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> | |
30 | ||
990570ed | 31 | #include <common/common.h> |
10a8a223 | 32 | #include <common/kernel-ctl/kernel-ctl.h> |
00e2e675 | 33 | #include <common/sessiond-comm/relayd.h> |
10a8a223 DG |
34 | #include <common/sessiond-comm/sessiond-comm.h> |
35 | #include <common/kernel-consumer/kernel-consumer.h> | |
00e2e675 | 36 | #include <common/relayd/relayd.h> |
10a8a223 DG |
37 | #include <common/ust-consumer/ust-consumer.h> |
38 | ||
39 | #include "consumer.h" | |
3bd1e081 MD |
40 | |
41 | struct lttng_consumer_global_data consumer_data = { | |
3bd1e081 MD |
42 | .stream_count = 0, |
43 | .need_update = 1, | |
44 | .type = LTTNG_CONSUMER_UNKNOWN, | |
45 | }; | |
46 | ||
47 | /* timeout parameter, to control the polling thread grace period. */ | |
48 | int consumer_poll_timeout = -1; | |
49 | ||
50 | /* | |
51 | * Flag to inform the polling thread to quit when all fd hung up. Updated by | |
52 | * the consumer_thread_receive_fds when it notices that all fds has hung up. | |
53 | * Also updated by the signal handler (consumer_should_exit()). Read by the | |
54 | * polling threads. | |
55 | */ | |
56 | volatile int consumer_quit = 0; | |
57 | ||
58 | /* | |
59 | * Find a stream. The consumer_data.lock must be locked during this | |
60 | * call. | |
61 | */ | |
62 | static struct lttng_consumer_stream *consumer_find_stream(int key) | |
63 | { | |
e4421fec DG |
64 | struct lttng_ht_iter iter; |
65 | struct lttng_ht_node_ulong *node; | |
66 | struct lttng_consumer_stream *stream = NULL; | |
3bd1e081 | 67 | |
7ad0a0cb MD |
68 | /* Negative keys are lookup failures */ |
69 | if (key < 0) | |
70 | return NULL; | |
e4421fec | 71 | |
6065ceec DG |
72 | rcu_read_lock(); |
73 | ||
e4421fec DG |
74 | lttng_ht_lookup(consumer_data.stream_ht, (void *)((unsigned long) key), |
75 | &iter); | |
76 | node = lttng_ht_iter_get_node_ulong(&iter); | |
77 | if (node != NULL) { | |
78 | stream = caa_container_of(node, struct lttng_consumer_stream, node); | |
3bd1e081 | 79 | } |
e4421fec | 80 | |
6065ceec DG |
81 | rcu_read_unlock(); |
82 | ||
e4421fec | 83 | return stream; |
3bd1e081 MD |
84 | } |
85 | ||
7ad0a0cb MD |
86 | static void consumer_steal_stream_key(int key) |
87 | { | |
88 | struct lttng_consumer_stream *stream; | |
89 | ||
04253271 | 90 | rcu_read_lock(); |
7ad0a0cb | 91 | stream = consumer_find_stream(key); |
04253271 | 92 | if (stream) { |
7ad0a0cb | 93 | stream->key = -1; |
04253271 MD |
94 | /* |
95 | * We don't want the lookup to match, but we still need | |
96 | * to iterate on this stream when iterating over the hash table. Just | |
97 | * change the node key. | |
98 | */ | |
99 | stream->node.key = -1; | |
100 | } | |
101 | rcu_read_unlock(); | |
7ad0a0cb MD |
102 | } |
103 | ||
3bd1e081 MD |
104 | static struct lttng_consumer_channel *consumer_find_channel(int key) |
105 | { | |
e4421fec DG |
106 | struct lttng_ht_iter iter; |
107 | struct lttng_ht_node_ulong *node; | |
108 | struct lttng_consumer_channel *channel = NULL; | |
3bd1e081 | 109 | |
7ad0a0cb MD |
110 | /* Negative keys are lookup failures */ |
111 | if (key < 0) | |
112 | return NULL; | |
e4421fec | 113 | |
6065ceec DG |
114 | rcu_read_lock(); |
115 | ||
e4421fec DG |
116 | lttng_ht_lookup(consumer_data.channel_ht, (void *)((unsigned long) key), |
117 | &iter); | |
118 | node = lttng_ht_iter_get_node_ulong(&iter); | |
119 | if (node != NULL) { | |
120 | channel = caa_container_of(node, struct lttng_consumer_channel, node); | |
3bd1e081 | 121 | } |
e4421fec | 122 | |
6065ceec DG |
123 | rcu_read_unlock(); |
124 | ||
e4421fec | 125 | return channel; |
3bd1e081 MD |
126 | } |
127 | ||
7ad0a0cb MD |
128 | static void consumer_steal_channel_key(int key) |
129 | { | |
130 | struct lttng_consumer_channel *channel; | |
131 | ||
04253271 | 132 | rcu_read_lock(); |
7ad0a0cb | 133 | channel = consumer_find_channel(key); |
04253271 | 134 | if (channel) { |
7ad0a0cb | 135 | channel->key = -1; |
04253271 MD |
136 | /* |
137 | * We don't want the lookup to match, but we still need | |
138 | * to iterate on this channel when iterating over the hash table. Just | |
139 | * change the node key. | |
140 | */ | |
141 | channel->node.key = -1; | |
142 | } | |
143 | rcu_read_unlock(); | |
7ad0a0cb MD |
144 | } |
145 | ||
702b1ea4 MD |
146 | static |
147 | void consumer_free_stream(struct rcu_head *head) | |
148 | { | |
149 | struct lttng_ht_node_ulong *node = | |
150 | caa_container_of(head, struct lttng_ht_node_ulong, head); | |
151 | struct lttng_consumer_stream *stream = | |
152 | caa_container_of(node, struct lttng_consumer_stream, node); | |
153 | ||
154 | free(stream); | |
155 | } | |
156 | ||
00e2e675 DG |
157 | /* |
158 | * RCU protected relayd socket pair free. | |
159 | */ | |
160 | static void consumer_rcu_free_relayd(struct rcu_head *head) | |
161 | { | |
162 | struct lttng_ht_node_ulong *node = | |
163 | caa_container_of(head, struct lttng_ht_node_ulong, head); | |
164 | struct consumer_relayd_sock_pair *relayd = | |
165 | caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
166 | ||
167 | free(relayd); | |
168 | } | |
169 | ||
170 | /* | |
171 | * Destroy and free relayd socket pair object. | |
172 | * | |
173 | * This function MUST be called with the consumer_data lock acquired. | |
174 | */ | |
175 | void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) | |
176 | { | |
177 | int ret; | |
178 | struct lttng_ht_iter iter; | |
179 | ||
173af62f DG |
180 | if (relayd == NULL) { |
181 | return; | |
182 | } | |
183 | ||
00e2e675 DG |
184 | DBG("Consumer destroy and close relayd socket pair"); |
185 | ||
186 | iter.iter.node = &relayd->node.node; | |
187 | ret = lttng_ht_del(consumer_data.relayd_ht, &iter); | |
173af62f DG |
188 | if (ret != 0) { |
189 | /* We assume the relayd was already destroyed */ | |
190 | return; | |
191 | } | |
00e2e675 DG |
192 | |
193 | /* Close all sockets */ | |
194 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
195 | (void) relayd_close(&relayd->control_sock); | |
196 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
197 | (void) relayd_close(&relayd->data_sock); | |
198 | ||
199 | /* RCU free() call */ | |
200 | call_rcu(&relayd->node.head, consumer_rcu_free_relayd); | |
201 | } | |
202 | ||
3bd1e081 MD |
203 | /* |
204 | * Remove a stream from the global list protected by a mutex. This | |
205 | * function is also responsible for freeing its data structures. | |
206 | */ | |
207 | void consumer_del_stream(struct lttng_consumer_stream *stream) | |
208 | { | |
209 | int ret; | |
e4421fec | 210 | struct lttng_ht_iter iter; |
3bd1e081 | 211 | struct lttng_consumer_channel *free_chan = NULL; |
00e2e675 DG |
212 | struct consumer_relayd_sock_pair *relayd; |
213 | ||
214 | assert(stream); | |
3bd1e081 MD |
215 | |
216 | pthread_mutex_lock(&consumer_data.lock); | |
217 | ||
218 | switch (consumer_data.type) { | |
219 | case LTTNG_CONSUMER_KERNEL: | |
220 | if (stream->mmap_base != NULL) { | |
221 | ret = munmap(stream->mmap_base, stream->mmap_len); | |
222 | if (ret != 0) { | |
223 | perror("munmap"); | |
224 | } | |
225 | } | |
226 | break; | |
7753dea8 MD |
227 | case LTTNG_CONSUMER32_UST: |
228 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
229 | lttng_ustconsumer_del_stream(stream); |
230 | break; | |
231 | default: | |
232 | ERR("Unknown consumer_data type"); | |
233 | assert(0); | |
234 | goto end; | |
235 | } | |
236 | ||
6065ceec | 237 | rcu_read_lock(); |
04253271 MD |
238 | iter.iter.node = &stream->node.node; |
239 | ret = lttng_ht_del(consumer_data.stream_ht, &iter); | |
240 | assert(!ret); | |
e4421fec | 241 | |
6065ceec DG |
242 | rcu_read_unlock(); |
243 | ||
3bd1e081 MD |
244 | if (consumer_data.stream_count <= 0) { |
245 | goto end; | |
246 | } | |
247 | consumer_data.stream_count--; | |
248 | if (!stream) { | |
249 | goto end; | |
250 | } | |
251 | if (stream->out_fd >= 0) { | |
4c462e79 MD |
252 | ret = close(stream->out_fd); |
253 | if (ret) { | |
254 | PERROR("close"); | |
255 | } | |
3bd1e081 | 256 | } |
b5c5fc29 | 257 | if (stream->wait_fd >= 0 && !stream->wait_fd_is_copy) { |
4c462e79 MD |
258 | ret = close(stream->wait_fd); |
259 | if (ret) { | |
260 | PERROR("close"); | |
261 | } | |
