Commit | Line | Data |
---|---|---|
3bd1e081 MD |
1 | /* |
2 | * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca> | |
3 | * Mathieu Desnoyers <mathieu.desnoyers@efficios.com> | |
00e2e675 | 4 | * 2012 - David Goulet <dgoulet@efficios.com> |
3bd1e081 | 5 | * |
d14d33bf AM |
6 | * This program is free software; you can redistribute it and/or modify |
7 | * it under the terms of the GNU General Public License, version 2 only, | |
8 | * as published by the Free Software Foundation. | |
3bd1e081 | 9 | * |
d14d33bf AM |
10 | * This program is distributed in the hope that it will be useful, but WITHOUT |
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
13 | * more details. | |
3bd1e081 | 14 | * |
d14d33bf AM |
15 | * You should have received a copy of the GNU General Public License along |
16 | * with this program; if not, write to the Free Software Foundation, Inc., | |
17 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. | |
3bd1e081 MD |
18 | */ |
19 | ||
20 | #define _GNU_SOURCE | |
21 | #include <assert.h> | |
3bd1e081 MD |
22 | #include <poll.h> |
23 | #include <pthread.h> | |
24 | #include <stdlib.h> | |
25 | #include <string.h> | |
26 | #include <sys/mman.h> | |
27 | #include <sys/socket.h> | |
28 | #include <sys/types.h> | |
29 | #include <unistd.h> | |
77c7c900 | 30 | #include <inttypes.h> |
331744e3 | 31 | #include <signal.h> |
3bd1e081 | 32 | |
990570ed | 33 | #include <common/common.h> |
fb3a43a9 DG |
34 | #include <common/utils.h> |
35 | #include <common/compat/poll.h> | |
10a8a223 | 36 | #include <common/kernel-ctl/kernel-ctl.h> |
00e2e675 | 37 | #include <common/sessiond-comm/relayd.h> |
10a8a223 DG |
38 | #include <common/sessiond-comm/sessiond-comm.h> |
39 | #include <common/kernel-consumer/kernel-consumer.h> | |
00e2e675 | 40 | #include <common/relayd/relayd.h> |
10a8a223 DG |
41 | #include <common/ust-consumer/ust-consumer.h> |
42 | ||
43 | #include "consumer.h" | |
3bd1e081 MD |
44 | |
45 | struct lttng_consumer_global_data consumer_data = { | |
3bd1e081 MD |
46 | .stream_count = 0, |
47 | .need_update = 1, | |
48 | .type = LTTNG_CONSUMER_UNKNOWN, | |
49 | }; | |
50 | ||
d8ef542d MD |
51 | enum consumer_channel_action { |
52 | CONSUMER_CHANNEL_ADD, | |
a0cbdd2e | 53 | CONSUMER_CHANNEL_DEL, |
d8ef542d MD |
54 | CONSUMER_CHANNEL_QUIT, |
55 | }; | |
56 | ||
57 | struct consumer_channel_msg { | |
58 | enum consumer_channel_action action; | |
a0cbdd2e MD |
59 | struct lttng_consumer_channel *chan; /* add */ |
60 | uint64_t key; /* del */ | |
d8ef542d MD |
61 | }; |
62 | ||
3bd1e081 MD |
63 | /* |
64 | * Flag to inform the polling thread to quit when all fd hung up. Updated by | |
65 | * the consumer_thread_receive_fds when it notices that all fds has hung up. | |
66 | * Also updated by the signal handler (consumer_should_exit()). Read by the | |
67 | * polling threads. | |
68 | */ | |
a98dae5f | 69 | volatile int consumer_quit; |
3bd1e081 | 70 | |
43c34bc3 | 71 | /* |
43c34bc3 DG |
72 | * Global hash table containing respectively metadata and data streams. The |
73 | * stream element in this ht should only be updated by the metadata poll thread | |
74 | * for the metadata and the data poll thread for the data. | |
75 | */ | |
40dc48e0 DG |
