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 | ||
61 | /* | |
62 | * Find a stream. The consumer_data.lock must be locked during this | |
63 | * call. | |
64 | */ | |
65 | static struct lttng_consumer_stream *consumer_find_stream(int key) | |
66 | { | |
e4421fec DG |
67 | struct lttng_ht_iter iter; |
68 | struct lttng_ht_node_ulong *node; | |
69 | struct lttng_consumer_stream *stream = NULL; | |
3bd1e081 | 70 | |
7ad0a0cb | 71 | /* Negative keys are lookup failures */ |
7a57cf92 | 72 | if (key < 0) { |
7ad0a0cb | 73 | return NULL; |
7a57cf92 | 74 | } |
e4421fec | 75 | |
6065ceec DG |
76 | rcu_read_lock(); |
77 | ||
e4421fec DG |
78 | lttng_ht_lookup(consumer_data.stream_ht, (void *)((unsigned long) key), |
79 | &iter); | |
80 | node = lttng_ht_iter_get_node_ulong(&iter); | |
81 | if (node != NULL) { | |
82 | stream = caa_container_of(node, struct lttng_consumer_stream, node); | |
3bd1e081 | 83 | } |
e4421fec | 84 | |
6065ceec DG |
85 | rcu_read_unlock(); |
86 | ||
e4421fec | 87 | return stream; |
3bd1e081 MD |
88 | } |
89 | ||
7ad0a0cb MD |
90 | static void consumer_steal_stream_key(int key) |
91 | { | |
92 | struct lttng_consumer_stream *stream; | |
93 | ||
04253271 | 94 | rcu_read_lock(); |
7ad0a0cb | 95 | stream = consumer_find_stream(key); |
04253271 | 96 | if (stream) { |
7ad0a0cb | 97 | stream->key = -1; |
04253271 MD |
98 | /* |
99 | * We don't want the lookup to match, but we still need | |
100 | * to iterate on this stream when iterating over the hash table. Just | |
101 | * change the node key. | |
102 | */ | |
103 | stream->node.key = -1; | |
104 | } | |
105 | rcu_read_unlock(); | |
7ad0a0cb MD |
106 | } |
107 | ||
3bd1e081 MD |
108 | static struct lttng_consumer_channel *consumer_find_channel(int key) |
109 | { | |
e4421fec DG |
110 | struct lttng_ht_iter iter; |
111 | struct lttng_ht_node_ulong *node; | |
112 | struct lttng_consumer_channel *channel = NULL; | |
3bd1e081 | 113 | |
7ad0a0cb | 114 | /* Negative keys are lookup failures */ |
7a57cf92 | 115 | if (key < 0) { |
7ad0a0cb | 116 | return NULL; |
7a57cf92 | 117 | } |
e4421fec | 118 | |
6065ceec DG |
119 | rcu_read_lock(); |
120 | ||
e4421fec DG |
121 | lttng_ht_lookup(consumer_data.channel_ht, (void *)((unsigned long) key), |
122 | &iter); | |
123 | node = lttng_ht_iter_get_node_ulong(&iter); | |
124 | if (node != NULL) { | |
125 | channel = caa_container_of(node, struct lttng_consumer_channel, node); | |
3bd1e081 | 126 | } |
e4421fec | 127 | |
6065ceec DG |
128 | rcu_read_unlock(); |
129 | ||
e4421fec | 130 | return channel; |
3bd1e081 MD |
131 | } |
132 | ||
7ad0a0cb MD |
133 | static void consumer_steal_channel_key(int key) |
134 | { | |
135 | struct lttng_consumer_channel *channel; | |
136 | ||
04253271 | 137 | rcu_read_lock(); |
7ad0a0cb | 138 | channel = consumer_find_channel(key); |
04253271 | 139 | if (channel) { |
7ad0a0cb | 140 | channel->key = -1; |
04253271 MD |
141 | /* |
142 | * We don't want the lookup to match, but we still need | |
143 | * to iterate on this channel when iterating over the hash table. Just | |
144 | * change the node key. | |
145 | */ | |
146 | channel->node.key = -1; | |
147 | } | |
148 | rcu_read_unlock(); | |
7ad0a0cb MD |
149 | } |
150 | ||
702b1ea4 MD |
151 | static |
152 | void consumer_free_stream(struct rcu_head *head) | |
153 | { | |
154 | struct lttng_ht_node_ulong *node = | |
155 | caa_container_of(head, struct lttng_ht_node_ulong, head); | |
156 | struct lttng_consumer_stream *stream = | |
157 | caa_container_of(node, struct lttng_consumer_stream, node); | |
158 | ||
159 | free(stream); | |
160 | } | |
161 | ||
00e2e675 DG |
162 | /* |
163 | * RCU protected relayd socket pair free. | |
164 | */ | |
165 | static void consumer_rcu_free_relayd(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 consumer_relayd_sock_pair *relayd = | |
170 | caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
171 | ||
172 | free(relayd); | |
173 | } | |
174 | ||
175 | /* | |
176 | * Destroy and free relayd socket pair object. | |
177 | * | |
178 | * This function MUST be called with the consumer_data lock acquired. | |
179 | */ | |
d09e1200 | 180 | static void destroy_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
181 | { |
182 | int ret; | |
183 | struct lttng_ht_iter iter; | |
184 | ||
173af62f DG |
185 | if (relayd == NULL) { |
186 | return; | |
187 | } | |
188 | ||
00e2e675 DG |
189 | DBG("Consumer destroy and close relayd socket pair"); |
190 | ||
191 | iter.iter.node = &relayd->node.node; | |
192 | ret = lttng_ht_del(consumer_data.relayd_ht, &iter); | |
173af62f DG |
193 | if (ret != 0) { |
194 | /* We assume the relayd was already destroyed */ | |
195 | return; | |
196 | } | |
00e2e675 DG |
197 | |
198 | /* Close all sockets */ | |
199 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
200 | (void) relayd_close(&relayd->control_sock); | |
201 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
202 | (void) relayd_close(&relayd->data_sock); | |
203 | ||
204 | /* RCU free() call */ | |
205 | call_rcu(&relayd->node.head, consumer_rcu_free_relayd); | |
206 | } | |
207 | ||
a6ba4fe1 DG |
208 | /* |
209 | * Flag a relayd socket pair for destruction. Destroy it if the refcount | |
210 | * reaches zero. | |
211 | * | |
212 | * RCU read side lock MUST be aquired before calling this function. | |
213 | */ | |
214 | void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) | |
215 | { | |
216 | assert(relayd); | |
217 | ||
218 | /* Set destroy flag for this object */ | |
219 | uatomic_set(&relayd->destroy_flag, 1); | |
220 | ||
221 | /* Destroy the relayd if refcount is 0 */ | |
222 | if (uatomic_read(&relayd->refcount) == 0) { | |
d09e1200 | 223 | destroy_relayd(relayd); |
a6ba4fe1 DG |
224 | } |
225 | } | |
226 | ||
3bd1e081 MD |
227 | /* |
228 | * Remove a stream from the global list protected by a mutex. This | |
229 | * function is also responsible for freeing its data structures. | |
230 | */ | |
231 | void consumer_del_stream(struct lttng_consumer_stream *stream) | |
232 | { | |
233 | int ret; | |
e4421fec | 234 | struct lttng_ht_iter iter; |
3bd1e081 | 235 | struct lttng_consumer_channel *free_chan = NULL; |
00e2e675 DG |
236 | struct consumer_relayd_sock_pair *relayd; |
237 | ||
238 | assert(stream); | |
3bd1e081 MD |
239 | |
240 | pthread_mutex_lock(&consumer_data.lock); | |
241 | ||
242 | switch (consumer_data.type) { | |
243 | case LTTNG_CONSUMER_KERNEL: | |
244 | if (stream->mmap_base != NULL) { | |
245 | ret = munmap(stream->mmap_base, stream->mmap_len); | |
246 | if (ret != 0) { | |
7a57cf92 | 247 | PERROR("munmap"); |
3bd1e081 MD |
248 | } |
249 | } | |
250 | break; | |
7753dea8 MD |
251 | case LTTNG_CONSUMER32_UST: |
252 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
253 | lttng_ustconsumer_del_stream(stream); |
254 | break; | |
255 | default: | |
256 | ERR("Unknown consumer_data type"); | |
257 | assert(0); | |
258 | goto end; | |
259 | } | |
260 | ||
6065ceec | 261 | rcu_read_lock(); |
04253271 MD |
262 | iter.iter.node = &stream->node.node; |
263 | ret = lttng_ht_del(consumer_data.stream_ht, &iter); | |
264 | assert(!ret); | |
e4421fec | 265 | |
6065ceec DG |
266 | rcu_read_unlock(); |
267 | ||
3bd1e081 MD |
268 | if (consumer_data.stream_count <= 0) { |
269 | goto end; | |
270 | } | |
271 | consumer_data.stream_count--; | |
272 | if (!stream) { | |
273 | goto end; | |
274 | } | |
275 | if (stream->out_fd >= 0) { | |
4c462e79 MD |
276 | ret = close(stream->out_fd); |
277 | if (ret) { | |
278 | PERROR("close"); | |
279 | } | |
3bd1e081 | 280 | } |
b5c5fc29 | 281 | if (stream->wait_fd >= 0 && !stream->wait_fd_is_copy) { |
4c462e79 MD |
282 | ret = close(stream->wait_fd); |
283 | if (ret) { | |
284 | PERROR("close"); | |
285 | } | |
3bd1e081 | 286 | } |
2c1dd183 | 287 | if (stream->shm_fd >= 0 && stream->wait_fd != stream->shm_fd) { |
4c462e79 MD |
288 | ret = close(stream->shm_fd); |
289 | if (ret) { | |
290 | PERROR("close"); | |
291 | } | |
3bd1e081 | 292 | } |
00e2e675 DG |
293 | |
294 | /* Check and cleanup relayd */ | |
b0b335c8 | 295 | rcu_read_lock(); |
00e2e675 DG |
296 | relayd = consumer_find_relayd(stream->net_seq_idx); |
297 | if (relayd != NULL) { | |
b0b335c8 MD |
298 | uatomic_dec(&relayd->refcount); |
299 | assert(uatomic_read(&relayd->refcount) >= 0); | |
173af62f | 300 | |
3f8e211f DG |
301 | /* Closing streams requires to lock the control socket. */ |
302 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
173af62f DG |
303 | ret = relayd_send_close_stream(&relayd->control_sock, |
304 | stream->relayd_stream_id, | |
305 | stream->next_net_seq_num - 1); | |
3f8e211f | 306 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
173af62f | 307 | if (ret < 0) { |
a4b92340 DG |
308 | DBG("Unable to close stream on the relayd. Continuing"); |
309 | /* | |
310 | * Continue here. There is nothing we can do for the relayd. | |
311 | * Chances are that the relayd has closed the socket so we just | |
312 | * continue cleaning up. | |
313 | */ | |
173af62f DG |
314 | } |
315 | ||
316 | /* Both conditions are met, we destroy the relayd. */ | |
317 | if (uatomic_read(&relayd->refcount) == 0 && | |
318 | uatomic_read(&relayd->destroy_flag)) { | |
d09e1200 | 319 | destroy_relayd(relayd); |
00e2e675 | 320 | } |
00e2e675 | 321 | } |
b0b335c8 | 322 | rcu_read_unlock(); |
00e2e675 | 323 | |
c30aaa51 MD |
324 | uatomic_dec(&stream->chan->refcount); |
325 | if (!uatomic_read(&stream->chan->refcount) | |
326 | && !uatomic_read(&stream->chan->nb_init_streams)) { | |
3bd1e081 | 327 | free_chan = stream->chan; |
00e2e675 DG |
328 | } |
329 | ||
702b1ea4 | 330 | call_rcu(&stream->node.head, consumer_free_stream); |
3bd1e081 MD |
331 | end: |
332 | consumer_data.need_update = 1; | |
333 | pthread_mutex_unlock(&consumer_data.lock); | |
334 | ||
c30aaa51 | 335 | if (free_chan) { |
3bd1e081 | 336 | consumer_del_channel(free_chan); |
c30aaa51 | 337 | } |
3bd1e081 MD |
338 | } |
339 | ||
340 | struct lttng_consumer_stream *consumer_allocate_stream( | |
341 | int channel_key, int stream_key, | |
342 | int shm_fd, int wait_fd, | |
343 | enum lttng_consumer_stream_state state, | |
344 | uint64_t mmap_len, | |
345 | enum lttng_event_output output, | |
6df2e2c9 MD |
346 | const char *path_name, |
347 | uid_t uid, | |
00e2e675 DG |
348 | gid_t gid, |
349 | int net_index, | |
c80048c6 MD |
350 | int metadata_flag, |
351 | int *alloc_ret) | |
3bd1e081 MD |
