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> |
10a8a223 | 33 | #include <common/kernel-ctl/kernel-ctl.h> |
00e2e675 | 34 | #include <common/sessiond-comm/relayd.h> |
10a8a223 DG |
35 | #include <common/sessiond-comm/sessiond-comm.h> |
36 | #include <common/kernel-consumer/kernel-consumer.h> | |
00e2e675 | 37 | #include <common/relayd/relayd.h> |
10a8a223 DG |
38 | #include <common/ust-consumer/ust-consumer.h> |
39 | ||
40 | #include "consumer.h" | |
3bd1e081 MD |
41 | |
42 | struct lttng_consumer_global_data consumer_data = { | |
3bd1e081 MD |
43 | .stream_count = 0, |
44 | .need_update = 1, | |
45 | .type = LTTNG_CONSUMER_UNKNOWN, | |
46 | }; | |
47 | ||
48 | /* timeout parameter, to control the polling thread grace period. */ | |
49 | int consumer_poll_timeout = -1; | |
50 | ||
51 | /* | |
52 | * Flag to inform the polling thread to quit when all fd hung up. Updated by | |
53 | * the consumer_thread_receive_fds when it notices that all fds has hung up. | |
54 | * Also updated by the signal handler (consumer_should_exit()). Read by the | |
55 | * polling threads. | |
56 | */ | |
57 | volatile int consumer_quit = 0; | |
58 | ||
59 | /* | |
60 | * Find a stream. The consumer_data.lock must be locked during this | |
61 | * call. | |
62 | */ | |
63 | static struct lttng_consumer_stream *consumer_find_stream(int key) | |
64 | { | |
e4421fec DG |
65 | struct lttng_ht_iter iter; |
66 | struct lttng_ht_node_ulong *node; | |
67 | struct lttng_consumer_stream *stream = NULL; | |
3bd1e081 | 68 | |
7ad0a0cb MD |
69 | /* Negative keys are lookup failures */ |
70 | if (key < 0) | |
71 | return NULL; | |
e4421fec | 72 | |
6065ceec DG |
73 | rcu_read_lock(); |
74 | ||
e4421fec DG |
75 | lttng_ht_lookup(consumer_data.stream_ht, (void *)((unsigned long) key), |
76 | &iter); | |
77 | node = lttng_ht_iter_get_node_ulong(&iter); | |
78 | if (node != NULL) { | |
79 | stream = caa_container_of(node, struct lttng_consumer_stream, node); | |
3bd1e081 | 80 | } |
e4421fec | 81 | |
6065ceec DG |
82 | rcu_read_unlock(); |
83 | ||
e4421fec | 84 | return stream; |
3bd1e081 MD |
85 | } |
86 | ||
7ad0a0cb MD |
87 | static void consumer_steal_stream_key(int key) |
88 | { | |
89 | struct lttng_consumer_stream *stream; | |
90 | ||
04253271 | 91 | rcu_read_lock(); |
7ad0a0cb | 92 | stream = consumer_find_stream(key); |
04253271 | 93 | if (stream) { |
7ad0a0cb | 94 | stream->key = -1; |
04253271 MD |
95 | /* |
96 | * We don't want the lookup to match, but we still need | |
97 | * to iterate on this stream when iterating over the hash table. Just | |
98 | * change the node key. | |
99 | */ | |
100 | stream->node.key = -1; | |
101 | } | |
102 | rcu_read_unlock(); | |
7ad0a0cb MD |
103 | } |
104 | ||
3bd1e081 MD |
105 | static struct lttng_consumer_channel *consumer_find_channel(int key) |
106 | { | |
e4421fec DG |
107 | struct lttng_ht_iter iter; |
108 | struct lttng_ht_node_ulong *node; | |
109 | struct lttng_consumer_channel *channel = NULL; | |
3bd1e081 | 110 | |
7ad0a0cb MD |
111 | /* Negative keys are lookup failures */ |
112 | if (key < 0) | |
113 | return NULL; | |
e4421fec | 114 | |
6065ceec DG |
115 | rcu_read_lock(); |
116 | ||
e4421fec DG |
117 | lttng_ht_lookup(consumer_data.channel_ht, (void *)((unsigned long) key), |
118 | &iter); | |
119 | node = lttng_ht_iter_get_node_ulong(&iter); | |
120 | if (node != NULL) { | |
121 | channel = caa_container_of(node, struct lttng_consumer_channel, node); | |
3bd1e081 | 122 | } |
e4421fec | 123 | |
6065ceec DG |
124 | rcu_read_unlock(); |
125 | ||
e4421fec | 126 | return channel; |
3bd1e081 MD |
127 | } |
128 | ||
7ad0a0cb MD |
129 | static void consumer_steal_channel_key(int key) |
130 | { | |
131 | struct lttng_consumer_channel *channel; | |
132 | ||
04253271 | 133 | rcu_read_lock(); |
7ad0a0cb | 134 | channel = consumer_find_channel(key); |
04253271 | 135 | if (channel) { |
7ad0a0cb | 136 | channel->key = -1; |
04253271 MD |
137 | /* |
138 | * We don't want the lookup to match, but we still need | |
139 | * to iterate on this channel when iterating over the hash table. Just | |
140 | * change the node key. | |
141 | */ | |
142 | channel->node.key = -1; | |
143 | } | |
144 | rcu_read_unlock(); | |
7ad0a0cb MD |
145 | } |
146 | ||
702b1ea4 MD |
147 | static |
148 | void consumer_free_stream(struct rcu_head *head) | |
149 | { | |
150 | struct lttng_ht_node_ulong *node = | |
151 | caa_container_of(head, struct lttng_ht_node_ulong, head); | |
152 | struct lttng_consumer_stream *stream = | |
153 | caa_container_of(node, struct lttng_consumer_stream, node); | |
154 | ||
155 | free(stream); | |
156 | } | |
157 | ||
00e2e675 DG |
158 | /* |
159 | * RCU protected relayd socket pair free. | |
160 | */ | |
161 | static void consumer_rcu_free_relayd(struct rcu_head *head) | |
162 | { | |
163 | struct lttng_ht_node_ulong *node = | |
164 | caa_container_of(head, struct lttng_ht_node_ulong, head); | |
165 | struct consumer_relayd_sock_pair *relayd = | |
166 | caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
167 | ||
168 | free(relayd); | |
169 | } | |
170 | ||
171 | /* | |
172 | * Destroy and free relayd socket pair object. | |
173 | * | |
174 | * This function MUST be called with the consumer_data lock acquired. | |
175 | */ | |
176 | void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) | |
177 | { | |
178 | int ret; | |
179 | struct lttng_ht_iter iter; | |
180 | ||
173af62f DG |
181 | if (relayd == NULL) { |
182 | return; | |
183 | } | |
184 | ||
00e2e675 DG |
185 | DBG("Consumer destroy and close relayd socket pair"); |
186 | ||
187 | iter.iter.node = &relayd->node.node; | |
188 | ret = lttng_ht_del(consumer_data.relayd_ht, &iter); | |
173af62f DG |
189 | if (ret != 0) { |
190 | /* We assume the relayd was already destroyed */ | |
191 | return; | |
192 | } | |
00e2e675 DG |
193 | |
194 | /* Close all sockets */ | |
195 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
196 | (void) relayd_close(&relayd->control_sock); | |
197 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
198 | (void) relayd_close(&relayd->data_sock); | |
199 | ||
200 | /* RCU free() call */ | |
201 | call_rcu(&relayd->node.head, consumer_rcu_free_relayd); | |
202 | } | |
203 | ||
a6ba4fe1 DG |
204 | /* |
205 | * Flag a relayd socket pair for destruction. Destroy it if the refcount | |
206 | * reaches zero. | |
207 | * | |
208 | * RCU read side lock MUST be aquired before calling this function. | |
209 | */ | |
210 | void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) | |
211 | { | |
212 | assert(relayd); | |
213 | ||
214 | /* Set destroy flag for this object */ | |
215 | uatomic_set(&relayd->destroy_flag, 1); | |
216 | ||
217 | /* Destroy the relayd if refcount is 0 */ | |
218 | if (uatomic_read(&relayd->refcount) == 0) { | |
219 | consumer_destroy_relayd(relayd); | |
220 | } | |
221 | } | |
222 | ||
3bd1e081 MD |
223 | /* |
224 | * Remove a stream from the global list protected by a mutex. This | |
225 | * function is also responsible for freeing its data structures. | |
226 | */ | |
227 | void consumer_del_stream(struct lttng_consumer_stream *stream) | |
228 | { | |
229 | int ret; | |
e4421fec | 230 | struct lttng_ht_iter iter; |
3bd1e081 | 231 | struct lttng_consumer_channel *free_chan = NULL; |
00e2e675 DG |
232 | struct consumer_relayd_sock_pair *relayd; |
233 | ||
234 | assert(stream); | |
3bd1e081 MD |
235 | |
236 | pthread_mutex_lock(&consumer_data.lock); | |
237 | ||
238 | switch (consumer_data.type) { | |
239 | case LTTNG_CONSUMER_KERNEL: | |
240 | if (stream->mmap_base != NULL) { | |
241 | ret = munmap(stream->mmap_base, stream->mmap_len); | |
242 | if (ret != 0) { | |
243 | perror("munmap"); | |
244 | } | |
245 | } | |
246 | break; | |
7753dea8 MD |
247 | case LTTNG_CONSUMER32_UST: |
248 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
249 | lttng_ustconsumer_del_stream(stream); |
250 | break; | |
251 | default: | |
252 | ERR("Unknown consumer_data type"); | |
253 | assert(0); | |
254 | goto end; | |
255 | } | |
256 | ||
6065ceec | 257 | rcu_read_lock(); |
04253271 MD |
258 | iter.iter.node = &stream->node.node; |
259 | ret = lttng_ht_del(consumer_data.stream_ht, &iter); | |
260 | assert(!ret); | |
e4421fec | 261 | |
6065ceec DG |
262 | rcu_read_unlock(); |
263 | ||
3bd1e081 MD |
264 | if (consumer_data.stream_count <= 0) { |
265 | goto end; | |
266 | } | |
267 | consumer_data.stream_count--; | |
268 | if (!stream) { | |
269 | goto end; | |
270 | } | |
271 | if (stream->out_fd >= 0) { | |
4c462e79 MD |
272 | ret = close(stream->out_fd); |
273 | if (ret) { | |
274 | PERROR("close"); | |
275 | } | |
3bd1e081 | 276 | } |
b5c5fc29 | 277 | if (stream->wait_fd >= 0 && !stream->wait_fd_is_copy) { |
4c462e79 MD |
278 | ret = close(stream->wait_fd); |
279 | if (ret) { | |