76 | static struct lttng_ht *metadata_ht; |
77 | static struct lttng_ht *data_ht; | |
43c34bc3 | 78 | |
acdb9057 DG |
79 | /* |
80 | * Notify a thread lttng pipe to poll back again. This usually means that some | |
81 | * global state has changed so we just send back the thread in a poll wait | |
82 | * call. | |
83 | */ | |
84 | static void notify_thread_lttng_pipe(struct lttng_pipe *pipe) | |
85 | { | |
86 | struct lttng_consumer_stream *null_stream = NULL; | |
87 | ||
88 | assert(pipe); | |
89 | ||
90 | (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream)); | |
91 | } | |
92 | ||
d8ef542d MD |
93 | static void notify_channel_pipe(struct lttng_consumer_local_data *ctx, |
94 | struct lttng_consumer_channel *chan, | |
a0cbdd2e | 95 | uint64_t key, |
d8ef542d MD |
96 | enum consumer_channel_action action) |
97 | { | |
98 | struct consumer_channel_msg msg; | |
99 | int ret; | |
100 | ||
e56251fc DG |
101 | memset(&msg, 0, sizeof(msg)); |
102 | ||
d8ef542d MD |
103 | msg.action = action; |
104 | msg.chan = chan; | |
105 | do { | |
106 | ret = write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg)); | |
107 | } while (ret < 0 && errno == EINTR); | |
108 | } | |
109 | ||
a0cbdd2e MD |
110 | void notify_thread_del_channel(struct lttng_consumer_local_data *ctx, |
111 | uint64_t key) | |
112 | { | |
113 | notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL); | |
114 | } | |
115 | ||
d8ef542d MD |
116 | static int read_channel_pipe(struct lttng_consumer_local_data *ctx, |
117 | struct lttng_consumer_channel **chan, | |
a0cbdd2e | 118 | uint64_t *key, |
d8ef542d MD |
119 | enum consumer_channel_action *action) |
120 | { | |
121 | struct consumer_channel_msg msg; | |
122 | int ret; | |
123 | ||
124 | do { | |
125 | ret = read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg)); | |
126 | } while (ret < 0 && errno == EINTR); | |
127 | if (ret > 0) { | |
128 | *action = msg.action; | |
129 | *chan = msg.chan; | |
a0cbdd2e | 130 | *key = msg.key; |
d8ef542d MD |
131 | } |
132 | return ret; | |
133 | } | |
134 | ||
3bd1e081 MD |
135 | /* |
136 | * Find a stream. The consumer_data.lock must be locked during this | |
137 | * call. | |
138 | */ | |
d88aee68 | 139 | static struct lttng_consumer_stream *find_stream(uint64_t key, |
8389e4f8 | 140 | struct lttng_ht *ht) |
3bd1e081 | 141 | { |
e4421fec | 142 | struct lttng_ht_iter iter; |
d88aee68 | 143 | struct lttng_ht_node_u64 *node; |
e4421fec | 144 | struct lttng_consumer_stream *stream = NULL; |
3bd1e081 | 145 | |
8389e4f8 DG |
146 | assert(ht); |
147 | ||
d88aee68 DG |
148 | /* -1ULL keys are lookup failures */ |
149 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 150 | return NULL; |
7a57cf92 | 151 | } |
e4421fec | 152 | |
6065ceec DG |
153 | rcu_read_lock(); |
154 | ||
d88aee68 DG |
155 | lttng_ht_lookup(ht, &key, &iter); |
156 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
157 | if (node != NULL) { |
158 | stream = caa_container_of(node, struct lttng_consumer_stream, node); | |
3bd1e081 | 159 | } |
e4421fec | 160 | |
6065ceec DG |
161 | rcu_read_unlock(); |
162 | ||
e4421fec | 163 | return stream; |
3bd1e081 MD |
164 | } |
165 | ||
ffe60014 | 166 | static void steal_stream_key(int key, struct lttng_ht *ht) |