3bd1e081 | 262 | } |
2c1dd183 | 263 | if (stream->shm_fd >= 0 && stream->wait_fd != stream->shm_fd) { |
4c462e79 MD |
264 | ret = close(stream->shm_fd); |
265 | if (ret) { | |
266 | PERROR("close"); | |
267 | } | |
3bd1e081 | 268 | } |
00e2e675 DG |
269 | |
270 | /* Check and cleanup relayd */ | |
b0b335c8 | 271 | rcu_read_lock(); |
00e2e675 DG |
272 | relayd = consumer_find_relayd(stream->net_seq_idx); |
273 | if (relayd != NULL) { | |
b0b335c8 MD |
274 | uatomic_dec(&relayd->refcount); |
275 | assert(uatomic_read(&relayd->refcount) >= 0); | |
173af62f DG |
276 | |
277 | ret = relayd_send_close_stream(&relayd->control_sock, | |
278 | stream->relayd_stream_id, | |
279 | stream->next_net_seq_num - 1); | |
280 | if (ret < 0) { | |
a4b92340 DG |
281 | DBG("Unable to close stream on the relayd. Continuing"); |
282 | /* | |
283 | * Continue here. There is nothing we can do for the relayd. | |
284 | * Chances are that the relayd has closed the socket so we just | |
285 | * continue cleaning up. | |
286 | */ | |
173af62f DG |
287 | } |
288 | ||
289 | /* Both conditions are met, we destroy the relayd. */ | |
290 | if (uatomic_read(&relayd->refcount) == 0 && | |
291 | uatomic_read(&relayd->destroy_flag)) { | |
00e2e675 DG |
292 | consumer_destroy_relayd(relayd); |
293 | } | |
00e2e675 | 294 | } |
b0b335c8 | 295 | rcu_read_unlock(); |
00e2e675 DG |
296 | |
297 | if (!--stream->chan->refcount) { | |
3bd1e081 | 298 | free_chan = stream->chan; |
00e2e675 DG |
299 | } |
300 | ||
702b1ea4 MD |
301 | |
302 | call_rcu(&stream->node.head, consumer_free_stream); | |
3bd1e081 MD |
303 | end: |
304 | consumer_data.need_update = 1; | |
305 | pthread_mutex_unlock(&consumer_data.lock); | |
306 | ||
307 | if (free_chan) | |
308 | consumer_del_channel(free_chan); | |
309 | } | |
310 | ||
311 | struct lttng_consumer_stream *consumer_allocate_stream( | |
312 | int channel_key, int stream_key, | |
313 | int shm_fd, int wait_fd, | |
314 | enum lttng_consumer_stream_state state, | |
315 | uint64_t mmap_len, | |
316 | enum lttng_event_output output, | |
6df2e2c9 MD |
317 | const char *path_name, |
318 | uid_t uid, | |
00e2e675 DG |
319 | gid_t gid, |
320 | int net_index, | |
321 | int metadata_flag) | |
3bd1e081 MD |
322 | { |
323 | struct lttng_consumer_stream *stream; | |
324 | int ret; | |
325 | ||
effcf122 | 326 | stream = zmalloc(sizeof(*stream)); |
3bd1e081 MD |
327 | if (stream == NULL) { |
328 | perror("malloc struct lttng_consumer_stream"); | |
329 | goto end; | |
330 | } | |
331 | stream->chan = consumer_find_channel(channel_key); | |
332 | if (!stream->chan) { | |
333 | perror("Unable to find channel key"); | |
334 | goto end; | |
335 | } | |
336 | stream->chan->refcount++; | |
337 | stream->key = stream_key; | |
338 | stream->shm_fd = shm_fd; | |
339 | stream->wait_fd = wait_fd; | |
340 | stream->out_fd = -1; | |
341 | stream->out_fd_offset = 0; | |
342 | stream->state = state; | |
343 | stream->mmap_len = mmap_len; | |
344 | stream->mmap_base = NULL; | |
345 | stream->output = output; | |
6df2e2c9 MD |
346 | stream->uid = uid; |
347 | stream->gid = gid; | |
00e2e675 DG |
348 | stream->net_seq_idx = net_index; |
349 | stream->metadata_flag = metadata_flag; | |
350 | strncpy(stream->path_name, path_name, sizeof(stream->path_name)); | |
351 | stream->path_name[sizeof(stream->path_name) - 1] = '\0'; | |
e4421fec | 352 | lttng_ht_node_init_ulong(&stream->node, stream->key); |
00e2e675 | 353 | lttng_ht_node_init_ulong(&stream->waitfd_node, stream->wait_fd); |
3bd1e081 MD |
354 | |
355 | switch (consumer_data.type) { | |
356 | case LTTNG_CONSUMER_KERNEL: | |
357 | break; | |
7753dea8 MD |
358 | case LTTNG_CONSUMER32_UST: |
359 | case LTTNG_CONSUMER64_UST: | |
5af2f756 | 360 | stream->cpu = stream->chan->cpucount++; |
3bd1e081 MD |
361 | ret = lttng_ustconsumer_allocate_stream(stream); |
362 | if (ret) { | |
363 | free(stream); | |
364 | return NULL; | |
365 | } | |
366 | break; | |
367 | default: | |
368 | ERR("Unknown consumer_data type"); | |
369 | assert(0); | |
370 | goto end; | |
371 | } | |
00e2e675 | 372 | DBG("Allocated stream %s (key %d, shm_fd %d, wait_fd %d, mmap_len %llu, out_fd %d, net_seq_idx %d)", |
3bd1e081 MD |
373 | stream->path_name, stream->key, |
374 | stream->shm_fd, | |
375 | stream->wait_fd, | |
376 | (unsigned long long) stream->mmap_len, | |
00e2e675 DG |
377 | stream->out_fd, |
378 | stream->net_seq_idx); | |
3bd1e081 MD |
379 | end: |
380 | return stream; | |
381 | } | |
382 | ||
383 | /* | |
384 | * Add a stream to the global list protected by a mutex. | |
385 | */ | |
386 | int consumer_add_stream(struct lttng_consumer_stream *stream) | |
387 | { | |
388 | int ret = 0; | |
c77fc10a DG |
389 | struct lttng_ht_node_ulong *node; |
390 | struct lttng_ht_iter iter; | |
00e2e675 | 391 | struct consumer_relayd_sock_pair *relayd; |
3bd1e081 MD |
392 | |
393 | pthread_mutex_lock(&consumer_data.lock); | |
7ad0a0cb MD |
394 | /* Steal stream identifier, for UST */ |
395 | consumer_steal_stream_key(stream->key); | |
c77fc10a | 396 | |
b0b335c8 | 397 | rcu_read_lock(); |
c77fc10a DG |
398 | lttng_ht_lookup(consumer_data.stream_ht, |
399 | (void *)((unsigned long) stream->key), &iter); | |
400 | node = lttng_ht_iter_get_node_ulong(&iter); | |
401 | if (node != NULL) { | |
402 | rcu_read_unlock(); | |
403 | /* Stream already exist. Ignore the insertion */ | |
404 | goto end; | |
405 | } | |
406 | ||
04253271 | 407 | lttng_ht_add_unique_ulong(consumer_data.stream_ht, &stream->node); |
00e2e675 DG |
408 | |
409 | /* Check and cleanup relayd */ | |
410 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
411 | if (relayd != NULL) { | |
b0b335c8 | 412 | uatomic_inc(&relayd->refcount); |
00e2e675 | 413 | } |
b0b335c8 | 414 | rcu_read_unlock(); |
00e2e675 DG |
415 | |
416 | /* Update consumer data */ | |
3bd1e081 MD |
417 | consumer_data.stream_count++; |
418 | consumer_data.need_update = 1; | |
419 | ||
420 | switch (consumer_data.type) { | |
421 | case LTTNG_CONSUMER_KERNEL: | |
422 | break; | |
7753dea8 MD |
423 | case LTTNG_CONSUMER32_UST: |
424 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
425 | /* Streams are in CPU number order (we rely on this) */ |
426 | stream->cpu = stream->chan->nr_streams++; | |
427 | break; | |
428 | default: | |
429 | ERR("Unknown consumer_data type"); | |
430 | assert(0); | |
431 | goto end; | |
432 | } | |
433 | ||
434 | end: | |
435 | pthread_mutex_unlock(&consumer_data.lock); | |
702b1ea4 | 436 | |
3bd1e081 MD |
437 | return ret; |
438 | } | |
439 | ||
00e2e675 DG |
440 | /* |
441 | * Add relayd socket to global consumer data hashtable. | |
442 | */ | |
443 | int consumer_add_relayd(struct consumer_relayd_sock_pair *relayd) | |
444 | { | |
445 | int ret = 0; | |
446 | struct lttng_ht_node_ulong *node; | |
447 | struct lttng_ht_iter iter; | |
448 | ||
449 | if (relayd == NULL) { | |
450 | ret = -1; | |
451 | goto end; | |
452 | } | |
453 | ||
454 | rcu_read_lock(); | |
455 | ||
456 | lttng_ht_lookup(consumer_data.relayd_ht, | |
457 | (void *)((unsigned long) relayd->net_seq_idx), &iter); | |
458 | node = lttng_ht_iter_get_node_ulong(&iter); | |
459 | if (node != NULL) { | |
460 | rcu_read_unlock(); | |
461 | /* Relayd already exist. Ignore the insertion */ | |
462 | goto end; | |
463 | } | |
464 | lttng_ht_add_unique_ulong(consumer_data.relayd_ht, &relayd->node); | |
465 | ||
466 | rcu_read_unlock(); | |
467 | ||
468 | end: | |
469 | return ret; | |
470 | } | |
471 | ||
472 | /* | |
473 | * Allocate and return a consumer relayd socket. | |
474 | */ | |
475 | struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( | |
476 | int net_seq_idx) | |
477 | { | |
478 | struct consumer_relayd_sock_pair *obj = NULL; | |
479 | ||
480 | /* Negative net sequence index is a failure */ | |
481 | if (net_seq_idx < 0) { | |
482 | goto error; | |
483 | } | |
484 | ||
485 | obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); | |
486 | if (obj == NULL) { | |
487 | PERROR("zmalloc relayd sock"); | |
488 | goto error; | |
489 | } | |
490 | ||
491 | obj->net_seq_idx = net_seq_idx; | |
492 | obj->refcount = 0; | |
173af62f | 493 | obj->destroy_flag = 0; |
00e2e675 DG |
494 | lttng_ht_node_init_ulong(&obj->node, obj->net_seq_idx); |