352 | { |
353 | struct lttng_consumer_stream *stream; | |
354 | int ret; | |
355 | ||
effcf122 | 356 | stream = zmalloc(sizeof(*stream)); |
3bd1e081 | 357 | if (stream == NULL) { |
7a57cf92 | 358 | PERROR("malloc struct lttng_consumer_stream"); |
c80048c6 | 359 | *alloc_ret = -ENOMEM; |
7a57cf92 | 360 | goto end; |
3bd1e081 | 361 | } |
7a57cf92 DG |
362 | |
363 | /* | |
364 | * Get stream's channel reference. Needed when adding the stream to the | |
365 | * global hash table. | |
366 | */ | |
3bd1e081 MD |
367 | stream->chan = consumer_find_channel(channel_key); |
368 | if (!stream->chan) { | |
c80048c6 | 369 | *alloc_ret = -ENOENT; |
7a57cf92 | 370 | ERR("Unable to find channel for stream %d", stream_key); |
c80048c6 | 371 | goto error; |
3bd1e081 MD |
372 | } |
373 | stream->chan->refcount++; | |
374 | stream->key = stream_key; | |
375 | stream->shm_fd = shm_fd; | |
376 | stream->wait_fd = wait_fd; | |
377 | stream->out_fd = -1; | |
378 | stream->out_fd_offset = 0; | |
379 | stream->state = state; | |
380 | stream->mmap_len = mmap_len; | |
381 | stream->mmap_base = NULL; | |
382 | stream->output = output; | |
6df2e2c9 MD |
383 | stream->uid = uid; |
384 | stream->gid = gid; | |
00e2e675 DG |
385 | stream->net_seq_idx = net_index; |
386 | stream->metadata_flag = metadata_flag; | |
387 | strncpy(stream->path_name, path_name, sizeof(stream->path_name)); | |
388 | stream->path_name[sizeof(stream->path_name) - 1] = '\0'; | |
e4421fec | 389 | lttng_ht_node_init_ulong(&stream->node, stream->key); |
00e2e675 | 390 | lttng_ht_node_init_ulong(&stream->waitfd_node, stream->wait_fd); |
3bd1e081 MD |
391 | |
392 | switch (consumer_data.type) { | |
393 | case LTTNG_CONSUMER_KERNEL: | |
394 | break; | |
7753dea8 MD |
395 | case LTTNG_CONSUMER32_UST: |
396 | case LTTNG_CONSUMER64_UST: | |
5af2f756 | 397 | stream->cpu = stream->chan->cpucount++; |
3bd1e081 MD |
398 | ret = lttng_ustconsumer_allocate_stream(stream); |
399 | if (ret) { | |
c80048c6 MD |
400 | *alloc_ret = -EINVAL; |
401 | goto error; | |
3bd1e081 MD |
402 | } |
403 | break; | |
404 | default: | |
405 | ERR("Unknown consumer_data type"); | |
c80048c6 MD |
406 | *alloc_ret = -EINVAL; |
407 | goto error; | |
3bd1e081 | 408 | } |
c30aaa51 MD |
409 | |
410 | /* | |
411 | * When nb_init_streams reaches 0, we don't need to trigger any action in | |
412 | * terms of destroying the associated channel, because the action that | |
413 | * causes the count to become 0 also causes a stream to be added. The | |
414 | * channel deletion will thus be triggered by the following removal of this | |
415 | * stream. | |
416 | */ | |
417 | if (uatomic_read(&stream->chan->nb_init_streams) > 0) { | |
418 | uatomic_dec(&stream->chan->nb_init_streams); | |
419 | } | |
420 | ||
421 | DBG3("Allocated stream %s (key %d, shm_fd %d, wait_fd %d, mmap_len %llu," | |
422 | " out_fd %d, net_seq_idx %d)", stream->path_name, stream->key, | |
423 | stream->shm_fd, stream->wait_fd, | |
424 | (unsigned long long) stream->mmap_len, stream->out_fd, | |
00e2e675 | 425 | stream->net_seq_idx); |
3bd1e081 | 426 | return stream; |
c80048c6 MD |
427 | |
428 | error: | |
429 | free(stream); | |
7a57cf92 | 430 | end: |
c80048c6 | 431 | return NULL; |
3bd1e081 MD |
432 | } |
433 | ||
434 | /* | |
435 | * Add a stream to the global list protected by a mutex. | |
436 | */ | |
437 | int consumer_add_stream(struct lttng_consumer_stream *stream) | |
438 | { | |
439 | int ret = 0; | |
c77fc10a DG |
440 | struct lttng_ht_node_ulong *node; |
441 | struct lttng_ht_iter iter; | |
00e2e675 | 442 | struct consumer_relayd_sock_pair *relayd; |
3bd1e081 MD |
443 | |
444 | pthread_mutex_lock(&consumer_data.lock); | |
7ad0a0cb MD |
445 | /* Steal stream identifier, for UST */ |
446 | consumer_steal_stream_key(stream->key); | |
c77fc10a | 447 | |
b0b335c8 | 448 | rcu_read_lock(); |
c77fc10a DG |
449 | lttng_ht_lookup(consumer_data.stream_ht, |
450 | (void *)((unsigned long) stream->key), &iter); | |
451 | node = lttng_ht_iter_get_node_ulong(&iter); | |
452 | if (node != NULL) { | |
453 | rcu_read_unlock(); | |
454 | /* Stream already exist. Ignore the insertion */ | |
455 | goto end; | |
456 | } | |
457 | ||
04253271 | 458 | lttng_ht_add_unique_ulong(consumer_data.stream_ht, &stream->node); |
00e2e675 DG |
459 | |
460 | /* Check and cleanup relayd */ | |
461 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
462 | if (relayd != NULL) { | |
b0b335c8 | 463 | uatomic_inc(&relayd->refcount); |
00e2e675 | 464 | } |
b0b335c8 | 465 | rcu_read_unlock(); |
00e2e675 DG |
466 | |
467 | /* Update consumer data */ | |
3bd1e081 MD |
468 | consumer_data.stream_count++; |
469 | consumer_data.need_update = 1; | |
470 | ||
3bd1e081 MD |
471 | end: |
472 | pthread_mutex_unlock(&consumer_data.lock); | |
702b1ea4 | 473 | |
3bd1e081 MD |
474 | return ret; |
475 | } | |
476 | ||
00e2e675 | 477 | /* |
3f8e211f DG |
478 | * Add relayd socket to global consumer data hashtable. RCU read side lock MUST |
479 | * be acquired before calling this. | |
00e2e675 | 480 | */ |
d09e1200 | 481 | static int add_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
482 | { |
483 | int ret = 0; | |
484 | struct lttng_ht_node_ulong *node; | |
485 | struct lttng_ht_iter iter; | |
486 | ||
487 | if (relayd == NULL) { | |
488 | ret = -1; | |
489 | goto end; | |
490 | } | |
491 | ||
00e2e675 DG |
492 | lttng_ht_lookup(consumer_data.relayd_ht, |
493 | (void *)((unsigned long) relayd->net_seq_idx), &iter); | |
494 | node = lttng_ht_iter_get_node_ulong(&iter); | |
495 | if (node != NULL) { | |
00e2e675 DG |
496 | /* Relayd already exist. Ignore the insertion */ |
497 | goto end; | |
498 | } | |
499 | lttng_ht_add_unique_ulong(consumer_data.relayd_ht, &relayd->node); | |
500 | ||
00e2e675 DG |
501 | end: |
502 | return ret; | |
503 | } | |
504 | ||
505 | /* | |
506 | * Allocate and return a consumer relayd socket. | |
507 | */ | |
508 | struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( | |
509 | int net_seq_idx) | |
510 | { | |
511 | struct consumer_relayd_sock_pair *obj = NULL; | |
512 | ||
513 | /* Negative net sequence index is a failure */ | |
514 | if (net_seq_idx < 0) { | |
515 | goto error; | |
516 | } | |
517 | ||
518 | obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); | |
519 | if (obj == NULL) { | |
520 | PERROR("zmalloc relayd sock"); | |
521 | goto error; | |
522 | } | |
523 | ||
524 | obj->net_seq_idx = net_seq_idx; | |
525 | obj->refcount = 0; | |
173af62f | 526 | obj->destroy_flag = 0; |
00e2e675 DG |
527 | lttng_ht_node_init_ulong(&obj->node, obj->net_seq_idx); |
528 | pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); | |
529 | ||
530 | error: | |
531 | return obj; | |
532 | } | |
533 | ||
534 | /* | |
535 | * Find a relayd socket pair in the global consumer data. | |
536 | * | |
537 | * Return the object if found else NULL. | |
b0b335c8 MD |
538 | * RCU read-side lock must be held across this call and while using the |
539 | * returned object. | |
00e2e675 DG |
540 | */ |
541 | struct consumer_relayd_sock_pair *consumer_find_relayd(int key) | |
542 | { | |
543 | struct lttng_ht_iter iter; | |
544 | struct lttng_ht_node_ulong *node; | |
545 | struct consumer_relayd_sock_pair *relayd = NULL; | |
546 | ||
547 | /* Negative keys are lookup failures */ | |
548 | if (key < 0) { | |
549 | goto error; | |
550 | } | |
551 | ||
00e2e675 DG |
552 | lttng_ht_lookup(consumer_data.relayd_ht, (void *)((unsigned long) key), |
553 | &iter); | |
554 | node = lttng_ht_iter_get_node_ulong(&iter); | |
555 | if (node != NULL) { | |
556 | relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
557 | } | |
558 | ||
00e2e675 DG |
559 | error: |
560 | return relayd; | |
561 | } | |
562 | ||
563 | /* | |
564 | * Handle stream for relayd transmission if the stream applies for network | |
565 | * streaming where the net sequence index is set. | |
566 | * | |
567 | * Return destination file descriptor or negative value on error. | |
568 | */ | |
6197aea7 | 569 | static int write_relayd_stream_header(struct lttng_consumer_stream *stream, |
1d4dfdef DG |
570 | size_t data_size, unsigned long padding, |
571 | struct consumer_relayd_sock_pair *relayd) | |
00e2e675 DG |
572 | { |
573 | int outfd = -1, ret; | |
00e2e675 DG |
574 | struct lttcomm_relayd_data_hdr data_hdr; |
575 | ||
576 | /* Safety net */ | |
577 | assert(stream); | |
6197aea7 | 578 | assert(relayd); |
00e2e675 DG |
579 | |
580 | /* Reset data header */ | |
581 | memset(&data_hdr, 0, sizeof(data_hdr)); | |
582 | ||
00e2e675 DG |
583 | if (stream->metadata_flag) { |
584 | /* Caller MUST acquire the relayd control socket lock */ | |
585 | ret = relayd_send_metadata(&relayd->control_sock, data_size); | |
586 | if (ret < 0) { | |
587 | goto error; | |
588 | } | |
589 | ||
590 | /* Metadata are always sent on the control socket. */ | |
591 | outfd = relayd->control_sock.fd; | |
592 | } else { | |
593 | /* Set header with stream information */ | |
594 | data_hdr.stream_id = htobe64(stream->relayd_stream_id); | |
595 | data_hdr.data_size = htobe32(data_size); | |
1d4dfdef | 596 | data_hdr.padding_size = htobe32(padding); |
173af62f | 597 | data_hdr.net_seq_num = htobe64(stream->next_net_seq_num++); |
00e2e675 DG |
598 | /* Other fields are zeroed previously */ |
599 | ||
600 | ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, | |
601 | sizeof(data_hdr)); | |
602 | if (ret < 0) { | |
603 | goto error; | |
604 | } | |
605 | ||
606 | /* Set to go on data socket */ | |
607 | outfd = relayd->data_sock.fd; | |
608 | } | |
609 | ||
610 | error: | |
611 | return outfd; | |
612 | } | |
613 | ||
3bd1e081 MD |
614 | /* |
615 | * Update a stream according to what we just received. | |
616 | */ | |
617 | void consumer_change_stream_state(int stream_key, | |
618 | enum lttng_consumer_stream_state state) | |
619 | { | |
620 | struct lttng_consumer_stream *stream; | |
621 | ||
622 | pthread_mutex_lock(&consumer_data.lock); | |
623 | stream = consumer_find_stream(stream_key); | |
624 | if (stream) { | |
625 | stream->state = state; | |
626 | } | |
627 | consumer_data.need_update = 1; | |
628 | pthread_mutex_unlock(&consumer_data.lock); | |
629 | } | |
630 | ||
702b1ea4 MD |
631 | static |
632 | void consumer_free_channel(struct rcu_head *head) | |
633 | { | |
634 | struct lttng_ht_node_ulong *node = | |
635 | caa_container_of(head, struct lttng_ht_node_ulong, head); | |
