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> |
3bd1e081 | 31 | |
990570ed | 32 | #include <common/common.h> |
fb3a43a9 DG |
33 | #include <common/utils.h> |
34 | #include <common/compat/poll.h> | |
10a8a223 | 35 | #include <common/kernel-ctl/kernel-ctl.h> |
00e2e675 | 36 | #include <common/sessiond-comm/relayd.h> |
10a8a223 DG |
37 | #include <common/sessiond-comm/sessiond-comm.h> |
38 | #include <common/kernel-consumer/kernel-consumer.h> | |
00e2e675 | 39 | #include <common/relayd/relayd.h> |
10a8a223 DG |
40 | #include <common/ust-consumer/ust-consumer.h> |
41 | ||
42 | #include "consumer.h" | |
3bd1e081 MD |
43 | |
44 | struct lttng_consumer_global_data consumer_data = { | |
3bd1e081 MD |
45 | .stream_count = 0, |
46 | .need_update = 1, | |
47 | .type = LTTNG_CONSUMER_UNKNOWN, | |
48 | }; | |
49 | ||
50 | /* timeout parameter, to control the polling thread grace period. */ | |
51 | int consumer_poll_timeout = -1; | |
52 | ||
53 | /* | |
54 | * Flag to inform the polling thread to quit when all fd hung up. Updated by | |
55 | * the consumer_thread_receive_fds when it notices that all fds has hung up. | |
56 | * Also updated by the signal handler (consumer_should_exit()). Read by the | |
57 | * polling threads. | |
58 | */ | |
59 | volatile int consumer_quit = 0; | |
60 | ||
43c34bc3 DG |
61 | /* |
62 | * The following two hash tables are visible by all threads which are separated | |
63 | * in different source files. | |
64 | * | |
65 | * Global hash table containing respectively metadata and data streams. The | |
66 | * stream element in this ht should only be updated by the metadata poll thread | |
67 | * for the metadata and the data poll thread for the data. | |
68 | */ | |
69 | struct lttng_ht *metadata_ht = NULL; | |
70 | struct lttng_ht *data_ht = NULL; | |
71 | ||
3bd1e081 MD |
72 | /* |
73 | * Find a stream. The consumer_data.lock must be locked during this | |
74 | * call. | |
75 | */ | |
8389e4f8 DG |
76 | static struct lttng_consumer_stream *consumer_find_stream(int key, |
77 | struct lttng_ht *ht) | |
3bd1e081 | 78 | { |
e4421fec DG |
79 | struct lttng_ht_iter iter; |
80 | struct lttng_ht_node_ulong *node; | |
81 | struct lttng_consumer_stream *stream = NULL; | |
3bd1e081 | 82 | |
8389e4f8 DG |
83 | assert(ht); |
84 | ||
7ad0a0cb | 85 | /* Negative keys are lookup failures */ |
7a57cf92 | 86 | if (key < 0) { |
7ad0a0cb | 87 | return NULL; |
7a57cf92 | 88 | } |
e4421fec | 89 | |
6065ceec DG |
90 | rcu_read_lock(); |
91 | ||
8389e4f8 | 92 | lttng_ht_lookup(ht, (void *)((unsigned long) key), &iter); |
e4421fec DG |
93 | node = lttng_ht_iter_get_node_ulong(&iter); |
94 | if (node != NULL) { | |
95 | stream = caa_container_of(node, struct lttng_consumer_stream, node); | |
3bd1e081 | 96 | } |
e4421fec | 97 | |
6065ceec DG |
98 | rcu_read_unlock(); |
99 | ||
e4421fec | 100 | return stream; |
3bd1e081 MD |
101 | } |
102 | ||
c869f647 | 103 | void consumer_steal_stream_key(int key, struct lttng_ht *ht) |
7ad0a0cb MD |
104 | { |
105 | struct lttng_consumer_stream *stream; | |
106 | ||
04253271 | 107 | rcu_read_lock(); |
8389e4f8 | 108 | stream = consumer_find_stream(key, ht); |
04253271 | 109 | if (stream) { |
7ad0a0cb | 110 | stream->key = -1; |
04253271 MD |
111 | /* |
112 | * We don't want the lookup to match, but we still need | |
113 | * to iterate on this stream when iterating over the hash table. Just | |
114 | * change the node key. | |
115 | */ | |
116 | stream->node.key = -1; | |
117 | } | |
118 | rcu_read_unlock(); | |
7ad0a0cb MD |
119 | } |
120 | ||
3bd1e081 MD |
121 | static struct lttng_consumer_channel *consumer_find_channel(int key) |
122 | { | |
e4421fec DG |
123 | struct lttng_ht_iter iter; |
124 | struct lttng_ht_node_ulong *node; | |
125 | struct lttng_consumer_channel *channel = NULL; | |
3bd1e081 | 126 | |
7ad0a0cb | 127 | /* Negative keys are lookup failures */ |
7a57cf92 | 128 | if (key < 0) { |
7ad0a0cb | 129 | return NULL; |
7a57cf92 | 130 | } |
e4421fec | 131 | |
6065ceec DG |
132 | rcu_read_lock(); |
133 | ||
e4421fec DG |
134 | lttng_ht_lookup(consumer_data.channel_ht, (void *)((unsigned long) key), |
135 | &iter); | |
136 | node = lttng_ht_iter_get_node_ulong(&iter); | |
137 | if (node != NULL) { | |
138 | channel = caa_container_of(node, struct lttng_consumer_channel, node); | |
3bd1e081 | 139 | } |
e4421fec | 140 | |
6065ceec DG |
141 | rcu_read_unlock(); |
142 | ||
e4421fec | 143 | return channel; |
3bd1e081 MD |
144 | } |
145 | ||
7ad0a0cb MD |
146 | static void consumer_steal_channel_key(int key) |
147 | { | |
148 | struct lttng_consumer_channel *channel; | |
149 | ||
04253271 | 150 | rcu_read_lock(); |
7ad0a0cb | 151 | channel = consumer_find_channel(key); |
04253271 | 152 | if (channel) { |
7ad0a0cb | 153 | channel->key = -1; |
04253271 MD |
154 | /* |
155 | * We don't want the lookup to match, but we still need | |
156 | * to iterate on this channel when iterating over the hash table. Just | |
157 | * change the node key. | |
158 | */ | |
159 | channel->node.key = -1; | |
160 | } | |
161 | rcu_read_unlock(); | |
7ad0a0cb MD |
162 | } |
163 | ||
702b1ea4 MD |
164 | static |
165 | void consumer_free_stream(struct rcu_head *head) | |
166 | { | |
167 | struct lttng_ht_node_ulong *node = | |
168 | caa_container_of(head, struct lttng_ht_node_ulong, head); | |
169 | struct lttng_consumer_stream *stream = | |
170 | caa_container_of(node, struct lttng_consumer_stream, node); | |
171 | ||
172 | free(stream); | |
173 | } | |
174 | ||
00e2e675 DG |
175 | /* |
176 | * RCU protected relayd socket pair free. | |
177 | */ | |
178 | static void consumer_rcu_free_relayd(struct rcu_head *head) | |
179 | { | |
180 | struct lttng_ht_node_ulong *node = | |
181 | caa_container_of(head, struct lttng_ht_node_ulong, head); | |
182 | struct consumer_relayd_sock_pair *relayd = | |
183 | caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
184 | ||
185 | free(relayd); | |
186 | } | |
187 | ||
188 | /* | |
189 | * Destroy and free relayd socket pair object. | |
190 | * | |
191 | * This function MUST be called with the consumer_data lock acquired. | |
192 | */ | |
d09e1200 | 193 | static void destroy_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
194 | { |
195 | int ret; | |
196 | struct lttng_ht_iter iter; | |
197 | ||
173af62f DG |
198 | if (relayd == NULL) { |
199 | return; | |
200 | } | |
201 | ||
00e2e675 DG |
202 | DBG("Consumer destroy and close relayd socket pair"); |
203 | ||
204 | iter.iter.node = &relayd->node.node; | |
205 | ret = lttng_ht_del(consumer_data.relayd_ht, &iter); | |
173af62f DG |
206 | if (ret != 0) { |
207 | /* We assume the relayd was already destroyed */ | |
208 | return; | |
209 | } | |
00e2e675 DG |
210 | |
211 | /* Close all sockets */ | |
212 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
213 | (void) relayd_close(&relayd->control_sock); | |
214 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
215 | (void) relayd_close(&relayd->data_sock); | |
216 | ||
217 | /* RCU free() call */ | |
218 | call_rcu(&relayd->node.head, consumer_rcu_free_relayd); | |
219 | } | |
220 | ||
a6ba4fe1 DG |
221 | /* |
222 | * Flag a relayd socket pair for destruction. Destroy it if the refcount | |
223 | * reaches zero. | |
224 | * | |
225 | * RCU read side lock MUST be aquired before calling this function. | |
226 | */ | |
227 | void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) | |
228 | { | |
229 | assert(relayd); | |
230 | ||
231 | /* Set destroy flag for this object */ | |
232 | uatomic_set(&relayd->destroy_flag, 1); | |
233 | ||
234 | /* Destroy the relayd if refcount is 0 */ | |
235 | if (uatomic_read(&relayd->refcount) == 0) { | |
d09e1200 | 236 | destroy_relayd(relayd); |
a6ba4fe1 DG |
237 | } |
238 | } | |
239 | ||
3bd1e081 MD |
240 | /* |
241 | * Remove a stream from the global list protected by a mutex. This | |
242 | * function is also responsible for freeing its data structures. | |
243 | */ | |
e316aad5 DG |
244 | void consumer_del_stream(struct lttng_consumer_stream *stream, |
245 | struct lttng_ht *ht) | |
3bd1e081 MD |
246 | { |
247 | int ret; | |
e4421fec | 248 | struct lttng_ht_iter iter; |
3bd1e081 | 249 | struct lttng_consumer_channel *free_chan = NULL; |
00e2e675 DG |
250 | struct consumer_relayd_sock_pair *relayd; |
251 | ||
252 | assert(stream); | |
3bd1e081 | 253 | |
e316aad5 DG |
254 | if (ht == NULL) { |
255 | /* Means the stream was allocated but not successfully added */ | |
256 | goto free_stream; | |
257 | } | |
258 | ||
3bd1e081 MD |
259 | pthread_mutex_lock(&consumer_data.lock); |
260 | ||
261 | switch (consumer_data.type) { | |
262 | case LTTNG_CONSUMER_KERNEL: | |
263 | if (stream->mmap_base != NULL) { | |
264 | ret = munmap(stream->mmap_base, stream->mmap_len); | |
265 | if (ret != 0) { | |
7a57cf92 | 266 | PERROR("munmap"); |
3bd1e081 MD |
267 | } |
268 | } | |
269 | break; | |
7753dea8 MD |
270 | case LTTNG_CONSUMER32_UST: |
271 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
272 | lttng_ustconsumer_del_stream(stream); |
273 | break; | |
274 | default: | |
275 | ERR("Unknown consumer_data type"); | |
276 | assert(0); | |
277 | goto end; | |
278 | } | |
279 | ||
6065ceec | 280 | rcu_read_lock(); |
04253271 | 281 | iter.iter.node = &stream->node.node; |
e316aad5 | 282 | ret = lttng_ht_del(ht, &iter); |
04253271 | 283 | assert(!ret); |
6065ceec DG |
284 | rcu_read_unlock(); |
285 | ||
50f8ae69 | 286 | assert(consumer_data.stream_count > 0); |
3bd1e081 | 287 | consumer_data.stream_count--; |
50f8ae69 | 288 | |
3bd1e081 | 289 | if (stream->out_fd >= 0) { |
4c462e79 MD |
290 | ret = close(stream->out_fd); |
291 | if (ret) { | |
292 | PERROR("close"); | |
293 | } | |
3bd1e081 | 294 | } |
b5c5fc29 | 295 | if (stream->wait_fd >= 0 && !stream->wait_fd_is_copy) { |
4c462e79 MD |
296 | ret = close(stream->wait_fd); |
297 | if (ret) { | |
298 | PERROR("close"); | |
299 | } | |
3bd1e081 | 300 | } |
2c1dd183 | 301 | if (stream->shm_fd >= 0 && stream->wait_fd != stream->shm_fd) { |
4c462e79 MD |
302 | ret = close(stream->shm_fd); |
303 | if (ret) { | |
304 | PERROR("close"); | |
305 | } | |
3bd1e081 | 306 | } |
00e2e675 DG |
307 | |
308 | /* Check and cleanup relayd */ | |
b0b335c8 | 309 | rcu_read_lock(); |
00e2e675 DG |
310 | relayd = consumer_find_relayd(stream->net_seq_idx); |
311 | if (relayd != NULL) { | |
b0b335c8 MD |
312 | uatomic_dec(&relayd->refcount); |
313 | assert(uatomic_read(&relayd->refcount) >= 0); | |
173af62f | 314 | |
3f8e211f DG |
315 | /* Closing streams requires to lock the control socket. */ |
316 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
173af62f DG |
317 | ret = relayd_send_close_stream(&relayd->control_sock, |
318 | stream->relayd_stream_id, | |
319 | stream->next_net_seq_num - 1); | |
3f8e211f | 320 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
173af62f | 321 | if (ret < 0) { |
a4b92340 DG |
322 | DBG("Unable to close stream on the relayd. Continuing"); |
323 | /* | |
324 | * Continue here. There is nothing we can do for the relayd. | |
325 | * Chances are that the relayd has closed the socket so we just | |