495 | pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); | |
496 | ||
497 | error: | |
498 | return obj; | |
499 | } | |
500 | ||
501 | /* | |
502 | * Find a relayd socket pair in the global consumer data. | |
503 | * | |
504 | * Return the object if found else NULL. | |
b0b335c8 MD |
505 | * RCU read-side lock must be held across this call and while using the |
506 | * returned object. | |
00e2e675 DG |
507 | */ |
508 | struct consumer_relayd_sock_pair *consumer_find_relayd(int key) | |
509 | { | |
510 | struct lttng_ht_iter iter; | |
511 | struct lttng_ht_node_ulong *node; | |
512 | struct consumer_relayd_sock_pair *relayd = NULL; | |
513 | ||
514 | /* Negative keys are lookup failures */ | |
515 | if (key < 0) { | |
516 | goto error; | |
517 | } | |
518 | ||
00e2e675 DG |
519 | lttng_ht_lookup(consumer_data.relayd_ht, (void *)((unsigned long) key), |
520 | &iter); | |
521 | node = lttng_ht_iter_get_node_ulong(&iter); | |
522 | if (node != NULL) { | |
523 | relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
524 | } | |
525 | ||
00e2e675 DG |
526 | error: |
527 | return relayd; | |
528 | } | |
529 | ||
530 | /* | |
531 | * Handle stream for relayd transmission if the stream applies for network | |
532 | * streaming where the net sequence index is set. | |
533 | * | |
534 | * Return destination file descriptor or negative value on error. | |
535 | */ | |
6197aea7 DG |
536 | static int write_relayd_stream_header(struct lttng_consumer_stream *stream, |
537 | size_t data_size, struct consumer_relayd_sock_pair *relayd) | |
00e2e675 DG |
538 | { |
539 | int outfd = -1, ret; | |
00e2e675 DG |
540 | struct lttcomm_relayd_data_hdr data_hdr; |
541 | ||
542 | /* Safety net */ | |
543 | assert(stream); | |
6197aea7 | 544 | assert(relayd); |
00e2e675 DG |
545 | |
546 | /* Reset data header */ | |
547 | memset(&data_hdr, 0, sizeof(data_hdr)); | |
548 | ||
00e2e675 DG |
549 | if (stream->metadata_flag) { |
550 | /* Caller MUST acquire the relayd control socket lock */ | |
551 | ret = relayd_send_metadata(&relayd->control_sock, data_size); | |
552 | if (ret < 0) { | |
553 | goto error; | |
554 | } | |
555 | ||
556 | /* Metadata are always sent on the control socket. */ | |
557 | outfd = relayd->control_sock.fd; | |
558 | } else { | |
559 | /* Set header with stream information */ | |
560 | data_hdr.stream_id = htobe64(stream->relayd_stream_id); | |
561 | data_hdr.data_size = htobe32(data_size); | |
173af62f | 562 | data_hdr.net_seq_num = htobe64(stream->next_net_seq_num++); |
00e2e675 DG |
563 | /* Other fields are zeroed previously */ |
564 | ||
565 | ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, | |
566 | sizeof(data_hdr)); | |
567 | if (ret < 0) { | |
568 | goto error; | |
569 | } | |
570 | ||
571 | /* Set to go on data socket */ | |
572 | outfd = relayd->data_sock.fd; | |
573 | } | |
574 | ||
575 | error: | |
576 | return outfd; | |
577 | } | |
578 | ||
3bd1e081 MD |
579 | /* |
580 | * Update a stream according to what we just received. | |
581 | */ | |
582 | void consumer_change_stream_state(int stream_key, | |
583 | enum lttng_consumer_stream_state state) | |
584 | { | |
585 | struct lttng_consumer_stream *stream; | |
586 | ||
587 | pthread_mutex_lock(&consumer_data.lock); | |
588 | stream = consumer_find_stream(stream_key); | |
589 | if (stream) { | |
590 | stream->state = state; | |
591 | } | |
592 | consumer_data.need_update = 1; | |
593 | pthread_mutex_unlock(&consumer_data.lock); | |
594 | } | |
595 | ||
702b1ea4 MD |
596 | static |
597 | void consumer_free_channel(struct rcu_head *head) | |
598 | { | |
599 | struct lttng_ht_node_ulong *node = | |
600 | caa_container_of(head, struct lttng_ht_node_ulong, head); | |
601 | struct lttng_consumer_channel *channel = | |
602 | caa_container_of(node, struct lttng_consumer_channel, node); | |
603 | ||
604 | free(channel); | |
605 | } | |
606 | ||
3bd1e081 MD |
607 | /* |
608 | * Remove a channel from the global list protected by a mutex. This | |
609 | * function is also responsible for freeing its data structures. | |
610 | */ | |
611 | void consumer_del_channel(struct lttng_consumer_channel *channel) | |
612 | { | |
613 | int ret; | |
e4421fec | 614 | struct lttng_ht_iter iter; |
3bd1e081 MD |
615 | |
616 | pthread_mutex_lock(&consumer_data.lock); | |
617 | ||
618 | switch (consumer_data.type) { | |
619 | case LTTNG_CONSUMER_KERNEL: | |
620 | break; | |
7753dea8 MD |
621 | case LTTNG_CONSUMER32_UST: |
622 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
623 | lttng_ustconsumer_del_channel(channel); |
624 | break; | |
625 | default: | |
626 | ERR("Unknown consumer_data type"); | |
627 | assert(0); | |
628 | goto end; | |
629 | } | |
630 | ||
6065ceec | 631 | rcu_read_lock(); |
04253271 MD |
632 | iter.iter.node = &channel->node.node; |
633 | ret = lttng_ht_del(consumer_data.channel_ht, &iter); | |
634 | assert(!ret); | |
6065ceec DG |
635 | rcu_read_unlock(); |
636 | ||
3bd1e081 MD |
637 | if (channel->mmap_base != NULL) { |
638 | ret = munmap(channel->mmap_base, channel->mmap_len); | |
639 | if (ret != 0) { | |
640 | perror("munmap"); | |
641 | } | |
642 | } | |
b5c5fc29 | 643 | if (channel->wait_fd >= 0 && !channel->wait_fd_is_copy) { |
4c462e79 MD |
644 | ret = close(channel->wait_fd); |
645 | if (ret) { | |
646 | PERROR("close"); | |
647 | } | |
3bd1e081 | 648 | } |
2c1dd183 | 649 | if (channel->shm_fd >= 0 && channel->wait_fd != channel->shm_fd) { |
4c462e79 MD |
650 | ret = close(channel->shm_fd); |
651 | if (ret) { | |
652 | PERROR("close"); | |
653 | } | |
3bd1e081 | 654 | } |
702b1ea4 MD |
655 | |
656 | call_rcu(&channel->node.head, consumer_free_channel); | |
3bd1e081 MD |
657 | end: |
658 | pthread_mutex_unlock(&consumer_data.lock); | |
659 | } | |
660 | ||
661 | struct lttng_consumer_channel *consumer_allocate_channel( | |
662 | int channel_key, | |
663 | int shm_fd, int wait_fd, | |
664 | uint64_t mmap_len, | |
665 | uint64_t max_sb_size) | |
666 | { | |
667 | struct lttng_consumer_channel *channel; | |
668 | int ret; | |
669 | ||
276b26d1 | 670 | channel = zmalloc(sizeof(*channel)); |
3bd1e081 MD |
671 | if (channel == NULL) { |
672 | perror("malloc struct lttng_consumer_channel"); | |
673 | goto end; | |
674 | } | |
675 | channel->key = channel_key; | |
676 | channel->shm_fd = shm_fd; | |
677 | channel->wait_fd = wait_fd; | |
678 | channel->mmap_len = mmap_len; | |
679 | channel->max_sb_size = max_sb_size; | |
680 | channel->refcount = 0; | |
681 | channel->nr_streams = 0; | |
e4421fec | 682 | lttng_ht_node_init_ulong(&channel->node, channel->key); |
3bd1e081 MD |
683 | |
684 | switch (consumer_data.type) { | |
685 | case LTTNG_CONSUMER_KERNEL: | |
686 | channel->mmap_base = NULL; | |
687 | channel->mmap_len = 0; | |
688 | break; | |
7753dea8 MD |
689 | case LTTNG_CONSUMER32_UST: |
690 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
691 | ret = lttng_ustconsumer_allocate_channel(channel); |
692 | if (ret) { | |
693 | free(channel); | |
694 | return NULL; | |
695 | } | |
696 | break; | |
697 | default: | |
698 | ERR("Unknown consumer_data type"); | |
699 | assert(0); | |
700 | goto end; | |
701 | } | |
702 | DBG("Allocated channel (key %d, shm_fd %d, wait_fd %d, mmap_len %llu, max_sb_size %llu)", | |
00e2e675 | 703 | channel->key, channel->shm_fd, channel->wait_fd, |
3bd1e081 MD |
704 | (unsigned long long) channel->mmap_len, |
705 | (unsigned long long) channel->max_sb_size); | |
706 | end: | |
707 | return channel; | |
708 | } | |
709 | ||
710 | /* | |
711 | * Add a channel to the global list protected by a mutex. | |
712 | */ | |
713 | int consumer_add_channel(struct lttng_consumer_channel *channel) | |
714 | { | |
c77fc10a DG |
715 | struct lttng_ht_node_ulong *node; |
716 | struct lttng_ht_iter iter; | |
717 | ||
3bd1e081 | 718 | pthread_mutex_lock(&consumer_data.lock); |
7ad0a0cb MD |
719 | /* Steal channel identifier, for UST */ |
720 | consumer_steal_channel_key(channel->key); | |
6065ceec | 721 | rcu_read_lock(); |
c77fc10a DG |
722 | |
723 | lttng_ht_lookup(consumer_data.channel_ht, | |
724 | (void *)((unsigned long) channel->key), &iter); | |
725 | node = lttng_ht_iter_get_node_ulong(&iter); | |
726 | if (node != NULL) { | |