636 | struct lttng_consumer_channel *channel = | |
637 | caa_container_of(node, struct lttng_consumer_channel, node); | |
638 | ||
639 | free(channel); | |
640 | } | |
641 | ||
3bd1e081 MD |
642 | /* |
643 | * Remove a channel from the global list protected by a mutex. This | |
644 | * function is also responsible for freeing its data structures. | |
645 | */ | |
646 | void consumer_del_channel(struct lttng_consumer_channel *channel) | |
647 | { | |
648 | int ret; | |
e4421fec | 649 | struct lttng_ht_iter iter; |
3bd1e081 MD |
650 | |
651 | pthread_mutex_lock(&consumer_data.lock); | |
652 | ||
653 | switch (consumer_data.type) { | |
654 | case LTTNG_CONSUMER_KERNEL: | |
655 | break; | |
7753dea8 MD |
656 | case LTTNG_CONSUMER32_UST: |
657 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
658 | lttng_ustconsumer_del_channel(channel); |
659 | break; | |
660 | default: | |
661 | ERR("Unknown consumer_data type"); | |
662 | assert(0); | |
663 | goto end; | |
664 | } | |
665 | ||
6065ceec | 666 | rcu_read_lock(); |
04253271 MD |
667 | iter.iter.node = &channel->node.node; |
668 | ret = lttng_ht_del(consumer_data.channel_ht, &iter); | |
669 | assert(!ret); | |
6065ceec DG |
670 | rcu_read_unlock(); |
671 | ||
3bd1e081 MD |
672 | if (channel->mmap_base != NULL) { |
673 | ret = munmap(channel->mmap_base, channel->mmap_len); | |
674 | if (ret != 0) { | |
7a57cf92 | 675 | PERROR("munmap"); |
3bd1e081 MD |
676 | } |
677 | } | |
b5c5fc29 | 678 | if (channel->wait_fd >= 0 && !channel->wait_fd_is_copy) { |
4c462e79 MD |
679 | ret = close(channel->wait_fd); |
680 | if (ret) { | |
681 | PERROR("close"); | |
682 | } | |
3bd1e081 | 683 | } |
2c1dd183 | 684 | if (channel->shm_fd >= 0 && channel->wait_fd != channel->shm_fd) { |
4c462e79 MD |
685 | ret = close(channel->shm_fd); |
686 | if (ret) { | |
687 | PERROR("close"); | |
688 | } | |
3bd1e081 | 689 | } |
702b1ea4 MD |
690 | |
691 | call_rcu(&channel->node.head, consumer_free_channel); | |
3bd1e081 MD |
692 | end: |
693 | pthread_mutex_unlock(&consumer_data.lock); | |
694 | } | |
695 | ||
696 | struct lttng_consumer_channel *consumer_allocate_channel( | |
697 | int channel_key, | |
698 | int shm_fd, int wait_fd, | |
699 | uint64_t mmap_len, | |
c30aaa51 MD |
700 | uint64_t max_sb_size, |
701 | unsigned int nb_init_streams) | |
3bd1e081 MD |
702 | { |
703 | struct lttng_consumer_channel *channel; | |
704 | int ret; | |
705 | ||
276b26d1 | 706 | channel = zmalloc(sizeof(*channel)); |
3bd1e081 | 707 | if (channel == NULL) { |
7a57cf92 | 708 | PERROR("malloc struct lttng_consumer_channel"); |
3bd1e081 MD |
709 | goto end; |
710 | } | |
711 | channel->key = channel_key; | |
712 | channel->shm_fd = shm_fd; | |
713 | channel->wait_fd = wait_fd; | |
714 | channel->mmap_len = mmap_len; | |
715 | channel->max_sb_size = max_sb_size; | |
716 | channel->refcount = 0; | |
c30aaa51 | 717 | channel->nb_init_streams = nb_init_streams; |
e4421fec | 718 | lttng_ht_node_init_ulong(&channel->node, channel->key); |
3bd1e081 MD |
719 | |
720 | switch (consumer_data.type) { | |
721 | case LTTNG_CONSUMER_KERNEL: | |
722 | channel->mmap_base = NULL; | |
723 | channel->mmap_len = 0; | |
724 | break; | |
7753dea8 MD |
725 | case LTTNG_CONSUMER32_UST: |
726 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
727 | ret = lttng_ustconsumer_allocate_channel(channel); |
728 | if (ret) { | |
729 | free(channel); | |
730 | return NULL; | |
731 | } | |
732 | break; | |
733 | default: | |
734 | ERR("Unknown consumer_data type"); | |
735 | assert(0); | |
736 | goto end; | |
737 | } | |
738 | DBG("Allocated channel (key %d, shm_fd %d, wait_fd %d, mmap_len %llu, max_sb_size %llu)", | |
00e2e675 | 739 | channel->key, channel->shm_fd, channel->wait_fd, |
3bd1e081 MD |
740 | (unsigned long long) channel->mmap_len, |
741 | (unsigned long long) channel->max_sb_size); | |
742 | end: | |
743 | return channel; | |
744 | } | |
745 | ||
746 | /* | |
747 | * Add a channel to the global list protected by a mutex. | |
748 | */ | |
749 | int consumer_add_channel(struct lttng_consumer_channel *channel) | |
750 | { | |
c77fc10a DG |
751 | struct lttng_ht_node_ulong *node; |
752 | struct lttng_ht_iter iter; | |
753 | ||
3bd1e081 | 754 | pthread_mutex_lock(&consumer_data.lock); |
7ad0a0cb MD |
755 | /* Steal channel identifier, for UST */ |
756 | consumer_steal_channel_key(channel->key); | |
6065ceec | 757 | rcu_read_lock(); |
c77fc10a DG |
758 | |
759 | lttng_ht_lookup(consumer_data.channel_ht, | |
760 | (void *)((unsigned long) channel->key), &iter); | |
761 | node = lttng_ht_iter_get_node_ulong(&iter); | |
762 | if (node != NULL) { | |
763 | /* Channel already exist. Ignore the insertion */ | |
764 | goto end; | |
765 | } | |
766 | ||
04253271 | 767 | lttng_ht_add_unique_ulong(consumer_data.channel_ht, &channel->node); |
c77fc10a DG |
768 | |
769 | end: | |
6065ceec | 770 | rcu_read_unlock(); |
3bd1e081 | 771 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 772 | |
7ad0a0cb | 773 | return 0; |
3bd1e081 MD |
774 | } |
775 | ||
776 | /* | |
777 | * Allocate the pollfd structure and the local view of the out fds to avoid | |
778 | * doing a lookup in the linked list and concurrency issues when writing is | |
779 | * needed. Called with consumer_data.lock held. | |
780 | * | |
781 | * Returns the number of fds in the structures. | |
782 | */ | |
783 | int consumer_update_poll_array( | |
784 | struct lttng_consumer_local_data *ctx, struct pollfd **pollfd, | |
fb3a43a9 | 785 | struct lttng_consumer_stream **local_stream) |
3bd1e081 | 786 | { |
3bd1e081 | 787 | int i = 0; |
e4421fec DG |
788 | struct lttng_ht_iter iter; |
789 | struct lttng_consumer_stream *stream; | |
3bd1e081 MD |
790 | |
791 | DBG("Updating poll fd array"); | |
481d6c57 | 792 | rcu_read_lock(); |
e4421fec DG |
793 | cds_lfht_for_each_entry(consumer_data.stream_ht->ht, &iter.iter, stream, |
794 | node.node) { | |
795 | if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM) { | |
3bd1e081 MD |
796 | continue; |
797 | } | |
e4421fec DG |
798 | DBG("Active FD %d", stream->wait_fd); |
799 | (*pollfd)[i].fd = stream->wait_fd; | |
3bd1e081 | 800 | (*pollfd)[i].events = POLLIN | POLLPRI; |
e4421fec | 801 | local_stream[i] = stream; |
3bd1e081 MD |
802 | i++; |
803 | } | |
481d6c57 | 804 | rcu_read_unlock(); |
3bd1e081 MD |
805 | |
806 | /* | |
807 | * Insert the consumer_poll_pipe at the end of the array and don't | |
808 | * increment i so nb_fd is the number of real FD. | |
809 | */ | |
810 | (*pollfd)[i].fd = ctx->consumer_poll_pipe[0]; | |
509bb1cf | 811 | (*pollfd)[i].events = POLLIN | POLLPRI; |
3bd1e081 MD |
812 | return i; |
813 | } | |
814 | ||
815 | /* | |
816 | * Poll on the should_quit pipe and the command socket return -1 on error and | |
817 | * should exit, 0 if data is available on the command socket | |
818 | */ | |
819 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) | |
820 | { | |
821 | int num_rdy; | |
822 | ||
88f2b785 | 823 | restart: |
3bd1e081 MD |
824 | num_rdy = poll(consumer_sockpoll, 2, -1); |
825 | if (num_rdy == -1) { | |
88f2b785 MD |
826 | /* |
827 | * Restart interrupted system call. | |
828 | */ | |
829 | if (errno == EINTR) { | |
830 | goto restart; | |
831 | } | |
7a57cf92 | 832 | PERROR("Poll error"); |
3bd1e081 MD |
833 | goto exit; |
834 | } | |
509bb1cf | 835 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
3bd1e081 MD |
836 | DBG("consumer_should_quit wake up"); |
837 | goto exit; | |
838 | } | |
839 | return 0; | |
840 | ||
841 | exit: | |
842 | return -1; | |
843 | } | |
844 | ||
845 | /* | |
846 | * Set the error socket. | |
847 | */ | |
848 | void lttng_consumer_set_error_sock( | |
849 | struct lttng_consumer_local_data *ctx, int sock) | |
850 | { | |
851 | ctx->consumer_error_socket = sock; | |
852 | } | |
853 | ||
854 | /* | |
855 | * Set the command socket path. | |
856 | */ | |
3bd1e081 MD |
857 | void lttng_consumer_set_command_sock_path( |
858 | struct lttng_consumer_local_data *ctx, char *sock) | |
859 | { | |
860 | ctx->consumer_command_sock_path = sock; | |
861 | } | |
862 | ||
863 | /* | |
864 | * Send return code to the session daemon. | |
865 | * If the socket is not defined, we return 0, it is not a fatal error | |
866 | */ | |
867 | int lttng_consumer_send_error( | |
868 | struct lttng_consumer_local_data *ctx, int cmd) | |
869 | { | |
870 | if (ctx->consumer_error_socket > 0) { | |
871 | return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd, | |
872 | sizeof(enum lttcomm_sessiond_command)); | |
873 | } | |
874 | ||
875 | return 0; | |
876 | } | |
877 | ||
878 | /* | |
879 | * Close all the tracefiles and stream fds, should be called when all instances | |
880 | * are destroyed. | |
881 | */ | |
882 | void lttng_consumer_cleanup(void) | |
883 | { | |
e4421fec | 884 | struct lttng_ht_iter iter; |
6065ceec DG |
885 | struct lttng_ht_node_ulong *node; |
886 | ||
887 | rcu_read_lock(); | |
3bd1e081 MD |
888 | |
889 | /* | |
6065ceec DG |
890 | * close all outfd. Called when there are no more threads running (after |
891 | * joining on the threads), no need to protect list iteration with mutex. | |
3bd1e081 | 892 | */ |
6065ceec DG |
893 | cds_lfht_for_each_entry(consumer_data.stream_ht->ht, &iter.iter, node, |
894 | node) { | |
702b1ea4 MD |
895 | struct lttng_consumer_stream *stream = |
896 | caa_container_of(node, struct lttng_consumer_stream, node); | |
897 | consumer_del_stream(stream); | |
3bd1e081 | 898 | } |
e4421fec | 899 | |
6065ceec DG |
900 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, node, |
901 | node) { | |
702b1ea4 MD |
902 | struct lttng_consumer_channel *channel = |
903 | caa_container_of(node, struct lttng_consumer_channel, node); | |
904 | consumer_del_channel(channel); | |
3bd1e081 | 905 | } |
6065ceec DG |
906 | |
907 | rcu_read_unlock(); | |
d6ce1df2 MD |
908 | |
909 | lttng_ht_destroy(consumer_data.stream_ht); | |
910 | lttng_ht_destroy(consumer_data.channel_ht); | |
3bd1e081 MD |
911 | } |
912 | ||
913 | /* | |
914 | * Called from signal handler. | |
915 | */ | |
916 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) | |
917 | { | |
918 | int ret; | |
919 | consumer_quit = 1; | |
6f94560a MD |
920 | do { |
921 | ret = write(ctx->consumer_should_quit[1], "4", 1); | |
922 | } while (ret < 0 && errno == EINTR); | |
3bd1e081 | 923 | if (ret < 0) { |
7a57cf92 | 924 | PERROR("write consumer quit"); |
3bd1e081 MD |
925 | } |
926 | } | |