326 | * continue cleaning up. | |
327 | */ | |
173af62f DG |
328 | } |
329 | ||
330 | /* Both conditions are met, we destroy the relayd. */ | |
331 | if (uatomic_read(&relayd->refcount) == 0 && | |
332 | uatomic_read(&relayd->destroy_flag)) { | |
d09e1200 | 333 | destroy_relayd(relayd); |
00e2e675 | 334 | } |
00e2e675 | 335 | } |
b0b335c8 | 336 | rcu_read_unlock(); |
00e2e675 | 337 | |
c30aaa51 MD |
338 | uatomic_dec(&stream->chan->refcount); |
339 | if (!uatomic_read(&stream->chan->refcount) | |
340 | && !uatomic_read(&stream->chan->nb_init_streams)) { | |
3bd1e081 | 341 | free_chan = stream->chan; |
00e2e675 DG |
342 | } |
343 | ||
3bd1e081 MD |
344 | end: |
345 | consumer_data.need_update = 1; | |
346 | pthread_mutex_unlock(&consumer_data.lock); | |
347 | ||
c30aaa51 | 348 | if (free_chan) { |
3bd1e081 | 349 | consumer_del_channel(free_chan); |
c30aaa51 | 350 | } |
e316aad5 DG |
351 | |
352 | free_stream: | |
353 | call_rcu(&stream->node.head, consumer_free_stream); | |
3bd1e081 MD |
354 | } |
355 | ||
356 | struct lttng_consumer_stream *consumer_allocate_stream( | |
357 | int channel_key, int stream_key, | |
358 | int shm_fd, int wait_fd, | |
359 | enum lttng_consumer_stream_state state, | |
360 | uint64_t mmap_len, | |
361 | enum lttng_event_output output, | |
6df2e2c9 MD |
362 | const char *path_name, |
363 | uid_t uid, | |
00e2e675 DG |
364 | gid_t gid, |
365 | int net_index, | |
c80048c6 MD |
366 | int metadata_flag, |
367 | int *alloc_ret) | |
3bd1e081 MD |
368 | { |
369 | struct lttng_consumer_stream *stream; | |
3bd1e081 | 370 | |
effcf122 | 371 | stream = zmalloc(sizeof(*stream)); |
3bd1e081 | 372 | if (stream == NULL) { |
7a57cf92 | 373 | PERROR("malloc struct lttng_consumer_stream"); |
c80048c6 | 374 | *alloc_ret = -ENOMEM; |
7a57cf92 | 375 | goto end; |
3bd1e081 | 376 | } |
7a57cf92 DG |
377 | |
378 | /* | |
379 | * Get stream's channel reference. Needed when adding the stream to the | |
380 | * global hash table. | |
381 | */ | |
3bd1e081 MD |
382 | stream->chan = consumer_find_channel(channel_key); |
383 | if (!stream->chan) { | |
c80048c6 | 384 | *alloc_ret = -ENOENT; |
7a57cf92 | 385 | ERR("Unable to find channel for stream %d", stream_key); |
c80048c6 | 386 | goto error; |
3bd1e081 | 387 | } |
e316aad5 | 388 | |
3bd1e081 MD |
389 | stream->key = stream_key; |
390 | stream->shm_fd = shm_fd; | |
391 | stream->wait_fd = wait_fd; | |
392 | stream->out_fd = -1; | |
393 | stream->out_fd_offset = 0; | |
394 | stream->state = state; | |
395 | stream->mmap_len = mmap_len; | |
396 | stream->mmap_base = NULL; | |
397 | stream->output = output; | |
6df2e2c9 MD |
398 | stream->uid = uid; |
399 | stream->gid = gid; | |
00e2e675 DG |
400 | stream->net_seq_idx = net_index; |
401 | stream->metadata_flag = metadata_flag; | |
402 | strncpy(stream->path_name, path_name, sizeof(stream->path_name)); | |
403 | stream->path_name[sizeof(stream->path_name) - 1] = '\0'; | |
58b1f425 DG |
404 | |
405 | /* | |
406 | * Index differently the metadata node because the thread is using an | |
407 | * internal hash table to match streams in the metadata_ht to the epoll set | |
408 | * file descriptor. | |
409 | */ | |
410 | if (metadata_flag) { | |
411 | lttng_ht_node_init_ulong(&stream->node, stream->wait_fd); | |
412 | } else { | |
413 | lttng_ht_node_init_ulong(&stream->node, stream->key); | |
414 | } | |
c30aaa51 | 415 | |
c869f647 DG |
416 | /* |
417 | * The cpu number is needed before using any ustctl_* actions. Ignored for | |
418 | * the kernel so the value does not matter. | |
419 | */ | |
420 | pthread_mutex_lock(&consumer_data.lock); | |
421 | stream->cpu = stream->chan->cpucount++; | |
422 | pthread_mutex_unlock(&consumer_data.lock); | |
423 | ||
c30aaa51 MD |
424 | DBG3("Allocated stream %s (key %d, shm_fd %d, wait_fd %d, mmap_len %llu," |
425 | " out_fd %d, net_seq_idx %d)", stream->path_name, stream->key, | |
426 | stream->shm_fd, stream->wait_fd, | |
427 | (unsigned long long) stream->mmap_len, stream->out_fd, | |
00e2e675 | 428 | stream->net_seq_idx); |
3bd1e081 | 429 | return stream; |
c80048c6 MD |
430 | |
431 | error: | |
432 | free(stream); | |
7a57cf92 | 433 | end: |
c80048c6 | 434 | return NULL; |
3bd1e081 MD |
435 | } |
436 | ||
437 | /* | |
438 | * Add a stream to the global list protected by a mutex. | |
439 | */ | |
43c34bc3 DG |
440 | static int consumer_add_stream(struct lttng_consumer_stream *stream, |
441 | struct lttng_ht *ht) | |
3bd1e081 MD |
442 | { |
443 | int ret = 0; | |
00e2e675 | 444 | struct consumer_relayd_sock_pair *relayd; |
3bd1e081 | 445 | |
e316aad5 | 446 | assert(stream); |
43c34bc3 | 447 | assert(ht); |
c77fc10a | 448 | |
e316aad5 DG |
449 | DBG3("Adding consumer stream %d", stream->key); |
450 | ||
451 | pthread_mutex_lock(&consumer_data.lock); | |
b0b335c8 | 452 | rcu_read_lock(); |
e316aad5 | 453 | |
43c34bc3 DG |
454 | /* Steal stream identifier to avoid having streams with the same key */ |
455 | consumer_steal_stream_key(stream->key, ht); | |
456 | ||
457 | lttng_ht_add_unique_ulong(ht, &stream->node); | |
00e2e675 DG |
458 | |
459 | /* Check and cleanup relayd */ | |
460 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
461 | if (relayd != NULL) { | |
b0b335c8 | 462 | uatomic_inc(&relayd->refcount); |
00e2e675 DG |
463 | } |
464 | ||
e316aad5 DG |
465 | /* Update channel refcount once added without error(s). */ |
466 | uatomic_inc(&stream->chan->refcount); | |
467 | ||
468 | /* | |
469 | * When nb_init_streams reaches 0, we don't need to trigger any action in | |
470 | * terms of destroying the associated channel, because the action that | |
471 | * causes the count to become 0 also causes a stream to be added. The | |
472 | * channel deletion will thus be triggered by the following removal of this | |
473 | * stream. | |
474 | */ | |
475 | if (uatomic_read(&stream->chan->nb_init_streams) > 0) { | |
476 | uatomic_dec(&stream->chan->nb_init_streams); | |
477 | } | |
478 | ||
479 | /* Update consumer data once the node is inserted. */ | |
3bd1e081 MD |
480 | consumer_data.stream_count++; |
481 | consumer_data.need_update = 1; | |
482 | ||
e316aad5 | 483 | rcu_read_unlock(); |
3bd1e081 | 484 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 485 | |
3bd1e081 MD |
486 | return ret; |
487 | } | |
488 | ||
00e2e675 | 489 | /* |
3f8e211f DG |
490 | * Add relayd socket to global consumer data hashtable. RCU read side lock MUST |
491 | * be acquired before calling this. | |
00e2e675 | 492 | */ |
d09e1200 | 493 | static int add_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
494 | { |
495 | int ret = 0; | |
496 | struct lttng_ht_node_ulong *node; | |
497 | struct lttng_ht_iter iter; | |
498 | ||
499 | if (relayd == NULL) { | |
500 | ret = -1; | |
501 | goto end; | |
502 | } | |
503 | ||
00e2e675 DG |
504 | lttng_ht_lookup(consumer_data.relayd_ht, |
505 | (void *)((unsigned long) relayd->net_seq_idx), &iter); | |
506 | node = lttng_ht_iter_get_node_ulong(&iter); | |
507 | if (node != NULL) { | |
00e2e675 DG |
508 | /* Relayd already exist. Ignore the insertion */ |
509 | goto end; | |
510 | } | |
511 | lttng_ht_add_unique_ulong(consumer_data.relayd_ht, &relayd->node); | |
512 | ||
00e2e675 DG |
513 | end: |
514 | return ret; | |
515 | } | |
516 | ||
517 | /* | |
518 | * Allocate and return a consumer relayd socket. | |
519 | */ | |
520 | struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( | |
521 | int net_seq_idx) | |
522 | { | |
523 | struct consumer_relayd_sock_pair *obj = NULL; | |
524 | ||
525 | /* Negative net sequence index is a failure */ | |
526 | if (net_seq_idx < 0) { | |
527 | goto error; | |
528 | } | |
529 | ||
530 | obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); | |
531 | if (obj == NULL) { | |
532 | PERROR("zmalloc relayd sock"); | |
533 | goto error; | |
534 | } | |
535 | ||
536 | obj->net_seq_idx = net_seq_idx; | |
537 | obj->refcount = 0; | |
173af62f | 538 | obj->destroy_flag = 0; |
00e2e675 DG |
539 | lttng_ht_node_init_ulong(&obj->node, obj->net_seq_idx); |
540 | pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); | |
541 | ||
542 | error: | |
543 | return obj; | |
544 | } | |
545 | ||
546 | /* | |
547 | * Find a relayd socket pair in the global consumer data. | |
548 | * | |
549 | * Return the object if found else NULL. | |
b0b335c8 MD |
550 | * RCU read-side lock must be held across this call and while using the |
551 | * returned object. | |
00e2e675 DG |
552 | */ |
553 | struct consumer_relayd_sock_pair *consumer_find_relayd(int key) | |
554 | { | |
555 | struct lttng_ht_iter iter; | |
556 | struct lttng_ht_node_ulong *node; | |
557 | struct consumer_relayd_sock_pair *relayd = NULL; | |
558 | ||
559 | /* Negative keys are lookup failures */ | |
560 | if (key < 0) { | |
561 | goto error; | |
562 | } | |
563 | ||
00e2e675 DG |
564 | lttng_ht_lookup(consumer_data.relayd_ht, (void *)((unsigned long) key), |
565 | &iter); | |
566 | node = lttng_ht_iter_get_node_ulong(&iter); | |
567 | if (node != NULL) { | |
568 | relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
569 | } | |
570 | ||
00e2e675 DG |
571 | error: |
572 | return relayd; | |
573 | } | |
574 | ||
575 | /* | |
576 | * Handle stream for relayd transmission if the stream applies for network | |
577 | * streaming where the net sequence index is set. | |
578 | * | |
579 | * Return destination file descriptor or negative value on error. | |
580 | */ | |
6197aea7 | 581 | static int write_relayd_stream_header(struct lttng_consumer_stream *stream, |
1d4dfdef DG |
582 | size_t data_size, unsigned long padding, |
583 | struct consumer_relayd_sock_pair *relayd) | |
00e2e675 DG |
584 | { |
585 | int outfd = -1, ret; | |
00e2e675 DG |
586 | struct lttcomm_relayd_data_hdr data_hdr; |
587 | ||
588 | /* Safety net */ | |
589 | assert(stream); | |
6197aea7 | 590 | assert(relayd); |
00e2e675 DG |
591 | |
592 | /* Reset data header */ | |
593 | memset(&data_hdr, 0, sizeof(data_hdr)); | |
594 | ||
00e2e675 DG |
595 | if (stream->metadata_flag) { |
596 | /* Caller MUST acquire the relayd control socket lock */ | |
597 | ret = relayd_send_metadata(&relayd->control_sock, data_size); | |
598 | if (ret < 0) { | |
599 | goto error; | |
600 | } | |
601 | ||
602 | /* Metadata are always sent on the control socket. */ | |
603 | outfd = relayd->control_sock.fd; | |
604 | } else { | |
605 | /* Set header with stream information */ | |
606 | data_hdr.stream_id = htobe64(stream->relayd_stream_id); | |
607 | data_hdr.data_size = htobe32(data_size); | |
1d4dfdef | 608 | data_hdr.padding_size = htobe32(padding); |
173af62f | 609 | data_hdr.net_seq_num = htobe64(stream->next_net_seq_num++); |
00e2e675 DG |
610 | /* Other fields are zeroed previously */ |
611 | ||
612 | ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, | |
613 | sizeof(data_hdr)); | |
614 | if (ret < 0) { | |
615 | goto error; | |
616 | } | |
617 | ||
618 | /* Set to go on data socket */ | |
619 | outfd = relayd->data_sock.fd; | |
620 | } | |
621 | ||
622 | error: | |
623 | return outfd; | |