727 | /* Channel already exist. Ignore the insertion */ | |
728 | goto end; | |
729 | } | |
730 | ||
04253271 | 731 | lttng_ht_add_unique_ulong(consumer_data.channel_ht, &channel->node); |
c77fc10a DG |
732 | |
733 | end: | |
6065ceec | 734 | rcu_read_unlock(); |
3bd1e081 | 735 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 736 | |
7ad0a0cb | 737 | return 0; |
3bd1e081 MD |
738 | } |
739 | ||
740 | /* | |
741 | * Allocate the pollfd structure and the local view of the out fds to avoid | |
742 | * doing a lookup in the linked list and concurrency issues when writing is | |
743 | * needed. Called with consumer_data.lock held. | |
744 | * | |
745 | * Returns the number of fds in the structures. | |
746 | */ | |
747 | int consumer_update_poll_array( | |
748 | struct lttng_consumer_local_data *ctx, struct pollfd **pollfd, | |
00e2e675 DG |
749 | struct lttng_consumer_stream **local_stream, |
750 | struct lttng_ht *metadata_ht) | |
3bd1e081 | 751 | { |
3bd1e081 | 752 | int i = 0; |
e4421fec DG |
753 | struct lttng_ht_iter iter; |
754 | struct lttng_consumer_stream *stream; | |
3bd1e081 MD |
755 | |
756 | DBG("Updating poll fd array"); | |
481d6c57 | 757 | rcu_read_lock(); |
e4421fec DG |
758 | cds_lfht_for_each_entry(consumer_data.stream_ht->ht, &iter.iter, stream, |
759 | node.node) { | |
760 | if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM) { | |
3bd1e081 MD |
761 | continue; |
762 | } | |
e4421fec DG |
763 | DBG("Active FD %d", stream->wait_fd); |
764 | (*pollfd)[i].fd = stream->wait_fd; | |
3bd1e081 | 765 | (*pollfd)[i].events = POLLIN | POLLPRI; |
00e2e675 DG |
766 | if (stream->metadata_flag && metadata_ht) { |
767 | lttng_ht_add_unique_ulong(metadata_ht, &stream->waitfd_node); | |
768 | DBG("Active FD added to metadata hash table"); | |
769 | } | |
e4421fec | 770 | local_stream[i] = stream; |
3bd1e081 MD |
771 | i++; |
772 | } | |
481d6c57 | 773 | rcu_read_unlock(); |
3bd1e081 MD |
774 | |
775 | /* | |
776 | * Insert the consumer_poll_pipe at the end of the array and don't | |
777 | * increment i so nb_fd is the number of real FD. | |
778 | */ | |
779 | (*pollfd)[i].fd = ctx->consumer_poll_pipe[0]; | |
509bb1cf | 780 | (*pollfd)[i].events = POLLIN | POLLPRI; |
3bd1e081 MD |
781 | return i; |
782 | } | |
783 | ||
784 | /* | |
785 | * Poll on the should_quit pipe and the command socket return -1 on error and | |
786 | * should exit, 0 if data is available on the command socket | |
787 | */ | |
788 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) | |
789 | { | |
790 | int num_rdy; | |
791 | ||
88f2b785 | 792 | restart: |
3bd1e081 MD |
793 | num_rdy = poll(consumer_sockpoll, 2, -1); |
794 | if (num_rdy == -1) { | |
88f2b785 MD |
795 | /* |
796 | * Restart interrupted system call. | |
797 | */ | |
798 | if (errno == EINTR) { | |
799 | goto restart; | |
800 | } | |
3bd1e081 MD |
801 | perror("Poll error"); |
802 | goto exit; | |
803 | } | |
509bb1cf | 804 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
3bd1e081 MD |
805 | DBG("consumer_should_quit wake up"); |
806 | goto exit; | |
807 | } | |
808 | return 0; | |
809 | ||
810 | exit: | |
811 | return -1; | |
812 | } | |
813 | ||
814 | /* | |
815 | * Set the error socket. | |
816 | */ | |
817 | void lttng_consumer_set_error_sock( | |
818 | struct lttng_consumer_local_data *ctx, int sock) | |
819 | { | |
820 | ctx->consumer_error_socket = sock; | |
821 | } | |
822 | ||
823 | /* | |
824 | * Set the command socket path. | |
825 | */ | |
3bd1e081 MD |
826 | void lttng_consumer_set_command_sock_path( |
827 | struct lttng_consumer_local_data *ctx, char *sock) | |
828 | { | |
829 | ctx->consumer_command_sock_path = sock; | |
830 | } | |
831 | ||
832 | /* | |
833 | * Send return code to the session daemon. | |
834 | * If the socket is not defined, we return 0, it is not a fatal error | |
835 | */ | |
836 | int lttng_consumer_send_error( | |
837 | struct lttng_consumer_local_data *ctx, int cmd) | |
838 | { | |
839 | if (ctx->consumer_error_socket > 0) { | |
840 | return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd, | |
841 | sizeof(enum lttcomm_sessiond_command)); | |
842 | } | |
843 | ||
844 | return 0; | |
845 | } | |
846 | ||
847 | /* | |
848 | * Close all the tracefiles and stream fds, should be called when all instances | |
849 | * are destroyed. | |
850 | */ | |
851 | void lttng_consumer_cleanup(void) | |
852 | { | |
e4421fec | 853 | struct lttng_ht_iter iter; |
6065ceec DG |
854 | struct lttng_ht_node_ulong *node; |
855 | ||
856 | rcu_read_lock(); | |
3bd1e081 MD |
857 | |
858 | /* | |
6065ceec DG |
859 | * close all outfd. Called when there are no more threads running (after |
860 | * joining on the threads), no need to protect list iteration with mutex. | |
3bd1e081 | 861 | */ |
6065ceec DG |
862 | cds_lfht_for_each_entry(consumer_data.stream_ht->ht, &iter.iter, node, |
863 | node) { | |
702b1ea4 MD |
864 | struct lttng_consumer_stream *stream = |
865 | caa_container_of(node, struct lttng_consumer_stream, node); | |
866 | consumer_del_stream(stream); | |
3bd1e081 | 867 | } |
e4421fec | 868 | |
6065ceec DG |
869 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, node, |
870 | node) { | |
702b1ea4 MD |
871 | struct lttng_consumer_channel *channel = |
872 | caa_container_of(node, struct lttng_consumer_channel, node); | |
873 | consumer_del_channel(channel); | |
3bd1e081 | 874 | } |
6065ceec DG |
875 | |
876 | rcu_read_unlock(); | |
d6ce1df2 MD |
877 | |
878 | lttng_ht_destroy(consumer_data.stream_ht); | |
879 | lttng_ht_destroy(consumer_data.channel_ht); | |
3bd1e081 MD |
880 | } |
881 | ||
882 | /* | |
883 | * Called from signal handler. | |
884 | */ | |
885 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) | |
886 | { | |
887 | int ret; | |
888 | consumer_quit = 1; | |
6f94560a MD |
889 | do { |
890 | ret = write(ctx->consumer_should_quit[1], "4", 1); | |
891 | } while (ret < 0 && errno == EINTR); | |
3bd1e081 MD |
892 | if (ret < 0) { |
893 | perror("write consumer quit"); | |
894 | } | |
895 | } | |
896 | ||
00e2e675 DG |
897 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
898 | off_t orig_offset) | |
3bd1e081 MD |
899 | { |
900 | int outfd = stream->out_fd; | |
901 | ||
902 | /* | |
903 | * This does a blocking write-and-wait on any page that belongs to the | |
904 | * subbuffer prior to the one we just wrote. | |
905 | * Don't care about error values, as these are just hints and ways to | |
906 | * limit the amount of page cache used. | |
907 | */ | |
908 | if (orig_offset < stream->chan->max_sb_size) { | |
909 | return; | |
910 | } | |
b9182dd9 | 911 | lttng_sync_file_range(outfd, orig_offset - stream->chan->max_sb_size, |
3bd1e081 MD |
912 | stream->chan->max_sb_size, |
913 | SYNC_FILE_RANGE_WAIT_BEFORE | |
914 | | SYNC_FILE_RANGE_WRITE | |
915 | | SYNC_FILE_RANGE_WAIT_AFTER); | |
916 | /* | |
917 | * Give hints to the kernel about how we access the file: | |
918 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after | |
919 | * we write it. | |
920 | * | |
921 | * We need to call fadvise again after the file grows because the | |
922 | * kernel does not seem to apply fadvise to non-existing parts of the | |
923 | * file. | |
924 | * | |
925 | * Call fadvise _after_ having waited for the page writeback to | |
926 | * complete because the dirty page writeback semantic is not well | |
927 | * defined. So it can be expected to lead to lower throughput in | |
928 | * streaming. | |
929 | */ | |
930 | posix_fadvise(outfd, orig_offset - stream->chan->max_sb_size, | |
931 | stream->chan->max_sb_size, POSIX_FADV_DONTNEED); | |
932 | } | |
933 | ||
934 | /* | |
935 | * Initialise the necessary environnement : | |
936 | * - create a new context | |
937 | * - create the poll_pipe | |
938 | * - create the should_quit pipe (for signal handler) | |
939 | * - create the thread pipe (for splice) | |
940 | * | |
941 | * Takes a function pointer as argument, this function is called when data is | |
942 | * available on a buffer. This function is responsible to do the | |
943 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the | |
944 | * buffer configuration and then kernctl_put_next_subbuf at the end. | |
945 | * | |
946 | * Returns a pointer to the new context or NULL on error. | |