927 | ||
00e2e675 DG |
928 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
929 | off_t orig_offset) | |
3bd1e081 MD |
930 | { |
931 | int outfd = stream->out_fd; | |
932 | ||
933 | /* | |
934 | * This does a blocking write-and-wait on any page that belongs to the | |
935 | * subbuffer prior to the one we just wrote. | |
936 | * Don't care about error values, as these are just hints and ways to | |
937 | * limit the amount of page cache used. | |
938 | */ | |
939 | if (orig_offset < stream->chan->max_sb_size) { | |
940 | return; | |
941 | } | |
b9182dd9 | 942 | lttng_sync_file_range(outfd, orig_offset - stream->chan->max_sb_size, |
3bd1e081 MD |
943 | stream->chan->max_sb_size, |
944 | SYNC_FILE_RANGE_WAIT_BEFORE | |
945 | | SYNC_FILE_RANGE_WRITE | |
946 | | SYNC_FILE_RANGE_WAIT_AFTER); | |
947 | /* | |
948 | * Give hints to the kernel about how we access the file: | |
949 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after | |
950 | * we write it. | |
951 | * | |
952 | * We need to call fadvise again after the file grows because the | |
953 | * kernel does not seem to apply fadvise to non-existing parts of the | |
954 | * file. | |
955 | * | |
956 | * Call fadvise _after_ having waited for the page writeback to | |
957 | * complete because the dirty page writeback semantic is not well | |
958 | * defined. So it can be expected to lead to lower throughput in | |
959 | * streaming. | |
960 | */ | |
961 | posix_fadvise(outfd, orig_offset - stream->chan->max_sb_size, | |
962 | stream->chan->max_sb_size, POSIX_FADV_DONTNEED); | |
963 | } | |
964 | ||
965 | /* | |
966 | * Initialise the necessary environnement : | |
967 | * - create a new context | |
968 | * - create the poll_pipe | |
969 | * - create the should_quit pipe (for signal handler) | |
970 | * - create the thread pipe (for splice) | |
971 | * | |
972 | * Takes a function pointer as argument, this function is called when data is | |
973 | * available on a buffer. This function is responsible to do the | |
974 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the | |
975 | * buffer configuration and then kernctl_put_next_subbuf at the end. | |
976 | * | |
977 | * Returns a pointer to the new context or NULL on error. | |
978 | */ | |
979 | struct lttng_consumer_local_data *lttng_consumer_create( | |
980 | enum lttng_consumer_type type, | |
4078b776 | 981 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
d41f73b7 | 982 | struct lttng_consumer_local_data *ctx), |
3bd1e081 MD |
983 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
984 | int (*recv_stream)(struct lttng_consumer_stream *stream), | |
985 | int (*update_stream)(int stream_key, uint32_t state)) | |
986 | { | |
987 | int ret, i; | |
988 | struct lttng_consumer_local_data *ctx; | |
989 | ||
990 | assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || | |
991 | consumer_data.type == type); | |
992 | consumer_data.type = type; | |
993 | ||
effcf122 | 994 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
3bd1e081 | 995 | if (ctx == NULL) { |
7a57cf92 | 996 | PERROR("allocating context"); |
3bd1e081 MD |
997 | goto error; |
998 | } | |
999 | ||
1000 | ctx->consumer_error_socket = -1; | |
1001 | /* assign the callbacks */ | |
1002 | ctx->on_buffer_ready = buffer_ready; | |
1003 | ctx->on_recv_channel = recv_channel; | |
1004 | ctx->on_recv_stream = recv_stream; | |
1005 | ctx->on_update_stream = update_stream; | |
1006 | ||
1007 | ret = pipe(ctx->consumer_poll_pipe); | |
1008 | if (ret < 0) { | |
7a57cf92 | 1009 | PERROR("Error creating poll pipe"); |
3bd1e081 MD |
1010 | goto error_poll_pipe; |
1011 | } | |
1012 | ||
04fdd819 MD |
1013 | /* set read end of the pipe to non-blocking */ |
1014 | ret = fcntl(ctx->consumer_poll_pipe[0], F_SETFL, O_NONBLOCK); | |
1015 | if (ret < 0) { | |
7a57cf92 | 1016 | PERROR("fcntl O_NONBLOCK"); |
04fdd819 MD |
1017 | goto error_poll_fcntl; |
1018 | } | |
1019 | ||
1020 | /* set write end of the pipe to non-blocking */ | |
1021 | ret = fcntl(ctx->consumer_poll_pipe[1], F_SETFL, O_NONBLOCK); | |
1022 | if (ret < 0) { | |
7a57cf92 | 1023 | PERROR("fcntl O_NONBLOCK"); |
04fdd819 MD |
1024 | goto error_poll_fcntl; |
1025 | } | |
1026 | ||
3bd1e081 MD |
1027 | ret = pipe(ctx->consumer_should_quit); |
1028 | if (ret < 0) { | |
7a57cf92 | 1029 | PERROR("Error creating recv pipe"); |
3bd1e081 MD |
1030 | goto error_quit_pipe; |
1031 | } | |
1032 | ||
1033 | ret = pipe(ctx->consumer_thread_pipe); | |
1034 | if (ret < 0) { | |
7a57cf92 | 1035 | PERROR("Error creating thread pipe"); |
3bd1e081 MD |
1036 | goto error_thread_pipe; |
1037 | } | |
1038 | ||
fb3a43a9 DG |
1039 | ret = utils_create_pipe(ctx->consumer_metadata_pipe); |
1040 | if (ret < 0) { | |
1041 | goto error_metadata_pipe; | |
1042 | } | |
3bd1e081 | 1043 | |
fb3a43a9 DG |
1044 | ret = utils_create_pipe(ctx->consumer_splice_metadata_pipe); |
1045 | if (ret < 0) { | |
1046 | goto error_splice_pipe; | |
1047 | } | |
1048 | ||
1049 | return ctx; | |
3bd1e081 | 1050 | |
fb3a43a9 DG |
1051 | error_splice_pipe: |
1052 | utils_close_pipe(ctx->consumer_metadata_pipe); | |
1053 | error_metadata_pipe: | |
1054 | utils_close_pipe(ctx->consumer_thread_pipe); | |
3bd1e081 MD |
1055 | error_thread_pipe: |
1056 | for (i = 0; i < 2; i++) { | |
1057 | int err; | |
1058 | ||
1059 | err = close(ctx->consumer_should_quit[i]); | |
4c462e79 MD |
1060 | if (err) { |
1061 | PERROR("close"); | |
1062 | } | |
3bd1e081 | 1063 | } |
04fdd819 | 1064 | error_poll_fcntl: |
3bd1e081 MD |
1065 | error_quit_pipe: |
1066 | for (i = 0; i < 2; i++) { | |
1067 | int err; | |
1068 | ||
1069 | err = close(ctx->consumer_poll_pipe[i]); | |
4c462e79 MD |
1070 | if (err) { |
1071 | PERROR("close"); | |
1072 | } | |
3bd1e081 MD |
1073 | } |
1074 | error_poll_pipe: | |
1075 | free(ctx); | |
1076 | error: | |
1077 | return NULL; | |
1078 | } | |
1079 | ||
1080 | /* | |
1081 | * Close all fds associated with the instance and free the context. | |
1082 | */ | |
1083 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) | |
1084 | { | |
4c462e79 MD |
1085 | int ret; |
1086 | ||
1087 | ret = close(ctx->consumer_error_socket); | |
1088 | if (ret) { | |
1089 | PERROR("close"); | |
1090 | } | |
1091 | ret = close(ctx->consumer_thread_pipe[0]); | |
1092 | if (ret) { | |
1093 | PERROR("close"); | |
1094 | } | |
1095 | ret = close(ctx->consumer_thread_pipe[1]); | |
1096 | if (ret) { | |
1097 | PERROR("close"); | |
1098 | } | |
1099 | ret = close(ctx->consumer_poll_pipe[0]); | |
1100 | if (ret) { | |
1101 | PERROR("close"); | |
1102 | } | |
1103 | ret = close(ctx->consumer_poll_pipe[1]); | |
1104 | if (ret) { | |
1105 | PERROR("close"); | |
1106 | } | |
1107 | ret = close(ctx->consumer_should_quit[0]); | |
1108 | if (ret) { | |
1109 | PERROR("close"); | |
1110 | } | |
1111 | ret = close(ctx->consumer_should_quit[1]); | |
1112 | if (ret) { | |
1113 | PERROR("close"); | |
1114 | } | |
fb3a43a9 DG |
1115 | utils_close_pipe(ctx->consumer_splice_metadata_pipe); |
1116 | ||
3bd1e081 MD |
1117 | unlink(ctx->consumer_command_sock_path); |
1118 | free(ctx); | |
1119 | } | |
1120 | ||
6197aea7 DG |
1121 | /* |
1122 | * Write the metadata stream id on the specified file descriptor. | |
1123 | */ | |
1124 | static int write_relayd_metadata_id(int fd, | |
1125 | struct lttng_consumer_stream *stream, | |
1d4dfdef DG |
1126 | struct consumer_relayd_sock_pair *relayd, |
1127 | unsigned long padding) | |
6197aea7 DG |
1128 | { |
1129 | int ret; | |
1d4dfdef | 1130 | struct lttcomm_relayd_metadata_payload hdr; |
6197aea7 | 1131 | |
1d4dfdef DG |
1132 | hdr.stream_id = htobe64(stream->relayd_stream_id); |
1133 | hdr.padding_size = htobe32(padding); | |
6197aea7 | 1134 | do { |
1d4dfdef | 1135 | ret = write(fd, (void *) &hdr, sizeof(hdr)); |
6197aea7 DG |
1136 | } while (ret < 0 && errno == EINTR); |
1137 | if (ret < 0) { | |
1138 | PERROR("write metadata stream id"); | |
1139 | goto end; | |
1140 | } | |
1d4dfdef DG |
1141 | DBG("Metadata stream id %" PRIu64 " with padding %lu written before data", |
1142 | stream->relayd_stream_id, padding); | |
6197aea7 DG |
1143 | |
1144 | end: | |
1145 | return ret; | |
1146 | } | |
1147 | ||
3bd1e081 | 1148 | /* |
09e26845 DG |
1149 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
1150 | * core function for writing trace buffers to either the local filesystem or | |
1151 | * the network. | |
1152 | * | |
1153 | * Careful review MUST be put if any changes occur! | |
3bd1e081 MD |
1154 | * |
1155 | * Returns the number of bytes written | |
1156 | */ | |
4078b776 | 1157 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
3bd1e081 | 1158 | struct lttng_consumer_local_data *ctx, |
1d4dfdef DG |
1159 | struct lttng_consumer_stream *stream, unsigned long len, |
1160 | unsigned long padding) | |
3bd1e081 | 1161 | { |
f02e1e8a DG |
1162 | unsigned long mmap_offset; |
1163 | ssize_t ret = 0, written = 0; | |
1164 | off_t orig_offset = stream->out_fd_offset; | |
1165 | /* Default is on the disk */ | |
1166 | int outfd = stream->out_fd; | |
f02e1e8a DG |
1167 | struct consumer_relayd_sock_pair *relayd = NULL; |
1168 | ||
1169 | /* RCU lock for the relayd pointer */ | |
1170 | rcu_read_lock(); | |
1171 | ||
1172 | /* Flag that the current stream if set for network streaming. */ | |
1173 | if (stream->net_seq_idx != -1) { | |
1174 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
1175 | if (relayd == NULL) { | |
1176 | goto end; | |
1177 | } | |
1178 | } | |
1179 | ||
1180 | /* get the offset inside the fd to mmap */ | |
3bd1e081 MD |
1181 | switch (consumer_data.type) { |
1182 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a DG |
1183 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); |
1184 | break; | |
7753dea8 MD |
1185 | case LTTNG_CONSUMER32_UST: |
1186 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a DG |
1187 | ret = lttng_ustctl_get_mmap_read_offset(stream->chan->handle, |
1188 | stream->buf, &mmap_offset); | |
1189 | break; | |
3bd1e081 MD |
1190 | default: |
1191 | ERR("Unknown consumer_data type"); | |
1192 | assert(0); | |
1193 | } | |
f02e1e8a DG |
1194 | if (ret != 0) { |
1195 | errno = -ret; | |
1196 | PERROR("tracer ctl get_mmap_read_offset"); | |
1197 | written = ret; | |
1198 | goto end; | |
1199 | } | |
b9182dd9 | 1200 | |
f02e1e8a DG |
1201 | /* Handle stream on the relayd if the output is on the network */ |
1202 | if (relayd) { | |
1203 | unsigned long netlen = len; | |
