * lttcomm_create_sock, so we can replace it by the one | |
3195 | * received from sessiond. | |
3196 | */ | |
3197 | if (close(relayd->control_sock.sock.fd)) { | |
3198 | PERROR("close"); | |
3199 | } | |
7735ef9e DG |
3200 | |
3201 | /* Assign new file descriptor */ | |
6151a90f | 3202 | relayd->control_sock.sock.fd = fd; |
f1dab979 | 3203 | fd = -1; /* For error path */ |
6151a90f JD |
3204 | /* Assign version values. */ |
3205 | relayd->control_sock.major = relayd_sock->major; | |
3206 | relayd->control_sock.minor = relayd_sock->minor; | |
c5b6f4f0 DG |
3207 | |
3208 | /* | |
59e71485 DG |
3209 | * Create a session on the relayd and store the returned id. Lock the |
3210 | * control socket mutex if the relayd was NOT created before. | |
c5b6f4f0 | 3211 | */ |
59e71485 DG |
3212 | if (!relayd_created) { |
3213 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3214 | } | |
c5b6f4f0 | 3215 | ret = relayd_create_session(&relayd->control_sock, |
f7079f67 | 3216 | &relayd->relayd_session_id); |
59e71485 DG |
3217 | if (!relayd_created) { |
3218 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3219 | } | |
c5b6f4f0 | 3220 | if (ret < 0) { |
ffe60014 DG |
3221 | /* |
3222 | * Close all sockets of a relayd object. It will be freed if it was | |
3223 | * created at the error code path or else it will be garbage | |
3224 | * collect. | |
3225 | */ | |
3226 | (void) relayd_close(&relayd->control_sock); | |
3227 | (void) relayd_close(&relayd->data_sock); | |
e124929f | 3228 | ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL; |
c5b6f4f0 DG |
3229 | goto error; |
3230 | } | |
3231 | ||
7735ef9e DG |
3232 | break; |
3233 | case LTTNG_STREAM_DATA: | |
3234 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3235 | lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock); |
3236 | ret = lttcomm_create_sock(&relayd->data_sock.sock); | |
4028eeb9 | 3237 | /* Handle create_sock error. */ |
f66c074c | 3238 | if (ret < 0) { |
e124929f | 3239 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3240 | goto error; |
f66c074c | 3241 | } |
98cf381a MD |
3242 | /* |
3243 | * Close the socket created internally by | |
3244 | * lttcomm_create_sock, so we can replace it by the one | |
3245 | * received from sessiond. | |
3246 | */ | |
3247 | if (close(relayd->data_sock.sock.fd)) { | |
3248 | PERROR("close"); | |
3249 | } | |
7735ef9e DG |
3250 | |
3251 | /* Assign new file descriptor */ | |
6151a90f | 3252 | relayd->data_sock.sock.fd = fd; |
f1dab979 | 3253 | fd = -1; /* for eventual error paths */ |
6151a90f JD |
3254 | /* Assign version values. */ |
3255 | relayd->data_sock.major = relayd_sock->major; | |
3256 | relayd->data_sock.minor = relayd_sock->minor; | |
7735ef9e DG |
3257 | break; |
3258 | default: | |
3259 | ERR("Unknown relayd socket type (%d)", sock_type); | |
59e71485 | 3260 | ret = -1; |
e124929f | 3261 | ret_code = LTTCOMM_CONSUMERD_FATAL; |
7735ef9e DG |
3262 | goto error; |
3263 | } | |
3264 | ||
d88aee68 | 3265 | DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)", |
7735ef9e DG |
3266 | sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", |
3267 | relayd->net_seq_idx, fd); | |
3268 | ||
e124929f MD |
3269 | /* We successfully added the socket. Send status back. */ |
3270 | ret = consumer_send_status_msg(sock, ret_code); | |
3271 | if (ret < 0) { | |
3272 | /* Somehow, the session daemon is not responding anymore. */ | |
3273 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3274 | goto error_nosignal; | |
3275 | } | |
3276 | ||
7735ef9e DG |
3277 | /* |
3278 | * Add relayd socket pair to consumer data hashtable. If object already | |
3279 | * exists or on error, the function gracefully returns. | |
3280 | */ | |
d09e1200 | 3281 | add_relayd(relayd); |
7735ef9e DG |
3282 | |
3283 | /* All good! */ | |
4028eeb9 | 3284 | return 0; |
7735ef9e DG |
3285 | |
3286 | error: | |
e124929f MD |
3287 | if (consumer_send_status_msg(sock, ret_code) < 0) { |
3288 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3289 | } | |
3290 | ||
3291 | error_nosignal: | |
4028eeb9 DG |
3292 | /* Close received socket if valid. */ |
3293 | if (fd >= 0) { | |
3294 | if (close(fd)) { | |
3295 | PERROR("close received socket"); | |
3296 | } | |
3297 | } | |
cd2b09ed DG |
3298 | |
3299 | if (relayd_created) { | |
cd2b09ed DG |