947 | */ | |
948 | struct lttng_consumer_local_data *lttng_consumer_create( | |
949 | enum lttng_consumer_type type, | |
4078b776 | 950 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
d41f73b7 | 951 | struct lttng_consumer_local_data *ctx), |
3bd1e081 MD |
952 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
953 | int (*recv_stream)(struct lttng_consumer_stream *stream), | |
954 | int (*update_stream)(int stream_key, uint32_t state)) | |
955 | { | |
956 | int ret, i; | |
957 | struct lttng_consumer_local_data *ctx; | |
958 | ||
959 | assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || | |
960 | consumer_data.type == type); | |
961 | consumer_data.type = type; | |
962 | ||
effcf122 | 963 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
3bd1e081 MD |
964 | if (ctx == NULL) { |
965 | perror("allocating context"); | |
966 | goto error; | |
967 | } | |
968 | ||
969 | ctx->consumer_error_socket = -1; | |
970 | /* assign the callbacks */ | |
971 | ctx->on_buffer_ready = buffer_ready; | |
972 | ctx->on_recv_channel = recv_channel; | |
973 | ctx->on_recv_stream = recv_stream; | |
974 | ctx->on_update_stream = update_stream; | |
975 | ||
976 | ret = pipe(ctx->consumer_poll_pipe); | |
977 | if (ret < 0) { | |
978 | perror("Error creating poll pipe"); | |
979 | goto error_poll_pipe; | |
980 | } | |
981 | ||
04fdd819 MD |
982 | /* set read end of the pipe to non-blocking */ |
983 | ret = fcntl(ctx->consumer_poll_pipe[0], F_SETFL, O_NONBLOCK); | |
984 | if (ret < 0) { | |
985 | perror("fcntl O_NONBLOCK"); | |
986 | goto error_poll_fcntl; | |
987 | } | |
988 | ||
989 | /* set write end of the pipe to non-blocking */ | |
990 | ret = fcntl(ctx->consumer_poll_pipe[1], F_SETFL, O_NONBLOCK); | |
991 | if (ret < 0) { | |
992 | perror("fcntl O_NONBLOCK"); | |
993 | goto error_poll_fcntl; | |
994 | } | |
995 | ||
3bd1e081 MD |
996 | ret = pipe(ctx->consumer_should_quit); |
997 | if (ret < 0) { | |
998 | perror("Error creating recv pipe"); | |
999 | goto error_quit_pipe; | |
1000 | } | |
1001 | ||
1002 | ret = pipe(ctx->consumer_thread_pipe); | |
1003 | if (ret < 0) { | |
1004 | perror("Error creating thread pipe"); | |
1005 | goto error_thread_pipe; | |
1006 | } | |
1007 | ||
1008 | return ctx; | |
1009 | ||
1010 | ||
1011 | error_thread_pipe: | |
1012 | for (i = 0; i < 2; i++) { | |
1013 | int err; | |
1014 | ||
1015 | err = close(ctx->consumer_should_quit[i]); | |
4c462e79 MD |
1016 | if (err) { |
1017 | PERROR("close"); | |
1018 | } | |
3bd1e081 | 1019 | } |
04fdd819 | 1020 | error_poll_fcntl: |
3bd1e081 MD |
1021 | error_quit_pipe: |
1022 | for (i = 0; i < 2; i++) { | |
1023 | int err; | |
1024 | ||
1025 | err = close(ctx->consumer_poll_pipe[i]); | |
4c462e79 MD |
1026 | if (err) { |
1027 | PERROR("close"); | |
1028 | } | |
3bd1e081 MD |
1029 | } |
1030 | error_poll_pipe: | |
1031 | free(ctx); | |
1032 | error: | |
1033 | return NULL; | |
1034 | } | |
1035 | ||
1036 | /* | |
1037 | * Close all fds associated with the instance and free the context. | |
1038 | */ | |
1039 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) | |
1040 | { | |
4c462e79 MD |
1041 | int ret; |
1042 | ||
1043 | ret = close(ctx->consumer_error_socket); | |
1044 | if (ret) { | |
1045 | PERROR("close"); | |
1046 | } | |
1047 | ret = close(ctx->consumer_thread_pipe[0]); | |
1048 | if (ret) { | |
1049 | PERROR("close"); | |
1050 | } | |
1051 | ret = close(ctx->consumer_thread_pipe[1]); | |
1052 | if (ret) { | |
1053 | PERROR("close"); | |
1054 | } | |
1055 | ret = close(ctx->consumer_poll_pipe[0]); | |
1056 | if (ret) { | |
1057 | PERROR("close"); | |
1058 | } | |
1059 | ret = close(ctx->consumer_poll_pipe[1]); | |
1060 | if (ret) { | |
1061 | PERROR("close"); | |
1062 | } | |
1063 | ret = close(ctx->consumer_should_quit[0]); | |
1064 | if (ret) { | |
1065 | PERROR("close"); | |
1066 | } | |
1067 | ret = close(ctx->consumer_should_quit[1]); | |
1068 | if (ret) { | |
1069 | PERROR("close"); | |
1070 | } | |
3bd1e081 MD |
1071 | unlink(ctx->consumer_command_sock_path); |
1072 | free(ctx); | |
1073 | } | |
1074 | ||
6197aea7 DG |
1075 | /* |
1076 | * Write the metadata stream id on the specified file descriptor. | |
1077 | */ | |
1078 | static int write_relayd_metadata_id(int fd, | |
1079 | struct lttng_consumer_stream *stream, | |
1080 | struct consumer_relayd_sock_pair *relayd) | |
1081 | { | |
1082 | int ret; | |
1083 | uint64_t metadata_id; | |
1084 | ||
1085 | metadata_id = htobe64(stream->relayd_stream_id); | |
1086 | do { | |
1087 | ret = write(fd, (void *) &metadata_id, | |
1088 | sizeof(stream->relayd_stream_id)); | |
1089 | } while (ret < 0 && errno == EINTR); | |
1090 | if (ret < 0) { | |
1091 | PERROR("write metadata stream id"); | |
1092 | goto end; | |
1093 | } | |
1094 | DBG("Metadata stream id %zu written before data", | |
1095 | stream->relayd_stream_id); | |
1096 | ||
1097 | end: | |
1098 | return ret; | |
1099 | } | |
1100 | ||
3bd1e081 | 1101 | /* |
09e26845 DG |
1102 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
1103 | * core function for writing trace buffers to either the local filesystem or | |
1104 | * the network. | |
1105 | * | |
1106 | * Careful review MUST be put if any changes occur! | |
3bd1e081 MD |
1107 | * |
1108 | * Returns the number of bytes written | |
1109 | */ | |
4078b776 | 1110 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
3bd1e081 MD |
1111 | struct lttng_consumer_local_data *ctx, |
1112 | struct lttng_consumer_stream *stream, unsigned long len) | |
1113 | { | |
f02e1e8a DG |
1114 | unsigned long mmap_offset; |
1115 | ssize_t ret = 0, written = 0; | |
1116 | off_t orig_offset = stream->out_fd_offset; | |
1117 | /* Default is on the disk */ | |
1118 | int outfd = stream->out_fd; | |
f02e1e8a DG |
1119 | struct consumer_relayd_sock_pair *relayd = NULL; |
1120 | ||
1121 | /* RCU lock for the relayd pointer */ | |
1122 | rcu_read_lock(); | |
1123 | ||
1124 | /* Flag that the current stream if set for network streaming. */ | |
1125 | if (stream->net_seq_idx != -1) { | |
1126 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
1127 | if (relayd == NULL) { | |
1128 | goto end; | |
1129 | } | |
1130 | } | |
1131 | ||
1132 | /* get the offset inside the fd to mmap */ | |
3bd1e081 MD |
1133 | switch (consumer_data.type) { |
1134 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a DG |
1135 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); |
1136 | break; | |
7753dea8 MD |
1137 | case LTTNG_CONSUMER32_UST: |
1138 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a DG |
1139 | ret = lttng_ustctl_get_mmap_read_offset(stream->chan->handle, |
1140 | stream->buf, &mmap_offset); | |
1141 | break; | |
3bd1e081 MD |
1142 | default: |
1143 | ERR("Unknown consumer_data type"); | |
1144 | assert(0); | |
1145 | } | |
f02e1e8a DG |
1146 | if (ret != 0) { |
1147 | errno = -ret; | |
1148 | PERROR("tracer ctl get_mmap_read_offset"); | |
1149 | written = ret; | |
1150 | goto end; | |
1151 | } | |
b9182dd9 | 1152 | |
f02e1e8a DG |
1153 | /* Handle stream on the relayd if the output is on the network */ |
1154 | if (relayd) { | |
1155 | unsigned long netlen = len; | |
1156 | ||
1157 | /* | |
1158 | * Lock the control socket for the complete duration of the function | |
1159 | * since from this point on we will use the socket. | |
1160 | */ | |
1161 | if (stream->metadata_flag) { | |
1162 | /* Metadata requires the control socket. */ | |
1163 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1164 | netlen += sizeof(stream->relayd_stream_id); | |
1165 | } | |
1166 | ||
6197aea7 | 1167 | ret = write_relayd_stream_header(stream, netlen, relayd); |
f02e1e8a DG |
1168 | if (ret >= 0) { |
1169 | /* Use the returned socket. */ | |
1170 | outfd = ret; | |
1171 | ||
1172 | /* Write metadata stream id before payload */ | |
1173 | if (stream->metadata_flag) { | |
6197aea7 | 1174 | ret = write_relayd_metadata_id(outfd, stream, relayd); |
f02e1e8a | 1175 | if (ret < 0) { |
f02e1e8a DG |
1176 | written = ret; |
1177 | goto end; | |
1178 | } | |
f02e1e8a DG |
1179 | } |
1180 | } | |
1181 | /* Else, use the default set before which is the filesystem. */ | |
1182 | } | |
1183 | ||
1184 | while (len > 0) { | |
1185 | do { | |