280 | PERROR("close"); | |
281 | } | |
3bd1e081 | 282 | } |
2c1dd183 | 283 | if (stream->shm_fd >= 0 && stream->wait_fd != stream->shm_fd) { |
4c462e79 MD |
284 | ret = close(stream->shm_fd); |
285 | if (ret) { | |
286 | PERROR("close"); | |
287 | } | |
3bd1e081 | 288 | } |
00e2e675 DG |
289 | |
290 | /* Check and cleanup relayd */ | |
b0b335c8 | 291 | rcu_read_lock(); |
00e2e675 DG |
292 | relayd = consumer_find_relayd(stream->net_seq_idx); |
293 | if (relayd != NULL) { | |
b0b335c8 MD |
294 | uatomic_dec(&relayd->refcount); |
295 | assert(uatomic_read(&relayd->refcount) >= 0); | |
173af62f | 296 | |
3f8e211f DG |
297 | /* Closing streams requires to lock the control socket. */ |
298 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
173af62f DG |
299 | ret = relayd_send_close_stream(&relayd->control_sock, |
300 | stream->relayd_stream_id, | |
301 | stream->next_net_seq_num - 1); | |
3f8e211f | 302 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
173af62f | 303 | if (ret < 0) { |
a4b92340 DG |
304 | DBG("Unable to close stream on the relayd. Continuing"); |
305 | /* | |
306 | * Continue here. There is nothing we can do for the relayd. | |
307 | * Chances are that the relayd has closed the socket so we just | |
308 | * continue cleaning up. | |
309 | */ | |
173af62f DG |
310 | } |
311 | ||
312 | /* Both conditions are met, we destroy the relayd. */ | |
313 | if (uatomic_read(&relayd->refcount) == 0 && | |
314 | uatomic_read(&relayd->destroy_flag)) { | |
00e2e675 DG |
315 | consumer_destroy_relayd(relayd); |
316 | } | |
00e2e675 | 317 | } |
b0b335c8 | 318 | rcu_read_unlock(); |
00e2e675 DG |
319 | |
320 | if (!--stream->chan->refcount) { | |
3bd1e081 | 321 | free_chan = stream->chan; |
00e2e675 DG |
322 | } |
323 | ||
702b1ea4 MD |
324 | |
325 | call_rcu(&stream->node.head, consumer_free_stream); | |
3bd1e081 MD |
326 | end: |
327 | consumer_data.need_update = 1; | |
328 | pthread_mutex_unlock(&consumer_data.lock); | |
329 | ||
330 | if (free_chan) | |
331 | consumer_del_channel(free_chan); | |
332 | } | |
333 | ||
334 | struct lttng_consumer_stream *consumer_allocate_stream( | |
335 | int channel_key, int stream_key, | |
336 | int shm_fd, int wait_fd, | |
337 | enum lttng_consumer_stream_state state, | |
338 | uint64_t mmap_len, | |
339 | enum lttng_event_output output, | |
6df2e2c9 MD |
340 | const char *path_name, |
341 | uid_t uid, | |
00e2e675 DG |
342 | gid_t gid, |
343 | int net_index, | |
344 | int metadata_flag) | |
3bd1e081 MD |
345 | { |
346 | struct lttng_consumer_stream *stream; | |
347 | int ret; | |
348 | ||
effcf122 | 349 | stream = zmalloc(sizeof(*stream)); |
3bd1e081 MD |
350 | if (stream == NULL) { |
351 | perror("malloc struct lttng_consumer_stream"); | |
352 | goto end; | |
353 | } | |
354 | stream->chan = consumer_find_channel(channel_key); | |
355 | if (!stream->chan) { | |
356 | perror("Unable to find channel key"); | |
357 | goto end; | |
358 | } | |
359 | stream->chan->refcount++; | |
360 | stream->key = stream_key; | |
361 | stream->shm_fd = shm_fd; | |
362 | stream->wait_fd = wait_fd; | |
363 | stream->out_fd = -1; | |
364 | stream->out_fd_offset = 0; | |
365 | stream->state = state; | |
366 | stream->mmap_len = mmap_len; | |
367 | stream->mmap_base = NULL; | |
368 | stream->output = output; | |
6df2e2c9 MD |
369 | stream->uid = uid; |
370 | stream->gid = gid; | |
00e2e675 DG |
371 | stream->net_seq_idx = net_index; |
372 | stream->metadata_flag = metadata_flag; | |
373 | strncpy(stream->path_name, path_name, sizeof(stream->path_name)); | |
374 | stream->path_name[sizeof(stream->path_name) - 1] = '\0'; | |
e4421fec | 375 | lttng_ht_node_init_ulong(&stream->node, stream->key); |
00e2e675 | 376 | lttng_ht_node_init_ulong(&stream->waitfd_node, stream->wait_fd); |
3bd1e081 MD |
377 | |
378 | switch (consumer_data.type) { | |
379 | case LTTNG_CONSUMER_KERNEL: | |
380 | break; | |
7753dea8 MD |
381 | case LTTNG_CONSUMER32_UST: |
382 | case LTTNG_CONSUMER64_UST: | |
5af2f756 | 383 | stream->cpu = stream->chan->cpucount++; |
3bd1e081 MD |
384 | ret = lttng_ustconsumer_allocate_stream(stream); |
385 | if (ret) { | |
386 | free(stream); | |
387 | return NULL; | |
388 | } | |
389 | break; | |
390 | default: | |
391 | ERR("Unknown consumer_data type"); | |
392 | assert(0); | |
393 | goto end; | |
394 | } | |
00e2e675 | 395 | DBG("Allocated stream %s (key %d, shm_fd %d, wait_fd %d, mmap_len %llu, out_fd %d, net_seq_idx %d)", |
3bd1e081 MD |
396 | stream->path_name, stream->key, |
397 | stream->shm_fd, | |
398 | stream->wait_fd, | |
399 | (unsigned long long) stream->mmap_len, | |
00e2e675 DG |
400 | stream->out_fd, |
401 | stream->net_seq_idx); | |
3bd1e081 MD |
402 | end: |
403 | return stream; | |
404 | } | |
405 | ||
406 | /* | |
407 | * Add a stream to the global list protected by a mutex. | |
408 | */ | |
409 | int consumer_add_stream(struct lttng_consumer_stream *stream) | |
410 | { | |
411 | int ret = 0; | |
c77fc10a DG |
412 | struct lttng_ht_node_ulong *node; |
413 | struct lttng_ht_iter iter; | |
00e2e675 | 414 | struct consumer_relayd_sock_pair *relayd; |
3bd1e081 MD |
415 | |
416 | pthread_mutex_lock(&consumer_data.lock); | |
7ad0a0cb MD |
417 | /* Steal stream identifier, for UST */ |
418 | consumer_steal_stream_key(stream->key); | |
c77fc10a | 419 | |
b0b335c8 | 420 | rcu_read_lock(); |
c77fc10a DG |
421 | lttng_ht_lookup(consumer_data.stream_ht, |
422 | (void *)((unsigned long) stream->key), &iter); | |
423 | node = lttng_ht_iter_get_node_ulong(&iter); | |
424 | if (node != NULL) { | |
425 | rcu_read_unlock(); | |
426 | /* Stream already exist. Ignore the insertion */ | |
427 | goto end; | |
428 | } | |
429 | ||
04253271 | 430 | lttng_ht_add_unique_ulong(consumer_data.stream_ht, &stream->node); |
00e2e675 DG |
431 | |
432 | /* Check and cleanup relayd */ | |
433 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
434 | if (relayd != NULL) { | |
b0b335c8 | 435 | uatomic_inc(&relayd->refcount); |
00e2e675 | 436 | } |
b0b335c8 | 437 | rcu_read_unlock(); |
00e2e675 DG |
438 | |
439 | /* Update consumer data */ | |
3bd1e081 MD |
440 | consumer_data.stream_count++; |
441 | consumer_data.need_update = 1; | |
442 | ||
443 | switch (consumer_data.type) { | |
444 | case LTTNG_CONSUMER_KERNEL: | |
445 | break; | |
7753dea8 MD |
446 | case LTTNG_CONSUMER32_UST: |
447 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
448 | /* Streams are in CPU number order (we rely on this) */ |
449 | stream->cpu = stream->chan->nr_streams++; | |
450 | break; | |
451 | default: | |
452 | ERR("Unknown consumer_data type"); | |
453 | assert(0); | |
454 | goto end; | |
455 | } | |
456 | ||
457 | end: | |
458 | pthread_mutex_unlock(&consumer_data.lock); | |
702b1ea4 | 459 | |
3bd1e081 MD |
460 | return ret; |
461 | } | |
462 | ||
00e2e675 | 463 | /* |
3f8e211f DG |
464 | * Add relayd socket to global consumer data hashtable. RCU read side lock MUST |
465 | * be acquired before calling this. | |
00e2e675 | 466 | */ |
3f8e211f | 467 | |
00e2e675 DG |
468 | int consumer_add_relayd(struct consumer_relayd_sock_pair *relayd) |
469 | { | |
470 | int ret = 0; | |
471 | struct lttng_ht_node_ulong *node; | |
472 | struct lttng_ht_iter iter; | |
473 | ||
474 | if (relayd == NULL) { | |
475 | ret = -1; | |
476 | goto end; | |
477 | } | |
478 | ||
00e2e675 DG |
479 | lttng_ht_lookup(consumer_data.relayd_ht, |
480 | (void *)((unsigned long) relayd->net_seq_idx), &iter); | |
481 | node = lttng_ht_iter_get_node_ulong(&iter); | |
482 | if (node != NULL) { | |
00e2e675 DG |
483 | /* Relayd already exist. Ignore the insertion */ |
484 | goto end; | |
485 | } | |
486 | lttng_ht_add_unique_ulong(consumer_data.relayd_ht, &relayd->node); | |
487 | ||
00e2e675 DG |
488 | end: |
489 | return ret; | |
490 | } | |
491 | ||
492 | /* | |
493 | * Allocate and return a consumer relayd socket. | |
494 | */ | |
495 | struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( | |
496 | int net_seq_idx) | |
497 | { | |
498 | struct consumer_relayd_sock_pair *obj = NULL; | |
499 | ||
500 | /* Negative net sequence index is a failure */ | |
501 | if (net_seq_idx < 0) { | |
502 | goto error; | |
503 | } | |
504 | ||
505 | obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); | |
506 | if (obj == NULL) { | |
507 | PERROR("zmalloc relayd sock"); | |
508 | goto error; | |
509 | } | |
510 | ||
511 | obj->net_seq_idx = net_seq_idx; | |
512 | obj->refcount = 0; | |
173af62f | 513 | obj->destroy_flag = 0; |
00e2e675 DG |
514 | lttng_ht_node_init_ulong(&obj->node, obj->net_seq_idx); |
515 | pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); | |
516 | ||
517 | error: | |
518 | return obj; | |
519 | } | |
520 | ||
521 | /* | |
522 | * Find a relayd socket pair in the global consumer data. | |
523 | * | |