7ad0a0cb MD |
167 | { |
168 | struct lttng_consumer_stream *stream; | |
169 | ||
04253271 | 170 | rcu_read_lock(); |
ffe60014 | 171 | stream = find_stream(key, ht); |
04253271 | 172 | if (stream) { |
d88aee68 | 173 | stream->key = -1ULL; |
04253271 MD |
174 | /* |
175 | * We don't want the lookup to match, but we still need | |
176 | * to iterate on this stream when iterating over the hash table. Just | |
177 | * change the node key. | |
178 | */ | |
d88aee68 | 179 | stream->node.key = -1ULL; |
04253271 MD |
180 | } |
181 | rcu_read_unlock(); | |
7ad0a0cb MD |
182 | } |
183 | ||
d56db448 DG |
184 | /* |
185 | * Return a channel object for the given key. | |
186 | * | |
187 | * RCU read side lock MUST be acquired before calling this function and | |
188 | * protects the channel ptr. | |
189 | */ | |
d88aee68 | 190 | struct lttng_consumer_channel *consumer_find_channel(uint64_t key) |
3bd1e081 | 191 | { |
e4421fec | 192 | struct lttng_ht_iter iter; |
d88aee68 | 193 | struct lttng_ht_node_u64 *node; |
e4421fec | 194 | struct lttng_consumer_channel *channel = NULL; |
3bd1e081 | 195 | |
d88aee68 DG |
196 | /* -1ULL keys are lookup failures */ |
197 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 198 | return NULL; |
7a57cf92 | 199 | } |
e4421fec | 200 | |
d88aee68 DG |
201 | lttng_ht_lookup(consumer_data.channel_ht, &key, &iter); |
202 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
203 | if (node != NULL) { |
204 | channel = caa_container_of(node, struct lttng_consumer_channel, node); | |
3bd1e081 | 205 | } |
e4421fec DG |
206 | |
207 | return channel; | |
3bd1e081 MD |
208 | } |
209 | ||
ffe60014 | 210 | static void free_stream_rcu(struct rcu_head *head) |
7ad0a0cb | 211 | { |
d88aee68 DG |
212 | struct lttng_ht_node_u64 *node = |
213 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
ffe60014 DG |
214 | struct lttng_consumer_stream *stream = |
215 | caa_container_of(node, struct lttng_consumer_stream, node); | |
7ad0a0cb | 216 | |
ffe60014 | 217 | free(stream); |
7ad0a0cb MD |
218 | } |
219 | ||
ffe60014 | 220 | static void free_channel_rcu(struct rcu_head *head) |
702b1ea4 | 221 | { |
d88aee68 DG |
222 | struct lttng_ht_node_u64 *node = |
223 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
ffe60014 DG |
224 | struct lttng_consumer_channel *channel = |
225 | caa_container_of(node, struct lttng_consumer_channel, node); | |
702b1ea4 | 226 | |
ffe60014 | 227 | free(channel); |
702b1ea4 MD |
228 | } |
229 | ||
00e2e675 DG |
230 | /* |
231 | * RCU protected relayd socket pair free. | |
232 | */ | |
ffe60014 | 233 | static void free_relayd_rcu(struct rcu_head *head) |
00e2e675 | 234 | { |
d88aee68 DG |
235 | struct lttng_ht_node_u64 *node = |
236 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
00e2e675 DG |
237 | struct consumer_relayd_sock_pair *relayd = |
238 | caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
239 | ||
8994307f DG |
240 | /* |
241 | * Close all sockets. This is done in the call RCU since we don't want the | |
242 | * socket fds to be reassigned thus potentially creating bad state of the | |
243 | * relayd object. | |
244 | * | |
245 | * We do not have to lock the control socket mutex here since at this stage | |