1186 | ret = write(outfd, stream->mmap_base + mmap_offset, len); | |
1187 | } while (ret < 0 && errno == EINTR); | |
1188 | if (ret < 0) { | |
1189 | PERROR("Error in file write"); | |
1190 | if (written == 0) { | |
1191 | written = ret; | |
1192 | } | |
1193 | goto end; | |
1194 | } else if (ret > len) { | |
1195 | PERROR("Error in file write (ret %ld > len %lu)", ret, len); | |
1196 | written += ret; | |
1197 | goto end; | |
1198 | } else { | |
1199 | len -= ret; | |
1200 | mmap_offset += ret; | |
1201 | } | |
1202 | DBG("Consumer mmap write() ret %ld (len %lu)", ret, len); | |
1203 | ||
1204 | /* This call is useless on a socket so better save a syscall. */ | |
1205 | if (!relayd) { | |
1206 | /* This won't block, but will start writeout asynchronously */ | |
1207 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret, | |
1208 | SYNC_FILE_RANGE_WRITE); | |
1209 | stream->out_fd_offset += ret; | |
1210 | } | |
1211 | written += ret; | |
1212 | } | |
1213 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1214 | ||
1215 | end: | |
1216 | /* Unlock only if ctrl socket used */ | |
1217 | if (relayd && stream->metadata_flag) { | |
1218 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1219 | } | |
1220 | ||
1221 | rcu_read_unlock(); | |
1222 | return written; | |
3bd1e081 MD |
1223 | } |
1224 | ||
1225 | /* | |
1226 | * Splice the data from the ring buffer to the tracefile. | |
1227 | * | |
1228 | * Returns the number of bytes spliced. | |
1229 | */ | |
4078b776 | 1230 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
3bd1e081 MD |
1231 | struct lttng_consumer_local_data *ctx, |
1232 | struct lttng_consumer_stream *stream, unsigned long len) | |
1233 | { | |
f02e1e8a DG |
1234 | ssize_t ret = 0, written = 0, ret_splice = 0; |
1235 | loff_t offset = 0; | |
1236 | off_t orig_offset = stream->out_fd_offset; | |
1237 | int fd = stream->wait_fd; | |
1238 | /* Default is on the disk */ | |
1239 | int outfd = stream->out_fd; | |
f02e1e8a DG |
1240 | struct consumer_relayd_sock_pair *relayd = NULL; |
1241 | ||
3bd1e081 MD |
1242 | switch (consumer_data.type) { |
1243 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a | 1244 | break; |
7753dea8 MD |
1245 | case LTTNG_CONSUMER32_UST: |
1246 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a | 1247 | /* Not supported for user space tracing */ |
3bd1e081 MD |
1248 | return -ENOSYS; |
1249 | default: | |
1250 | ERR("Unknown consumer_data type"); | |
1251 | assert(0); | |
3bd1e081 MD |
1252 | } |
1253 | ||
f02e1e8a DG |
1254 | /* RCU lock for the relayd pointer */ |
1255 | rcu_read_lock(); | |
1256 | ||
1257 | /* Flag that the current stream if set for network streaming. */ | |
1258 | if (stream->net_seq_idx != -1) { | |
1259 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
1260 | if (relayd == NULL) { | |
1261 | goto end; | |
1262 | } | |
1263 | } | |
1264 | ||
1265 | /* Write metadata stream id before payload */ | |
1266 | if (stream->metadata_flag && relayd) { | |
1267 | /* | |
1268 | * Lock the control socket for the complete duration of the function | |
1269 | * since from this point on we will use the socket. | |
1270 | */ | |
1271 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1272 | ||
6197aea7 DG |
1273 | ret = write_relayd_metadata_id(ctx->consumer_thread_pipe[1], |
1274 | stream, relayd); | |
f02e1e8a | 1275 | if (ret < 0) { |
f02e1e8a DG |
1276 | written = ret; |
1277 | goto end; | |
1278 | } | |
f02e1e8a DG |
1279 | } |
1280 | ||
1281 | while (len > 0) { | |
1282 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d)", | |
1283 | (unsigned long)offset, len, fd); | |
1284 | ret_splice = splice(fd, &offset, ctx->consumer_thread_pipe[1], NULL, len, | |
1285 | SPLICE_F_MOVE | SPLICE_F_MORE); | |
1286 | DBG("splice chan to pipe, ret %zd", ret_splice); | |
1287 | if (ret_splice < 0) { | |
1288 | PERROR("Error in relay splice"); | |
1289 | if (written == 0) { | |
1290 | written = ret_splice; | |
1291 | } | |
1292 | ret = errno; | |
1293 | goto splice_error; | |
1294 | } | |
1295 | ||
1296 | /* Handle stream on the relayd if the output is on the network */ | |
1297 | if (relayd) { | |
1298 | if (stream->metadata_flag) { | |
1299 | /* Update counter to fit the spliced data */ | |
1300 | ret_splice += sizeof(stream->relayd_stream_id); | |
1301 | len += sizeof(stream->relayd_stream_id); | |
1302 | /* | |
1303 | * We do this so the return value can match the len passed as | |
1304 | * argument to this function. | |
1305 | */ | |
1306 | written -= sizeof(stream->relayd_stream_id); | |
1307 | } | |
1308 | ||
6197aea7 | 1309 | ret = write_relayd_stream_header(stream, ret_splice, relayd); |
f02e1e8a DG |
1310 | if (ret >= 0) { |
1311 | /* Use the returned socket. */ | |
1312 | outfd = ret; | |
1313 | } else { | |
6197aea7 DG |
1314 | ERR("Remote relayd disconnected. Stopping"); |
1315 | goto end; | |
f02e1e8a DG |
1316 | } |
1317 | } | |
1318 | ||
1319 | /* Splice data out */ | |
1320 | ret_splice = splice(ctx->consumer_thread_pipe[0], NULL, outfd, NULL, | |
1321 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); | |
1322 | DBG("Kernel consumer splice pipe to file, ret %zd", ret_splice); | |
1323 | if (ret_splice < 0) { | |
1324 | PERROR("Error in file splice"); | |
1325 | if (written == 0) { | |
1326 | written = ret_splice; | |
1327 | } | |
1328 | ret = errno; | |
1329 | goto splice_error; | |
1330 | } else if (ret_splice > len) { | |
1331 | errno = EINVAL; | |
1332 | PERROR("Wrote more data than requested %zd (len: %lu)", | |
1333 | ret_splice, len); | |
1334 | written += ret_splice; | |
1335 | ret = errno; | |
1336 | goto splice_error; | |
1337 | } | |
1338 | len -= ret_splice; | |
1339 | ||
1340 | /* This call is useless on a socket so better save a syscall. */ | |
1341 | if (!relayd) { | |
1342 | /* This won't block, but will start writeout asynchronously */ | |
1343 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, | |
1344 | SYNC_FILE_RANGE_WRITE); | |
1345 | stream->out_fd_offset += ret_splice; | |
1346 | } | |
1347 | written += ret_splice; | |
1348 | } | |
1349 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1350 | ||
1351 | ret = ret_splice; | |
1352 | ||
1353 | goto end; | |
1354 | ||
1355 | splice_error: | |
1356 | /* send the appropriate error description to sessiond */ | |
1357 | switch (ret) { | |
1358 | case EBADF: | |
1359 | lttng_consumer_send_error(ctx, CONSUMERD_SPLICE_EBADF); | |
1360 | break; | |
1361 | case EINVAL: | |
1362 | lttng_consumer_send_error(ctx, CONSUMERD_SPLICE_EINVAL); | |
1363 | break; | |
1364 | case ENOMEM: | |
1365 | lttng_consumer_send_error(ctx, CONSUMERD_SPLICE_ENOMEM); | |
1366 | break; | |
1367 | case ESPIPE: | |
1368 | lttng_consumer_send_error(ctx, CONSUMERD_SPLICE_ESPIPE); | |
1369 | break; | |
1370 | } | |
1371 | ||
1372 | end: | |
1373 | if (relayd && stream->metadata_flag) { | |
1374 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1375 | } | |
1376 | ||
1377 | rcu_read_unlock(); | |
1378 | return written; | |
3bd1e081 MD |
1379 | } |
1380 | ||
1381 | /* | |
1382 | * Take a snapshot for a specific fd | |
1383 | * | |
1384 | * Returns 0 on success, < 0 on error | |
1385 | */ | |
1386 | int lttng_consumer_take_snapshot(struct lttng_consumer_local_data *ctx, | |
1387 | struct lttng_consumer_stream *stream) | |
1388 | { | |
1389 | switch (consumer_data.type) { | |
1390 | case LTTNG_CONSUMER_KERNEL: | |
1391 | return lttng_kconsumer_take_snapshot(ctx, stream); | |
7753dea8 MD |
1392 | case LTTNG_CONSUMER32_UST: |
1393 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1394 | return lttng_ustconsumer_take_snapshot(ctx, stream); |
1395 | default: | |
1396 | ERR("Unknown consumer_data type"); | |
1397 | assert(0); | |
1398 | return -ENOSYS; | |
1399 | } | |
1400 | ||
1401 | } | |
1402 | ||
1403 | /* | |
1404 | * Get the produced position | |
1405 | * | |
1406 | * Returns 0 on success, < 0 on error | |
1407 | */ | |
1408 | int lttng_consumer_get_produced_snapshot( | |
1409 | struct lttng_consumer_local_data *ctx, | |
1410 | struct lttng_consumer_stream *stream, | |
1411 | unsigned long *pos) | |
1412 | { | |
1413 | switch (consumer_data.type) { | |
1414 | case LTTNG_CONSUMER_KERNEL: | |
1415 | return lttng_kconsumer_get_produced_snapshot(ctx, stream, pos); | |