524 | * Return the object if found else NULL. | |
b0b335c8 MD |
525 | * RCU read-side lock must be held across this call and while using the |
526 | * returned object. | |
00e2e675 DG |
527 | */ |
528 | struct consumer_relayd_sock_pair *consumer_find_relayd(int key) | |
529 | { | |
530 | struct lttng_ht_iter iter; | |
531 | struct lttng_ht_node_ulong *node; | |
532 | struct consumer_relayd_sock_pair *relayd = NULL; | |
533 | ||
534 | /* Negative keys are lookup failures */ | |
535 | if (key < 0) { | |
536 | goto error; | |
537 | } | |
538 | ||
00e2e675 DG |
539 | lttng_ht_lookup(consumer_data.relayd_ht, (void *)((unsigned long) key), |
540 | &iter); | |
541 | node = lttng_ht_iter_get_node_ulong(&iter); | |
542 | if (node != NULL) { | |
543 | relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
544 | } | |
545 | ||
00e2e675 DG |
546 | error: |
547 | return relayd; | |
548 | } | |
549 | ||
550 | /* | |
551 | * Handle stream for relayd transmission if the stream applies for network | |
552 | * streaming where the net sequence index is set. | |
553 | * | |
554 | * Return destination file descriptor or negative value on error. | |
555 | */ | |
6197aea7 DG |
556 | static int write_relayd_stream_header(struct lttng_consumer_stream *stream, |
557 | size_t data_size, struct consumer_relayd_sock_pair *relayd) | |
00e2e675 DG |
558 | { |
559 | int outfd = -1, ret; | |
00e2e675 DG |
560 | struct lttcomm_relayd_data_hdr data_hdr; |
561 | ||
562 | /* Safety net */ | |
563 | assert(stream); | |
6197aea7 | 564 | assert(relayd); |
00e2e675 DG |
565 | |
566 | /* Reset data header */ | |
567 | memset(&data_hdr, 0, sizeof(data_hdr)); | |
568 | ||
00e2e675 DG |
569 | if (stream->metadata_flag) { |
570 | /* Caller MUST acquire the relayd control socket lock */ | |
571 | ret = relayd_send_metadata(&relayd->control_sock, data_size); | |
572 | if (ret < 0) { | |
573 | goto error; | |
574 | } | |
575 | ||
576 | /* Metadata are always sent on the control socket. */ | |
577 | outfd = relayd->control_sock.fd; | |
578 | } else { | |
579 | /* Set header with stream information */ | |
580 | data_hdr.stream_id = htobe64(stream->relayd_stream_id); | |
581 | data_hdr.data_size = htobe32(data_size); | |
173af62f | 582 | data_hdr.net_seq_num = htobe64(stream->next_net_seq_num++); |
00e2e675 DG |
583 | /* Other fields are zeroed previously */ |
584 | ||
585 | ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, | |
586 | sizeof(data_hdr)); | |
587 | if (ret < 0) { | |
588 | goto error; | |
589 | } | |
590 | ||
591 | /* Set to go on data socket */ | |
592 | outfd = relayd->data_sock.fd; | |
593 | } | |
594 | ||
595 | error: | |
596 | return outfd; | |
597 | } | |
598 | ||
3bd1e081 MD |
599 | /* |
600 | * Update a stream according to what we just received. | |
601 | */ | |
602 | void consumer_change_stream_state(int stream_key, | |
603 | enum lttng_consumer_stream_state state) | |
604 | { | |
605 | struct lttng_consumer_stream *stream; | |
606 | ||
607 | pthread_mutex_lock(&consumer_data.lock); | |
608 | stream = consumer_find_stream(stream_key); | |
609 | if (stream) { | |
610 | stream->state = state; | |
611 | } | |
612 | consumer_data.need_update = 1; | |
613 | pthread_mutex_unlock(&consumer_data.lock); | |
614 | } | |
615 | ||
702b1ea4 MD |
616 | static |
617 | void consumer_free_channel(struct rcu_head *head) | |
618 | { | |
619 | struct lttng_ht_node_ulong *node = | |
620 | caa_container_of(head, struct lttng_ht_node_ulong, head); | |
621 | struct lttng_consumer_channel *channel = | |
622 | caa_container_of(node, struct lttng_consumer_channel, node); | |
623 | ||
624 | free(channel); | |
625 | } | |
626 | ||
3bd1e081 MD |
627 | /* |
628 | * Remove a channel from the global list protected by a mutex. This | |
629 | * function is also responsible for freeing its data structures. | |
630 | */ | |
631 | void consumer_del_channel(struct lttng_consumer_channel *channel) | |
632 | { | |
633 | int ret; | |
e4421fec | 634 | struct lttng_ht_iter iter; |
3bd1e081 MD |
635 | |
636 | pthread_mutex_lock(&consumer_data.lock); | |
637 | ||
638 | switch (consumer_data.type) { | |
639 | case LTTNG_CONSUMER_KERNEL: | |
640 | break; | |
7753dea8 MD |
641 | case LTTNG_CONSUMER32_UST: |
642 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
643 | lttng_ustconsumer_del_channel(channel); |
644 | break; | |
645 | default: | |
646 | ERR("Unknown consumer_data type"); | |
647 | assert(0); | |
648 | goto end; | |
649 | } | |
650 | ||
6065ceec | 651 | rcu_read_lock(); |
04253271 MD |
652 | iter.iter.node = &channel->node.node; |
653 | ret = lttng_ht_del(consumer_data.channel_ht, &iter); | |
654 | assert(!ret); | |
6065ceec DG |
655 | rcu_read_unlock(); |
656 | ||
3bd1e081 MD |
657 | if (channel->mmap_base != NULL) { |
658 | ret = munmap(channel->mmap_base, channel->mmap_len); | |
659 | if (ret != 0) { | |
660 | perror("munmap"); | |
661 | } | |
662 | } | |
b5c5fc29 | 663 | if (channel->wait_fd >= 0 && !channel->wait_fd_is_copy) { |
4c462e79 MD |
664 | ret = close(channel->wait_fd); |
665 | if (ret) { | |
666 | PERROR("close"); | |
667 | } | |
3bd1e081 | 668 | } |
2c1dd183 | 669 | if (channel->shm_fd >= 0 && channel->wait_fd != channel->shm_fd) { |
4c462e79 MD |
670 | ret = close(channel->shm_fd); |
671 | if (ret) { | |
672 | PERROR("close"); | |
673 | } | |
3bd1e081 | 674 | } |
702b1ea4 MD |
675 | |
676 | call_rcu(&channel->node.head, consumer_free_channel); | |
3bd1e081 MD |
677 | end: |
678 | pthread_mutex_unlock(&consumer_data.lock); | |
679 | } | |
680 | ||
681 | struct lttng_consumer_channel *consumer_allocate_channel( | |
682 | int channel_key, | |
683 | int shm_fd, int wait_fd, | |
684 | uint64_t mmap_len, | |
685 | uint64_t max_sb_size) | |
686 | { | |
687 | struct lttng_consumer_channel *channel; | |
688 | int ret; | |
689 | ||
276b26d1 | 690 | channel = zmalloc(sizeof(*channel)); |
3bd1e081 MD |
691 | if (channel == NULL) { |
692 | perror("malloc struct lttng_consumer_channel"); | |
693 | goto end; | |
694 | } | |
695 | channel->key = channel_key; | |
696 | channel->shm_fd = shm_fd; | |
697 | channel->wait_fd = wait_fd; | |
698 | channel->mmap_len = mmap_len; | |
699 | channel->max_sb_size = max_sb_size; | |
700 | channel->refcount = 0; | |
701 | channel->nr_streams = 0; | |
e4421fec | 702 | lttng_ht_node_init_ulong(&channel->node, channel->key); |
3bd1e081 MD |
703 | |
704 | switch (consumer_data.type) { | |
705 | case LTTNG_CONSUMER_KERNEL: | |
706 | channel->mmap_base = NULL; | |
707 | channel->mmap_len = 0; | |
708 | break; | |
7753dea8 MD |
709 | case LTTNG_CONSUMER32_UST: |
710 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
711 | ret = lttng_ustconsumer_allocate_channel(channel); |
712 | if (ret) { | |
713 | free(channel); | |
714 | return NULL; | |
715 | } | |
716 | break; | |
717 | default: | |
718 | ERR("Unknown consumer_data type"); | |
719 | assert(0); | |
720 | goto end; | |
721 | } | |
722 | DBG("Allocated channel (key %d, shm_fd %d, wait_fd %d, mmap_len %llu, max_sb_size %llu)", | |
00e2e675 | 723 | channel->key, channel->shm_fd, channel->wait_fd, |
3bd1e081 MD |
724 | (unsigned long long) channel->mmap_len, |
725 | (unsigned long long) channel->max_sb_size); | |
726 | end: | |
727 | return channel; | |
728 | } | |
729 | ||
730 | /* | |
731 | * Add a channel to the global list protected by a mutex. | |
732 | */ | |
733 | int consumer_add_channel(struct lttng_consumer_channel *channel) | |
734 | { | |
c77fc10a DG |
735 | struct lttng_ht_node_ulong *node; |
736 | struct lttng_ht_iter iter; | |
737 | ||
3bd1e081 | 738 | pthread_mutex_lock(&consumer_data.lock); |
7ad0a0cb MD |
739 | /* Steal channel identifier, for UST */ |
740 | consumer_steal_channel_key(channel->key); | |
6065ceec | 741 | rcu_read_lock(); |
c77fc10a DG |
742 | |
743 | lttng_ht_lookup(consumer_data.channel_ht, | |
744 | (void *)((unsigned long) channel->key), &iter); | |
745 | node = lttng_ht_iter_get_node_ulong(&iter); | |
746 | if (node != NULL) { | |
747 | /* Channel already exist. Ignore the insertion */ | |
748 | goto end; | |
749 | } | |
750 | ||
04253271 | 751 | lttng_ht_add_unique_ulong(consumer_data.channel_ht, &channel->node); |
c77fc10a DG |
752 | |
753 | end: | |
6065ceec | 754 | rcu_read_unlock(); |
3bd1e081 | 755 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 756 | |
7ad0a0cb | 757 | return 0; |
3bd1e081 MD |
758 | } |
759 | ||
760 | /* | |
761 | * Allocate the pollfd structure and the local view of the out fds to avoid | |
762 | * doing a lookup in the linked list and concurrency issues when writing is | |
763 | * needed. Called with consumer_data.lock held. | |
764 | * | |
765 | * Returns the number of fds in the structures. | |
766 | */ | |
767 | int consumer_update_poll_array( | |
768 | struct lttng_consumer_local_data *ctx, struct pollfd **pollfd, | |