1204 | ||
1205 | /* | |
1206 | * Lock the control socket for the complete duration of the function | |
1207 | * since from this point on we will use the socket. | |
1208 | */ | |
1209 | if (stream->metadata_flag) { | |
1210 | /* Metadata requires the control socket. */ | |
1211 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1d4dfdef | 1212 | netlen += sizeof(struct lttcomm_relayd_metadata_payload); |
f02e1e8a DG |
1213 | } |
1214 | ||
1d4dfdef | 1215 | ret = write_relayd_stream_header(stream, netlen, padding, relayd); |
f02e1e8a DG |
1216 | if (ret >= 0) { |
1217 | /* Use the returned socket. */ | |
1218 | outfd = ret; | |
1219 | ||
1220 | /* Write metadata stream id before payload */ | |
1221 | if (stream->metadata_flag) { | |
1d4dfdef | 1222 | ret = write_relayd_metadata_id(outfd, stream, relayd, padding); |
f02e1e8a | 1223 | if (ret < 0) { |
f02e1e8a DG |
1224 | written = ret; |
1225 | goto end; | |
1226 | } | |
f02e1e8a DG |
1227 | } |
1228 | } | |
1229 | /* Else, use the default set before which is the filesystem. */ | |
1d4dfdef DG |
1230 | } else { |
1231 | /* No streaming, we have to set the len with the full padding */ | |
1232 | len += padding; | |
f02e1e8a DG |
1233 | } |
1234 | ||
1235 | while (len > 0) { | |
1236 | do { | |
1237 | ret = write(outfd, stream->mmap_base + mmap_offset, len); | |
1238 | } while (ret < 0 && errno == EINTR); | |
1d4dfdef | 1239 | DBG("Consumer mmap write() ret %zd (len %lu)", ret, len); |
f02e1e8a DG |
1240 | if (ret < 0) { |
1241 | PERROR("Error in file write"); | |
1242 | if (written == 0) { | |
1243 | written = ret; | |
1244 | } | |
1245 | goto end; | |
1246 | } else if (ret > len) { | |
77c7c900 | 1247 | PERROR("Error in file write (ret %zd > len %lu)", ret, len); |
f02e1e8a DG |
1248 | written += ret; |
1249 | goto end; | |
1250 | } else { | |
1251 | len -= ret; | |
1252 | mmap_offset += ret; | |
1253 | } | |
f02e1e8a DG |
1254 | |
1255 | /* This call is useless on a socket so better save a syscall. */ | |
1256 | if (!relayd) { | |
1257 | /* This won't block, but will start writeout asynchronously */ | |
1258 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret, | |
1259 | SYNC_FILE_RANGE_WRITE); | |
1260 | stream->out_fd_offset += ret; | |
1261 | } | |
1262 | written += ret; | |
1263 | } | |
1264 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1265 | ||
1266 | end: | |
1267 | /* Unlock only if ctrl socket used */ | |
1268 | if (relayd && stream->metadata_flag) { | |
1269 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1270 | } | |
1271 | ||
1272 | rcu_read_unlock(); | |
1273 | return written; | |
3bd1e081 MD |
1274 | } |
1275 | ||
1276 | /* | |
1277 | * Splice the data from the ring buffer to the tracefile. | |
1278 | * | |
1279 | * Returns the number of bytes spliced. | |
1280 | */ | |
4078b776 | 1281 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
3bd1e081 | 1282 | struct lttng_consumer_local_data *ctx, |
1d4dfdef DG |
1283 | struct lttng_consumer_stream *stream, unsigned long len, |
1284 | unsigned long padding) | |
3bd1e081 | 1285 | { |
f02e1e8a DG |
1286 | ssize_t ret = 0, written = 0, ret_splice = 0; |
1287 | loff_t offset = 0; | |
1288 | off_t orig_offset = stream->out_fd_offset; | |
1289 | int fd = stream->wait_fd; | |
1290 | /* Default is on the disk */ | |
1291 | int outfd = stream->out_fd; | |
f02e1e8a | 1292 | struct consumer_relayd_sock_pair *relayd = NULL; |
fb3a43a9 | 1293 | int *splice_pipe; |
f02e1e8a | 1294 | |
3bd1e081 MD |
1295 | switch (consumer_data.type) { |
1296 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a | 1297 | break; |
7753dea8 MD |
1298 | case LTTNG_CONSUMER32_UST: |
1299 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a | 1300 | /* Not supported for user space tracing */ |
3bd1e081 MD |
1301 | return -ENOSYS; |
1302 | default: | |
1303 | ERR("Unknown consumer_data type"); | |
1304 | assert(0); | |
3bd1e081 MD |
1305 | } |
1306 | ||
f02e1e8a DG |
1307 | /* RCU lock for the relayd pointer */ |
1308 | rcu_read_lock(); | |
1309 | ||
1310 | /* Flag that the current stream if set for network streaming. */ | |
1311 | if (stream->net_seq_idx != -1) { | |
1312 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
1313 | if (relayd == NULL) { | |
1314 | goto end; | |
1315 | } | |
1316 | } | |
1317 | ||
fb3a43a9 DG |
1318 | /* |
1319 | * Choose right pipe for splice. Metadata and trace data are handled by | |
1320 | * different threads hence the use of two pipes in order not to race or | |
1321 | * corrupt the written data. | |
1322 | */ | |
1323 | if (stream->metadata_flag) { | |
1324 | splice_pipe = ctx->consumer_splice_metadata_pipe; | |
1325 | } else { | |
1326 | splice_pipe = ctx->consumer_thread_pipe; | |
1327 | } | |
1328 | ||
f02e1e8a | 1329 | /* Write metadata stream id before payload */ |
1d4dfdef DG |
1330 | if (relayd) { |
1331 | int total_len = len; | |
f02e1e8a | 1332 | |
1d4dfdef DG |
1333 | if (stream->metadata_flag) { |
1334 | /* | |
1335 | * Lock the control socket for the complete duration of the function | |
1336 | * since from this point on we will use the socket. | |
1337 | */ | |
1338 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1339 | ||
1340 | ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd, | |
1341 | padding); | |
1342 | if (ret < 0) { | |
1343 | written = ret; | |
1344 | goto end; | |
1345 | } | |
1346 | ||
1347 | total_len += sizeof(struct lttcomm_relayd_metadata_payload); | |
1348 | } | |
1349 | ||
1350 | ret = write_relayd_stream_header(stream, total_len, padding, relayd); | |
1351 | if (ret >= 0) { | |
1352 | /* Use the returned socket. */ | |
1353 | outfd = ret; | |
1354 | } else { | |
1355 | ERR("Remote relayd disconnected. Stopping"); | |
f02e1e8a DG |
1356 | goto end; |
1357 | } | |
1d4dfdef DG |
1358 | } else { |
1359 | /* No streaming, we have to set the len with the full padding */ | |
1360 | len += padding; | |
f02e1e8a DG |
1361 | } |
1362 | ||
1363 | while (len > 0) { | |
1d4dfdef DG |
1364 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)", |
1365 | (unsigned long)offset, len, fd, splice_pipe[1]); | |
fb3a43a9 | 1366 | ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len, |
f02e1e8a DG |
1367 | SPLICE_F_MOVE | SPLICE_F_MORE); |
1368 | DBG("splice chan to pipe, ret %zd", ret_splice); | |
1369 | if (ret_splice < 0) { | |
1370 | PERROR("Error in relay splice"); | |
1371 | if (written == 0) { | |
1372 | written = ret_splice; | |
1373 | } | |
1374 | ret = errno; | |
1375 | goto splice_error; | |
1376 | } | |
1377 | ||
1378 | /* Handle stream on the relayd if the output is on the network */ | |
1379 | if (relayd) { | |
1380 | if (stream->metadata_flag) { | |
1d4dfdef DG |
1381 | size_t metadata_payload_size = |
1382 | sizeof(struct lttcomm_relayd_metadata_payload); | |
1383 | ||
f02e1e8a | 1384 | /* Update counter to fit the spliced data */ |
1d4dfdef DG |
1385 | ret_splice += metadata_payload_size; |
1386 | len += metadata_payload_size; | |
f02e1e8a DG |
1387 | /* |
1388 | * We do this so the return value can match the len passed as | |
1389 | * argument to this function. | |
1390 | */ | |
1d4dfdef | 1391 | written -= metadata_payload_size; |
f02e1e8a DG |
1392 | } |
1393 | } | |
1394 | ||
1395 | /* Splice data out */ | |
fb3a43a9 | 1396 | ret_splice = splice(splice_pipe[0], NULL, outfd, NULL, |
f02e1e8a | 1397 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); |
1d4dfdef | 1398 | DBG("Consumer splice pipe to file, ret %zd", ret_splice); |
f02e1e8a DG |
1399 | if (ret_splice < 0) { |
1400 | PERROR("Error in file splice"); | |
1401 | if (written == 0) { | |
1402 | written = ret_splice; | |
1403 | } | |
1404 | ret = errno; | |
1405 | goto splice_error; | |
1406 | } else if (ret_splice > len) { | |
1407 | errno = EINVAL; | |
1408 | PERROR("Wrote more data than requested %zd (len: %lu)", | |
1409 | ret_splice, len); | |
1410 | written += ret_splice; | |
1411 | ret = errno; | |
1412 | goto splice_error; | |
1413 | } | |
1414 | len -= ret_splice; | |
1415 | ||
1416 | /* This call is useless on a socket so better save a syscall. */ | |
1417 | if (!relayd) { | |
1418 | /* This won't block, but will start writeout asynchronously */ | |
1419 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, | |
1420 | SYNC_FILE_RANGE_WRITE); | |
1421 | stream->out_fd_offset += ret_splice; | |
1422 | } | |
1423 | written += ret_splice; | |
1424 | } | |
1425 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1426 | ||
1427 | ret = ret_splice; | |
1428 | ||
1429 | goto end; | |
1430 | ||
1431 | splice_error: | |
1432 | /* send the appropriate error description to sessiond */ | |
1433 | switch (ret) { | |
1434 | case EBADF: | |
f73fabfd | 1435 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EBADF); |
f02e1e8a DG |
1436 | break; |
1437 | case EINVAL: | |
f73fabfd | 1438 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); |
f02e1e8a DG |
1439 | break; |
1440 | case ENOMEM: | |
f73fabfd | 1441 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); |
f02e1e8a DG |
1442 | break; |
1443 | case ESPIPE: | |
f73fabfd | 1444 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); |
f02e1e8a DG |
1445 | break; |
1446 | } | |
1447 | ||
1448 | end: | |
1449 | if (relayd && stream->metadata_flag) { | |
1450 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1451 | } | |
1452 | ||
1453 | rcu_read_unlock(); | |
1454 | return written; | |
3bd1e081 MD |
1455 | } |
1456 | ||
1457 | /* | |
1458 | * Take a snapshot for a specific fd | |
1459 | * | |
1460 | * Returns 0 on success, < 0 on error | |
1461 | */ | |
1462 | int lttng_consumer_take_snapshot(struct lttng_consumer_local_data *ctx, | |
1463 | struct lttng_consumer_stream *stream) | |
1464 | { | |
1465 | switch (consumer_data.type) { | |
1466 | case LTTNG_CONSUMER_KERNEL: | |
1467 | return lttng_kconsumer_take_snapshot(ctx, stream); | |
7753dea8 MD |
1468 | case LTTNG_CONSUMER32_UST: |
1469 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1470 | return lttng_ustconsumer_take_snapshot(ctx, stream); |
1471 | default: | |
1472 | ERR("Unknown consumer_data type"); | |
1473 | assert(0); | |
1474 | return -ENOSYS; | |
1475 | } | |
1476 | ||
1477 | } | |
1478 | ||
1479 | /* | |
1480 | * Get the produced position | |
1481 | * | |
1482 | * Returns 0 on success, < 0 on error | |
1483 | */ | |
1484 | int lttng_consumer_get_produced_snapshot( | |
1485 | struct lttng_consumer_local_data *ctx, | |