7753dea8 MD |
1416 | case LTTNG_CONSUMER32_UST: |
1417 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1418 | return lttng_ustconsumer_get_produced_snapshot(ctx, stream, pos); |
1419 | default: | |
1420 | ERR("Unknown consumer_data type"); | |
1421 | assert(0); | |
1422 | return -ENOSYS; | |
1423 | } | |
1424 | } | |
1425 | ||
1426 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, | |
1427 | int sock, struct pollfd *consumer_sockpoll) | |
1428 | { | |
1429 | switch (consumer_data.type) { | |
1430 | case LTTNG_CONSUMER_KERNEL: | |
1431 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
7753dea8 MD |
1432 | case LTTNG_CONSUMER32_UST: |
1433 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1434 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
1435 | default: | |
1436 | ERR("Unknown consumer_data type"); | |
1437 | assert(0); | |
1438 | return -ENOSYS; | |
1439 | } | |
1440 | } | |
1441 | ||
1442 | /* | |
e4421fec | 1443 | * This thread polls the fds in the set to consume the data and write |
3bd1e081 MD |
1444 | * it to tracefile if necessary. |
1445 | */ | |
1446 | void *lttng_consumer_thread_poll_fds(void *data) | |
1447 | { | |
1448 | int num_rdy, num_hup, high_prio, ret, i; | |
1449 | struct pollfd *pollfd = NULL; | |
1450 | /* local view of the streams */ | |
1451 | struct lttng_consumer_stream **local_stream = NULL; | |
1452 | /* local view of consumer_data.fds_count */ | |
1453 | int nb_fd = 0; | |
3bd1e081 | 1454 | struct lttng_consumer_local_data *ctx = data; |
00e2e675 DG |
1455 | struct lttng_ht *metadata_ht; |
1456 | struct lttng_ht_iter iter; | |
1457 | struct lttng_ht_node_ulong *node; | |
1458 | struct lttng_consumer_stream *metadata_stream; | |
1459 | ssize_t len; | |
1460 | ||
1461 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
3bd1e081 | 1462 | |
e7b994a3 DG |
1463 | rcu_register_thread(); |
1464 | ||
effcf122 | 1465 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream)); |
3bd1e081 MD |
1466 | |
1467 | while (1) { | |
1468 | high_prio = 0; | |
1469 | num_hup = 0; | |
1470 | ||
1471 | /* | |
e4421fec | 1472 | * the fds set has been updated, we need to update our |
3bd1e081 MD |
1473 | * local array as well |
1474 | */ | |
1475 | pthread_mutex_lock(&consumer_data.lock); | |
1476 | if (consumer_data.need_update) { | |
1477 | if (pollfd != NULL) { | |
1478 | free(pollfd); | |
1479 | pollfd = NULL; | |
1480 | } | |
1481 | if (local_stream != NULL) { | |
1482 | free(local_stream); | |
1483 | local_stream = NULL; | |
1484 | } | |
1485 | ||
1486 | /* allocate for all fds + 1 for the consumer_poll_pipe */ | |
effcf122 | 1487 | pollfd = zmalloc((consumer_data.stream_count + 1) * sizeof(struct pollfd)); |
3bd1e081 MD |
1488 | if (pollfd == NULL) { |
1489 | perror("pollfd malloc"); | |
1490 | pthread_mutex_unlock(&consumer_data.lock); | |
1491 | goto end; | |
1492 | } | |
1493 | ||
1494 | /* allocate for all fds + 1 for the consumer_poll_pipe */ | |
effcf122 | 1495 | local_stream = zmalloc((consumer_data.stream_count + 1) * |
3bd1e081 MD |
1496 | sizeof(struct lttng_consumer_stream)); |
1497 | if (local_stream == NULL) { | |
1498 | perror("local_stream malloc"); | |
1499 | pthread_mutex_unlock(&consumer_data.lock); | |
1500 | goto end; | |
1501 | } | |
00e2e675 DG |
1502 | ret = consumer_update_poll_array(ctx, &pollfd, local_stream, |
1503 | metadata_ht); | |
3bd1e081 MD |
1504 | if (ret < 0) { |
1505 | ERR("Error in allocating pollfd or local_outfds"); | |
1506 | lttng_consumer_send_error(ctx, CONSUMERD_POLL_ERROR); | |
1507 | pthread_mutex_unlock(&consumer_data.lock); | |
1508 | goto end; | |
1509 | } | |
1510 | nb_fd = ret; | |
1511 | consumer_data.need_update = 0; | |
1512 | } | |
1513 | pthread_mutex_unlock(&consumer_data.lock); | |
1514 | ||
4078b776 MD |
1515 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
1516 | if (nb_fd == 0 && consumer_quit == 1) { | |
1517 | goto end; | |
1518 | } | |
3bd1e081 | 1519 | /* poll on the array of fds */ |
88f2b785 | 1520 | restart: |
3bd1e081 MD |
1521 | DBG("polling on %d fd", nb_fd + 1); |
1522 | num_rdy = poll(pollfd, nb_fd + 1, consumer_poll_timeout); | |
1523 | DBG("poll num_rdy : %d", num_rdy); | |
1524 | if (num_rdy == -1) { | |
88f2b785 MD |
1525 | /* |
1526 | * Restart interrupted system call. | |
1527 | */ | |
1528 | if (errno == EINTR) { | |
1529 | goto restart; | |
1530 | } | |
3bd1e081 MD |
1531 | perror("Poll error"); |
1532 | lttng_consumer_send_error(ctx, CONSUMERD_POLL_ERROR); | |
1533 | goto end; | |
1534 | } else if (num_rdy == 0) { | |
1535 | DBG("Polling thread timed out"); | |
1536 | goto end; | |
1537 | } | |
1538 | ||
3bd1e081 | 1539 | /* |
00e2e675 DG |
1540 | * If the consumer_poll_pipe triggered poll go directly to the |
1541 | * beginning of the loop to update the array. We want to prioritize | |
1542 | * array update over low-priority reads. | |
3bd1e081 | 1543 | */ |
509bb1cf | 1544 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
04fdd819 MD |
1545 | size_t pipe_readlen; |
1546 | char tmp; | |
1547 | ||
3bd1e081 | 1548 | DBG("consumer_poll_pipe wake up"); |
04fdd819 MD |
1549 | /* Consume 1 byte of pipe data */ |
1550 | do { | |
1551 | pipe_readlen = read(ctx->consumer_poll_pipe[0], &tmp, 1); | |
1552 | } while (pipe_readlen == -1 && errno == EINTR); | |
3bd1e081 MD |
1553 | continue; |
1554 | } | |
1555 | ||
1556 | /* Take care of high priority channels first. */ | |
1557 | for (i = 0; i < nb_fd; i++) { | |
00e2e675 DG |
1558 | /* Lookup for metadata which is the highest priority */ |
1559 | lttng_ht_lookup(metadata_ht, | |
1560 | (void *)((unsigned long) pollfd[i].fd), &iter); | |
1561 | node = lttng_ht_iter_get_node_ulong(&iter); | |
1562 | if (node != NULL && | |
1563 | (pollfd[i].revents & (POLLIN | POLLPRI))) { | |
1564 | DBG("Urgent metadata read on fd %d", pollfd[i].fd); | |
1565 | metadata_stream = caa_container_of(node, | |
1566 | struct lttng_consumer_stream, waitfd_node); | |
1567 | high_prio = 1; | |
1568 | len = ctx->on_buffer_ready(metadata_stream, ctx); | |
1569 | /* it's ok to have an unavailable sub-buffer */ | |
1570 | if (len < 0 && len != -EAGAIN) { | |
1571 | goto end; | |
1572 | } else if (len > 0) { | |
1573 | metadata_stream->data_read = 1; | |
1574 | } | |
1575 | } else if (pollfd[i].revents & POLLPRI) { | |
d41f73b7 MD |
1576 | DBG("Urgent read on fd %d", pollfd[i].fd); |
1577 | high_prio = 1; | |
4078b776 | 1578 | len = ctx->on_buffer_ready(local_stream[i], ctx); |
d41f73b7 | 1579 | /* it's ok to have an unavailable sub-buffer */ |
4078b776 MD |
1580 | if (len < 0 && len != -EAGAIN) { |
1581 | goto end; | |
1582 | } else if (len > 0) { | |
1583 | local_stream[i]->data_read = 1; | |
d41f73b7 | 1584 | } |
3bd1e081 MD |
1585 | } |
1586 | } | |
1587 | ||
4078b776 MD |
1588 | /* |
1589 | * If we read high prio channel in this loop, try again | |
1590 | * for more high prio data. | |
1591 | */ | |
1592 | if (high_prio) { | |
3bd1e081 MD |
1593 | continue; |
1594 | } | |
1595 | ||
1596 | /* Take care of low priority channels. */ | |
4078b776 MD |
1597 | for (i = 0; i < nb_fd; i++) { |
1598 | if ((pollfd[i].revents & POLLIN) || | |
1599 | local_stream[i]->hangup_flush_done) { | |
4078b776 MD |
1600 | DBG("Normal read on fd %d", pollfd[i].fd); |
1601 | len = ctx->on_buffer_ready(local_stream[i], ctx); | |
1602 | /* it's ok to have an unavailable sub-buffer */ | |
1603 | if (len < 0 && len != -EAGAIN) { | |
1604 | goto end; | |
1605 | } else if (len > 0) { | |
1606 | local_stream[i]->data_read = 1; | |
1607 | } | |
1608 | } | |
1609 | } | |
1610 | ||
1611 | /* Handle hangup and errors */ | |
1612 | for (i = 0; i < nb_fd; i++) { | |
1613 | if (!local_stream[i]->hangup_flush_done | |
1614 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) | |
1615 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
1616 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
1617 | DBG("fd %d is hup|err|nval. Attempting flush and read.", | |
1618 | pollfd[i].fd); | |
1619 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); | |
1620 | /* Attempt read again, for the data we just flushed. */ | |
1621 | local_stream[i]->data_read = 1; | |
1622 | } | |
1623 | /* | |
1624 | * If the poll flag is HUP/ERR/NVAL and we have | |