00e2e675 DG |
769 | struct lttng_consumer_stream **local_stream, |
770 | struct lttng_ht *metadata_ht) | |
3bd1e081 | 771 | { |
3bd1e081 | 772 | int i = 0; |
e4421fec DG |
773 | struct lttng_ht_iter iter; |
774 | struct lttng_consumer_stream *stream; | |
3bd1e081 MD |
775 | |
776 | DBG("Updating poll fd array"); | |
481d6c57 | 777 | rcu_read_lock(); |
e4421fec DG |
778 | cds_lfht_for_each_entry(consumer_data.stream_ht->ht, &iter.iter, stream, |
779 | node.node) { | |
780 | if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM) { | |
3bd1e081 MD |
781 | continue; |
782 | } | |
e4421fec DG |
783 | DBG("Active FD %d", stream->wait_fd); |
784 | (*pollfd)[i].fd = stream->wait_fd; | |
3bd1e081 | 785 | (*pollfd)[i].events = POLLIN | POLLPRI; |
00e2e675 DG |
786 | if (stream->metadata_flag && metadata_ht) { |
787 | lttng_ht_add_unique_ulong(metadata_ht, &stream->waitfd_node); | |
788 | DBG("Active FD added to metadata hash table"); | |
789 | } | |
e4421fec | 790 | local_stream[i] = stream; |
3bd1e081 MD |
791 | i++; |
792 | } | |
481d6c57 | 793 | rcu_read_unlock(); |
3bd1e081 MD |
794 | |
795 | /* | |
796 | * Insert the consumer_poll_pipe at the end of the array and don't | |
797 | * increment i so nb_fd is the number of real FD. | |
798 | */ | |
799 | (*pollfd)[i].fd = ctx->consumer_poll_pipe[0]; | |
509bb1cf | 800 | (*pollfd)[i].events = POLLIN | POLLPRI; |
3bd1e081 MD |
801 | return i; |
802 | } | |
803 | ||
804 | /* | |
805 | * Poll on the should_quit pipe and the command socket return -1 on error and | |
806 | * should exit, 0 if data is available on the command socket | |
807 | */ | |
808 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) | |
809 | { | |
810 | int num_rdy; | |
811 | ||
88f2b785 | 812 | restart: |
3bd1e081 MD |
813 | num_rdy = poll(consumer_sockpoll, 2, -1); |
814 | if (num_rdy == -1) { | |
88f2b785 MD |
815 | /* |
816 | * Restart interrupted system call. | |
817 | */ | |
818 | if (errno == EINTR) { | |
819 | goto restart; | |
820 | } | |
3bd1e081 MD |
821 | perror("Poll error"); |
822 | goto exit; | |
823 | } | |
509bb1cf | 824 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
3bd1e081 MD |
825 | DBG("consumer_should_quit wake up"); |
826 | goto exit; | |
827 | } | |
828 | return 0; | |
829 | ||
830 | exit: | |
831 | return -1; | |
832 | } | |
833 | ||
834 | /* | |
835 | * Set the error socket. | |
836 | */ | |
837 | void lttng_consumer_set_error_sock( | |
838 | struct lttng_consumer_local_data *ctx, int sock) | |
839 | { | |
840 | ctx->consumer_error_socket = sock; | |
841 | } | |
842 | ||
843 | /* | |
844 | * Set the command socket path. | |
845 | */ | |
3bd1e081 MD |
846 | void lttng_consumer_set_command_sock_path( |
847 | struct lttng_consumer_local_data *ctx, char *sock) | |
848 | { | |
849 | ctx->consumer_command_sock_path = sock; | |
850 | } | |
851 | ||
852 | /* | |
853 | * Send return code to the session daemon. | |
854 | * If the socket is not defined, we return 0, it is not a fatal error | |
855 | */ | |
856 | int lttng_consumer_send_error( | |
857 | struct lttng_consumer_local_data *ctx, int cmd) | |
858 | { | |
859 | if (ctx->consumer_error_socket > 0) { | |
860 | return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd, | |
861 | sizeof(enum lttcomm_sessiond_command)); | |
862 | } | |
863 | ||
864 | return 0; | |
865 | } | |
866 | ||
867 | /* | |
868 | * Close all the tracefiles and stream fds, should be called when all instances | |
869 | * are destroyed. | |
870 | */ | |
871 | void lttng_consumer_cleanup(void) | |
872 | { | |
e4421fec | 873 | struct lttng_ht_iter iter; |
6065ceec DG |
874 | struct lttng_ht_node_ulong *node; |
875 | ||
876 | rcu_read_lock(); | |
3bd1e081 MD |
877 | |
878 | /* | |
6065ceec DG |
879 | * close all outfd. Called when there are no more threads running (after |
880 | * joining on the threads), no need to protect list iteration with mutex. | |
3bd1e081 | 881 | */ |
6065ceec DG |
882 | cds_lfht_for_each_entry(consumer_data.stream_ht->ht, &iter.iter, node, |
883 | node) { | |
702b1ea4 MD |
884 | struct lttng_consumer_stream *stream = |
885 | caa_container_of(node, struct lttng_consumer_stream, node); | |
886 | consumer_del_stream(stream); | |
3bd1e081 | 887 | } |
e4421fec | 888 | |
6065ceec DG |
889 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, node, |
890 | node) { | |
702b1ea4 MD |
891 | struct lttng_consumer_channel *channel = |
892 | caa_container_of(node, struct lttng_consumer_channel, node); | |
893 | consumer_del_channel(channel); | |
3bd1e081 | 894 | } |
6065ceec DG |
895 | |
896 | rcu_read_unlock(); | |
d6ce1df2 MD |
897 | |
898 | lttng_ht_destroy(consumer_data.stream_ht); | |
899 | lttng_ht_destroy(consumer_data.channel_ht); | |
3bd1e081 MD |
900 | } |
901 | ||
902 | /* | |
903 | * Called from signal handler. | |
904 | */ | |
905 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) | |
906 | { | |
907 | int ret; | |
908 | consumer_quit = 1; | |
6f94560a MD |
909 | do { |
910 | ret = write(ctx->consumer_should_quit[1], "4", 1); | |
911 | } while (ret < 0 && errno == EINTR); | |
3bd1e081 MD |
912 | if (ret < 0) { |
913 | perror("write consumer quit"); | |
914 | } | |
915 | } | |
916 | ||
00e2e675 DG |
917 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
918 | off_t orig_offset) | |
3bd1e081 MD |
919 | { |
920 | int outfd = stream->out_fd; | |
921 | ||
922 | /* | |
923 | * This does a blocking write-and-wait on any page that belongs to the | |
924 | * subbuffer prior to the one we just wrote. | |
925 | * Don't care about error values, as these are just hints and ways to | |
926 | * limit the amount of page cache used. | |
927 | */ | |
928 | if (orig_offset < stream->chan->max_sb_size) { | |
929 | return; | |
930 | } | |
b9182dd9 | 931 | lttng_sync_file_range(outfd, orig_offset - stream->chan->max_sb_size, |
3bd1e081 MD |
932 | stream->chan->max_sb_size, |
933 | SYNC_FILE_RANGE_WAIT_BEFORE | |
934 | | SYNC_FILE_RANGE_WRITE | |
935 | | SYNC_FILE_RANGE_WAIT_AFTER); | |
936 | /* | |
937 | * Give hints to the kernel about how we access the file: | |
938 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after | |
939 | * we write it. | |
940 | * | |
941 | * We need to call fadvise again after the file grows because the | |
942 | * kernel does not seem to apply fadvise to non-existing parts of the | |
943 | * file. | |
944 | * | |
945 | * Call fadvise _after_ having waited for the page writeback to | |
946 | * complete because the dirty page writeback semantic is not well | |
947 | * defined. So it can be expected to lead to lower throughput in | |
948 | * streaming. | |
949 | */ | |
950 | posix_fadvise(outfd, orig_offset - stream->chan->max_sb_size, | |
951 | stream->chan->max_sb_size, POSIX_FADV_DONTNEED); | |
952 | } | |
953 | ||
954 | /* | |
955 | * Initialise the necessary environnement : | |
956 | * - create a new context | |
957 | * - create the poll_pipe | |
958 | * - create the should_quit pipe (for signal handler) | |
959 | * - create the thread pipe (for splice) | |
960 | * | |
961 | * Takes a function pointer as argument, this function is called when data is | |
962 | * available on a buffer. This function is responsible to do the | |
963 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the | |
964 | * buffer configuration and then kernctl_put_next_subbuf at the end. | |
965 | * | |
966 | * Returns a pointer to the new context or NULL on error. | |
967 | */ | |
968 | struct lttng_consumer_local_data *lttng_consumer_create( | |
969 | enum lttng_consumer_type type, | |
4078b776 | 970 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
d41f73b7 | 971 | struct lttng_consumer_local_data *ctx), |
3bd1e081 MD |
972 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
973 | int (*recv_stream)(struct lttng_consumer_stream *stream), | |
974 | int (*update_stream)(int stream_key, uint32_t state)) | |
975 | { | |
976 | int ret, i; | |
977 | struct lttng_consumer_local_data *ctx; | |
978 | ||
979 | assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || | |
980 | consumer_data.type == type); | |
981 | consumer_data.type = type; | |
982 | ||
effcf122 | 983 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
3bd1e081 MD |
984 | if (ctx == NULL) { |
985 | perror("allocating context"); | |
986 | goto error; | |
987 | } | |
988 | ||
989 | ctx->consumer_error_socket = -1; | |
990 | /* assign the callbacks */ | |
991 | ctx->on_buffer_ready = buffer_ready; | |
992 | ctx->on_recv_channel = recv_channel; | |
993 | ctx->on_recv_stream = recv_stream; | |
994 | ctx->on_update_stream = update_stream; | |
995 | ||
996 | ret = pipe(ctx->consumer_poll_pipe); | |
997 | if (ret < 0) { | |
998 | perror("Error creating poll pipe"); | |
999 | goto error_poll_pipe; | |