624 | } | |
625 | ||
702b1ea4 MD |
626 | static |
627 | void consumer_free_channel(struct rcu_head *head) | |
628 | { | |
629 | struct lttng_ht_node_ulong *node = | |
630 | caa_container_of(head, struct lttng_ht_node_ulong, head); | |
631 | struct lttng_consumer_channel *channel = | |
632 | caa_container_of(node, struct lttng_consumer_channel, node); | |
633 | ||
634 | free(channel); | |
635 | } | |
636 | ||
3bd1e081 MD |
637 | /* |
638 | * Remove a channel from the global list protected by a mutex. This | |
639 | * function is also responsible for freeing its data structures. | |
640 | */ | |
641 | void consumer_del_channel(struct lttng_consumer_channel *channel) | |
642 | { | |
643 | int ret; | |
e4421fec | 644 | struct lttng_ht_iter iter; |
3bd1e081 MD |
645 | |
646 | pthread_mutex_lock(&consumer_data.lock); | |
647 | ||
648 | switch (consumer_data.type) { | |
649 | case LTTNG_CONSUMER_KERNEL: | |
650 | break; | |
7753dea8 MD |
651 | case LTTNG_CONSUMER32_UST: |
652 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
653 | lttng_ustconsumer_del_channel(channel); |
654 | break; | |
655 | default: | |
656 | ERR("Unknown consumer_data type"); | |
657 | assert(0); | |
658 | goto end; | |
659 | } | |
660 | ||
6065ceec | 661 | rcu_read_lock(); |
04253271 MD |
662 | iter.iter.node = &channel->node.node; |
663 | ret = lttng_ht_del(consumer_data.channel_ht, &iter); | |
664 | assert(!ret); | |
6065ceec DG |
665 | rcu_read_unlock(); |
666 | ||
3bd1e081 MD |
667 | if (channel->mmap_base != NULL) { |
668 | ret = munmap(channel->mmap_base, channel->mmap_len); | |
669 | if (ret != 0) { | |
7a57cf92 | 670 | PERROR("munmap"); |
3bd1e081 MD |
671 | } |
672 | } | |
b5c5fc29 | 673 | if (channel->wait_fd >= 0 && !channel->wait_fd_is_copy) { |
4c462e79 MD |
674 | ret = close(channel->wait_fd); |
675 | if (ret) { | |
676 | PERROR("close"); | |
677 | } | |
3bd1e081 | 678 | } |
2c1dd183 | 679 | if (channel->shm_fd >= 0 && channel->wait_fd != channel->shm_fd) { |
4c462e79 MD |
680 | ret = close(channel->shm_fd); |
681 | if (ret) { | |
682 | PERROR("close"); | |
683 | } | |
3bd1e081 | 684 | } |
702b1ea4 MD |
685 | |
686 | call_rcu(&channel->node.head, consumer_free_channel); | |
3bd1e081 MD |
687 | end: |
688 | pthread_mutex_unlock(&consumer_data.lock); | |
689 | } | |
690 | ||
691 | struct lttng_consumer_channel *consumer_allocate_channel( | |
692 | int channel_key, | |
693 | int shm_fd, int wait_fd, | |
694 | uint64_t mmap_len, | |
c30aaa51 MD |
695 | uint64_t max_sb_size, |
696 | unsigned int nb_init_streams) | |
3bd1e081 MD |
697 | { |
698 | struct lttng_consumer_channel *channel; | |
699 | int ret; | |
700 | ||
276b26d1 | 701 | channel = zmalloc(sizeof(*channel)); |
3bd1e081 | 702 | if (channel == NULL) { |
7a57cf92 | 703 | PERROR("malloc struct lttng_consumer_channel"); |
3bd1e081 MD |
704 | goto end; |
705 | } | |
706 | channel->key = channel_key; | |
707 | channel->shm_fd = shm_fd; | |
708 | channel->wait_fd = wait_fd; | |
709 | channel->mmap_len = mmap_len; | |
710 | channel->max_sb_size = max_sb_size; | |
711 | channel->refcount = 0; | |
c30aaa51 | 712 | channel->nb_init_streams = nb_init_streams; |
e4421fec | 713 | lttng_ht_node_init_ulong(&channel->node, channel->key); |
3bd1e081 MD |
714 | |
715 | switch (consumer_data.type) { | |
716 | case LTTNG_CONSUMER_KERNEL: | |
717 | channel->mmap_base = NULL; | |
718 | channel->mmap_len = 0; | |
719 | break; | |
7753dea8 MD |
720 | case LTTNG_CONSUMER32_UST: |
721 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
722 | ret = lttng_ustconsumer_allocate_channel(channel); |
723 | if (ret) { | |
724 | free(channel); | |
725 | return NULL; | |
726 | } | |
727 | break; | |
728 | default: | |
729 | ERR("Unknown consumer_data type"); | |
730 | assert(0); | |
731 | goto end; | |
732 | } | |
733 | DBG("Allocated channel (key %d, shm_fd %d, wait_fd %d, mmap_len %llu, max_sb_size %llu)", | |
00e2e675 | 734 | channel->key, channel->shm_fd, channel->wait_fd, |
3bd1e081 MD |
735 | (unsigned long long) channel->mmap_len, |
736 | (unsigned long long) channel->max_sb_size); | |
737 | end: | |
738 | return channel; | |
739 | } | |
740 | ||
741 | /* | |
742 | * Add a channel to the global list protected by a mutex. | |
743 | */ | |
744 | int consumer_add_channel(struct lttng_consumer_channel *channel) | |
745 | { | |
c77fc10a DG |
746 | struct lttng_ht_node_ulong *node; |
747 | struct lttng_ht_iter iter; | |
748 | ||
3bd1e081 | 749 | pthread_mutex_lock(&consumer_data.lock); |
7ad0a0cb MD |
750 | /* Steal channel identifier, for UST */ |
751 | consumer_steal_channel_key(channel->key); | |
6065ceec | 752 | rcu_read_lock(); |
c77fc10a DG |
753 | |
754 | lttng_ht_lookup(consumer_data.channel_ht, | |
755 | (void *)((unsigned long) channel->key), &iter); | |
756 | node = lttng_ht_iter_get_node_ulong(&iter); | |
757 | if (node != NULL) { | |
758 | /* Channel already exist. Ignore the insertion */ | |
759 | goto end; | |
760 | } | |
761 | ||
04253271 | 762 | lttng_ht_add_unique_ulong(consumer_data.channel_ht, &channel->node); |
c77fc10a DG |
763 | |
764 | end: | |
6065ceec | 765 | rcu_read_unlock(); |
3bd1e081 | 766 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 767 | |
7ad0a0cb | 768 | return 0; |
3bd1e081 MD |
769 | } |
770 | ||
771 | /* | |
772 | * Allocate the pollfd structure and the local view of the out fds to avoid | |
773 | * doing a lookup in the linked list and concurrency issues when writing is | |
774 | * needed. Called with consumer_data.lock held. | |
775 | * | |
776 | * Returns the number of fds in the structures. | |
777 | */ | |
43c34bc3 | 778 | static int consumer_update_poll_array( |
3bd1e081 | 779 | struct lttng_consumer_local_data *ctx, struct pollfd **pollfd, |
43c34bc3 | 780 | struct lttng_consumer_stream **local_stream, struct lttng_ht *ht) |
3bd1e081 | 781 | { |
3bd1e081 | 782 | int i = 0; |
e4421fec DG |
783 | struct lttng_ht_iter iter; |
784 | struct lttng_consumer_stream *stream; | |
3bd1e081 MD |
785 | |
786 | DBG("Updating poll fd array"); | |
481d6c57 | 787 | rcu_read_lock(); |
43c34bc3 | 788 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
e4421fec | 789 | if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM) { |
3bd1e081 MD |
790 | continue; |
791 | } | |
e4421fec DG |
792 | DBG("Active FD %d", stream->wait_fd); |
793 | (*pollfd)[i].fd = stream->wait_fd; | |
3bd1e081 | 794 | (*pollfd)[i].events = POLLIN | POLLPRI; |
e4421fec | 795 | local_stream[i] = stream; |
3bd1e081 MD |
796 | i++; |
797 | } | |
481d6c57 | 798 | rcu_read_unlock(); |
3bd1e081 MD |
799 | |
800 | /* | |
50f8ae69 | 801 | * Insert the consumer_data_pipe at the end of the array and don't |
3bd1e081 MD |
802 | * increment i so nb_fd is the number of real FD. |
803 | */ | |
50f8ae69 | 804 | (*pollfd)[i].fd = ctx->consumer_data_pipe[0]; |
509bb1cf | 805 | (*pollfd)[i].events = POLLIN | POLLPRI; |
3bd1e081 MD |
806 | return i; |
807 | } | |
808 | ||
809 | /* | |
810 | * Poll on the should_quit pipe and the command socket return -1 on error and | |
811 | * should exit, 0 if data is available on the command socket | |
812 | */ | |
813 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) | |
814 | { | |
815 | int num_rdy; | |
816 | ||
88f2b785 | 817 | restart: |
3bd1e081 MD |
818 | num_rdy = poll(consumer_sockpoll, 2, -1); |
819 | if (num_rdy == -1) { | |
88f2b785 MD |
820 | /* |
821 | * Restart interrupted system call. | |
822 | */ | |
823 | if (errno == EINTR) { | |
824 | goto restart; | |
825 | } | |
7a57cf92 | 826 | PERROR("Poll error"); |
3bd1e081 MD |
827 | goto exit; |
828 | } | |
509bb1cf | 829 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
3bd1e081 MD |
830 | DBG("consumer_should_quit wake up"); |
831 | goto exit; | |
832 | } | |
833 | return 0; | |
834 | ||
835 | exit: | |
836 | return -1; | |
837 | } | |
838 | ||
839 | /* | |
840 | * Set the error socket. | |
841 | */ | |
842 | void lttng_consumer_set_error_sock( | |
843 | struct lttng_consumer_local_data *ctx, int sock) | |
844 | { | |
845 | ctx->consumer_error_socket = sock; | |
846 | } | |
847 | ||
848 | /* | |
849 | * Set the command socket path. | |
850 | */ | |
3bd1e081 MD |
851 | void lttng_consumer_set_command_sock_path( |
852 | struct lttng_consumer_local_data *ctx, char *sock) | |
853 | { | |
854 | ctx->consumer_command_sock_path = sock; | |
855 | } | |
856 | ||
857 | /* | |
858 | * Send return code to the session daemon. | |
859 | * If the socket is not defined, we return 0, it is not a fatal error | |
860 | */ | |
861 | int lttng_consumer_send_error( | |
862 | struct lttng_consumer_local_data *ctx, int cmd) | |
863 | { | |
864 | if (ctx->consumer_error_socket > 0) { | |
865 | return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd, | |
866 | sizeof(enum lttcomm_sessiond_command)); | |
867 | } | |
868 | ||
869 | return 0; | |
870 | } | |
871 | ||
872 | /* | |
873 | * Close all the tracefiles and stream fds, should be called when all instances | |
874 | * are destroyed. | |
875 | */ | |
876 | void lttng_consumer_cleanup(void) | |
877 | { | |
e4421fec | 878 | struct lttng_ht_iter iter; |
6065ceec DG |
879 | struct lttng_ht_node_ulong *node; |
880 | ||
881 | rcu_read_lock(); | |
3bd1e081 | 882 | |
6065ceec DG |
883 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, node, |
884 | node) { | |
702b1ea4 MD |
885 | struct lttng_consumer_channel *channel = |
886 | caa_container_of(node, struct lttng_consumer_channel, node); | |
887 | consumer_del_channel(channel); | |
3bd1e081 | 888 | } |
6065ceec DG |
889 | |
890 | rcu_read_unlock(); | |
d6ce1df2 | 891 | |
d6ce1df2 | 892 | lttng_ht_destroy(consumer_data.channel_ht); |
3bd1e081 MD |
893 | } |
894 | ||
895 | /* | |
896 | * Called from signal handler. | |
897 | */ | |
898 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) | |
899 | { | |
900 | int ret; | |
901 | consumer_quit = 1; | |
6f94560a MD |
902 | do { |
903 | ret = write(ctx->consumer_should_quit[1], "4", 1); | |
904 | } while (ret < 0 && errno == EINTR); | |
3bd1e081 | 905 | if (ret < 0) { |
7a57cf92 | 906 | PERROR("write consumer quit"); |
3bd1e081 MD |
907 | } |
908 | } | |
909 | ||
00e2e675 DG |
910 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
911 | off_t orig_offset) | |
3bd1e081 MD |
912 | { |
913 | int outfd = stream->out_fd; | |
914 | ||
915 | /* | |
916 | * This does a blocking write-and-wait on any page that belongs to the | |
917 | * subbuffer prior to the one we just wrote. | |
918 | * Don't care about error values, as these are just hints and ways to | |
919 | * limit the amount of page cache used. | |
920 | */ | |
921 | if (orig_offset < stream->chan->max_sb_size) { | |
922 | return; | |
923 | } | |
b9182dd9 | 924 | lttng_sync_file_range(outfd, orig_offset - stream->chan->max_sb_size, |
3bd1e081 MD |
925 | stream->chan->max_sb_size, |
926 | SYNC_FILE_RANGE_WAIT_BEFORE | |
927 | | SYNC_FILE_RANGE_WRITE | |
928 | | SYNC_FILE_RANGE_WAIT_AFTER); | |
929 | /* | |
930 | * Give hints to the kernel about how we access the file: | |
931 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after | |