246 | * there is no one referencing to this relayd object. | |
247 | */ | |
248 | (void) relayd_close(&relayd->control_sock); | |
249 | (void) relayd_close(&relayd->data_sock); | |
250 | ||
00e2e675 DG |
251 | free(relayd); |
252 | } | |
253 | ||
254 | /* | |
255 | * Destroy and free relayd socket pair object. | |
256 | * | |
257 | * This function MUST be called with the consumer_data lock acquired. | |
258 | */ | |
d09e1200 | 259 | static void destroy_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
260 | { |
261 | int ret; | |
262 | struct lttng_ht_iter iter; | |
263 | ||
173af62f DG |
264 | if (relayd == NULL) { |
265 | return; | |
266 | } | |
267 | ||
00e2e675 DG |
268 | DBG("Consumer destroy and close relayd socket pair"); |
269 | ||
270 | iter.iter.node = &relayd->node.node; | |
271 | ret = lttng_ht_del(consumer_data.relayd_ht, &iter); | |
173af62f | 272 | if (ret != 0) { |
8994307f | 273 | /* We assume the relayd is being or is destroyed */ |
173af62f DG |
274 | return; |
275 | } | |
00e2e675 | 276 | |
00e2e675 | 277 | /* RCU free() call */ |
ffe60014 DG |
278 | call_rcu(&relayd->node.head, free_relayd_rcu); |
279 | } | |
280 | ||
281 | /* | |
282 | * Remove a channel from the global list protected by a mutex. This function is | |
283 | * also responsible for freeing its data structures. | |
284 | */ | |
285 | void consumer_del_channel(struct lttng_consumer_channel *channel) | |
286 | { | |
287 | int ret; | |
288 | struct lttng_ht_iter iter; | |
289 | ||
d88aee68 | 290 | DBG("Consumer delete channel key %" PRIu64, channel->key); |
ffe60014 DG |
291 | |
292 | pthread_mutex_lock(&consumer_data.lock); | |
293 | ||
294 | switch (consumer_data.type) { | |
295 | case LTTNG_CONSUMER_KERNEL: | |
296 | break; | |
297 | case LTTNG_CONSUMER32_UST: | |
298 | case LTTNG_CONSUMER64_UST: | |
299 | lttng_ustconsumer_del_channel(channel); | |
300 | break; | |
301 | default: | |
302 | ERR("Unknown consumer_data type"); | |
303 | assert(0); | |
304 | goto end; | |
305 | } | |
306 | ||
307 | rcu_read_lock(); | |
308 | iter.iter.node = &channel->node.node; | |
309 | ret = lttng_ht_del(consumer_data.channel_ht, &iter); | |
310 | assert(!ret); | |
311 | rcu_read_unlock(); | |
312 | ||
313 | call_rcu(&channel->node.head, free_channel_rcu); | |
314 | end: | |
315 | pthread_mutex_unlock(&consumer_data.lock); | |
00e2e675 DG |
316 | } |
317 | ||
228b5bf7 DG |
318 | /* |
319 | * Iterate over the relayd hash table and destroy each element. Finally, | |
320 | * destroy the whole hash table. | |
321 | */ | |
322 | static void cleanup_relayd_ht(void) | |
323 | { | |
324 | struct lttng_ht_iter iter; | |
325 | struct consumer_relayd_sock_pair *relayd; | |
326 | ||
327 | rcu_read_lock(); | |
328 | ||
329 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
330 | node.node) { | |
331 | destroy_relayd(relayd); | |
332 | } | |
333 | ||
228b5bf7 | 334 | rcu_read_unlock(); |
36b588ed MD |
335 | |
336 | lttng_ht_destroy(consumer_data.relayd_ht); | |
228b5bf7 DG |
337 | } |
338 | ||
8994307f DG |
339 | /* |
340 | * Update the end point status of all streams having the given network sequence | |
341 | * index (relayd index). | |
342 | * | |
343 | * It's atomically set without having the stream mutex locked which is fine | |