1000 | } | |
1001 | ||
04fdd819 MD |
1002 | /* set read end of the pipe to non-blocking */ |
1003 | ret = fcntl(ctx->consumer_poll_pipe[0], F_SETFL, O_NONBLOCK); | |
1004 | if (ret < 0) { | |
1005 | perror("fcntl O_NONBLOCK"); | |
1006 | goto error_poll_fcntl; | |
1007 | } | |
1008 | ||
1009 | /* set write end of the pipe to non-blocking */ | |
1010 | ret = fcntl(ctx->consumer_poll_pipe[1], F_SETFL, O_NONBLOCK); | |
1011 | if (ret < 0) { | |
1012 | perror("fcntl O_NONBLOCK"); | |
1013 | goto error_poll_fcntl; | |
1014 | } | |
1015 | ||
3bd1e081 MD |
1016 | ret = pipe(ctx->consumer_should_quit); |
1017 | if (ret < 0) { | |
1018 | perror("Error creating recv pipe"); | |
1019 | goto error_quit_pipe; | |
1020 | } | |
1021 | ||
1022 | ret = pipe(ctx->consumer_thread_pipe); | |
1023 | if (ret < 0) { | |
1024 | perror("Error creating thread pipe"); | |
1025 | goto error_thread_pipe; | |
1026 | } | |
1027 | ||
1028 | return ctx; | |
1029 | ||
1030 | ||
1031 | error_thread_pipe: | |
1032 | for (i = 0; i < 2; i++) { | |
1033 | int err; | |
1034 | ||
1035 | err = close(ctx->consumer_should_quit[i]); | |
4c462e79 MD |
1036 | if (err) { |
1037 | PERROR("close"); | |
1038 | } | |
3bd1e081 | 1039 | } |
04fdd819 | 1040 | error_poll_fcntl: |
3bd1e081 MD |
1041 | error_quit_pipe: |
1042 | for (i = 0; i < 2; i++) { | |
1043 | int err; | |
1044 | ||
1045 | err = close(ctx->consumer_poll_pipe[i]); | |
4c462e79 MD |
1046 | if (err) { |
1047 | PERROR("close"); | |
1048 | } | |
3bd1e081 MD |
1049 | } |
1050 | error_poll_pipe: | |
1051 | free(ctx); | |
1052 | error: | |
1053 | return NULL; | |
1054 | } | |
1055 | ||
1056 | /* | |
1057 | * Close all fds associated with the instance and free the context. | |
1058 | */ | |
1059 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) | |
1060 | { | |
4c462e79 MD |
1061 | int ret; |
1062 | ||
1063 | ret = close(ctx->consumer_error_socket); | |
1064 | if (ret) { | |
1065 | PERROR("close"); | |
1066 | } | |
1067 | ret = close(ctx->consumer_thread_pipe[0]); | |
1068 | if (ret) { | |
1069 | PERROR("close"); | |
1070 | } | |
1071 | ret = close(ctx->consumer_thread_pipe[1]); | |
1072 | if (ret) { | |
1073 | PERROR("close"); | |
1074 | } | |
1075 | ret = close(ctx->consumer_poll_pipe[0]); | |
1076 | if (ret) { | |
1077 | PERROR("close"); | |
1078 | } | |
1079 | ret = close(ctx->consumer_poll_pipe[1]); | |
1080 | if (ret) { | |
1081 | PERROR("close"); | |
1082 | } | |
1083 | ret = close(ctx->consumer_should_quit[0]); | |
1084 | if (ret) { | |
1085 | PERROR("close"); | |
1086 | } | |
1087 | ret = close(ctx->consumer_should_quit[1]); | |
1088 | if (ret) { | |
1089 | PERROR("close"); | |
1090 | } | |
3bd1e081 MD |
1091 | unlink(ctx->consumer_command_sock_path); |
1092 | free(ctx); | |
1093 | } | |
1094 | ||
6197aea7 DG |
1095 | /* |
1096 | * Write the metadata stream id on the specified file descriptor. | |
1097 | */ | |
1098 | static int write_relayd_metadata_id(int fd, | |
1099 | struct lttng_consumer_stream *stream, | |
1100 | struct consumer_relayd_sock_pair *relayd) | |
1101 | { | |
1102 | int ret; | |
1103 | uint64_t metadata_id; | |
1104 | ||
1105 | metadata_id = htobe64(stream->relayd_stream_id); | |
1106 | do { | |
1107 | ret = write(fd, (void *) &metadata_id, | |
1108 | sizeof(stream->relayd_stream_id)); | |
1109 | } while (ret < 0 && errno == EINTR); | |
1110 | if (ret < 0) { | |
1111 | PERROR("write metadata stream id"); | |
1112 | goto end; | |
1113 | } | |
77c7c900 | 1114 | DBG("Metadata stream id %" PRIu64 " written before data", |
6197aea7 DG |
1115 | stream->relayd_stream_id); |
1116 | ||
1117 | end: | |
1118 | return ret; | |
1119 | } | |
1120 | ||
3bd1e081 | 1121 | /* |
09e26845 DG |
1122 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
1123 | * core function for writing trace buffers to either the local filesystem or | |
1124 | * the network. | |
1125 | * | |
1126 | * Careful review MUST be put if any changes occur! | |
3bd1e081 MD |
1127 | * |
1128 | * Returns the number of bytes written | |
1129 | */ | |
4078b776 | 1130 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
3bd1e081 MD |
1131 | struct lttng_consumer_local_data *ctx, |
1132 | struct lttng_consumer_stream *stream, unsigned long len) | |
1133 | { | |
f02e1e8a DG |
1134 | unsigned long mmap_offset; |
1135 | ssize_t ret = 0, written = 0; | |
1136 | off_t orig_offset = stream->out_fd_offset; | |
1137 | /* Default is on the disk */ | |
1138 | int outfd = stream->out_fd; | |
f02e1e8a DG |
1139 | struct consumer_relayd_sock_pair *relayd = NULL; |
1140 | ||
1141 | /* RCU lock for the relayd pointer */ | |
1142 | rcu_read_lock(); | |
1143 | ||
1144 | /* Flag that the current stream if set for network streaming. */ | |
1145 | if (stream->net_seq_idx != -1) { | |
1146 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
1147 | if (relayd == NULL) { | |
1148 | goto end; | |
1149 | } | |
1150 | } | |
1151 | ||
1152 | /* get the offset inside the fd to mmap */ | |
3bd1e081 MD |
1153 | switch (consumer_data.type) { |
1154 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a DG |
1155 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); |
1156 | break; | |
7753dea8 MD |
1157 | case LTTNG_CONSUMER32_UST: |
1158 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a DG |
1159 | ret = lttng_ustctl_get_mmap_read_offset(stream->chan->handle, |
1160 | stream->buf, &mmap_offset); | |
1161 | break; | |
3bd1e081 MD |
1162 | default: |
1163 | ERR("Unknown consumer_data type"); | |
1164 | assert(0); | |
1165 | } | |
f02e1e8a DG |
1166 | if (ret != 0) { |
1167 | errno = -ret; | |
1168 | PERROR("tracer ctl get_mmap_read_offset"); | |
1169 | written = ret; | |
1170 | goto end; | |
1171 | } | |
b9182dd9 | 1172 | |
f02e1e8a DG |
1173 | /* Handle stream on the relayd if the output is on the network */ |
1174 | if (relayd) { | |
1175 | unsigned long netlen = len; | |
1176 | ||
1177 | /* | |
1178 | * Lock the control socket for the complete duration of the function | |
1179 | * since from this point on we will use the socket. | |
1180 | */ | |
1181 | if (stream->metadata_flag) { | |
1182 | /* Metadata requires the control socket. */ | |
1183 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1184 | netlen += sizeof(stream->relayd_stream_id); | |
1185 | } | |
1186 | ||
6197aea7 | 1187 | ret = write_relayd_stream_header(stream, netlen, relayd); |
f02e1e8a DG |
1188 | if (ret >= 0) { |
1189 | /* Use the returned socket. */ | |
1190 | outfd = ret; | |
1191 | ||
1192 | /* Write metadata stream id before payload */ | |
1193 | if (stream->metadata_flag) { | |
6197aea7 | 1194 | ret = write_relayd_metadata_id(outfd, stream, relayd); |
f02e1e8a | 1195 | if (ret < 0) { |
f02e1e8a DG |
1196 | written = ret; |
1197 | goto end; | |
1198 | } | |
f02e1e8a DG |
1199 | } |
1200 | } | |
1201 | /* Else, use the default set before which is the filesystem. */ | |
1202 | } | |
1203 | ||
1204 | while (len > 0) { | |
1205 | do { | |
1206 | ret = write(outfd, stream->mmap_base + mmap_offset, len); | |
1207 | } while (ret < 0 && errno == EINTR); | |
1208 | if (ret < 0) { | |
1209 | PERROR("Error in file write"); | |
1210 | if (written == 0) { | |
1211 | written = ret; | |
1212 | } | |
1213 | goto end; | |
1214 | } else if (ret > len) { | |
77c7c900 | 1215 | PERROR("Error in file write (ret %zd > len %lu)", ret, len); |
f02e1e8a DG |
1216 | written += ret; |
1217 | goto end; | |
1218 | } else { | |
1219 | len -= ret; | |
1220 | mmap_offset += ret; | |
1221 | } | |
77c7c900 | 1222 | DBG("Consumer mmap write() ret %zd (len %lu)", ret, len); |
f02e1e8a DG |
1223 | |
1224 | /* This call is useless on a socket so better save a syscall. */ | |
1225 | if (!relayd) { | |
1226 | /* This won't block, but will start writeout asynchronously */ | |
1227 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret, | |
1228 | SYNC_FILE_RANGE_WRITE); | |
1229 | stream->out_fd_offset += ret; | |
1230 | } | |
1231 | written += ret; | |
1232 | } | |
1233 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1234 | ||
1235 | end: | |
1236 | /* Unlock only if ctrl socket used */ | |
1237 | if (relayd && stream->metadata_flag) { | |
1238 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1239 | } | |
1240 | ||
1241 | rcu_read_unlock(); | |
1242 | return written; | |
3bd1e081 MD |
1243 | } |
1244 | ||
1245 | /* | |
1246 | * Splice the data from the ring buffer to the tracefile. | |
1247 | * | |
1248 | * Returns the number of bytes spliced. | |
1249 | */ | |
4078b776 | 1250 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
3bd1e081 MD |
1251 | struct lttng_consumer_local_data *ctx, |
1252 | struct lttng_consumer_stream *stream, unsigned long len) | |
1253 | { | |
f02e1e8a DG |
1254 | ssize_t ret = 0, written = 0, ret_splice = 0; |
1255 | loff_t offset = 0; | |