932 | * we write it. | |
933 | * | |
934 | * We need to call fadvise again after the file grows because the | |
935 | * kernel does not seem to apply fadvise to non-existing parts of the | |
936 | * file. | |
937 | * | |
938 | * Call fadvise _after_ having waited for the page writeback to | |
939 | * complete because the dirty page writeback semantic is not well | |
940 | * defined. So it can be expected to lead to lower throughput in | |
941 | * streaming. | |
942 | */ | |
943 | posix_fadvise(outfd, orig_offset - stream->chan->max_sb_size, | |
944 | stream->chan->max_sb_size, POSIX_FADV_DONTNEED); | |
945 | } | |
946 | ||
947 | /* | |
948 | * Initialise the necessary environnement : | |
949 | * - create a new context | |
950 | * - create the poll_pipe | |
951 | * - create the should_quit pipe (for signal handler) | |
952 | * - create the thread pipe (for splice) | |
953 | * | |
954 | * Takes a function pointer as argument, this function is called when data is | |
955 | * available on a buffer. This function is responsible to do the | |
956 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the | |
957 | * buffer configuration and then kernctl_put_next_subbuf at the end. | |
958 | * | |
959 | * Returns a pointer to the new context or NULL on error. | |
960 | */ | |
961 | struct lttng_consumer_local_data *lttng_consumer_create( | |
962 | enum lttng_consumer_type type, | |
4078b776 | 963 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
d41f73b7 | 964 | struct lttng_consumer_local_data *ctx), |
3bd1e081 MD |
965 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
966 | int (*recv_stream)(struct lttng_consumer_stream *stream), | |
967 | int (*update_stream)(int stream_key, uint32_t state)) | |
968 | { | |
969 | int ret, i; | |
970 | struct lttng_consumer_local_data *ctx; | |
971 | ||
972 | assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || | |
973 | consumer_data.type == type); | |
974 | consumer_data.type = type; | |
975 | ||
effcf122 | 976 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
3bd1e081 | 977 | if (ctx == NULL) { |
7a57cf92 | 978 | PERROR("allocating context"); |
3bd1e081 MD |
979 | goto error; |
980 | } | |
981 | ||
982 | ctx->consumer_error_socket = -1; | |
983 | /* assign the callbacks */ | |
984 | ctx->on_buffer_ready = buffer_ready; | |
985 | ctx->on_recv_channel = recv_channel; | |
986 | ctx->on_recv_stream = recv_stream; | |
987 | ctx->on_update_stream = update_stream; | |
988 | ||
50f8ae69 | 989 | ret = pipe(ctx->consumer_data_pipe); |
3bd1e081 | 990 | if (ret < 0) { |
7a57cf92 | 991 | PERROR("Error creating poll pipe"); |
3bd1e081 MD |
992 | goto error_poll_pipe; |
993 | } | |
994 | ||
04fdd819 | 995 | /* set read end of the pipe to non-blocking */ |
50f8ae69 | 996 | ret = fcntl(ctx->consumer_data_pipe[0], F_SETFL, O_NONBLOCK); |
04fdd819 | 997 | if (ret < 0) { |
7a57cf92 | 998 | PERROR("fcntl O_NONBLOCK"); |
04fdd819 MD |
999 | goto error_poll_fcntl; |
1000 | } | |
1001 | ||
1002 | /* set write end of the pipe to non-blocking */ | |
50f8ae69 | 1003 | ret = fcntl(ctx->consumer_data_pipe[1], F_SETFL, O_NONBLOCK); |
04fdd819 | 1004 | if (ret < 0) { |
7a57cf92 | 1005 | PERROR("fcntl O_NONBLOCK"); |
04fdd819 MD |
1006 | goto error_poll_fcntl; |
1007 | } | |
1008 | ||
3bd1e081 MD |
1009 | ret = pipe(ctx->consumer_should_quit); |
1010 | if (ret < 0) { | |
7a57cf92 | 1011 | PERROR("Error creating recv pipe"); |
3bd1e081 MD |
1012 | goto error_quit_pipe; |
1013 | } | |
1014 | ||
1015 | ret = pipe(ctx->consumer_thread_pipe); | |
1016 | if (ret < 0) { | |
7a57cf92 | 1017 | PERROR("Error creating thread pipe"); |
3bd1e081 MD |
1018 | goto error_thread_pipe; |
1019 | } | |
1020 | ||
fb3a43a9 DG |
1021 | ret = utils_create_pipe(ctx->consumer_metadata_pipe); |
1022 | if (ret < 0) { | |
1023 | goto error_metadata_pipe; | |
1024 | } | |
3bd1e081 | 1025 | |
fb3a43a9 DG |
1026 | ret = utils_create_pipe(ctx->consumer_splice_metadata_pipe); |
1027 | if (ret < 0) { | |
1028 | goto error_splice_pipe; | |
1029 | } | |
1030 | ||
1031 | return ctx; | |
3bd1e081 | 1032 | |
fb3a43a9 DG |
1033 | error_splice_pipe: |
1034 | utils_close_pipe(ctx->consumer_metadata_pipe); | |
1035 | error_metadata_pipe: | |
1036 | utils_close_pipe(ctx->consumer_thread_pipe); | |
3bd1e081 MD |
1037 | error_thread_pipe: |
1038 | for (i = 0; i < 2; i++) { | |
1039 | int err; | |
1040 | ||
1041 | err = close(ctx->consumer_should_quit[i]); | |
4c462e79 MD |
1042 | if (err) { |
1043 | PERROR("close"); | |
1044 | } | |
3bd1e081 | 1045 | } |
04fdd819 | 1046 | error_poll_fcntl: |
3bd1e081 MD |
1047 | error_quit_pipe: |
1048 | for (i = 0; i < 2; i++) { | |
1049 | int err; | |
1050 | ||
50f8ae69 | 1051 | err = close(ctx->consumer_data_pipe[i]); |
4c462e79 MD |
1052 | if (err) { |
1053 | PERROR("close"); | |
1054 | } | |
3bd1e081 MD |
1055 | } |
1056 | error_poll_pipe: | |
1057 | free(ctx); | |
1058 | error: | |
1059 | return NULL; | |
1060 | } | |
1061 | ||
1062 | /* | |
1063 | * Close all fds associated with the instance and free the context. | |
1064 | */ | |
1065 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) | |
1066 | { | |
4c462e79 MD |
1067 | int ret; |
1068 | ||
1069 | ret = close(ctx->consumer_error_socket); | |
1070 | if (ret) { | |
1071 | PERROR("close"); | |
1072 | } | |
1073 | ret = close(ctx->consumer_thread_pipe[0]); | |
1074 | if (ret) { | |
1075 | PERROR("close"); | |
1076 | } | |
1077 | ret = close(ctx->consumer_thread_pipe[1]); | |
1078 | if (ret) { | |
1079 | PERROR("close"); | |
1080 | } | |
50f8ae69 | 1081 | ret = close(ctx->consumer_data_pipe[0]); |
4c462e79 MD |
1082 | if (ret) { |
1083 | PERROR("close"); | |
1084 | } | |
50f8ae69 | 1085 | ret = close(ctx->consumer_data_pipe[1]); |
4c462e79 MD |
1086 | if (ret) { |
1087 | PERROR("close"); | |
1088 | } | |
1089 | ret = close(ctx->consumer_should_quit[0]); | |
1090 | if (ret) { | |
1091 | PERROR("close"); | |
1092 | } | |
1093 | ret = close(ctx->consumer_should_quit[1]); | |
1094 | if (ret) { | |
1095 | PERROR("close"); | |
1096 | } | |
fb3a43a9 DG |
1097 | utils_close_pipe(ctx->consumer_splice_metadata_pipe); |
1098 | ||
3bd1e081 MD |
1099 | unlink(ctx->consumer_command_sock_path); |
1100 | free(ctx); | |
1101 | } | |
1102 | ||
6197aea7 DG |
1103 | /* |
1104 | * Write the metadata stream id on the specified file descriptor. | |
1105 | */ | |
1106 | static int write_relayd_metadata_id(int fd, | |
1107 | struct lttng_consumer_stream *stream, | |
1d4dfdef DG |
1108 | struct consumer_relayd_sock_pair *relayd, |
1109 | unsigned long padding) | |
6197aea7 DG |
1110 | { |
1111 | int ret; | |
1d4dfdef | 1112 | struct lttcomm_relayd_metadata_payload hdr; |
6197aea7 | 1113 | |
1d4dfdef DG |
1114 | hdr.stream_id = htobe64(stream->relayd_stream_id); |
1115 | hdr.padding_size = htobe32(padding); | |
6197aea7 | 1116 | do { |
1d4dfdef | 1117 | ret = write(fd, (void *) &hdr, sizeof(hdr)); |
6197aea7 DG |
1118 | } while (ret < 0 && errno == EINTR); |
1119 | if (ret < 0) { | |
1120 | PERROR("write metadata stream id"); | |
1121 | goto end; | |
1122 | } | |
1d4dfdef DG |
1123 | DBG("Metadata stream id %" PRIu64 " with padding %lu written before data", |
1124 | stream->relayd_stream_id, padding); | |
6197aea7 DG |
1125 | |
1126 | end: | |
1127 | return ret; | |
1128 | } | |
1129 | ||
3bd1e081 | 1130 | /* |
09e26845 DG |
1131 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
1132 | * core function for writing trace buffers to either the local filesystem or | |
1133 | * the network. | |
1134 | * | |
1135 | * Careful review MUST be put if any changes occur! | |
3bd1e081 MD |
1136 | * |
1137 | * Returns the number of bytes written | |
1138 | */ | |
4078b776 | 1139 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
3bd1e081 | 1140 | struct lttng_consumer_local_data *ctx, |
1d4dfdef DG |
1141 | struct lttng_consumer_stream *stream, unsigned long len, |
1142 | unsigned long padding) | |
3bd1e081 | 1143 | { |
f02e1e8a DG |
1144 | unsigned long mmap_offset; |
1145 | ssize_t ret = 0, written = 0; | |
1146 | off_t orig_offset = stream->out_fd_offset; | |
1147 | /* Default is on the disk */ | |
1148 | int outfd = stream->out_fd; | |
f02e1e8a DG |
1149 | struct consumer_relayd_sock_pair *relayd = NULL; |
1150 | ||
1151 | /* RCU lock for the relayd pointer */ | |
1152 | rcu_read_lock(); | |
1153 | ||
1154 | /* Flag that the current stream if set for network streaming. */ | |
1155 | if (stream->net_seq_idx != -1) { | |
1156 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
1157 | if (relayd == NULL) { | |
1158 | goto end; | |
1159 | } | |
1160 | } | |
1161 | ||
1162 | /* get the offset inside the fd to mmap */ | |
3bd1e081 MD |
1163 | switch (consumer_data.type) { |
1164 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a DG |
1165 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); |
1166 | break; | |
7753dea8 MD |
1167 | case LTTNG_CONSUMER32_UST: |
1168 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a DG |
1169 | ret = lttng_ustctl_get_mmap_read_offset(stream->chan->handle, |
1170 | stream->buf, &mmap_offset); | |
1171 | break; | |
3bd1e081 MD |
1172 | default: |
1173 | ERR("Unknown consumer_data type"); | |
1174 | assert(0); | |
1175 | } | |
f02e1e8a DG |
1176 | if (ret != 0) { |
1177 | errno = -ret; | |
1178 | PERROR("tracer ctl get_mmap_read_offset"); | |
1179 | written = ret; | |
1180 | goto end; | |
1181 | } | |
b9182dd9 | 1182 | |
f02e1e8a DG |
1183 | /* Handle stream on the relayd if the output is on the network */ |
1184 | if (relayd) { | |
1185 | unsigned long netlen = len; | |
1186 | ||
1187 | /* | |
1188 | * Lock the control socket for the complete duration of the function | |
1189 | * since from this point on we will use the socket. | |
1190 | */ | |
1191 | if (stream->metadata_flag) { | |
1192 | /* Metadata requires the control socket. */ | |
1193 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1d4dfdef | 1194 | netlen += sizeof(struct lttcomm_relayd_metadata_payload); |
f02e1e8a DG |
1195 | } |
1196 | ||
1d4dfdef | 1197 | ret = write_relayd_stream_header(stream, netlen, padding, relayd); |
f02e1e8a DG |
1198 | if (ret >= 0) { |
1199 | /* Use the returned socket. */ | |
1200 | outfd = ret; | |
1201 | ||
1202 | /* Write metadata stream id before payload */ | |
1203 | if (stream->metadata_flag) { | |
1d4dfdef | 1204 | ret = write_relayd_metadata_id(outfd, stream, relayd, padding); |
f02e1e8a | 1205 | if (ret < 0) { |
f02e1e8a DG |
1206 | written = ret; |
1207 | goto end; | |
1208 | } | |
f02e1e8a DG |
1209 | } |
1210 | } | |
1211 | /* Else, use the default set before which is the filesystem. */ | |
1d4dfdef DG |
1212 | } else { |
1213 | /* No streaming, we have to set the len with the full padding */ | |
1214 | len += padding; | |
f02e1e8a DG |
1215 | } |
1216 | ||
1217 | while (len > 0) { | |
1218 | do { | |
1219 | ret = write(outfd, stream->mmap_base + mmap_offset, len); | |
1220 | } while (ret < 0 && errno == EINTR); | |
1d4dfdef | 1221 | DBG("Consumer mmap write() ret %zd (len %lu)", ret, len); |
f02e1e8a DG |
1222 | if (ret < 0) { |