344 | * because we handle the write/read race with a pipe wakeup for each thread. | |
345 | */ | |
346 | static void update_endpoint_status_by_netidx(int net_seq_idx, | |
347 | enum consumer_endpoint_status status) | |
348 | { | |
349 | struct lttng_ht_iter iter; | |
350 | struct lttng_consumer_stream *stream; | |
351 | ||
352 | DBG("Consumer set delete flag on stream by idx %d", net_seq_idx); | |
353 | ||
354 | rcu_read_lock(); | |
355 | ||
356 | /* Let's begin with metadata */ | |
357 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
358 | if (stream->net_seq_idx == net_seq_idx) { | |
359 | uatomic_set(&stream->endpoint_status, status); | |
360 | DBG("Delete flag set to metadata stream %d", stream->wait_fd); | |
361 | } | |
362 | } | |
363 | ||
364 | /* Follow up by the data streams */ | |
365 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
366 | if (stream->net_seq_idx == net_seq_idx) { | |
367 | uatomic_set(&stream->endpoint_status, status); | |
368 | DBG("Delete flag set to data stream %d", stream->wait_fd); | |
369 | } | |
370 | } | |
371 | rcu_read_unlock(); | |
372 | } | |
373 | ||
374 | /* | |
375 | * Cleanup a relayd object by flagging every associated streams for deletion, | |
376 | * destroying the object meaning removing it from the relayd hash table, | |
377 | * closing the sockets and freeing the memory in a RCU call. | |
378 | * | |
379 | * If a local data context is available, notify the threads that the streams' | |
380 | * state have changed. | |
381 | */ | |
382 | static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd, | |
383 | struct lttng_consumer_local_data *ctx) | |
384 | { | |
385 | int netidx; | |
386 | ||
387 | assert(relayd); | |
388 | ||
9617607b DG |
389 | DBG("Cleaning up relayd sockets"); |
390 | ||
8994307f DG |
391 | /* Save the net sequence index before destroying the object */ |
392 | netidx = relayd->net_seq_idx; | |
393 | ||
394 | /* | |
395 | * Delete the relayd from the relayd hash table, close the sockets and free | |
396 | * the object in a RCU call. | |
397 | */ | |
398 | destroy_relayd(relayd); | |
399 | ||
400 | /* Set inactive endpoint to all streams */ | |
401 | update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE); | |
402 | ||
403 | /* | |
404 | * With a local data context, notify the threads that the streams' state | |
405 | * have changed. The write() action on the pipe acts as an "implicit" | |
406 | * memory barrier ordering the updates of the end point status from the | |
407 | * read of this status which happens AFTER receiving this notify. | |
408 | */ | |
409 | if (ctx) { | |
acdb9057 | 410 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
13886d2d | 411 | notify_thread_lttng_pipe(ctx->consumer_metadata_pipe); |
8994307f DG |
412 | } |
413 | } | |
414 | ||
a6ba4fe1 DG |
415 | /* |
416 | * Flag a relayd socket pair for destruction. Destroy it if the refcount | |
417 | * reaches zero. | |
418 | * | |
419 | * RCU read side lock MUST be aquired before calling this function. | |
420 | */ | |
421 | void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) | |
422 | { | |
423 | assert(relayd); | |
424 | ||
425 | /* Set destroy flag for this object */ | |
426 | uatomic_set(&relayd->destroy_flag, 1); | |