1256 | off_t orig_offset = stream->out_fd_offset; | |
1257 | int fd = stream->wait_fd; | |
1258 | /* Default is on the disk */ | |
1259 | int outfd = stream->out_fd; | |
f02e1e8a DG |
1260 | struct consumer_relayd_sock_pair *relayd = NULL; |
1261 | ||
3bd1e081 MD |
1262 | switch (consumer_data.type) { |
1263 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a | 1264 | break; |
7753dea8 MD |
1265 | case LTTNG_CONSUMER32_UST: |
1266 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a | 1267 | /* Not supported for user space tracing */ |
3bd1e081 MD |
1268 | return -ENOSYS; |
1269 | default: | |
1270 | ERR("Unknown consumer_data type"); | |
1271 | assert(0); | |
3bd1e081 MD |
1272 | } |
1273 | ||
f02e1e8a DG |
1274 | /* RCU lock for the relayd pointer */ |
1275 | rcu_read_lock(); | |
1276 | ||
1277 | /* Flag that the current stream if set for network streaming. */ | |
1278 | if (stream->net_seq_idx != -1) { | |
1279 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
1280 | if (relayd == NULL) { | |
1281 | goto end; | |
1282 | } | |
1283 | } | |
1284 | ||
1285 | /* Write metadata stream id before payload */ | |
1286 | if (stream->metadata_flag && relayd) { | |
1287 | /* | |
1288 | * Lock the control socket for the complete duration of the function | |
1289 | * since from this point on we will use the socket. | |
1290 | */ | |
1291 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1292 | ||
6197aea7 DG |
1293 | ret = write_relayd_metadata_id(ctx->consumer_thread_pipe[1], |
1294 | stream, relayd); | |
f02e1e8a | 1295 | if (ret < 0) { |
f02e1e8a DG |
1296 | written = ret; |
1297 | goto end; | |
1298 | } | |
f02e1e8a DG |
1299 | } |
1300 | ||
1301 | while (len > 0) { | |
1302 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d)", | |
1303 | (unsigned long)offset, len, fd); | |
1304 | ret_splice = splice(fd, &offset, ctx->consumer_thread_pipe[1], NULL, len, | |
1305 | SPLICE_F_MOVE | SPLICE_F_MORE); | |
1306 | DBG("splice chan to pipe, ret %zd", ret_splice); | |
1307 | if (ret_splice < 0) { | |
1308 | PERROR("Error in relay splice"); | |
1309 | if (written == 0) { | |
1310 | written = ret_splice; | |
1311 | } | |
1312 | ret = errno; | |
1313 | goto splice_error; | |
1314 | } | |
1315 | ||
1316 | /* Handle stream on the relayd if the output is on the network */ | |
1317 | if (relayd) { | |
1318 | if (stream->metadata_flag) { | |
1319 | /* Update counter to fit the spliced data */ | |
1320 | ret_splice += sizeof(stream->relayd_stream_id); | |
1321 | len += sizeof(stream->relayd_stream_id); | |
1322 | /* | |
1323 | * We do this so the return value can match the len passed as | |
1324 | * argument to this function. | |
1325 | */ | |
1326 | written -= sizeof(stream->relayd_stream_id); | |
1327 | } | |
1328 | ||
6197aea7 | 1329 | ret = write_relayd_stream_header(stream, ret_splice, relayd); |
f02e1e8a DG |
1330 | if (ret >= 0) { |
1331 | /* Use the returned socket. */ | |
1332 | outfd = ret; | |
1333 | } else { | |
6197aea7 DG |
1334 | ERR("Remote relayd disconnected. Stopping"); |
1335 | goto end; | |
f02e1e8a DG |
1336 | } |
1337 | } | |
1338 | ||
1339 | /* Splice data out */ | |
1340 | ret_splice = splice(ctx->consumer_thread_pipe[0], NULL, outfd, NULL, | |
1341 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); | |
1342 | DBG("Kernel consumer splice pipe to file, ret %zd", ret_splice); | |
1343 | if (ret_splice < 0) { | |
1344 | PERROR("Error in file splice"); | |
1345 | if (written == 0) { | |
1346 | written = ret_splice; | |
1347 | } | |
1348 | ret = errno; | |
1349 | goto splice_error; | |
1350 | } else if (ret_splice > len) { | |
1351 | errno = EINVAL; | |
1352 | PERROR("Wrote more data than requested %zd (len: %lu)", | |
1353 | ret_splice, len); | |
1354 | written += ret_splice; | |
1355 | ret = errno; | |
1356 | goto splice_error; | |
1357 | } | |
1358 | len -= ret_splice; | |
1359 | ||
1360 | /* This call is useless on a socket so better save a syscall. */ | |
1361 | if (!relayd) { | |
1362 | /* This won't block, but will start writeout asynchronously */ | |
1363 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, | |
1364 | SYNC_FILE_RANGE_WRITE); | |
1365 | stream->out_fd_offset += ret_splice; | |
1366 | } | |
1367 | written += ret_splice; | |
1368 | } | |
1369 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1370 | ||
1371 | ret = ret_splice; | |
1372 | ||
1373 | goto end; | |
1374 | ||
1375 | splice_error: | |
1376 | /* send the appropriate error description to sessiond */ | |
1377 | switch (ret) { | |
1378 | case EBADF: | |
f73fabfd | 1379 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EBADF); |
f02e1e8a DG |
1380 | break; |
1381 | case EINVAL: | |
f73fabfd | 1382 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); |
f02e1e8a DG |
1383 | break; |
1384 | case ENOMEM: | |
f73fabfd | 1385 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); |
f02e1e8a DG |
1386 | break; |
1387 | case ESPIPE: | |
f73fabfd | 1388 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); |
f02e1e8a DG |
1389 | break; |
1390 | } | |
1391 | ||
1392 | end: | |
1393 | if (relayd && stream->metadata_flag) { | |
1394 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1395 | } | |
1396 | ||
1397 | rcu_read_unlock(); | |
1398 | return written; | |
3bd1e081 MD |
1399 | } |
1400 | ||
1401 | /* | |
1402 | * Take a snapshot for a specific fd | |
1403 | * | |
1404 | * Returns 0 on success, < 0 on error | |
1405 | */ | |
1406 | int lttng_consumer_take_snapshot(struct lttng_consumer_local_data *ctx, | |
1407 | struct lttng_consumer_stream *stream) | |
1408 | { | |
1409 | switch (consumer_data.type) { | |
1410 | case LTTNG_CONSUMER_KERNEL: | |
1411 | return lttng_kconsumer_take_snapshot(ctx, stream); | |
7753dea8 MD |
1412 | case LTTNG_CONSUMER32_UST: |
1413 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1414 | return lttng_ustconsumer_take_snapshot(ctx, stream); |
1415 | default: | |
1416 | ERR("Unknown consumer_data type"); | |
1417 | assert(0); | |
1418 | return -ENOSYS; | |
1419 | } | |
1420 | ||
1421 | } | |
1422 | ||
1423 | /* | |
1424 | * Get the produced position | |
1425 | * | |
1426 | * Returns 0 on success, < 0 on error | |
1427 | */ | |
1428 | int lttng_consumer_get_produced_snapshot( | |
1429 | struct lttng_consumer_local_data *ctx, | |
1430 | struct lttng_consumer_stream *stream, | |
1431 | unsigned long *pos) | |
1432 | { | |
1433 | switch (consumer_data.type) { | |
1434 | case LTTNG_CONSUMER_KERNEL: | |
1435 | return lttng_kconsumer_get_produced_snapshot(ctx, stream, pos); | |
7753dea8 MD |
1436 | case LTTNG_CONSUMER32_UST: |
1437 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1438 | return lttng_ustconsumer_get_produced_snapshot(ctx, stream, pos); |
1439 | default: | |
1440 | ERR("Unknown consumer_data type"); | |
1441 | assert(0); | |
1442 | return -ENOSYS; | |
1443 | } | |
1444 | } | |
1445 | ||
1446 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, | |
1447 | int sock, struct pollfd *consumer_sockpoll) | |
1448 | { | |
1449 | switch (consumer_data.type) { | |
1450 | case LTTNG_CONSUMER_KERNEL: | |
1451 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
7753dea8 MD |
1452 | case LTTNG_CONSUMER32_UST: |
1453 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1454 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
1455 | default: | |
1456 | ERR("Unknown consumer_data type"); | |
1457 | assert(0); | |
1458 | return -ENOSYS; | |
1459 | } | |
1460 | } | |
1461 | ||
1462 | /* | |
e4421fec | 1463 | * This thread polls the fds in the set to consume the data and write |
3bd1e081 MD |
1464 | * it to tracefile if necessary. |
1465 | */ | |
1466 | void *lttng_consumer_thread_poll_fds(void *data) | |
1467 | { | |
1468 | int num_rdy, num_hup, high_prio, ret, i; | |
1469 | struct pollfd *pollfd = NULL; | |
1470 | /* local view of the streams */ | |
1471 | struct lttng_consumer_stream **local_stream = NULL; | |
1472 | /* local view of consumer_data.fds_count */ | |
1473 | int nb_fd = 0; | |
3bd1e081 | 1474 | struct lttng_consumer_local_data *ctx = data; |
00e2e675 DG |
1475 | struct lttng_ht *metadata_ht; |
1476 | struct lttng_ht_iter iter; | |
1477 | struct lttng_ht_node_ulong *node; | |
1478 | struct lttng_consumer_stream *metadata_stream; | |
1479 | ssize_t len; | |
1480 | ||
1481 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
3bd1e081 | 1482 | |
e7b994a3 DG |
1483 | rcu_register_thread(); |
1484 | ||
effcf122 | 1485 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream)); |
3bd1e081 MD |
1486 | |
1487 | while (1) { | |
1488 | high_prio = 0; | |
1489 | num_hup = 0; | |
1490 | ||
1491 | /* | |
e4421fec | 1492 | * the fds set has been updated, we need to update our |
3bd1e081 MD |
1493 | * local array as well |
1494 | */ | |