1223 | PERROR("Error in file write"); | |
1224 | if (written == 0) { | |
1225 | written = ret; | |
1226 | } | |
1227 | goto end; | |
1228 | } else if (ret > len) { | |
77c7c900 | 1229 | PERROR("Error in file write (ret %zd > len %lu)", ret, len); |
f02e1e8a DG |
1230 | written += ret; |
1231 | goto end; | |
1232 | } else { | |
1233 | len -= ret; | |
1234 | mmap_offset += ret; | |
1235 | } | |
f02e1e8a DG |
1236 | |
1237 | /* This call is useless on a socket so better save a syscall. */ | |
1238 | if (!relayd) { | |
1239 | /* This won't block, but will start writeout asynchronously */ | |
1240 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret, | |
1241 | SYNC_FILE_RANGE_WRITE); | |
1242 | stream->out_fd_offset += ret; | |
1243 | } | |
1244 | written += ret; | |
1245 | } | |
1246 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1247 | ||
1248 | end: | |
1249 | /* Unlock only if ctrl socket used */ | |
1250 | if (relayd && stream->metadata_flag) { | |
1251 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1252 | } | |
1253 | ||
1254 | rcu_read_unlock(); | |
1255 | return written; | |
3bd1e081 MD |
1256 | } |
1257 | ||
1258 | /* | |
1259 | * Splice the data from the ring buffer to the tracefile. | |
1260 | * | |
1261 | * Returns the number of bytes spliced. | |
1262 | */ | |
4078b776 | 1263 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
3bd1e081 | 1264 | struct lttng_consumer_local_data *ctx, |
1d4dfdef DG |
1265 | struct lttng_consumer_stream *stream, unsigned long len, |
1266 | unsigned long padding) | |
3bd1e081 | 1267 | { |
f02e1e8a DG |
1268 | ssize_t ret = 0, written = 0, ret_splice = 0; |
1269 | loff_t offset = 0; | |
1270 | off_t orig_offset = stream->out_fd_offset; | |
1271 | int fd = stream->wait_fd; | |
1272 | /* Default is on the disk */ | |
1273 | int outfd = stream->out_fd; | |
f02e1e8a | 1274 | struct consumer_relayd_sock_pair *relayd = NULL; |
fb3a43a9 | 1275 | int *splice_pipe; |
f02e1e8a | 1276 | |
3bd1e081 MD |
1277 | switch (consumer_data.type) { |
1278 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a | 1279 | break; |
7753dea8 MD |
1280 | case LTTNG_CONSUMER32_UST: |
1281 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a | 1282 | /* Not supported for user space tracing */ |
3bd1e081 MD |
1283 | return -ENOSYS; |
1284 | default: | |
1285 | ERR("Unknown consumer_data type"); | |
1286 | assert(0); | |
3bd1e081 MD |
1287 | } |
1288 | ||
f02e1e8a DG |
1289 | /* RCU lock for the relayd pointer */ |
1290 | rcu_read_lock(); | |
1291 | ||
1292 | /* Flag that the current stream if set for network streaming. */ | |
1293 | if (stream->net_seq_idx != -1) { | |
1294 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
1295 | if (relayd == NULL) { | |
1296 | goto end; | |
1297 | } | |
1298 | } | |
1299 | ||
fb3a43a9 DG |
1300 | /* |
1301 | * Choose right pipe for splice. Metadata and trace data are handled by | |
1302 | * different threads hence the use of two pipes in order not to race or | |
1303 | * corrupt the written data. | |
1304 | */ | |
1305 | if (stream->metadata_flag) { | |
1306 | splice_pipe = ctx->consumer_splice_metadata_pipe; | |
1307 | } else { | |
1308 | splice_pipe = ctx->consumer_thread_pipe; | |
1309 | } | |
1310 | ||
f02e1e8a | 1311 | /* Write metadata stream id before payload */ |
1d4dfdef DG |
1312 | if (relayd) { |
1313 | int total_len = len; | |
f02e1e8a | 1314 | |
1d4dfdef DG |
1315 | if (stream->metadata_flag) { |
1316 | /* | |
1317 | * Lock the control socket for the complete duration of the function | |
1318 | * since from this point on we will use the socket. | |
1319 | */ | |
1320 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1321 | ||
1322 | ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd, | |
1323 | padding); | |
1324 | if (ret < 0) { | |
1325 | written = ret; | |
1326 | goto end; | |
1327 | } | |
1328 | ||
1329 | total_len += sizeof(struct lttcomm_relayd_metadata_payload); | |
1330 | } | |
1331 | ||
1332 | ret = write_relayd_stream_header(stream, total_len, padding, relayd); | |
1333 | if (ret >= 0) { | |
1334 | /* Use the returned socket. */ | |
1335 | outfd = ret; | |
1336 | } else { | |
1337 | ERR("Remote relayd disconnected. Stopping"); | |
f02e1e8a DG |
1338 | goto end; |
1339 | } | |
1d4dfdef DG |
1340 | } else { |
1341 | /* No streaming, we have to set the len with the full padding */ | |
1342 | len += padding; | |
f02e1e8a DG |
1343 | } |
1344 | ||
1345 | while (len > 0) { | |
1d4dfdef DG |
1346 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)", |
1347 | (unsigned long)offset, len, fd, splice_pipe[1]); | |
fb3a43a9 | 1348 | ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len, |
f02e1e8a DG |
1349 | SPLICE_F_MOVE | SPLICE_F_MORE); |
1350 | DBG("splice chan to pipe, ret %zd", ret_splice); | |
1351 | if (ret_splice < 0) { | |
1352 | PERROR("Error in relay splice"); | |
1353 | if (written == 0) { | |
1354 | written = ret_splice; | |
1355 | } | |
1356 | ret = errno; | |
1357 | goto splice_error; | |
1358 | } | |
1359 | ||
1360 | /* Handle stream on the relayd if the output is on the network */ | |
1361 | if (relayd) { | |
1362 | if (stream->metadata_flag) { | |
1d4dfdef DG |
1363 | size_t metadata_payload_size = |
1364 | sizeof(struct lttcomm_relayd_metadata_payload); | |
1365 | ||
f02e1e8a | 1366 | /* Update counter to fit the spliced data */ |
1d4dfdef DG |
1367 | ret_splice += metadata_payload_size; |
1368 | len += metadata_payload_size; | |
f02e1e8a DG |
1369 | /* |
1370 | * We do this so the return value can match the len passed as | |
1371 | * argument to this function. | |
1372 | */ | |
1d4dfdef | 1373 | written -= metadata_payload_size; |
f02e1e8a DG |
1374 | } |
1375 | } | |
1376 | ||
1377 | /* Splice data out */ | |
fb3a43a9 | 1378 | ret_splice = splice(splice_pipe[0], NULL, outfd, NULL, |
f02e1e8a | 1379 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); |
1d4dfdef | 1380 | DBG("Consumer splice pipe to file, ret %zd", ret_splice); |
f02e1e8a DG |
1381 | if (ret_splice < 0) { |
1382 | PERROR("Error in file splice"); | |
1383 | if (written == 0) { | |
1384 | written = ret_splice; | |
1385 | } | |
1386 | ret = errno; | |
1387 | goto splice_error; | |
1388 | } else if (ret_splice > len) { | |
1389 | errno = EINVAL; | |
1390 | PERROR("Wrote more data than requested %zd (len: %lu)", | |
1391 | ret_splice, len); | |
1392 | written += ret_splice; | |
1393 | ret = errno; | |
1394 | goto splice_error; | |
1395 | } | |
1396 | len -= ret_splice; | |
1397 | ||
1398 | /* This call is useless on a socket so better save a syscall. */ | |
1399 | if (!relayd) { | |
1400 | /* This won't block, but will start writeout asynchronously */ | |
1401 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, | |
1402 | SYNC_FILE_RANGE_WRITE); | |
1403 | stream->out_fd_offset += ret_splice; | |
1404 | } | |
1405 | written += ret_splice; | |
1406 | } | |
1407 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1408 | ||
1409 | ret = ret_splice; | |
1410 | ||
1411 | goto end; | |
1412 | ||
1413 | splice_error: | |
1414 | /* send the appropriate error description to sessiond */ | |
1415 | switch (ret) { | |
1416 | case EBADF: | |
f73fabfd | 1417 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EBADF); |
f02e1e8a DG |
1418 | break; |
1419 | case EINVAL: | |
f73fabfd | 1420 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); |
f02e1e8a DG |
1421 | break; |
1422 | case ENOMEM: | |
f73fabfd | 1423 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); |
f02e1e8a DG |
1424 | break; |
1425 | case ESPIPE: | |
f73fabfd | 1426 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); |
f02e1e8a DG |
1427 | break; |
1428 | } | |
1429 | ||
1430 | end: | |
1431 | if (relayd && stream->metadata_flag) { | |
1432 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1433 | } | |
1434 | ||
1435 | rcu_read_unlock(); | |
1436 | return written; | |
3bd1e081 MD |
1437 | } |
1438 | ||
1439 | /* | |
1440 | * Take a snapshot for a specific fd | |
1441 | * | |
1442 | * Returns 0 on success, < 0 on error | |
1443 | */ | |
1444 | int lttng_consumer_take_snapshot(struct lttng_consumer_local_data *ctx, | |
1445 | struct lttng_consumer_stream *stream) | |
1446 | { | |
1447 | switch (consumer_data.type) { | |
1448 | case LTTNG_CONSUMER_KERNEL: | |
1449 | return lttng_kconsumer_take_snapshot(ctx, stream); | |
7753dea8 MD |
1450 | case LTTNG_CONSUMER32_UST: |
1451 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1452 | return lttng_ustconsumer_take_snapshot(ctx, stream); |
1453 | default: | |
1454 | ERR("Unknown consumer_data type"); | |
1455 | assert(0); | |
1456 | return -ENOSYS; | |
1457 | } | |
1458 | ||
1459 | } | |
1460 | ||
1461 | /* | |
1462 | * Get the produced position | |
1463 | * | |
1464 | * Returns 0 on success, < 0 on error | |
1465 | */ | |
1466 | int lttng_consumer_get_produced_snapshot( | |
1467 | struct lttng_consumer_local_data *ctx, | |
1468 | struct lttng_consumer_stream *stream, | |
1469 | unsigned long *pos) | |
1470 | { | |
1471 | switch (consumer_data.type) { | |
1472 | case LTTNG_CONSUMER_KERNEL: | |
1473 | return lttng_kconsumer_get_produced_snapshot(ctx, stream, pos); | |
7753dea8 MD |
1474 | case LTTNG_CONSUMER32_UST: |
1475 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1476 | return lttng_ustconsumer_get_produced_snapshot(ctx, stream, pos); |
1477 | default: | |
1478 | ERR("Unknown consumer_data type"); | |
1479 | assert(0); | |
1480 | return -ENOSYS; | |
1481 | } | |
1482 | } | |
1483 | ||
1484 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, | |
1485 | int sock, struct pollfd *consumer_sockpoll) | |
1486 | { | |
1487 | switch (consumer_data.type) { | |
1488 | case LTTNG_CONSUMER_KERNEL: | |
1489 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
7753dea8 MD |
1490 | case LTTNG_CONSUMER32_UST: |
1491 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1492 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
1493 | default: | |
1494 | ERR("Unknown consumer_data type"); | |
1495 | assert(0); | |
1496 | return -ENOSYS; | |
1497 | } | |
1498 | } | |
1499 | ||
43c34bc3 DG |
1500 | /* |
1501 | * Iterate over all streams of the hashtable and free them properly. | |
1502 | * | |
1503 | * WARNING: *MUST* be used with data stream only. | |
1504 | */ | |
1505 | static void destroy_data_stream_ht(struct lttng_ht *ht) | |
1506 | { | |
1507 | int ret; | |
1508 | struct lttng_ht_iter iter; | |
1509 | struct lttng_consumer_stream *stream; | |
1510 | ||
1511 | if (ht == NULL) { | |
1512 | return; | |
1513 | } | |
1514 | ||
1515 | rcu_read_lock(); | |
1516 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1517 | ret = lttng_ht_del(ht, &iter); | |
1518 | assert(!ret); | |
1519 | ||
1520 | call_rcu(&stream->node.head, consumer_free_stream); | |
1521 | } | |
1522 | rcu_read_unlock(); | |
1523 | ||
1524 | lttng_ht_destroy(ht); | |
1525 | } | |
1526 | ||
fb3a43a9 | 1527 | /* |
f724d81e | 1528 | * Iterate over all streams of the hashtable and free them properly. |
e316aad5 DG |
1529 | * |