3300 | free(relayd); |
3301 | } | |
3302 | ||
7735ef9e DG |
3303 | return ret; |
3304 | } | |
ca22feea | 3305 | |
4e9a4686 DG |
3306 | /* |
3307 | * Try to lock the stream mutex. | |
3308 | * | |
3309 | * On success, 1 is returned else 0 indicating that the mutex is NOT lock. | |
3310 | */ | |
3311 | static int stream_try_lock(struct lttng_consumer_stream *stream) | |
3312 | { | |
3313 | int ret; | |
3314 | ||
3315 | assert(stream); | |
3316 | ||
3317 | /* | |
3318 | * Try to lock the stream mutex. On failure, we know that the stream is | |
3319 | * being used else where hence there is data still being extracted. | |
3320 | */ | |
3321 | ret = pthread_mutex_trylock(&stream->lock); | |
3322 | if (ret) { | |
3323 | /* For both EBUSY and EINVAL error, the mutex is NOT locked. */ | |
3324 | ret = 0; | |
3325 | goto end; | |
3326 | } | |
3327 | ||
3328 | ret = 1; | |
3329 | ||
3330 | end: | |
3331 | return ret; | |
3332 | } | |
3333 | ||
f7079f67 DG |
3334 | /* |
3335 | * Search for a relayd associated to the session id and return the reference. | |
3336 | * | |
3337 | * A rcu read side lock MUST be acquire before calling this function and locked | |
3338 | * until the relayd object is no longer necessary. | |
3339 | */ | |
3340 | static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id) | |
3341 | { | |
3342 | struct lttng_ht_iter iter; | |
f7079f67 | 3343 | struct consumer_relayd_sock_pair *relayd = NULL; |
f7079f67 DG |
3344 | |
3345 | /* Iterate over all relayd since they are indexed by net_seq_idx. */ | |
3346 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
3347 | node.node) { | |
18261bd1 DG |
3348 | /* |
3349 | * Check by sessiond id which is unique here where the relayd session | |
3350 | * id might not be when having multiple relayd. | |
3351 | */ | |
3352 | if (relayd->sessiond_session_id == id) { | |
f7079f67 | 3353 | /* Found the relayd. There can be only one per id. */ |
18261bd1 | 3354 | goto found; |
f7079f67 DG |
3355 | } |
3356 | } | |
3357 | ||
18261bd1 DG |
3358 | return NULL; |
3359 | ||
3360 | found: | |
f7079f67 DG |
3361 | return relayd; |
3362 | } | |
3363 | ||
ca22feea DG |
3364 | /* |
3365 | * Check if for a given session id there is still data needed to be extract | |
3366 | * from the buffers. | |
3367 | * | |
6d805429 | 3368 | * Return 1 if data is pending or else 0 meaning ready to be read. |
ca22feea | 3369 | */ |
6d805429 | 3370 | int consumer_data_pending(uint64_t id) |
ca22feea DG |
3371 | { |
3372 | int ret; | |
3373 | struct lttng_ht_iter iter; | |
3374 | struct lttng_ht *ht; | |
3375 | struct lttng_consumer_stream *stream; | |
f7079f67 | 3376 | struct consumer_relayd_sock_pair *relayd = NULL; |
6d805429 | 3377 | int (*data_pending)(struct lttng_consumer_stream *); |
ca22feea | 3378 | |
6d805429 | 3379 | DBG("Consumer data pending command on session id %" PRIu64, id); |
ca22feea | 3380 | |
6f6eda74 | 3381 | rcu_read_lock(); |
ca22feea DG |
3382 | pthread_mutex_lock(&consumer_data.lock); |
3383 | ||
3384 | switch (consumer_data.type) { | |
3385 | case LTTNG_CONSUMER_KERNEL: | |
6d805429 | 3386 | data_pending = lttng_kconsumer_data_pending; |
ca22feea DG |
3387 | break; |
3388 | case LTTNG_CONSUMER32_UST: | |
3389 | case LTTNG_CONSUMER64_UST: | |
6d805429 | 3390 | data_pending = lttng_ustconsumer_data_pending; |
ca22feea DG |
3391 | break; |
3392 | default: | |
3393 | ERR("Unknown consumer data type"); | |
3394 | assert(0); | |
3395 | } | |
3396 | ||
3397 | /* Ease our life a bit */ | |
3398 | ht = consumer_data.stream_list_ht; | |
3399 | ||
f7079f67 DG |
3400 | relayd = find_relayd_by_session_id(id); |
3401 | if (relayd) { | |
3402 | /* Send init command for data pending. */ | |
3403 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3404 | ret = relayd_begin_data_pending(&relayd->control_sock, | |
3405 | relayd->relayd_session_id); | |
3406 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3407 | if (ret < 0) { | |
3408 | /* Communication error thus the relayd so no data pending. */ | |
3409 | goto data_not_pending; | |
3410 | } | |
3411 | } | |
3412 | ||
c8f59ee5 | 3413 | cds_lfht_for_each_entry_duplicate(ht->ht, |
d88aee68 DG |
3414 | ht->hash_fct(&id, lttng_ht_seed), |
3415 | ht->match_fct, &id, | |
ca22feea | 3416 | &iter.iter, stream, node_session_id.node) { |
4e9a4686 DG |