1486 | struct lttng_consumer_stream *stream, | |
1487 | unsigned long *pos) | |
1488 | { | |
1489 | switch (consumer_data.type) { | |
1490 | case LTTNG_CONSUMER_KERNEL: | |
1491 | return lttng_kconsumer_get_produced_snapshot(ctx, stream, pos); | |
7753dea8 MD |
1492 | case LTTNG_CONSUMER32_UST: |
1493 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1494 | return lttng_ustconsumer_get_produced_snapshot(ctx, stream, pos); |
1495 | default: | |
1496 | ERR("Unknown consumer_data type"); | |
1497 | assert(0); | |
1498 | return -ENOSYS; | |
1499 | } | |
1500 | } | |
1501 | ||
1502 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, | |
1503 | int sock, struct pollfd *consumer_sockpoll) | |
1504 | { | |
1505 | switch (consumer_data.type) { | |
1506 | case LTTNG_CONSUMER_KERNEL: | |
1507 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
7753dea8 MD |
1508 | case LTTNG_CONSUMER32_UST: |
1509 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1510 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
1511 | default: | |
1512 | ERR("Unknown consumer_data type"); | |
1513 | assert(0); | |
1514 | return -ENOSYS; | |
1515 | } | |
1516 | } | |
1517 | ||
fb3a43a9 | 1518 | /* |
f724d81e | 1519 | * Iterate over all streams of the hashtable and free them properly. |
fb3a43a9 DG |
1520 | */ |
1521 | static void destroy_stream_ht(struct lttng_ht *ht) | |
1522 | { | |
1523 | int ret; | |
1524 | struct lttng_ht_iter iter; | |
1525 | struct lttng_consumer_stream *stream; | |
1526 | ||
1527 | if (ht == NULL) { | |
1528 | return; | |
1529 | } | |
1530 | ||
d09e1200 | 1531 | rcu_read_lock(); |
fb3a43a9 DG |
1532 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
1533 | ret = lttng_ht_del(ht, &iter); | |
1534 | assert(!ret); | |
1535 | ||
f724d81e | 1536 | call_rcu(&stream->node.head, consumer_free_stream); |
fb3a43a9 | 1537 | } |
d09e1200 | 1538 | rcu_read_unlock(); |
fb3a43a9 DG |
1539 | |
1540 | lttng_ht_destroy(ht); | |
1541 | } | |
1542 | ||
1543 | /* | |
1544 | * Clean up a metadata stream and free its memory. | |
1545 | */ | |
1546 | static void consumer_del_metadata_stream(struct lttng_consumer_stream *stream) | |
1547 | { | |
1548 | int ret; | |
fb3a43a9 DG |
1549 | struct consumer_relayd_sock_pair *relayd; |
1550 | ||
1551 | assert(stream); | |
1552 | /* | |
1553 | * This call should NEVER receive regular stream. It must always be | |
1554 | * metadata stream and this is crucial for data structure synchronization. | |
1555 | */ | |
1556 | assert(stream->metadata_flag); | |
1557 | ||
1558 | pthread_mutex_lock(&consumer_data.lock); | |
1559 | switch (consumer_data.type) { | |
1560 | case LTTNG_CONSUMER_KERNEL: | |
1561 | if (stream->mmap_base != NULL) { | |
1562 | ret = munmap(stream->mmap_base, stream->mmap_len); | |
1563 | if (ret != 0) { | |
1564 | PERROR("munmap metadata stream"); | |
1565 | } | |
1566 | } | |
1567 | break; | |
1568 | case LTTNG_CONSUMER32_UST: | |
1569 | case LTTNG_CONSUMER64_UST: | |
1570 | lttng_ustconsumer_del_stream(stream); | |
1571 | break; | |
1572 | default: | |
1573 | ERR("Unknown consumer_data type"); | |
1574 | assert(0); | |
1575 | } | |
1576 | pthread_mutex_unlock(&consumer_data.lock); | |
1577 | ||
1578 | if (stream->out_fd >= 0) { | |
1579 | ret = close(stream->out_fd); | |
1580 | if (ret) { | |
1581 | PERROR("close"); | |
1582 | } | |
1583 | } | |
1584 | ||
1585 | if (stream->wait_fd >= 0 && !stream->wait_fd_is_copy) { | |
1586 | ret = close(stream->wait_fd); | |
1587 | if (ret) { | |
1588 | PERROR("close"); | |
1589 | } | |
1590 | } | |
1591 | ||
1592 | if (stream->shm_fd >= 0 && stream->wait_fd != stream->shm_fd) { | |
1593 | ret = close(stream->shm_fd); | |
1594 | if (ret) { | |
1595 | PERROR("close"); | |
1596 | } | |
1597 | } | |
1598 | ||
1599 | /* Check and cleanup relayd */ | |
1600 | rcu_read_lock(); | |
1601 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
1602 | if (relayd != NULL) { | |
1603 | uatomic_dec(&relayd->refcount); | |
1604 | assert(uatomic_read(&relayd->refcount) >= 0); | |
1605 | ||
1606 | /* Closing streams requires to lock the control socket. */ | |
1607 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1608 | ret = relayd_send_close_stream(&relayd->control_sock, | |
1609 | stream->relayd_stream_id, stream->next_net_seq_num - 1); | |
1610 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1611 | if (ret < 0) { | |
1612 | DBG("Unable to close stream on the relayd. Continuing"); | |
1613 | /* | |
1614 | * Continue here. There is nothing we can do for the relayd. | |
1615 | * Chances are that the relayd has closed the socket so we just | |
1616 | * continue cleaning up. | |
1617 | */ | |
1618 | } | |
1619 | ||
1620 | /* Both conditions are met, we destroy the relayd. */ | |
1621 | if (uatomic_read(&relayd->refcount) == 0 && | |
1622 | uatomic_read(&relayd->destroy_flag)) { | |
d09e1200 | 1623 | destroy_relayd(relayd); |
fb3a43a9 DG |
1624 | } |
1625 | } | |
1626 | rcu_read_unlock(); | |
1627 | ||
1628 | /* Atomically decrement channel refcount since other threads can use it. */ | |
1629 | uatomic_dec(&stream->chan->refcount); | |
c30aaa51 MD |
1630 | if (!uatomic_read(&stream->chan->refcount) |
1631 | && !uatomic_read(&stream->chan->nb_init_streams)) { | |
1632 | /* Go for channel deletion! */ | |
1633 | consumer_del_channel(stream->chan); | |
fb3a43a9 DG |
1634 | } |
1635 | ||
f724d81e | 1636 | call_rcu(&stream->node.head, consumer_free_stream); |
fb3a43a9 DG |
1637 | } |
1638 | ||
1639 | /* | |
1640 | * Action done with the metadata stream when adding it to the consumer internal | |
1641 | * data structures to handle it. | |
1642 | */ | |
1643 | static void consumer_add_metadata_stream(struct lttng_consumer_stream *stream) | |
1644 | { | |
1645 | struct consumer_relayd_sock_pair *relayd; | |
1646 | ||
1647 | /* Find relayd and, if one is found, increment refcount. */ | |
1648 | rcu_read_lock(); | |
1649 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
1650 | if (relayd != NULL) { | |
1651 | uatomic_inc(&relayd->refcount); | |
1652 | } | |
1653 | rcu_read_unlock(); | |
1654 | } | |
1655 | ||
1656 | /* | |
1657 | * Thread polls on metadata file descriptor and write them on disk or on the | |
1658 | * network. | |
1659 | */ | |
1660 | void *lttng_consumer_thread_poll_metadata(void *data) | |
1661 | { | |
1662 | int ret, i, pollfd; | |
1663 | uint32_t revents, nb_fd; | |
1664 | struct lttng_consumer_stream *stream; | |
1665 | struct lttng_ht_iter iter; | |
1666 | struct lttng_ht_node_ulong *node; | |
1667 | struct lttng_ht *metadata_ht = NULL; | |
1668 | struct lttng_poll_event events; | |
1669 | struct lttng_consumer_local_data *ctx = data; | |
1670 | ssize_t len; | |
1671 | ||
1672 | rcu_register_thread(); | |
1673 | ||
1674 | DBG("Thread metadata poll started"); | |
1675 | ||
1676 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
1677 | if (metadata_ht == NULL) { | |
1678 | goto end; | |
1679 | } | |
1680 | ||
1681 | /* Size is set to 1 for the consumer_metadata pipe */ | |
1682 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
1683 | if (ret < 0) { | |
1684 | ERR("Poll set creation failed"); | |
1685 | goto end; | |
1686 | } | |
1687 | ||
1688 | ret = lttng_poll_add(&events, ctx->consumer_metadata_pipe[0], LPOLLIN); | |
1689 | if (ret < 0) { | |
1690 | goto end; | |
1691 | } | |
1692 | ||
1693 | /* Main loop */ | |
1694 | DBG("Metadata main loop started"); | |
1695 | ||
1696 | while (1) { | |
1697 | lttng_poll_reset(&events); | |
1698 | ||
1699 | nb_fd = LTTNG_POLL_GETNB(&events); | |
1700 | ||
1701 | /* Only the metadata pipe is set */ | |
1702 | if (nb_fd == 0 && consumer_quit == 1) { | |
1703 | goto end; | |
1704 | } | |
1705 | ||
1706 | restart: | |
1707 | DBG("Metadata poll wait with %d fd(s)", nb_fd); | |
1708 | ret = lttng_poll_wait(&events, -1); | |
1709 | DBG("Metadata event catched in thread"); | |
1710 | if (ret < 0) { | |
1711 | if (errno == EINTR) { | |
1712 | goto restart; | |
1713 | } | |
1714 | goto error; | |
1715 | } | |
1716 | ||
1717 | for (i = 0; i < nb_fd; i++) { | |
1718 | revents = LTTNG_POLL_GETEV(&events, i); | |
1719 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
1720 | ||
1721 | /* Check the metadata pipe for incoming metadata. */ | |
1722 | if (pollfd == ctx->consumer_metadata_pipe[0]) { | |
4adabd61 | 1723 | if (revents & (LPOLLERR | LPOLLHUP )) { |
fb3a43a9 DG |
1724 | DBG("Metadata thread pipe hung up"); |
1725 | /* | |
1726 | * Remove the pipe from the poll set and continue the loop | |
1727 | * since their might be data to consume. | |
1728 | */ | |
1729 | lttng_poll_del(&events, ctx->consumer_metadata_pipe[0]); | |
1730 | close(ctx->consumer_metadata_pipe[0]); | |
1731 | continue; | |
1732 | } else if (revents & LPOLLIN) { | |
fb3a43a9 | 1733 | do { |
633d0084 DG |
1734 | /* Get the stream pointer received */ |
1735 | ret = read(pollfd, &stream, sizeof(stream)); | |
fb3a43a9 | 1736 | } while (ret < 0 && errno == EINTR); |
633d0084 DG |
1737 | if (ret < 0 || |
1738 | ret < sizeof(struct lttng_consumer_stream *)) { | |
fb3a43a9 | 1739 | PERROR("read metadata stream"); |
fb3a43a9 DG |
1740 | /* |
1741 | * Let's continue here and hope we can still work | |
1742 | * without stopping the consumer. XXX: Should we? | |
1743 | */ | |
1744 | continue; | |
1745 | } | |
1746 | ||
1747 | DBG("Adding metadata stream %d to poll set", | |
1748 | stream->wait_fd); | |
1749 | ||
d09e1200 | 1750 | rcu_read_lock(); |
fb3a43a9 DG |
1751 | /* The node should be init at this point */ |
1752 | lttng_ht_add_unique_ulong(metadata_ht, | |
1753 | &stream->waitfd_node); | |
d09e1200 | 1754 | rcu_read_unlock(); |
fb3a43a9 DG |
1755 | |
1756 | /* Add metadata stream to the global poll events list */ | |
1757 | lttng_poll_add(&events, stream->wait_fd, | |
1758 | LPOLLIN | LPOLLPRI); | |
1759 | ||
1760 | consumer_add_metadata_stream(stream); | |
1761 | } | |
1762 | ||
1763 | /* Metadata pipe handled. Continue handling the others */ | |
1764 | continue; | |
1765 | } | |
1766 | ||
1767 | /* From here, the event is a metadata wait fd */ | |
1768 | ||
d09e1200 | 1769 | rcu_read_lock(); |
fb3a43a9 DG |
1770 | lttng_ht_lookup(metadata_ht, (void *)((unsigned long) pollfd), |
1771 | &iter); | |
1772 | node = lttng_ht_iter_get_node_ulong(&iter); | |
1773 | if (node == NULL) { | |
1774 | /* FD not found, continue loop */ | |
d09e1200 | 1775 | rcu_read_unlock(); |
fb3a43a9 DG |