1530 | * XXX: Should not be only for metadata stream or else use an other name. | |
fb3a43a9 DG |
1531 | */ |
1532 | static void destroy_stream_ht(struct lttng_ht *ht) | |
1533 | { | |
1534 | int ret; | |
1535 | struct lttng_ht_iter iter; | |
1536 | struct lttng_consumer_stream *stream; | |
1537 | ||
1538 | if (ht == NULL) { | |
1539 | return; | |
1540 | } | |
1541 | ||
d09e1200 | 1542 | rcu_read_lock(); |
58b1f425 | 1543 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
fb3a43a9 DG |
1544 | ret = lttng_ht_del(ht, &iter); |
1545 | assert(!ret); | |
1546 | ||
58b1f425 | 1547 | call_rcu(&stream->node.head, consumer_free_stream); |
fb3a43a9 | 1548 | } |
d09e1200 | 1549 | rcu_read_unlock(); |
fb3a43a9 DG |
1550 | |
1551 | lttng_ht_destroy(ht); | |
1552 | } | |
1553 | ||
1554 | /* | |
1555 | * Clean up a metadata stream and free its memory. | |
1556 | */ | |
e316aad5 DG |
1557 | void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, |
1558 | struct lttng_ht *ht) | |
fb3a43a9 DG |
1559 | { |
1560 | int ret; | |
e316aad5 DG |
1561 | struct lttng_ht_iter iter; |
1562 | struct lttng_consumer_channel *free_chan = NULL; | |
fb3a43a9 DG |
1563 | struct consumer_relayd_sock_pair *relayd; |
1564 | ||
1565 | assert(stream); | |
1566 | /* | |
1567 | * This call should NEVER receive regular stream. It must always be | |
1568 | * metadata stream and this is crucial for data structure synchronization. | |
1569 | */ | |
1570 | assert(stream->metadata_flag); | |
1571 | ||
e316aad5 DG |
1572 | DBG3("Consumer delete metadata stream %d", stream->wait_fd); |
1573 | ||
1574 | if (ht == NULL) { | |
1575 | /* Means the stream was allocated but not successfully added */ | |
1576 | goto free_stream; | |
1577 | } | |
1578 | ||
fb3a43a9 DG |
1579 | pthread_mutex_lock(&consumer_data.lock); |
1580 | switch (consumer_data.type) { | |
1581 | case LTTNG_CONSUMER_KERNEL: | |
1582 | if (stream->mmap_base != NULL) { | |
1583 | ret = munmap(stream->mmap_base, stream->mmap_len); | |
1584 | if (ret != 0) { | |
1585 | PERROR("munmap metadata stream"); | |
1586 | } | |
1587 | } | |
1588 | break; | |
1589 | case LTTNG_CONSUMER32_UST: | |
1590 | case LTTNG_CONSUMER64_UST: | |
1591 | lttng_ustconsumer_del_stream(stream); | |
1592 | break; | |
1593 | default: | |
1594 | ERR("Unknown consumer_data type"); | |
1595 | assert(0); | |
e316aad5 | 1596 | goto end; |
fb3a43a9 | 1597 | } |
fb3a43a9 | 1598 | |
c869f647 | 1599 | rcu_read_lock(); |
58b1f425 | 1600 | iter.iter.node = &stream->node.node; |
c869f647 DG |
1601 | ret = lttng_ht_del(ht, &iter); |
1602 | assert(!ret); | |
1603 | rcu_read_unlock(); | |
1604 | ||
fb3a43a9 DG |
1605 | if (stream->out_fd >= 0) { |
1606 | ret = close(stream->out_fd); | |
1607 | if (ret) { | |
1608 | PERROR("close"); | |
1609 | } | |
1610 | } | |
1611 | ||
1612 | if (stream->wait_fd >= 0 && !stream->wait_fd_is_copy) { | |
1613 | ret = close(stream->wait_fd); | |
1614 | if (ret) { | |
1615 | PERROR("close"); | |
1616 | } | |
1617 | } | |
1618 | ||
1619 | if (stream->shm_fd >= 0 && stream->wait_fd != stream->shm_fd) { | |
1620 | ret = close(stream->shm_fd); | |
1621 | if (ret) { | |
1622 | PERROR("close"); | |
1623 | } | |
1624 | } | |
1625 | ||
1626 | /* Check and cleanup relayd */ | |
1627 | rcu_read_lock(); | |
1628 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
1629 | if (relayd != NULL) { | |
1630 | uatomic_dec(&relayd->refcount); | |
1631 | assert(uatomic_read(&relayd->refcount) >= 0); | |
1632 | ||
1633 | /* Closing streams requires to lock the control socket. */ | |
1634 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1635 | ret = relayd_send_close_stream(&relayd->control_sock, | |
1636 | stream->relayd_stream_id, stream->next_net_seq_num - 1); | |
1637 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1638 | if (ret < 0) { | |
1639 | DBG("Unable to close stream on the relayd. Continuing"); | |
1640 | /* | |
1641 | * Continue here. There is nothing we can do for the relayd. | |
1642 | * Chances are that the relayd has closed the socket so we just | |
1643 | * continue cleaning up. | |
1644 | */ | |
1645 | } | |
1646 | ||
1647 | /* Both conditions are met, we destroy the relayd. */ | |
1648 | if (uatomic_read(&relayd->refcount) == 0 && | |
1649 | uatomic_read(&relayd->destroy_flag)) { | |
d09e1200 | 1650 | destroy_relayd(relayd); |
fb3a43a9 DG |
1651 | } |
1652 | } | |
1653 | rcu_read_unlock(); | |
1654 | ||
1655 | /* Atomically decrement channel refcount since other threads can use it. */ | |
1656 | uatomic_dec(&stream->chan->refcount); | |
c30aaa51 MD |
1657 | if (!uatomic_read(&stream->chan->refcount) |
1658 | && !uatomic_read(&stream->chan->nb_init_streams)) { | |
1659 | /* Go for channel deletion! */ | |
e316aad5 | 1660 | free_chan = stream->chan; |
fb3a43a9 DG |
1661 | } |
1662 | ||
e316aad5 DG |
1663 | end: |
1664 | pthread_mutex_unlock(&consumer_data.lock); | |
1665 | ||
1666 | if (free_chan) { | |
1667 | consumer_del_channel(free_chan); | |
1668 | } | |
1669 | ||
1670 | free_stream: | |
58b1f425 | 1671 | call_rcu(&stream->node.head, consumer_free_stream); |
fb3a43a9 DG |
1672 | } |
1673 | ||
1674 | /* | |
1675 | * Action done with the metadata stream when adding it to the consumer internal | |
1676 | * data structures to handle it. | |
1677 | */ | |
e316aad5 DG |
1678 | static int consumer_add_metadata_stream(struct lttng_consumer_stream *stream, |
1679 | struct lttng_ht *ht) | |
fb3a43a9 | 1680 | { |
e316aad5 | 1681 | int ret = 0; |
fb3a43a9 DG |
1682 | struct consumer_relayd_sock_pair *relayd; |
1683 | ||
e316aad5 DG |
1684 | assert(stream); |
1685 | assert(ht); | |
1686 | ||
1687 | DBG3("Adding metadata stream %d to hash table", stream->wait_fd); | |
1688 | ||
1689 | pthread_mutex_lock(&consumer_data.lock); | |
1690 | ||
e316aad5 DG |
1691 | /* |
1692 | * From here, refcounts are updated so be _careful_ when returning an error | |
1693 | * after this point. | |
1694 | */ | |
1695 | ||
fb3a43a9 | 1696 | rcu_read_lock(); |
e316aad5 | 1697 | /* Find relayd and, if one is found, increment refcount. */ |
fb3a43a9 DG |
1698 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1699 | if (relayd != NULL) { | |
1700 | uatomic_inc(&relayd->refcount); | |
1701 | } | |
e316aad5 DG |
1702 | |
1703 | /* Update channel refcount once added without error(s). */ | |
1704 | uatomic_inc(&stream->chan->refcount); | |
1705 | ||
1706 | /* | |
1707 | * When nb_init_streams reaches 0, we don't need to trigger any action in | |
1708 | * terms of destroying the associated channel, because the action that | |
1709 | * causes the count to become 0 also causes a stream to be added. The | |
1710 | * channel deletion will thus be triggered by the following removal of this | |
1711 | * stream. | |
1712 | */ | |
1713 | if (uatomic_read(&stream->chan->nb_init_streams) > 0) { | |
1714 | uatomic_dec(&stream->chan->nb_init_streams); | |
1715 | } | |
1716 | ||
43c34bc3 DG |
1717 | /* Steal stream identifier to avoid having streams with the same key */ |
1718 | consumer_steal_stream_key(stream->key, ht); | |
1719 | ||
58b1f425 | 1720 | lttng_ht_add_unique_ulong(ht, &stream->node); |
fb3a43a9 | 1721 | rcu_read_unlock(); |
e316aad5 | 1722 | |
e316aad5 DG |
1723 | pthread_mutex_unlock(&consumer_data.lock); |
1724 | return ret; | |
fb3a43a9 DG |
1725 | } |
1726 | ||
1727 | /* | |
1728 | * Thread polls on metadata file descriptor and write them on disk or on the | |
1729 | * network. | |
1730 | */ | |
7d980def | 1731 | void *consumer_thread_metadata_poll(void *data) |
fb3a43a9 DG |
1732 | { |
1733 | int ret, i, pollfd; | |
1734 | uint32_t revents, nb_fd; | |
e316aad5 | 1735 | struct lttng_consumer_stream *stream = NULL; |
fb3a43a9 DG |
1736 | struct lttng_ht_iter iter; |
1737 | struct lttng_ht_node_ulong *node; | |
fb3a43a9 DG |
1738 | struct lttng_poll_event events; |
1739 | struct lttng_consumer_local_data *ctx = data; | |
1740 | ssize_t len; | |
1741 | ||
1742 | rcu_register_thread(); | |
1743 | ||
1744 | DBG("Thread metadata poll started"); | |
1745 | ||
fb3a43a9 DG |
1746 | /* Size is set to 1 for the consumer_metadata pipe */ |
1747 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
1748 | if (ret < 0) { | |
1749 | ERR("Poll set creation failed"); | |
1750 | goto end; | |
1751 | } | |
1752 | ||
1753 | ret = lttng_poll_add(&events, ctx->consumer_metadata_pipe[0], LPOLLIN); | |
1754 | if (ret < 0) { | |
1755 | goto end; | |
1756 | } | |
1757 | ||
1758 | /* Main loop */ | |
1759 | DBG("Metadata main loop started"); | |
1760 | ||
1761 | while (1) { | |
1762 | lttng_poll_reset(&events); | |
1763 | ||
1764 | nb_fd = LTTNG_POLL_GETNB(&events); | |
1765 | ||
1766 | /* Only the metadata pipe is set */ | |
1767 | if (nb_fd == 0 && consumer_quit == 1) { | |
1768 | goto end; | |
1769 | } | |
1770 | ||
1771 | restart: | |
1772 | DBG("Metadata poll wait with %d fd(s)", nb_fd); | |
1773 | ret = lttng_poll_wait(&events, -1); | |
1774 | DBG("Metadata event catched in thread"); | |
1775 | if (ret < 0) { | |
1776 | if (errno == EINTR) { | |
e316aad5 | 1777 | ERR("Poll EINTR catched"); |
fb3a43a9 DG |
1778 | goto restart; |
1779 | } | |
1780 | goto error; | |
1781 | } | |
1782 | ||
e316aad5 | 1783 | /* From here, the event is a metadata wait fd */ |
fb3a43a9 DG |
1784 | for (i = 0; i < nb_fd; i++) { |
1785 | revents = LTTNG_POLL_GETEV(&events, i); | |
1786 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
1787 | ||
e316aad5 DG |
1788 | /* Just don't waste time if no returned events for the fd */ |
1789 | if (!revents) { | |
1790 | continue; | |
1791 | } | |
1792 | ||
fb3a43a9 | 1793 | if (pollfd == ctx->consumer_metadata_pipe[0]) { |
4adabd61 | 1794 | if (revents & (LPOLLERR | LPOLLHUP )) { |
fb3a43a9 DG |
1795 | DBG("Metadata thread pipe hung up"); |
1796 | /* | |
1797 | * Remove the pipe from the poll set and continue the loop | |
1798 | * since their might be data to consume. | |
1799 | */ | |
1800 | lttng_poll_del(&events, ctx->consumer_metadata_pipe[0]); | |
1801 | close(ctx->consumer_metadata_pipe[0]); | |
1802 | continue; | |
1803 | } else if (revents & LPOLLIN) { | |
fb3a43a9 | 1804 | do { |
633d0084 DG |
1805 | /* Get the stream pointer received */ |
1806 | ret = read(pollfd, &stream, sizeof(stream)); | |
fb3a43a9 | 1807 | } while (ret < 0 && errno == EINTR); |
633d0084 DG |
1808 | if (ret < 0 || |
1809 | ret < sizeof(struct lttng_consumer_stream *)) { | |
fb3a43a9 | 1810 | PERROR("read metadata stream"); |
fb3a43a9 DG |
1811 | /* |
1812 | * Let's continue here and hope we can still work | |
1813 | * without stopping the consumer. XXX: Should we? | |
1814 | */ | |
1815 | continue; | |
1816 | } | |
1817 | ||
1818 | DBG("Adding metadata stream %d to poll set", | |
1819 | stream->wait_fd); | |
1820 | ||
e316aad5 DG |
1821 | ret = consumer_add_metadata_stream(stream, metadata_ht); |
1822 | if (ret) { | |
1823 | ERR("Unable to add metadata stream"); | |
1824 | /* Stream was not setup properly. Continuing. */ | |
1825 | consumer_del_metadata_stream(stream, NULL); | |
1826 | continue; | |
1827 | } | |
fb3a43a9 DG |
1828 | |
1829 | /* Add metadata stream to the global poll events list */ | |