3417 | /* If this call fails, the stream is being used hence data pending. */ |
3418 | ret = stream_try_lock(stream); | |
3419 | if (!ret) { | |
f7079f67 | 3420 | goto data_pending; |
ca22feea | 3421 | } |
ca22feea | 3422 | |
4e9a4686 DG |
3423 | /* |
3424 | * A removed node from the hash table indicates that the stream has | |
3425 | * been deleted thus having a guarantee that the buffers are closed | |
3426 | * on the consumer side. However, data can still be transmitted | |
3427 | * over the network so don't skip the relayd check. | |
3428 | */ | |
3429 | ret = cds_lfht_is_node_deleted(&stream->node.node); | |
3430 | if (!ret) { | |
3431 | /* Check the stream if there is data in the buffers. */ | |
6d805429 DG |
3432 | ret = data_pending(stream); |
3433 | if (ret == 1) { | |
4e9a4686 | 3434 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3435 | goto data_pending; |
4e9a4686 DG |
3436 | } |
3437 | } | |
3438 | ||
3439 | /* Relayd check */ | |
f7079f67 | 3440 | if (relayd) { |
c8f59ee5 DG |
3441 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
3442 | if (stream->metadata_flag) { | |
ad7051c0 DG |
3443 | ret = relayd_quiescent_control(&relayd->control_sock, |
3444 | stream->relayd_stream_id); | |
c8f59ee5 | 3445 | } else { |
6d805429 | 3446 | ret = relayd_data_pending(&relayd->control_sock, |
39df6d9f DG |
3447 | stream->relayd_stream_id, |
3448 | stream->next_net_seq_num - 1); | |
c8f59ee5 DG |
3449 | } |
3450 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
6d805429 | 3451 | if (ret == 1) { |
4e9a4686 | 3452 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3453 | goto data_pending; |
c8f59ee5 DG |
3454 | } |
3455 | } | |
4e9a4686 | 3456 | pthread_mutex_unlock(&stream->lock); |
c8f59ee5 | 3457 | } |
ca22feea | 3458 | |
f7079f67 DG |
3459 | if (relayd) { |
3460 | unsigned int is_data_inflight = 0; | |
3461 | ||
3462 | /* Send init command for data pending. */ | |
3463 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3464 | ret = relayd_end_data_pending(&relayd->control_sock, | |
3465 | relayd->relayd_session_id, &is_data_inflight); | |
3466 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
bdd88757 | 3467 | if (ret < 0) { |
f7079f67 DG |
3468 | goto data_not_pending; |
3469 | } | |
bdd88757 DG |
3470 | if (is_data_inflight) { |
3471 | goto data_pending; | |
3472 | } | |
f7079f67 DG |
3473 | } |
3474 | ||
ca22feea | 3475 | /* |
f7079f67 DG |
3476 | * Finding _no_ node in the hash table and no inflight data means that the |
3477 | * stream(s) have been removed thus data is guaranteed to be available for | |
3478 | * analysis from the trace files. | |
ca22feea DG |
3479 | */ |
3480 | ||
f7079f67 | 3481 | data_not_pending: |
ca22feea DG |
3482 | /* Data is available to be read by a viewer. */ |
3483 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3484 | rcu_read_unlock(); |
6d805429 | 3485 | return 0; |
ca22feea | 3486 | |
f7079f67 | 3487 | data_pending: |
ca22feea DG |
3488 | /* Data is still being extracted from buffers. */ |
3489 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3490 | rcu_read_unlock(); |
6d805429 | 3491 | return 1; |
ca22feea | 3492 | } |
f50f23d9 DG |
3493 | |
3494 | /* | |
3495 | * Send a ret code status message to the sessiond daemon. | |
3496 | * | |
3497 | * Return the sendmsg() return value. | |
3498 | */ | |
3499 | int consumer_send_status_msg(int sock, int ret_code) | |
3500 | { | |
3501 | struct lttcomm_consumer_status_msg msg; | |
3502 | ||
3503 | msg.ret_code = ret_code; | |
3504 | ||
3505 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3506 | } | |
ffe60014 DG |
3507 | |
3508 | /* | |
3509 | * Send a channel status message to the sessiond daemon. | |
3510 | * | |
3511 | * Return the sendmsg() return value. | |
3512 | */ | |
3513 | int consumer_send_status_channel(int sock, | |
3514 | struct lttng_consumer_channel *channel) | |
3515 | { | |
3516 | struct lttcomm_consumer_status_channel msg; | |
3517 | ||
3518 | assert(sock >= 0); | |
3519 | ||
3520 | if (!channel) { | |
3521 | msg.ret_code = -LTTNG_ERR_UST_CHAN_FAIL; | |
3522 | } else { | |
3523 | msg.ret_code = LTTNG_OK; | |
3524 | msg.key = channel->key; | |
3525 | msg.stream_count = channel->streams.count; | |
3526 | } | |
3527 | ||
3528 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3529 | } |