1495 | pthread_mutex_lock(&consumer_data.lock); | |
1496 | if (consumer_data.need_update) { | |
1497 | if (pollfd != NULL) { | |
1498 | free(pollfd); | |
1499 | pollfd = NULL; | |
1500 | } | |
1501 | if (local_stream != NULL) { | |
1502 | free(local_stream); | |
1503 | local_stream = NULL; | |
1504 | } | |
1505 | ||
1506 | /* allocate for all fds + 1 for the consumer_poll_pipe */ | |
effcf122 | 1507 | pollfd = zmalloc((consumer_data.stream_count + 1) * sizeof(struct pollfd)); |
3bd1e081 MD |
1508 | if (pollfd == NULL) { |
1509 | perror("pollfd malloc"); | |
1510 | pthread_mutex_unlock(&consumer_data.lock); | |
1511 | goto end; | |
1512 | } | |
1513 | ||
1514 | /* allocate for all fds + 1 for the consumer_poll_pipe */ | |
effcf122 | 1515 | local_stream = zmalloc((consumer_data.stream_count + 1) * |
3bd1e081 MD |
1516 | sizeof(struct lttng_consumer_stream)); |
1517 | if (local_stream == NULL) { | |
1518 | perror("local_stream malloc"); | |
1519 | pthread_mutex_unlock(&consumer_data.lock); | |
1520 | goto end; | |
1521 | } | |
00e2e675 DG |
1522 | ret = consumer_update_poll_array(ctx, &pollfd, local_stream, |
1523 | metadata_ht); | |
3bd1e081 MD |
1524 | if (ret < 0) { |
1525 | ERR("Error in allocating pollfd or local_outfds"); | |
f73fabfd | 1526 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
1527 | pthread_mutex_unlock(&consumer_data.lock); |
1528 | goto end; | |
1529 | } | |
1530 | nb_fd = ret; | |
1531 | consumer_data.need_update = 0; | |
1532 | } | |
1533 | pthread_mutex_unlock(&consumer_data.lock); | |
1534 | ||
4078b776 MD |
1535 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
1536 | if (nb_fd == 0 && consumer_quit == 1) { | |
1537 | goto end; | |
1538 | } | |
3bd1e081 | 1539 | /* poll on the array of fds */ |
88f2b785 | 1540 | restart: |
3bd1e081 MD |
1541 | DBG("polling on %d fd", nb_fd + 1); |
1542 | num_rdy = poll(pollfd, nb_fd + 1, consumer_poll_timeout); | |
1543 | DBG("poll num_rdy : %d", num_rdy); | |
1544 | if (num_rdy == -1) { | |
88f2b785 MD |
1545 | /* |
1546 | * Restart interrupted system call. | |
1547 | */ | |
1548 | if (errno == EINTR) { | |
1549 | goto restart; | |
1550 | } | |
3bd1e081 | 1551 | perror("Poll error"); |
f73fabfd | 1552 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
1553 | goto end; |
1554 | } else if (num_rdy == 0) { | |
1555 | DBG("Polling thread timed out"); | |
1556 | goto end; | |
1557 | } | |
1558 | ||
3bd1e081 | 1559 | /* |
00e2e675 DG |
1560 | * If the consumer_poll_pipe triggered poll go directly to the |
1561 | * beginning of the loop to update the array. We want to prioritize | |
1562 | * array update over low-priority reads. | |
3bd1e081 | 1563 | */ |
509bb1cf | 1564 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
04fdd819 MD |
1565 | size_t pipe_readlen; |
1566 | char tmp; | |
1567 | ||
3bd1e081 | 1568 | DBG("consumer_poll_pipe wake up"); |
04fdd819 MD |
1569 | /* Consume 1 byte of pipe data */ |
1570 | do { | |
1571 | pipe_readlen = read(ctx->consumer_poll_pipe[0], &tmp, 1); | |
1572 | } while (pipe_readlen == -1 && errno == EINTR); | |
3bd1e081 MD |
1573 | continue; |
1574 | } | |
1575 | ||
1576 | /* Take care of high priority channels first. */ | |
1577 | for (i = 0; i < nb_fd; i++) { | |
00e2e675 DG |
1578 | /* Lookup for metadata which is the highest priority */ |
1579 | lttng_ht_lookup(metadata_ht, | |
1580 | (void *)((unsigned long) pollfd[i].fd), &iter); | |
1581 | node = lttng_ht_iter_get_node_ulong(&iter); | |
1582 | if (node != NULL && | |
1583 | (pollfd[i].revents & (POLLIN | POLLPRI))) { | |
1584 | DBG("Urgent metadata read on fd %d", pollfd[i].fd); | |
1585 | metadata_stream = caa_container_of(node, | |
1586 | struct lttng_consumer_stream, waitfd_node); | |
1587 | high_prio = 1; | |
1588 | len = ctx->on_buffer_ready(metadata_stream, ctx); | |
1589 | /* it's ok to have an unavailable sub-buffer */ | |
1590 | if (len < 0 && len != -EAGAIN) { | |
1591 | goto end; | |
1592 | } else if (len > 0) { | |
1593 | metadata_stream->data_read = 1; | |
1594 | } | |
1595 | } else if (pollfd[i].revents & POLLPRI) { | |
d41f73b7 MD |
1596 | DBG("Urgent read on fd %d", pollfd[i].fd); |
1597 | high_prio = 1; | |
4078b776 | 1598 | len = ctx->on_buffer_ready(local_stream[i], ctx); |
d41f73b7 | 1599 | /* it's ok to have an unavailable sub-buffer */ |
4078b776 MD |
1600 | if (len < 0 && len != -EAGAIN) { |
1601 | goto end; | |
1602 | } else if (len > 0) { | |
1603 | local_stream[i]->data_read = 1; | |
d41f73b7 | 1604 | } |
3bd1e081 MD |
1605 | } |
1606 | } | |
1607 | ||
4078b776 MD |
1608 | /* |
1609 | * If we read high prio channel in this loop, try again | |
1610 | * for more high prio data. | |
1611 | */ | |
1612 | if (high_prio) { | |
3bd1e081 MD |
1613 | continue; |
1614 | } | |
1615 | ||
1616 | /* Take care of low priority channels. */ | |
4078b776 MD |
1617 | for (i = 0; i < nb_fd; i++) { |
1618 | if ((pollfd[i].revents & POLLIN) || | |
1619 | local_stream[i]->hangup_flush_done) { | |
4078b776 MD |
1620 | DBG("Normal read on fd %d", pollfd[i].fd); |
1621 | len = ctx->on_buffer_ready(local_stream[i], ctx); | |
1622 | /* it's ok to have an unavailable sub-buffer */ | |
1623 | if (len < 0 && len != -EAGAIN) { | |
1624 | goto end; | |
1625 | } else if (len > 0) { | |
1626 | local_stream[i]->data_read = 1; | |
1627 | } | |
1628 | } | |
1629 | } | |
1630 | ||
1631 | /* Handle hangup and errors */ | |
1632 | for (i = 0; i < nb_fd; i++) { | |
1633 | if (!local_stream[i]->hangup_flush_done | |
1634 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) | |
1635 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
1636 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
1637 | DBG("fd %d is hup|err|nval. Attempting flush and read.", | |
1638 | pollfd[i].fd); | |
1639 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); | |
1640 | /* Attempt read again, for the data we just flushed. */ | |
1641 | local_stream[i]->data_read = 1; | |
1642 | } | |
1643 | /* | |
1644 | * If the poll flag is HUP/ERR/NVAL and we have | |
1645 | * read no data in this pass, we can remove the | |
1646 | * stream from its hash table. | |
1647 | */ | |
1648 | if ((pollfd[i].revents & POLLHUP)) { | |
1649 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); | |
1650 | if (!local_stream[i]->data_read) { | |
00e2e675 DG |
1651 | if (local_stream[i]->metadata_flag) { |
1652 | iter.iter.node = &local_stream[i]->waitfd_node.node; | |
1653 | ret = lttng_ht_del(metadata_ht, &iter); | |
1654 | assert(!ret); | |
1655 | } | |
702b1ea4 | 1656 | consumer_del_stream(local_stream[i]); |
4078b776 MD |
1657 | num_hup++; |
1658 | } | |
1659 | } else if (pollfd[i].revents & POLLERR) { | |
1660 | ERR("Error returned in polling fd %d.", pollfd[i].fd); | |
1661 | if (!local_stream[i]->data_read) { | |
00e2e675 DG |
1662 | if (local_stream[i]->metadata_flag) { |
1663 | iter.iter.node = &local_stream[i]->waitfd_node.node; | |
1664 | ret = lttng_ht_del(metadata_ht, &iter); | |
1665 | assert(!ret); | |
1666 | } | |
702b1ea4 | 1667 | consumer_del_stream(local_stream[i]); |
4078b776 MD |
1668 | num_hup++; |
1669 | } | |
1670 | } else if (pollfd[i].revents & POLLNVAL) { | |
1671 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); | |
1672 | if (!local_stream[i]->data_read) { | |
00e2e675 DG |
1673 | if (local_stream[i]->metadata_flag) { |
1674 | iter.iter.node = &local_stream[i]->waitfd_node.node; | |
1675 | ret = lttng_ht_del(metadata_ht, &iter); | |
1676 | assert(!ret); | |
1677 | } | |
702b1ea4 | 1678 | consumer_del_stream(local_stream[i]); |
4078b776 | 1679 | num_hup++; |
3bd1e081 MD |
1680 | } |
1681 | } | |
4078b776 | 1682 | local_stream[i]->data_read = 0; |
3bd1e081 MD |
1683 | } |
1684 | } | |
1685 | end: | |
1686 | DBG("polling thread exiting"); | |
1687 | if (pollfd != NULL) { | |
1688 | free(pollfd); | |
1689 | pollfd = NULL; | |
1690 | } | |
1691 | if (local_stream != NULL) { | |
1692 | free(local_stream); | |
1693 | local_stream = NULL; | |
1694 | } | |
e7b994a3 | 1695 | rcu_unregister_thread(); |
3bd1e081 MD |
1696 | return NULL; |
1697 | } | |
1698 | ||
1699 | /* | |
1700 | * This thread listens on the consumerd socket and receives the file | |
1701 | * descriptors from the session daemon. | |
1702 | */ | |
1703 | void *lttng_consumer_thread_receive_fds(void *data) | |
1704 | { | |
1705 | int sock, client_socket, ret; | |
1706 | /* | |
1707 | * structure to poll for incoming data on communication socket avoids | |
1708 | * making blocking sockets. | |
1709 | */ | |
1710 | struct pollfd consumer_sockpoll[2]; | |
1711 | struct lttng_consumer_local_data *ctx = data; | |
1712 | ||
e7b994a3 DG |
1713 | rcu_register_thread(); |
1714 | ||
3bd1e081 MD |
1715 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
1716 | unlink(ctx->consumer_command_sock_path); | |