1625 | * read no data in this pass, we can remove the | |
1626 | * stream from its hash table. | |
1627 | */ | |
1628 | if ((pollfd[i].revents & POLLHUP)) { | |
1629 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); | |
1630 | if (!local_stream[i]->data_read) { | |
00e2e675 DG |
1631 | if (local_stream[i]->metadata_flag) { |
1632 | iter.iter.node = &local_stream[i]->waitfd_node.node; | |
1633 | ret = lttng_ht_del(metadata_ht, &iter); | |
1634 | assert(!ret); | |
1635 | } | |
702b1ea4 | 1636 | consumer_del_stream(local_stream[i]); |
4078b776 MD |
1637 | num_hup++; |
1638 | } | |
1639 | } else if (pollfd[i].revents & POLLERR) { | |
1640 | ERR("Error returned in polling fd %d.", pollfd[i].fd); | |
1641 | if (!local_stream[i]->data_read) { | |
00e2e675 DG |
1642 | if (local_stream[i]->metadata_flag) { |
1643 | iter.iter.node = &local_stream[i]->waitfd_node.node; | |
1644 | ret = lttng_ht_del(metadata_ht, &iter); | |
1645 | assert(!ret); | |
1646 | } | |
702b1ea4 | 1647 | consumer_del_stream(local_stream[i]); |
4078b776 MD |
1648 | num_hup++; |
1649 | } | |
1650 | } else if (pollfd[i].revents & POLLNVAL) { | |
1651 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); | |
1652 | if (!local_stream[i]->data_read) { | |
00e2e675 DG |
1653 | if (local_stream[i]->metadata_flag) { |
1654 | iter.iter.node = &local_stream[i]->waitfd_node.node; | |
1655 | ret = lttng_ht_del(metadata_ht, &iter); | |
1656 | assert(!ret); | |
1657 | } | |
702b1ea4 | 1658 | consumer_del_stream(local_stream[i]); |
4078b776 | 1659 | num_hup++; |
3bd1e081 MD |
1660 | } |
1661 | } | |
4078b776 | 1662 | local_stream[i]->data_read = 0; |
3bd1e081 MD |
1663 | } |
1664 | } | |
1665 | end: | |
1666 | DBG("polling thread exiting"); | |
1667 | if (pollfd != NULL) { | |
1668 | free(pollfd); | |
1669 | pollfd = NULL; | |
1670 | } | |
1671 | if (local_stream != NULL) { | |
1672 | free(local_stream); | |
1673 | local_stream = NULL; | |
1674 | } | |
e7b994a3 | 1675 | rcu_unregister_thread(); |
3bd1e081 MD |
1676 | return NULL; |
1677 | } | |
1678 | ||
1679 | /* | |
1680 | * This thread listens on the consumerd socket and receives the file | |
1681 | * descriptors from the session daemon. | |
1682 | */ | |
1683 | void *lttng_consumer_thread_receive_fds(void *data) | |
1684 | { | |
1685 | int sock, client_socket, ret; | |
1686 | /* | |
1687 | * structure to poll for incoming data on communication socket avoids | |
1688 | * making blocking sockets. | |
1689 | */ | |
1690 | struct pollfd consumer_sockpoll[2]; | |
1691 | struct lttng_consumer_local_data *ctx = data; | |
1692 | ||
e7b994a3 DG |
1693 | rcu_register_thread(); |
1694 | ||
3bd1e081 MD |
1695 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
1696 | unlink(ctx->consumer_command_sock_path); | |
1697 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); | |
1698 | if (client_socket < 0) { | |
1699 | ERR("Cannot create command socket"); | |
1700 | goto end; | |
1701 | } | |
1702 | ||
1703 | ret = lttcomm_listen_unix_sock(client_socket); | |
1704 | if (ret < 0) { | |
1705 | goto end; | |
1706 | } | |
1707 | ||
32258573 | 1708 | DBG("Sending ready command to lttng-sessiond"); |
3bd1e081 MD |
1709 | ret = lttng_consumer_send_error(ctx, CONSUMERD_COMMAND_SOCK_READY); |
1710 | /* return < 0 on error, but == 0 is not fatal */ | |
1711 | if (ret < 0) { | |
32258573 | 1712 | ERR("Error sending ready command to lttng-sessiond"); |
3bd1e081 MD |
1713 | goto end; |
1714 | } | |
1715 | ||
1716 | ret = fcntl(client_socket, F_SETFL, O_NONBLOCK); | |
1717 | if (ret < 0) { | |
1718 | perror("fcntl O_NONBLOCK"); | |
1719 | goto end; | |
1720 | } | |
1721 | ||
1722 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ | |
1723 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; | |
1724 | consumer_sockpoll[0].events = POLLIN | POLLPRI; | |
1725 | consumer_sockpoll[1].fd = client_socket; | |
1726 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
1727 | ||
1728 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
1729 | goto end; | |
1730 | } | |
1731 | DBG("Connection on client_socket"); | |
1732 | ||
1733 | /* Blocking call, waiting for transmission */ | |
1734 | sock = lttcomm_accept_unix_sock(client_socket); | |
1735 | if (sock <= 0) { | |
1736 | WARN("On accept"); | |
1737 | goto end; | |
1738 | } | |
1739 | ret = fcntl(sock, F_SETFL, O_NONBLOCK); | |
1740 | if (ret < 0) { | |
1741 | perror("fcntl O_NONBLOCK"); | |
1742 | goto end; | |
1743 | } | |
1744 | ||
1745 | /* update the polling structure to poll on the established socket */ | |
1746 | consumer_sockpoll[1].fd = sock; | |
1747 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
1748 | ||
1749 | while (1) { | |
1750 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
1751 | goto end; | |
1752 | } | |
1753 | DBG("Incoming command on sock"); | |
1754 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
1755 | if (ret == -ENOENT) { | |
1756 | DBG("Received STOP command"); | |
1757 | goto end; | |
1758 | } | |
1759 | if (ret < 0) { | |
1760 | ERR("Communication interrupted on command socket"); | |
1761 | goto end; | |
1762 | } | |
1763 | if (consumer_quit) { | |
1764 | DBG("consumer_thread_receive_fds received quit from signal"); | |
1765 | goto end; | |
1766 | } | |
1767 | DBG("received fds on sock"); | |
1768 | } | |
1769 | end: | |
1770 | DBG("consumer_thread_receive_fds exiting"); | |
1771 | ||
1772 | /* | |
1773 | * when all fds have hung up, the polling thread | |
1774 | * can exit cleanly | |
1775 | */ | |
1776 | consumer_quit = 1; | |
1777 | ||
1778 | /* | |
1779 | * 2s of grace period, if no polling events occur during | |
1780 | * this period, the polling thread will exit even if there | |
1781 | * are still open FDs (should not happen, but safety mechanism). | |
1782 | */ | |
1783 | consumer_poll_timeout = LTTNG_CONSUMER_POLL_TIMEOUT; | |
1784 | ||
04fdd819 MD |
1785 | /* |
1786 | * Wake-up the other end by writing a null byte in the pipe | |
1787 | * (non-blocking). Important note: Because writing into the | |
1788 | * pipe is non-blocking (and therefore we allow dropping wakeup | |
1789 | * data, as long as there is wakeup data present in the pipe | |
1790 | * buffer to wake up the other end), the other end should | |
1791 | * perform the following sequence for waiting: | |
1792 | * 1) empty the pipe (reads). | |
1793 | * 2) perform update operation. | |
1794 | * 3) wait on the pipe (poll). | |
1795 | */ | |
1796 | do { | |
1797 | ret = write(ctx->consumer_poll_pipe[1], "", 1); | |
6f94560a | 1798 | } while (ret < 0 && errno == EINTR); |
e7b994a3 | 1799 | rcu_unregister_thread(); |
3bd1e081 MD |
1800 | return NULL; |
1801 | } | |
d41f73b7 | 1802 | |
4078b776 | 1803 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
d41f73b7 MD |
1804 | struct lttng_consumer_local_data *ctx) |
1805 | { | |
1806 | switch (consumer_data.type) { | |
1807 | case LTTNG_CONSUMER_KERNEL: | |
1808 | return lttng_kconsumer_read_subbuffer(stream, ctx); | |
7753dea8 MD |
1809 | case LTTNG_CONSUMER32_UST: |
1810 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
1811 | return lttng_ustconsumer_read_subbuffer(stream, ctx); |
1812 | default: | |
1813 | ERR("Unknown consumer_data type"); | |
1814 | assert(0); | |
1815 | return -ENOSYS; | |
1816 | } | |
1817 | } | |
1818 | ||
1819 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
1820 | { | |
1821 | switch (consumer_data.type) { | |
1822 | case LTTNG_CONSUMER_KERNEL: | |
1823 | return lttng_kconsumer_on_recv_stream(stream); | |
7753dea8 MD |
1824 | case LTTNG_CONSUMER32_UST: |
1825 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
1826 | return lttng_ustconsumer_on_recv_stream(stream); |
1827 | default: | |
1828 | ERR("Unknown consumer_data type"); | |
1829 | assert(0); | |
1830 | return -ENOSYS; | |
1831 | } | |
1832 | } | |
e4421fec DG |
1833 | |
1834 | /* | |
1835 | * Allocate and set consumer data hash tables. | |
1836 | */ | |
1837 | void lttng_consumer_init(void) | |
1838 | { | |
1839 | consumer_data.stream_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
1840 | consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
00e2e675 | 1841 | consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); |
e4421fec | 1842 | } |