1776 | continue; |
1777 | } | |
1778 | ||
1779 | stream = caa_container_of(node, struct lttng_consumer_stream, | |
1780 | waitfd_node); | |
1781 | ||
1782 | /* Get the data out of the metadata file descriptor */ | |
1783 | if (revents & (LPOLLIN | LPOLLPRI)) { | |
1784 | DBG("Metadata available on fd %d", pollfd); | |
1785 | assert(stream->wait_fd == pollfd); | |
1786 | ||
1787 | len = ctx->on_buffer_ready(stream, ctx); | |
1788 | /* It's ok to have an unavailable sub-buffer */ | |
1789 | if (len < 0 && len != -EAGAIN) { | |
d09e1200 | 1790 | rcu_read_unlock(); |
fb3a43a9 DG |
1791 | goto end; |
1792 | } else if (len > 0) { | |
1793 | stream->data_read = 1; | |
1794 | } | |
1795 | } | |
1796 | ||
1797 | /* | |
1798 | * Remove the stream from the hash table since there is no data | |
1799 | * left on the fd because we previously did a read on the buffer. | |
1800 | */ | |
4adabd61 | 1801 | if (revents & (LPOLLERR | LPOLLHUP)) { |
fb3a43a9 DG |
1802 | DBG("Metadata fd %d is hup|err|nval.", pollfd); |
1803 | if (!stream->hangup_flush_done | |
1804 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
1805 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
1806 | DBG("Attempting to flush and consume the UST buffers"); | |
1807 | lttng_ustconsumer_on_stream_hangup(stream); | |
1808 | ||
1809 | /* We just flushed the stream now read it. */ | |
1810 | len = ctx->on_buffer_ready(stream, ctx); | |
1811 | /* It's ok to have an unavailable sub-buffer */ | |
1812 | if (len < 0 && len != -EAGAIN) { | |
d09e1200 | 1813 | rcu_read_unlock(); |
fb3a43a9 DG |
1814 | goto end; |
1815 | } | |
1816 | } | |
1817 | ||
1818 | /* Removing it from hash table, poll set and free memory */ | |
1819 | lttng_ht_del(metadata_ht, &iter); | |
d09e1200 | 1820 | |
fb3a43a9 DG |
1821 | lttng_poll_del(&events, stream->wait_fd); |
1822 | consumer_del_metadata_stream(stream); | |
1823 | } | |
d09e1200 | 1824 | rcu_read_unlock(); |
fb3a43a9 DG |
1825 | } |
1826 | } | |
1827 | ||
1828 | error: | |
1829 | end: | |
1830 | DBG("Metadata poll thread exiting"); | |
1831 | lttng_poll_clean(&events); | |
1832 | ||
1833 | if (metadata_ht) { | |
1834 | destroy_stream_ht(metadata_ht); | |
1835 | } | |
1836 | ||
1837 | rcu_unregister_thread(); | |
1838 | return NULL; | |
1839 | } | |
1840 | ||
3bd1e081 | 1841 | /* |
e4421fec | 1842 | * This thread polls the fds in the set to consume the data and write |
3bd1e081 MD |
1843 | * it to tracefile if necessary. |
1844 | */ | |
1845 | void *lttng_consumer_thread_poll_fds(void *data) | |
1846 | { | |
1847 | int num_rdy, num_hup, high_prio, ret, i; | |
1848 | struct pollfd *pollfd = NULL; | |
1849 | /* local view of the streams */ | |
1850 | struct lttng_consumer_stream **local_stream = NULL; | |
1851 | /* local view of consumer_data.fds_count */ | |
1852 | int nb_fd = 0; | |
3bd1e081 | 1853 | struct lttng_consumer_local_data *ctx = data; |
00e2e675 | 1854 | ssize_t len; |
fb3a43a9 DG |
1855 | pthread_t metadata_thread; |
1856 | void *status; | |
3bd1e081 | 1857 | |
e7b994a3 DG |
1858 | rcu_register_thread(); |
1859 | ||
fb3a43a9 DG |
1860 | /* Start metadata polling thread */ |
1861 | ret = pthread_create(&metadata_thread, NULL, | |
1862 | lttng_consumer_thread_poll_metadata, (void *) ctx); | |
1863 | if (ret < 0) { | |
1864 | PERROR("pthread_create metadata thread"); | |
1865 | goto end; | |
1866 | } | |
1867 | ||
effcf122 | 1868 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream)); |
3bd1e081 MD |
1869 | |
1870 | while (1) { | |
1871 | high_prio = 0; | |
1872 | num_hup = 0; | |
1873 | ||
1874 | /* | |
e4421fec | 1875 | * the fds set has been updated, we need to update our |
3bd1e081 MD |
1876 | * local array as well |
1877 | */ | |
1878 | pthread_mutex_lock(&consumer_data.lock); | |
1879 | if (consumer_data.need_update) { | |
1880 | if (pollfd != NULL) { | |
1881 | free(pollfd); | |
1882 | pollfd = NULL; | |
1883 | } | |
1884 | if (local_stream != NULL) { | |
1885 | free(local_stream); | |
1886 | local_stream = NULL; | |
1887 | } | |
1888 | ||
1889 | /* allocate for all fds + 1 for the consumer_poll_pipe */ | |
effcf122 | 1890 | pollfd = zmalloc((consumer_data.stream_count + 1) * sizeof(struct pollfd)); |
3bd1e081 | 1891 | if (pollfd == NULL) { |
7a57cf92 | 1892 | PERROR("pollfd malloc"); |
3bd1e081 MD |
1893 | pthread_mutex_unlock(&consumer_data.lock); |
1894 | goto end; | |
1895 | } | |
1896 | ||
1897 | /* allocate for all fds + 1 for the consumer_poll_pipe */ | |
effcf122 | 1898 | local_stream = zmalloc((consumer_data.stream_count + 1) * |
3bd1e081 MD |
1899 | sizeof(struct lttng_consumer_stream)); |
1900 | if (local_stream == NULL) { | |
7a57cf92 | 1901 | PERROR("local_stream malloc"); |
3bd1e081 MD |
1902 | pthread_mutex_unlock(&consumer_data.lock); |
1903 | goto end; | |
1904 | } | |
fb3a43a9 | 1905 | ret = consumer_update_poll_array(ctx, &pollfd, local_stream); |
3bd1e081 MD |
1906 | if (ret < 0) { |
1907 | ERR("Error in allocating pollfd or local_outfds"); | |
f73fabfd | 1908 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
1909 | pthread_mutex_unlock(&consumer_data.lock); |
1910 | goto end; | |
1911 | } | |
1912 | nb_fd = ret; | |
1913 | consumer_data.need_update = 0; | |
1914 | } | |
1915 | pthread_mutex_unlock(&consumer_data.lock); | |
1916 | ||
4078b776 MD |
1917 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
1918 | if (nb_fd == 0 && consumer_quit == 1) { | |
1919 | goto end; | |
1920 | } | |
3bd1e081 | 1921 | /* poll on the array of fds */ |
88f2b785 | 1922 | restart: |
3bd1e081 MD |
1923 | DBG("polling on %d fd", nb_fd + 1); |
1924 | num_rdy = poll(pollfd, nb_fd + 1, consumer_poll_timeout); | |
1925 | DBG("poll num_rdy : %d", num_rdy); | |
1926 | if (num_rdy == -1) { | |
88f2b785 MD |
1927 | /* |
1928 | * Restart interrupted system call. | |
1929 | */ | |
1930 | if (errno == EINTR) { | |
1931 | goto restart; | |
1932 | } | |
7a57cf92 | 1933 | PERROR("Poll error"); |
f73fabfd | 1934 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
1935 | goto end; |
1936 | } else if (num_rdy == 0) { | |
1937 | DBG("Polling thread timed out"); | |
1938 | goto end; | |
1939 | } | |
1940 | ||
3bd1e081 | 1941 | /* |
00e2e675 DG |
1942 | * If the consumer_poll_pipe triggered poll go directly to the |
1943 | * beginning of the loop to update the array. We want to prioritize | |
1944 | * array update over low-priority reads. | |
3bd1e081 | 1945 | */ |
509bb1cf | 1946 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
04fdd819 MD |
1947 | size_t pipe_readlen; |
1948 | char tmp; | |
1949 | ||
3bd1e081 | 1950 | DBG("consumer_poll_pipe wake up"); |
04fdd819 MD |
1951 | /* Consume 1 byte of pipe data */ |
1952 | do { | |
1953 | pipe_readlen = read(ctx->consumer_poll_pipe[0], &tmp, 1); | |
1954 | } while (pipe_readlen == -1 && errno == EINTR); | |
3bd1e081 MD |
1955 | continue; |
1956 | } | |
1957 | ||
1958 | /* Take care of high priority channels first. */ | |
1959 | for (i = 0; i < nb_fd; i++) { | |
fb3a43a9 | 1960 | if (pollfd[i].revents & POLLPRI) { |
d41f73b7 MD |
1961 | DBG("Urgent read on fd %d", pollfd[i].fd); |
1962 | high_prio = 1; | |
4078b776 | 1963 | len = ctx->on_buffer_ready(local_stream[i], ctx); |
d41f73b7 | 1964 | /* it's ok to have an unavailable sub-buffer */ |
4078b776 MD |
1965 | if (len < 0 && len != -EAGAIN) { |
1966 | goto end; | |
1967 | } else if (len > 0) { | |
1968 | local_stream[i]->data_read = 1; | |
d41f73b7 | 1969 | } |
3bd1e081 MD |
1970 | } |
1971 | } | |
1972 | ||
4078b776 MD |
1973 | /* |
1974 | * If we read high prio channel in this loop, try again | |
1975 | * for more high prio data. | |
1976 | */ | |
1977 | if (high_prio) { | |
3bd1e081 MD |
1978 | continue; |
1979 | } | |
1980 | ||
1981 | /* Take care of low priority channels. */ | |
4078b776 MD |
1982 | for (i = 0; i < nb_fd; i++) { |
1983 | if ((pollfd[i].revents & POLLIN) || | |
1984 | local_stream[i]->hangup_flush_done) { | |
4078b776 MD |
1985 | DBG("Normal read on fd %d", pollfd[i].fd); |
1986 | len = ctx->on_buffer_ready(local_stream[i], ctx); | |
1987 | /* it's ok to have an unavailable sub-buffer */ | |
1988 | if (len < 0 && len != -EAGAIN) { | |
1989 | goto end; | |
1990 | } else if (len > 0) { | |
1991 | local_stream[i]->data_read = 1; | |
1992 | } | |
1993 | } | |
1994 | } | |
1995 | ||
1996 | /* Handle hangup and errors */ | |
1997 | for (i = 0; i < nb_fd; i++) { | |
1998 | if (!local_stream[i]->hangup_flush_done | |
1999 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) | |
2000 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2001 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2002 | DBG("fd %d is hup|err|nval. Attempting flush and read.", | |
2003 | pollfd[i].fd); | |
2004 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); | |
2005 | /* Attempt read again, for the data we just flushed. */ | |
2006 | local_stream[i]->data_read = 1; | |
2007 | } | |
2008 | /* | |
2009 | * If the poll flag is HUP/ERR/NVAL and we have | |
2010 | * read no data in this pass, we can remove the | |
2011 | * stream from its hash table. | |
2012 | */ | |
2013 | if ((pollfd[i].revents & POLLHUP)) { | |
2014 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); | |
2015 | if (!local_stream[i]->data_read) { | |
702b1ea4 | 2016 | consumer_del_stream(local_stream[i]); |
4078b776 MD |
2017 | num_hup++; |
2018 | } | |
2019 | } else if (pollfd[i].revents & POLLERR) { | |
2020 | ERR("Error returned in polling fd %d.", pollfd[i].fd); | |
2021 | if (!local_stream[i]->data_read) { | |
702b1ea4 | 2022 | consumer_del_stream(local_stream[i]); |
4078b776 MD |
2023 | num_hup++; |
2024 | } | |
2025 | } else if (pollfd[i].revents & POLLNVAL) { | |
2026 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); | |
2027 | if (!local_stream[i]->data_read) { | |
702b1ea4 | 2028 | consumer_del_stream(local_stream[i]); |
4078b776 | 2029 | num_hup++; |
3bd1e081 MD |
2030 | } |
2031 | } | |
4078b776 | 2032 | local_stream[i]->data_read = 0; |
3bd1e081 MD |
2033 | } |
2034 | } | |
2035 | end: | |
2036 | DBG("polling thread exiting"); | |
2037 | if (pollfd != NULL) { | |
2038 | free(pollfd); | |
2039 | pollfd = NULL; | |
2040 | } | |
2041 | if (local_stream != NULL) { | |
2042 | free(local_stream); | |
2043 | local_stream = NULL; | |
2044 | } | |
fb3a43a9 DG |
2045 | |
2046 | /* | |
2047 | * Close the write side of the pipe so epoll_wait() in | |
2048 | * lttng_consumer_thread_poll_metadata can catch it. The thread is | |