1717 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); | |
1718 | if (client_socket < 0) { | |
1719 | ERR("Cannot create command socket"); | |
1720 | goto end; | |
1721 | } | |
1722 | ||
1723 | ret = lttcomm_listen_unix_sock(client_socket); | |
1724 | if (ret < 0) { | |
1725 | goto end; | |
1726 | } | |
1727 | ||
32258573 | 1728 | DBG("Sending ready command to lttng-sessiond"); |
f73fabfd | 1729 | ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); |
3bd1e081 MD |
1730 | /* return < 0 on error, but == 0 is not fatal */ |
1731 | if (ret < 0) { | |
32258573 | 1732 | ERR("Error sending ready command to lttng-sessiond"); |
3bd1e081 MD |
1733 | goto end; |
1734 | } | |
1735 | ||
1736 | ret = fcntl(client_socket, F_SETFL, O_NONBLOCK); | |
1737 | if (ret < 0) { | |
1738 | perror("fcntl O_NONBLOCK"); | |
1739 | goto end; | |
1740 | } | |
1741 | ||
1742 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ | |
1743 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; | |
1744 | consumer_sockpoll[0].events = POLLIN | POLLPRI; | |
1745 | consumer_sockpoll[1].fd = client_socket; | |
1746 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
1747 | ||
1748 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
1749 | goto end; | |
1750 | } | |
1751 | DBG("Connection on client_socket"); | |
1752 | ||
1753 | /* Blocking call, waiting for transmission */ | |
1754 | sock = lttcomm_accept_unix_sock(client_socket); | |
1755 | if (sock <= 0) { | |
1756 | WARN("On accept"); | |
1757 | goto end; | |
1758 | } | |
1759 | ret = fcntl(sock, F_SETFL, O_NONBLOCK); | |
1760 | if (ret < 0) { | |
1761 | perror("fcntl O_NONBLOCK"); | |
1762 | goto end; | |
1763 | } | |
1764 | ||
1765 | /* update the polling structure to poll on the established socket */ | |
1766 | consumer_sockpoll[1].fd = sock; | |
1767 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
1768 | ||
1769 | while (1) { | |
1770 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
1771 | goto end; | |
1772 | } | |
1773 | DBG("Incoming command on sock"); | |
1774 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
1775 | if (ret == -ENOENT) { | |
1776 | DBG("Received STOP command"); | |
1777 | goto end; | |
1778 | } | |
1779 | if (ret < 0) { | |
1780 | ERR("Communication interrupted on command socket"); | |
1781 | goto end; | |
1782 | } | |
1783 | if (consumer_quit) { | |
1784 | DBG("consumer_thread_receive_fds received quit from signal"); | |
1785 | goto end; | |
1786 | } | |
1787 | DBG("received fds on sock"); | |
1788 | } | |
1789 | end: | |
1790 | DBG("consumer_thread_receive_fds exiting"); | |
1791 | ||
1792 | /* | |
1793 | * when all fds have hung up, the polling thread | |
1794 | * can exit cleanly | |
1795 | */ | |
1796 | consumer_quit = 1; | |
1797 | ||
1798 | /* | |
1799 | * 2s of grace period, if no polling events occur during | |
1800 | * this period, the polling thread will exit even if there | |
1801 | * are still open FDs (should not happen, but safety mechanism). | |
1802 | */ | |
1803 | consumer_poll_timeout = LTTNG_CONSUMER_POLL_TIMEOUT; | |
1804 | ||
04fdd819 MD |
1805 | /* |
1806 | * Wake-up the other end by writing a null byte in the pipe | |
1807 | * (non-blocking). Important note: Because writing into the | |
1808 | * pipe is non-blocking (and therefore we allow dropping wakeup | |
1809 | * data, as long as there is wakeup data present in the pipe | |
1810 | * buffer to wake up the other end), the other end should | |
1811 | * perform the following sequence for waiting: | |
1812 | * 1) empty the pipe (reads). | |
1813 | * 2) perform update operation. | |
1814 | * 3) wait on the pipe (poll). | |
1815 | */ | |
1816 | do { | |
1817 | ret = write(ctx->consumer_poll_pipe[1], "", 1); | |
6f94560a | 1818 | } while (ret < 0 && errno == EINTR); |
e7b994a3 | 1819 | rcu_unregister_thread(); |
3bd1e081 MD |
1820 | return NULL; |
1821 | } | |
d41f73b7 | 1822 | |
4078b776 | 1823 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
d41f73b7 MD |
1824 | struct lttng_consumer_local_data *ctx) |
1825 | { | |
1826 | switch (consumer_data.type) { | |
1827 | case LTTNG_CONSUMER_KERNEL: | |
1828 | return lttng_kconsumer_read_subbuffer(stream, ctx); | |
7753dea8 MD |
1829 | case LTTNG_CONSUMER32_UST: |
1830 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
1831 | return lttng_ustconsumer_read_subbuffer(stream, ctx); |
1832 | default: | |
1833 | ERR("Unknown consumer_data type"); | |
1834 | assert(0); | |
1835 | return -ENOSYS; | |
1836 | } | |
1837 | } | |
1838 | ||
1839 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
1840 | { | |
1841 | switch (consumer_data.type) { | |
1842 | case LTTNG_CONSUMER_KERNEL: | |
1843 | return lttng_kconsumer_on_recv_stream(stream); | |
7753dea8 MD |
1844 | case LTTNG_CONSUMER32_UST: |
1845 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
1846 | return lttng_ustconsumer_on_recv_stream(stream); |
1847 | default: | |
1848 | ERR("Unknown consumer_data type"); | |
1849 | assert(0); | |
1850 | return -ENOSYS; | |
1851 | } | |
1852 | } | |
e4421fec DG |
1853 | |
1854 | /* | |
1855 | * Allocate and set consumer data hash tables. | |
1856 | */ | |
1857 | void lttng_consumer_init(void) | |
1858 | { | |
1859 | consumer_data.stream_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
1860 | consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
00e2e675 | 1861 | consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); |
e4421fec | 1862 | } |
7735ef9e DG |
1863 | |
1864 | /* | |
1865 | * Process the ADD_RELAYD command receive by a consumer. | |
1866 | * | |
1867 | * This will create a relayd socket pair and add it to the relayd hash table. | |
1868 | * The caller MUST acquire a RCU read side lock before calling it. | |
1869 | */ | |
1870 | int consumer_add_relayd_socket(int net_seq_idx, int sock_type, | |
1871 | struct lttng_consumer_local_data *ctx, int sock, | |
1872 | struct pollfd *consumer_sockpoll, struct lttcomm_sock *relayd_sock) | |
1873 | { | |
1874 | int fd, ret = -1; | |
1875 | struct consumer_relayd_sock_pair *relayd; | |
1876 | ||
1877 | DBG("Consumer adding relayd socket (idx: %d)", net_seq_idx); | |
1878 | ||
1879 | /* Get relayd reference if exists. */ | |
1880 | relayd = consumer_find_relayd(net_seq_idx); | |
1881 | if (relayd == NULL) { | |
1882 | /* Not found. Allocate one. */ | |
1883 | relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); | |
1884 | if (relayd == NULL) { | |
f73fabfd | 1885 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); |
7735ef9e DG |
1886 | goto error; |
1887 | } | |
1888 | } | |
1889 | ||
1890 | /* Poll on consumer socket. */ | |
1891 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
1892 | ret = -EINTR; | |
1893 | goto error; | |
1894 | } | |
1895 | ||
1896 | /* Get relayd socket from session daemon */ | |
1897 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); | |
1898 | if (ret != sizeof(fd)) { | |
f73fabfd | 1899 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); |
7735ef9e DG |
1900 | ret = -1; |
1901 | goto error; | |
1902 | } | |
1903 | ||
1904 | /* Copy socket information and received FD */ | |
1905 | switch (sock_type) { | |
1906 | case LTTNG_STREAM_CONTROL: | |
1907 | /* Copy received lttcomm socket */ | |
1908 | lttcomm_copy_sock(&relayd->control_sock, relayd_sock); | |
1909 | ret = lttcomm_create_sock(&relayd->control_sock); | |
1910 | if (ret < 0) { | |
1911 | goto error; | |
1912 | } | |
1913 | ||
1914 | /* Close the created socket fd which is useless */ | |
1915 | close(relayd->control_sock.fd); | |
1916 | ||
1917 | /* Assign new file descriptor */ | |
1918 | relayd->control_sock.fd = fd; | |
1919 | break; | |
1920 | case LTTNG_STREAM_DATA: | |
1921 | /* Copy received lttcomm socket */ | |
1922 | lttcomm_copy_sock(&relayd->data_sock, relayd_sock); | |
1923 | ret = lttcomm_create_sock(&relayd->data_sock); | |
1924 | if (ret < 0) { | |
1925 | goto error; | |
1926 | } | |
1927 | ||
1928 | /* Close the created socket fd which is useless */ | |
1929 | close(relayd->data_sock.fd); | |
1930 | ||
1931 | /* Assign new file descriptor */ | |
1932 | relayd->data_sock.fd = fd; | |
1933 | break; | |
1934 | default: | |
1935 | ERR("Unknown relayd socket type (%d)", sock_type); | |
1936 | goto error; | |
1937 | } | |
1938 | ||
1939 | DBG("Consumer %s socket created successfully with net idx %d (fd: %d)", | |
1940 | sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", | |
1941 | relayd->net_seq_idx, fd); | |
1942 | ||
1943 | /* | |
1944 | * Add relayd socket pair to consumer data hashtable. If object already | |
1945 | * exists or on error, the function gracefully returns. | |
1946 | */ | |
1947 | consumer_add_relayd(relayd); | |
1948 | ||
1949 | /* All good! */ | |
1950 | ret = 0; | |
1951 | ||
1952 | error: | |
1953 | return ret; | |
1954 | } |