1830 | lttng_poll_add(&events, stream->wait_fd, | |
1831 | LPOLLIN | LPOLLPRI); | |
fb3a43a9 DG |
1832 | } |
1833 | ||
e316aad5 | 1834 | /* Handle other stream */ |
fb3a43a9 DG |
1835 | continue; |
1836 | } | |
1837 | ||
d09e1200 | 1838 | rcu_read_lock(); |
fb3a43a9 DG |
1839 | lttng_ht_lookup(metadata_ht, (void *)((unsigned long) pollfd), |
1840 | &iter); | |
1841 | node = lttng_ht_iter_get_node_ulong(&iter); | |
e316aad5 | 1842 | assert(node); |
fb3a43a9 DG |
1843 | |
1844 | stream = caa_container_of(node, struct lttng_consumer_stream, | |
58b1f425 | 1845 | node); |
fb3a43a9 | 1846 | |
e316aad5 | 1847 | /* Check for error event */ |
4adabd61 | 1848 | if (revents & (LPOLLERR | LPOLLHUP)) { |
e316aad5 | 1849 | DBG("Metadata fd %d is hup|err.", pollfd); |
fb3a43a9 DG |
1850 | if (!stream->hangup_flush_done |
1851 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
1852 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
1853 | DBG("Attempting to flush and consume the UST buffers"); | |
1854 | lttng_ustconsumer_on_stream_hangup(stream); | |
1855 | ||
1856 | /* We just flushed the stream now read it. */ | |
4bb94b75 DG |
1857 | do { |
1858 | len = ctx->on_buffer_ready(stream, ctx); | |
1859 | /* | |
1860 | * We don't check the return value here since if we get | |
1861 | * a negative len, it means an error occured thus we | |
1862 | * simply remove it from the poll set and free the | |
1863 | * stream. | |
1864 | */ | |
1865 | } while (len > 0); | |
fb3a43a9 DG |
1866 | } |
1867 | ||
fb3a43a9 | 1868 | lttng_poll_del(&events, stream->wait_fd); |
e316aad5 DG |
1869 | /* |
1870 | * This call update the channel states, closes file descriptors | |
1871 | * and securely free the stream. | |
1872 | */ | |
1873 | consumer_del_metadata_stream(stream, metadata_ht); | |
1874 | } else if (revents & (LPOLLIN | LPOLLPRI)) { | |
1875 | /* Get the data out of the metadata file descriptor */ | |
1876 | DBG("Metadata available on fd %d", pollfd); | |
1877 | assert(stream->wait_fd == pollfd); | |
1878 | ||
1879 | len = ctx->on_buffer_ready(stream, ctx); | |
1880 | /* It's ok to have an unavailable sub-buffer */ | |
b64403e3 | 1881 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
e316aad5 DG |
1882 | rcu_read_unlock(); |
1883 | goto end; | |
1884 | } else if (len > 0) { | |
1885 | stream->data_read = 1; | |
1886 | } | |
fb3a43a9 | 1887 | } |
e316aad5 DG |
1888 | |
1889 | /* Release RCU lock for the stream looked up */ | |
d09e1200 | 1890 | rcu_read_unlock(); |
fb3a43a9 DG |
1891 | } |
1892 | } | |
1893 | ||
1894 | error: | |
1895 | end: | |
1896 | DBG("Metadata poll thread exiting"); | |
1897 | lttng_poll_clean(&events); | |
1898 | ||
1899 | if (metadata_ht) { | |
1900 | destroy_stream_ht(metadata_ht); | |
1901 | } | |
1902 | ||
1903 | rcu_unregister_thread(); | |
1904 | return NULL; | |
1905 | } | |
1906 | ||
3bd1e081 | 1907 | /* |
e4421fec | 1908 | * This thread polls the fds in the set to consume the data and write |
3bd1e081 MD |
1909 | * it to tracefile if necessary. |
1910 | */ | |
7d980def | 1911 | void *consumer_thread_data_poll(void *data) |
3bd1e081 MD |
1912 | { |
1913 | int num_rdy, num_hup, high_prio, ret, i; | |
1914 | struct pollfd *pollfd = NULL; | |
1915 | /* local view of the streams */ | |
c869f647 | 1916 | struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; |
3bd1e081 MD |
1917 | /* local view of consumer_data.fds_count */ |
1918 | int nb_fd = 0; | |
3bd1e081 | 1919 | struct lttng_consumer_local_data *ctx = data; |
00e2e675 | 1920 | ssize_t len; |
3bd1e081 | 1921 | |
e7b994a3 DG |
1922 | rcu_register_thread(); |
1923 | ||
43c34bc3 DG |
1924 | data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); |
1925 | if (data_ht == NULL) { | |
1926 | goto end; | |
1927 | } | |
1928 | ||
effcf122 | 1929 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream)); |
3bd1e081 MD |
1930 | |
1931 | while (1) { | |
1932 | high_prio = 0; | |
1933 | num_hup = 0; | |
1934 | ||
1935 | /* | |
e4421fec | 1936 | * the fds set has been updated, we need to update our |
3bd1e081 MD |
1937 | * local array as well |
1938 | */ | |
1939 | pthread_mutex_lock(&consumer_data.lock); | |
1940 | if (consumer_data.need_update) { | |
1941 | if (pollfd != NULL) { | |
1942 | free(pollfd); | |
1943 | pollfd = NULL; | |
1944 | } | |
1945 | if (local_stream != NULL) { | |
1946 | free(local_stream); | |
1947 | local_stream = NULL; | |
1948 | } | |
1949 | ||
50f8ae69 | 1950 | /* allocate for all fds + 1 for the consumer_data_pipe */ |
effcf122 | 1951 | pollfd = zmalloc((consumer_data.stream_count + 1) * sizeof(struct pollfd)); |
3bd1e081 | 1952 | if (pollfd == NULL) { |
7a57cf92 | 1953 | PERROR("pollfd malloc"); |
3bd1e081 MD |
1954 | pthread_mutex_unlock(&consumer_data.lock); |
1955 | goto end; | |
1956 | } | |
1957 | ||
50f8ae69 | 1958 | /* allocate for all fds + 1 for the consumer_data_pipe */ |
effcf122 | 1959 | local_stream = zmalloc((consumer_data.stream_count + 1) * |
3bd1e081 MD |
1960 | sizeof(struct lttng_consumer_stream)); |
1961 | if (local_stream == NULL) { | |
7a57cf92 | 1962 | PERROR("local_stream malloc"); |
3bd1e081 MD |
1963 | pthread_mutex_unlock(&consumer_data.lock); |
1964 | goto end; | |
1965 | } | |
43c34bc3 DG |
1966 | ret = consumer_update_poll_array(ctx, &pollfd, local_stream, |
1967 | data_ht); | |
3bd1e081 MD |
1968 | if (ret < 0) { |
1969 | ERR("Error in allocating pollfd or local_outfds"); | |
f73fabfd | 1970 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
1971 | pthread_mutex_unlock(&consumer_data.lock); |
1972 | goto end; | |
1973 | } | |
1974 | nb_fd = ret; | |
1975 | consumer_data.need_update = 0; | |
1976 | } | |
1977 | pthread_mutex_unlock(&consumer_data.lock); | |
1978 | ||
4078b776 MD |
1979 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
1980 | if (nb_fd == 0 && consumer_quit == 1) { | |
1981 | goto end; | |
1982 | } | |
3bd1e081 | 1983 | /* poll on the array of fds */ |
88f2b785 | 1984 | restart: |
3bd1e081 MD |
1985 | DBG("polling on %d fd", nb_fd + 1); |
1986 | num_rdy = poll(pollfd, nb_fd + 1, consumer_poll_timeout); | |
1987 | DBG("poll num_rdy : %d", num_rdy); | |
1988 | if (num_rdy == -1) { | |
88f2b785 MD |
1989 | /* |
1990 | * Restart interrupted system call. | |
1991 | */ | |
1992 | if (errno == EINTR) { | |
1993 | goto restart; | |
1994 | } | |
7a57cf92 | 1995 | PERROR("Poll error"); |
f73fabfd | 1996 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
1997 | goto end; |
1998 | } else if (num_rdy == 0) { | |
1999 | DBG("Polling thread timed out"); | |
2000 | goto end; | |
2001 | } | |
2002 | ||
3bd1e081 | 2003 | /* |
50f8ae69 | 2004 | * If the consumer_data_pipe triggered poll go directly to the |
00e2e675 DG |
2005 | * beginning of the loop to update the array. We want to prioritize |
2006 | * array update over low-priority reads. | |
3bd1e081 | 2007 | */ |
509bb1cf | 2008 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
04fdd819 | 2009 | size_t pipe_readlen; |
04fdd819 | 2010 | |
50f8ae69 | 2011 | DBG("consumer_data_pipe wake up"); |
04fdd819 MD |
2012 | /* Consume 1 byte of pipe data */ |
2013 | do { | |
50f8ae69 | 2014 | pipe_readlen = read(ctx->consumer_data_pipe[0], &new_stream, |
c869f647 | 2015 | sizeof(new_stream)); |
04fdd819 | 2016 | } while (pipe_readlen == -1 && errno == EINTR); |
c869f647 DG |
2017 | |
2018 | /* | |
2019 | * If the stream is NULL, just ignore it. It's also possible that | |
2020 | * the sessiond poll thread changed the consumer_quit state and is | |
2021 | * waking us up to test it. | |
2022 | */ | |
2023 | if (new_stream == NULL) { | |
2024 | continue; | |
2025 | } | |
2026 | ||
43c34bc3 | 2027 | ret = consumer_add_stream(new_stream, data_ht); |
c869f647 DG |
2028 | if (ret) { |
2029 | ERR("Consumer add stream %d failed. Continuing", | |
2030 | new_stream->key); | |
2031 | /* | |
2032 | * At this point, if the add_stream fails, it is not in the | |
2033 | * hash table thus passing the NULL value here. | |
2034 | */ | |
2035 | consumer_del_stream(new_stream, NULL); | |
2036 | } | |
2037 | ||
2038 | /* Continue to update the local streams and handle prio ones */ | |
3bd1e081 MD |
2039 | continue; |
2040 | } | |
2041 | ||
2042 | /* Take care of high priority channels first. */ | |
2043 | for (i = 0; i < nb_fd; i++) { | |
fb3a43a9 | 2044 | if (pollfd[i].revents & POLLPRI) { |
d41f73b7 MD |
2045 | DBG("Urgent read on fd %d", pollfd[i].fd); |
2046 | high_prio = 1; | |
4078b776 | 2047 | len = ctx->on_buffer_ready(local_stream[i], ctx); |
d41f73b7 | 2048 | /* it's ok to have an unavailable sub-buffer */ |
b64403e3 | 2049 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
4078b776 MD |
2050 | goto end; |
2051 | } else if (len > 0) { | |
2052 | local_stream[i]->data_read = 1; | |
d41f73b7 | 2053 | } |
3bd1e081 MD |
2054 | } |
2055 | } | |
2056 | ||
4078b776 MD |
2057 | /* |
2058 | * If we read high prio channel in this loop, try again | |
2059 | * for more high prio data. | |
2060 | */ | |
2061 | if (high_prio) { | |
3bd1e081 MD |
2062 | continue; |
2063 | } | |
2064 | ||
2065 | /* Take care of low priority channels. */ | |
4078b776 MD |
2066 | for (i = 0; i < nb_fd; i++) { |
2067 | if ((pollfd[i].revents & POLLIN) || | |
2068 | local_stream[i]->hangup_flush_done) { | |
4078b776 MD |
2069 | DBG("Normal read on fd %d", pollfd[i].fd); |
2070 | len = ctx->on_buffer_ready(local_stream[i], ctx); | |
2071 | /* it's ok to have an unavailable sub-buffer */ | |
b64403e3 | 2072 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
4078b776 MD |
2073 | goto end; |
2074 | } else if (len > 0) { | |
2075 | local_stream[i]->data_read = 1; | |
2076 | } | |
2077 | } | |
2078 | } | |
2079 | ||
2080 | /* Handle hangup and errors */ | |
2081 | for (i = 0; i < nb_fd; i++) { | |
2082 | if (!local_stream[i]->hangup_flush_done | |
2083 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) | |
2084 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2085 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2086 | DBG("fd %d is hup|err|nval. Attempting flush and read.", | |
2087 | pollfd[i].fd); | |
2088 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); | |
2089 | /* Attempt read again, for the data we just flushed. */ | |
2090 | local_stream[i]->data_read = 1; | |
2091 | } | |
2092 | /* | |
2093 | * If the poll flag is HUP/ERR/NVAL and we have | |
2094 | * read no data in this pass, we can remove the | |
2095 | * stream from its hash table. | |
2096 | */ | |
2097 | if ((pollfd[i].revents & POLLHUP)) { | |
2098 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); | |
2099 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2100 | consumer_del_stream(local_stream[i], data_ht); |
4078b776 MD |
2101 | num_hup++; |
2102 | } | |
2103 | } else if (pollfd[i].revents & POLLERR) { | |
2104 | ERR("Error returned in polling fd %d.", pollfd[i].fd); | |
2105 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2106 | consumer_del_stream(local_stream[i], data_ht); |
4078b776 MD |
2107 | num_hup++; |
2108 | } | |
2109 | } else if (pollfd[i].revents & POLLNVAL) { | |