2049 | * monitoring the read side of the pipe. If we close them both, epoll_wait | |
2050 | * strangely does not return and could create a endless wait period if the | |
2051 | * pipe is the only tracked fd in the poll set. The thread will take care | |
2052 | * of closing the read side. | |
2053 | */ | |
2054 | close(ctx->consumer_metadata_pipe[1]); | |
2055 | if (ret) { | |
2056 | ret = pthread_join(metadata_thread, &status); | |
2057 | if (ret < 0) { | |
2058 | PERROR("pthread_join metadata thread"); | |
2059 | } | |
2060 | } | |
2061 | ||
e7b994a3 | 2062 | rcu_unregister_thread(); |
3bd1e081 MD |
2063 | return NULL; |
2064 | } | |
2065 | ||
2066 | /* | |
2067 | * This thread listens on the consumerd socket and receives the file | |
2068 | * descriptors from the session daemon. | |
2069 | */ | |
2070 | void *lttng_consumer_thread_receive_fds(void *data) | |
2071 | { | |
2072 | int sock, client_socket, ret; | |
2073 | /* | |
2074 | * structure to poll for incoming data on communication socket avoids | |
2075 | * making blocking sockets. | |
2076 | */ | |
2077 | struct pollfd consumer_sockpoll[2]; | |
2078 | struct lttng_consumer_local_data *ctx = data; | |
2079 | ||
e7b994a3 DG |
2080 | rcu_register_thread(); |
2081 | ||
3bd1e081 MD |
2082 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
2083 | unlink(ctx->consumer_command_sock_path); | |
2084 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); | |
2085 | if (client_socket < 0) { | |
2086 | ERR("Cannot create command socket"); | |
2087 | goto end; | |
2088 | } | |
2089 | ||
2090 | ret = lttcomm_listen_unix_sock(client_socket); | |
2091 | if (ret < 0) { | |
2092 | goto end; | |
2093 | } | |
2094 | ||
32258573 | 2095 | DBG("Sending ready command to lttng-sessiond"); |
f73fabfd | 2096 | ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); |
3bd1e081 MD |
2097 | /* return < 0 on error, but == 0 is not fatal */ |
2098 | if (ret < 0) { | |
32258573 | 2099 | ERR("Error sending ready command to lttng-sessiond"); |
3bd1e081 MD |
2100 | goto end; |
2101 | } | |
2102 | ||
2103 | ret = fcntl(client_socket, F_SETFL, O_NONBLOCK); | |
2104 | if (ret < 0) { | |
7a57cf92 | 2105 | PERROR("fcntl O_NONBLOCK"); |
3bd1e081 MD |
2106 | goto end; |
2107 | } | |
2108 | ||
2109 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ | |
2110 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; | |
2111 | consumer_sockpoll[0].events = POLLIN | POLLPRI; | |
2112 | consumer_sockpoll[1].fd = client_socket; | |
2113 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
2114 | ||
2115 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
2116 | goto end; | |
2117 | } | |
2118 | DBG("Connection on client_socket"); | |
2119 | ||
2120 | /* Blocking call, waiting for transmission */ | |
2121 | sock = lttcomm_accept_unix_sock(client_socket); | |
2122 | if (sock <= 0) { | |
2123 | WARN("On accept"); | |
2124 | goto end; | |
2125 | } | |
2126 | ret = fcntl(sock, F_SETFL, O_NONBLOCK); | |
2127 | if (ret < 0) { | |
7a57cf92 | 2128 | PERROR("fcntl O_NONBLOCK"); |
3bd1e081 MD |
2129 | goto end; |
2130 | } | |
2131 | ||
2132 | /* update the polling structure to poll on the established socket */ | |
2133 | consumer_sockpoll[1].fd = sock; | |
2134 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
2135 | ||
2136 | while (1) { | |
2137 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
2138 | goto end; | |
2139 | } | |
2140 | DBG("Incoming command on sock"); | |
2141 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
2142 | if (ret == -ENOENT) { | |
2143 | DBG("Received STOP command"); | |
2144 | goto end; | |
2145 | } | |
4cbc1a04 DG |
2146 | if (ret <= 0) { |
2147 | /* | |
2148 | * This could simply be a session daemon quitting. Don't output | |
2149 | * ERR() here. | |
2150 | */ | |
2151 | DBG("Communication interrupted on command socket"); | |
3bd1e081 MD |
2152 | goto end; |
2153 | } | |
2154 | if (consumer_quit) { | |
2155 | DBG("consumer_thread_receive_fds received quit from signal"); | |
2156 | goto end; | |
2157 | } | |
2158 | DBG("received fds on sock"); | |
2159 | } | |
2160 | end: | |
2161 | DBG("consumer_thread_receive_fds exiting"); | |
2162 | ||
2163 | /* | |
2164 | * when all fds have hung up, the polling thread | |
2165 | * can exit cleanly | |
2166 | */ | |
2167 | consumer_quit = 1; | |
2168 | ||
2169 | /* | |
2170 | * 2s of grace period, if no polling events occur during | |
2171 | * this period, the polling thread will exit even if there | |
2172 | * are still open FDs (should not happen, but safety mechanism). | |
2173 | */ | |
2174 | consumer_poll_timeout = LTTNG_CONSUMER_POLL_TIMEOUT; | |
2175 | ||
04fdd819 MD |
2176 | /* |
2177 | * Wake-up the other end by writing a null byte in the pipe | |
2178 | * (non-blocking). Important note: Because writing into the | |
2179 | * pipe is non-blocking (and therefore we allow dropping wakeup | |
2180 | * data, as long as there is wakeup data present in the pipe | |
2181 | * buffer to wake up the other end), the other end should | |
2182 | * perform the following sequence for waiting: | |
2183 | * 1) empty the pipe (reads). | |
2184 | * 2) perform update operation. | |
2185 | * 3) wait on the pipe (poll). | |
2186 | */ | |
2187 | do { | |
2188 | ret = write(ctx->consumer_poll_pipe[1], "", 1); | |
6f94560a | 2189 | } while (ret < 0 && errno == EINTR); |
e7b994a3 | 2190 | rcu_unregister_thread(); |
3bd1e081 MD |
2191 | return NULL; |
2192 | } | |
d41f73b7 | 2193 | |
4078b776 | 2194 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
d41f73b7 MD |
2195 | struct lttng_consumer_local_data *ctx) |
2196 | { | |
2197 | switch (consumer_data.type) { | |
2198 | case LTTNG_CONSUMER_KERNEL: | |
2199 | return lttng_kconsumer_read_subbuffer(stream, ctx); | |
7753dea8 MD |
2200 | case LTTNG_CONSUMER32_UST: |
2201 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
2202 | return lttng_ustconsumer_read_subbuffer(stream, ctx); |
2203 | default: | |
2204 | ERR("Unknown consumer_data type"); | |
2205 | assert(0); | |
2206 | return -ENOSYS; | |
2207 | } | |
2208 | } | |
2209 | ||
2210 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
2211 | { | |
2212 | switch (consumer_data.type) { | |
2213 | case LTTNG_CONSUMER_KERNEL: | |
2214 | return lttng_kconsumer_on_recv_stream(stream); | |
7753dea8 MD |
2215 | case LTTNG_CONSUMER32_UST: |
2216 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
2217 | return lttng_ustconsumer_on_recv_stream(stream); |
2218 | default: | |
2219 | ERR("Unknown consumer_data type"); | |
2220 | assert(0); | |
2221 | return -ENOSYS; | |
2222 | } | |
2223 | } | |
e4421fec DG |
2224 | |
2225 | /* | |
2226 | * Allocate and set consumer data hash tables. | |
2227 | */ | |
2228 | void lttng_consumer_init(void) | |
2229 | { | |
2230 | consumer_data.stream_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
2231 | consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
00e2e675 | 2232 | consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); |
e4421fec | 2233 | } |
7735ef9e DG |
2234 | |
2235 | /* | |
2236 | * Process the ADD_RELAYD command receive by a consumer. | |
2237 | * | |
2238 | * This will create a relayd socket pair and add it to the relayd hash table. | |
2239 | * The caller MUST acquire a RCU read side lock before calling it. | |
2240 | */ | |
2241 | int consumer_add_relayd_socket(int net_seq_idx, int sock_type, | |
2242 | struct lttng_consumer_local_data *ctx, int sock, | |
2243 | struct pollfd *consumer_sockpoll, struct lttcomm_sock *relayd_sock) | |
2244 | { | |
2245 | int fd, ret = -1; | |
2246 | struct consumer_relayd_sock_pair *relayd; | |
2247 | ||
2248 | DBG("Consumer adding relayd socket (idx: %d)", net_seq_idx); | |
2249 | ||
2250 | /* Get relayd reference if exists. */ | |
2251 | relayd = consumer_find_relayd(net_seq_idx); | |
2252 | if (relayd == NULL) { | |
2253 | /* Not found. Allocate one. */ | |
2254 | relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); | |
2255 | if (relayd == NULL) { | |
f73fabfd | 2256 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); |
7735ef9e DG |
2257 | goto error; |
2258 | } | |
2259 | } | |
2260 | ||
2261 | /* Poll on consumer socket. */ | |
2262 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
2263 | ret = -EINTR; | |
2264 | goto error; | |
2265 | } | |
2266 | ||
2267 | /* Get relayd socket from session daemon */ | |
2268 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); | |
2269 | if (ret != sizeof(fd)) { | |
f73fabfd | 2270 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); |
7735ef9e DG |
2271 | ret = -1; |
2272 | goto error; | |
2273 | } | |
2274 | ||
2275 | /* Copy socket information and received FD */ | |
2276 | switch (sock_type) { | |
2277 | case LTTNG_STREAM_CONTROL: | |
2278 | /* Copy received lttcomm socket */ | |
2279 | lttcomm_copy_sock(&relayd->control_sock, relayd_sock); | |
2280 | ret = lttcomm_create_sock(&relayd->control_sock); | |
2281 | if (ret < 0) { | |
2282 | goto error; | |
2283 | } | |
2284 | ||
2285 | /* Close the created socket fd which is useless */ | |
2286 | close(relayd->control_sock.fd); | |
2287 | ||
2288 | /* Assign new file descriptor */ | |
2289 | relayd->control_sock.fd = fd; | |
2290 | break; | |
2291 | case LTTNG_STREAM_DATA: | |
2292 | /* Copy received lttcomm socket */ | |
2293 | lttcomm_copy_sock(&relayd->data_sock, relayd_sock); | |
2294 | ret = lttcomm_create_sock(&relayd->data_sock); | |
2295 | if (ret < 0) { | |
2296 | goto error; | |
2297 | } | |
2298 | ||
2299 | /* Close the created socket fd which is useless */ | |
2300 | close(relayd->data_sock.fd); | |
2301 | ||
2302 | /* Assign new file descriptor */ | |
2303 | relayd->data_sock.fd = fd; | |
2304 | break; | |
2305 | default: | |
2306 | ERR("Unknown relayd socket type (%d)", sock_type); | |
2307 | goto error; | |
2308 | } | |
2309 | ||
2310 | DBG("Consumer %s socket created successfully with net idx %d (fd: %d)", | |
2311 | sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", | |
2312 | relayd->net_seq_idx, fd); | |
2313 | ||
2314 | /* | |
2315 | * Add relayd socket pair to consumer data hashtable. If object already | |
2316 | * exists or on error, the function gracefully returns. | |
2317 | */ | |
d09e1200 | 2318 | add_relayd(relayd); |
7735ef9e DG |
2319 | |
2320 | /* All good! */ | |
2321 | ret = 0; | |
2322 | ||
2323 | error: | |
2324 | return ret; | |
2325 | } |