2110 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); | |
2111 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2112 | consumer_del_stream(local_stream[i], data_ht); |
4078b776 | 2113 | num_hup++; |
3bd1e081 MD |
2114 | } |
2115 | } | |
4078b776 | 2116 | local_stream[i]->data_read = 0; |
3bd1e081 MD |
2117 | } |
2118 | } | |
2119 | end: | |
2120 | DBG("polling thread exiting"); | |
2121 | if (pollfd != NULL) { | |
2122 | free(pollfd); | |
2123 | pollfd = NULL; | |
2124 | } | |
2125 | if (local_stream != NULL) { | |
2126 | free(local_stream); | |
2127 | local_stream = NULL; | |
2128 | } | |
fb3a43a9 DG |
2129 | |
2130 | /* | |
2131 | * Close the write side of the pipe so epoll_wait() in | |
7d980def DG |
2132 | * consumer_thread_metadata_poll can catch it. The thread is monitoring the |
2133 | * read side of the pipe. If we close them both, epoll_wait strangely does | |
2134 | * not return and could create a endless wait period if the pipe is the | |
2135 | * only tracked fd in the poll set. The thread will take care of closing | |
2136 | * the read side. | |
fb3a43a9 DG |
2137 | */ |
2138 | close(ctx->consumer_metadata_pipe[1]); | |
fb3a43a9 | 2139 | |
43c34bc3 DG |
2140 | if (data_ht) { |
2141 | destroy_data_stream_ht(data_ht); | |
2142 | } | |
2143 | ||
e7b994a3 | 2144 | rcu_unregister_thread(); |
3bd1e081 MD |
2145 | return NULL; |
2146 | } | |
2147 | ||
2148 | /* | |
2149 | * This thread listens on the consumerd socket and receives the file | |
2150 | * descriptors from the session daemon. | |
2151 | */ | |
7d980def | 2152 | void *consumer_thread_sessiond_poll(void *data) |
3bd1e081 MD |
2153 | { |
2154 | int sock, client_socket, ret; | |
2155 | /* | |
2156 | * structure to poll for incoming data on communication socket avoids | |
2157 | * making blocking sockets. | |
2158 | */ | |
2159 | struct pollfd consumer_sockpoll[2]; | |
2160 | struct lttng_consumer_local_data *ctx = data; | |
2161 | ||
e7b994a3 DG |
2162 | rcu_register_thread(); |
2163 | ||
3bd1e081 MD |
2164 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
2165 | unlink(ctx->consumer_command_sock_path); | |
2166 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); | |
2167 | if (client_socket < 0) { | |
2168 | ERR("Cannot create command socket"); | |
2169 | goto end; | |
2170 | } | |
2171 | ||
2172 | ret = lttcomm_listen_unix_sock(client_socket); | |
2173 | if (ret < 0) { | |
2174 | goto end; | |
2175 | } | |
2176 | ||
32258573 | 2177 | DBG("Sending ready command to lttng-sessiond"); |
f73fabfd | 2178 | ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); |
3bd1e081 MD |
2179 | /* return < 0 on error, but == 0 is not fatal */ |
2180 | if (ret < 0) { | |
32258573 | 2181 | ERR("Error sending ready command to lttng-sessiond"); |
3bd1e081 MD |
2182 | goto end; |
2183 | } | |
2184 | ||
2185 | ret = fcntl(client_socket, F_SETFL, O_NONBLOCK); | |
2186 | if (ret < 0) { | |
7a57cf92 | 2187 | PERROR("fcntl O_NONBLOCK"); |
3bd1e081 MD |
2188 | goto end; |
2189 | } | |
2190 | ||
2191 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ | |
2192 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; | |
2193 | consumer_sockpoll[0].events = POLLIN | POLLPRI; | |
2194 | consumer_sockpoll[1].fd = client_socket; | |
2195 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
2196 | ||
2197 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
2198 | goto end; | |
2199 | } | |
2200 | DBG("Connection on client_socket"); | |
2201 | ||
2202 | /* Blocking call, waiting for transmission */ | |
2203 | sock = lttcomm_accept_unix_sock(client_socket); | |
2204 | if (sock <= 0) { | |
2205 | WARN("On accept"); | |
2206 | goto end; | |
2207 | } | |
2208 | ret = fcntl(sock, F_SETFL, O_NONBLOCK); | |
2209 | if (ret < 0) { | |
7a57cf92 | 2210 | PERROR("fcntl O_NONBLOCK"); |
3bd1e081 MD |
2211 | goto end; |
2212 | } | |
2213 | ||
2214 | /* update the polling structure to poll on the established socket */ | |
2215 | consumer_sockpoll[1].fd = sock; | |
2216 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
2217 | ||
2218 | while (1) { | |
2219 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
2220 | goto end; | |
2221 | } | |
2222 | DBG("Incoming command on sock"); | |
2223 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
2224 | if (ret == -ENOENT) { | |
2225 | DBG("Received STOP command"); | |
2226 | goto end; | |
2227 | } | |
4cbc1a04 DG |
2228 | if (ret <= 0) { |
2229 | /* | |
2230 | * This could simply be a session daemon quitting. Don't output | |
2231 | * ERR() here. | |
2232 | */ | |
2233 | DBG("Communication interrupted on command socket"); | |
3bd1e081 MD |
2234 | goto end; |
2235 | } | |
2236 | if (consumer_quit) { | |
2237 | DBG("consumer_thread_receive_fds received quit from signal"); | |
2238 | goto end; | |
2239 | } | |
2240 | DBG("received fds on sock"); | |
2241 | } | |
2242 | end: | |
2243 | DBG("consumer_thread_receive_fds exiting"); | |
2244 | ||
2245 | /* | |
2246 | * when all fds have hung up, the polling thread | |
2247 | * can exit cleanly | |
2248 | */ | |
2249 | consumer_quit = 1; | |
2250 | ||
2251 | /* | |
2252 | * 2s of grace period, if no polling events occur during | |
2253 | * this period, the polling thread will exit even if there | |
2254 | * are still open FDs (should not happen, but safety mechanism). | |
2255 | */ | |
2256 | consumer_poll_timeout = LTTNG_CONSUMER_POLL_TIMEOUT; | |
2257 | ||
04fdd819 | 2258 | /* |
c869f647 DG |
2259 | * Notify the data poll thread to poll back again and test the |
2260 | * consumer_quit state to quit gracefully. | |
04fdd819 MD |
2261 | */ |
2262 | do { | |
c869f647 DG |
2263 | struct lttng_consumer_stream *null_stream = NULL; |
2264 | ||
50f8ae69 | 2265 | ret = write(ctx->consumer_data_pipe[1], &null_stream, |
c869f647 | 2266 | sizeof(null_stream)); |
6f94560a | 2267 | } while (ret < 0 && errno == EINTR); |
c869f647 | 2268 | |
e7b994a3 | 2269 | rcu_unregister_thread(); |
3bd1e081 MD |
2270 | return NULL; |
2271 | } | |
d41f73b7 | 2272 | |
4078b776 | 2273 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
d41f73b7 MD |
2274 | struct lttng_consumer_local_data *ctx) |
2275 | { | |
2276 | switch (consumer_data.type) { | |
2277 | case LTTNG_CONSUMER_KERNEL: | |
2278 | return lttng_kconsumer_read_subbuffer(stream, ctx); | |
7753dea8 MD |
2279 | case LTTNG_CONSUMER32_UST: |
2280 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
2281 | return lttng_ustconsumer_read_subbuffer(stream, ctx); |
2282 | default: | |
2283 | ERR("Unknown consumer_data type"); | |
2284 | assert(0); | |
2285 | return -ENOSYS; | |
2286 | } | |
2287 | } | |
2288 | ||
2289 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
2290 | { | |
2291 | switch (consumer_data.type) { | |
2292 | case LTTNG_CONSUMER_KERNEL: | |
2293 | return lttng_kconsumer_on_recv_stream(stream); | |
7753dea8 MD |
2294 | case LTTNG_CONSUMER32_UST: |
2295 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
2296 | return lttng_ustconsumer_on_recv_stream(stream); |
2297 | default: | |
2298 | ERR("Unknown consumer_data type"); | |
2299 | assert(0); | |
2300 | return -ENOSYS; | |
2301 | } | |
2302 | } | |
e4421fec DG |
2303 | |
2304 | /* | |
2305 | * Allocate and set consumer data hash tables. | |
2306 | */ | |
2307 | void lttng_consumer_init(void) | |
2308 | { | |
e4421fec | 2309 | consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); |
00e2e675 | 2310 | consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); |
43c34bc3 DG |
2311 | |
2312 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
2313 | assert(metadata_ht); | |
2314 | data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
2315 | assert(data_ht); | |
e4421fec | 2316 | } |
7735ef9e DG |
2317 | |
2318 | /* | |
2319 | * Process the ADD_RELAYD command receive by a consumer. | |
2320 | * | |
2321 | * This will create a relayd socket pair and add it to the relayd hash table. | |
2322 | * The caller MUST acquire a RCU read side lock before calling it. | |
2323 | */ | |
2324 | int consumer_add_relayd_socket(int net_seq_idx, int sock_type, | |
2325 | struct lttng_consumer_local_data *ctx, int sock, | |
2326 | struct pollfd *consumer_sockpoll, struct lttcomm_sock *relayd_sock) | |
2327 | { | |
2328 | int fd, ret = -1; | |
2329 | struct consumer_relayd_sock_pair *relayd; | |
2330 | ||
2331 | DBG("Consumer adding relayd socket (idx: %d)", net_seq_idx); | |
2332 | ||
2333 | /* Get relayd reference if exists. */ | |
2334 | relayd = consumer_find_relayd(net_seq_idx); | |
2335 | if (relayd == NULL) { | |
2336 | /* Not found. Allocate one. */ | |
2337 | relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); | |
2338 | if (relayd == NULL) { | |
f73fabfd | 2339 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); |
7735ef9e DG |
2340 | goto error; |
2341 | } | |
2342 | } | |
2343 | ||
2344 | /* Poll on consumer socket. */ | |
2345 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
2346 | ret = -EINTR; | |
2347 | goto error; | |
2348 | } | |
2349 | ||
2350 | /* Get relayd socket from session daemon */ | |
2351 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); | |
2352 | if (ret != sizeof(fd)) { | |
f73fabfd | 2353 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); |
7735ef9e DG |
2354 | ret = -1; |
2355 | goto error; | |
2356 | } | |
2357 | ||
2358 | /* Copy socket information and received FD */ | |
2359 | switch (sock_type) { | |
2360 | case LTTNG_STREAM_CONTROL: | |
2361 | /* Copy received lttcomm socket */ | |
2362 | lttcomm_copy_sock(&relayd->control_sock, relayd_sock); | |
2363 | ret = lttcomm_create_sock(&relayd->control_sock); | |
2364 | if (ret < 0) { | |
2365 | goto error; | |
2366 | } | |
2367 | ||
2368 | /* Close the created socket fd which is useless */ | |
2369 | close(relayd->control_sock.fd); | |
2370 | ||
2371 | /* Assign new file descriptor */ | |
2372 | relayd->control_sock.fd = fd; | |
2373 | break; | |
2374 | case LTTNG_STREAM_DATA: | |
2375 | /* Copy received lttcomm socket */ | |
2376 | lttcomm_copy_sock(&relayd->data_sock, relayd_sock); | |
2377 | ret = lttcomm_create_sock(&relayd->data_sock); | |
2378 | if (ret < 0) { | |
2379 | goto error; | |
2380 | } | |
2381 | ||
2382 | /* Close the created socket fd which is useless */ | |
2383 | close(relayd->data_sock.fd); | |
2384 | ||
2385 | /* Assign new file descriptor */ | |
2386 | relayd->data_sock.fd = fd; | |
2387 | break; | |
2388 | default: | |
2389 | ERR("Unknown relayd socket type (%d)", sock_type); | |
2390 | goto error; | |
2391 | } | |
2392 | ||
2393 | DBG("Consumer %s socket created successfully with net idx %d (fd: %d)", | |
2394 | sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", | |
2395 | relayd->net_seq_idx, fd); | |
2396 | ||
2397 | /* | |
2398 | * Add relayd socket pair to consumer data hashtable. If object already | |
2399 | * exists or on error, the function gracefully returns. | |
2400 | */ | |
d09e1200 | 2401 | add_relayd(relayd); |
7735ef9e DG |
2402 | |
2403 | /* All good! */ | |
2404 | ret = 0; | |
2405 | ||
2406 | error: | |
2407 | return ret; | |
2408 | } |