427 | ||
428 | /* Destroy the relayd if refcount is 0 */ | |
429 | if (uatomic_read(&relayd->refcount) == 0) { | |
d09e1200 | 430 | destroy_relayd(relayd); |
a6ba4fe1 DG |
431 | } |
432 | } | |
433 | ||
3bd1e081 MD |
434 | /* |
435 | * Remove a stream from the global list protected by a mutex. This | |
436 | * function is also responsible for freeing its data structures. | |
437 | */ | |
e316aad5 DG |
438 | void consumer_del_stream(struct lttng_consumer_stream *stream, |
439 | struct lttng_ht *ht) | |
3bd1e081 MD |
440 | { |
441 | int ret; | |
e4421fec | 442 | struct lttng_ht_iter iter; |
3bd1e081 | 443 | struct lttng_consumer_channel *free_chan = NULL; |
00e2e675 DG |
444 | struct consumer_relayd_sock_pair *relayd; |
445 | ||
446 | assert(stream); | |
3bd1e081 | 447 | |
8994307f DG |
448 | DBG("Consumer del stream %d", stream->wait_fd); |
449 | ||
e316aad5 DG |
450 | if (ht == NULL) { |
451 | /* Means the stream was allocated but not successfully added */ | |
ffe60014 | 452 | goto free_stream_rcu; |
e316aad5 DG |
453 | } |
454 | ||
3bd1e081 | 455 | pthread_mutex_lock(&consumer_data.lock); |
74251bb8 | 456 | pthread_mutex_lock(&stream->lock); |
3bd1e081 MD |
457 | |
458 | switch (consumer_data.type) { | |
459 | case LTTNG_CONSUMER_KERNEL: | |
460 | if (stream->mmap_base != NULL) { | |
461 | ret = munmap(stream->mmap_base, stream->mmap_len); | |
462 | if (ret != 0) { | |
7a57cf92 | 463 | PERROR("munmap"); |
3bd1e081 MD |
464 | } |
465 | } | |
4c95e622 JD |
466 | |
467 | if (stream->wait_fd >= 0) { | |
468 | ret = close(stream->wait_fd); | |
469 | if (ret) { | |
470 | PERROR("close"); | |
471 | } | |
472 | } | |
3bd1e081 | 473 | break; |
7753dea8 MD |
474 | case LTTNG_CONSUMER32_UST: |
475 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
476 | lttng_ustconsumer_del_stream(stream); |
477 | break; | |
478 | default: | |
479 | ERR("Unknown consumer_data type"); | |
480 | assert(0); | |
481 | goto end; | |
482 | } | |
483 | ||
6065ceec | 484 | rcu_read_lock(); |
04253271 | 485 | iter.iter.node = &stream->node.node; |
e316aad5 | 486 | ret = lttng_ht_del(ht, &iter); |
04253271 | 487 | assert(!ret); |
ca22feea | 488 | |
d8ef542d MD |
489 | iter.iter.node = &stream->node_channel_id.node; |
490 | ret = lttng_ht_del(consumer_data.stream_per_chan_id_ht, &iter); | |
491 | assert(!ret); | |
492 | ||
ca22feea DG |
493 | iter.iter.node = &stream->node_session_id.node; |
494 | ret = lttng_ht_del(consumer_data.stream_list_ht, &iter); | |
495 | assert(!ret); | |
6065ceec DG |
496 | rcu_read_unlock(); |
497 | ||
50f8ae69 | 498 | assert(consumer_data.stream_count > 0); |
3bd1e081 | 499 | consumer_data.stream_count--; |
50f8ae69 | 500 | |
3bd1e081 | 501 | if (stream->out_fd >= 0) { |
4c462e79 MD |
502 | ret = close(stream->out_fd); |
503 | if (ret) { | |
504 | PERROR("close"); | |
505 | } | |
3bd1e081 | 506 | } |
00e2e675 DG |
507 | |
508 | /* Check and cleanup relayd */ | |
b0b335c8 | 509 | rcu_read_lock(); |
00e2e675 DG |
510 | relayd = consumer_find_relayd(stream->net_seq_idx); |
511 | if (relayd != NULL) { | |
b0b335c8 MD |
512 | uatomic_dec(&relayd->refcount); |
513 | assert(uatomic_read(&relayd->refcount) >= 0); | |
173af62f | 514 | |
3f8e211f DG |
515 | /* Closing streams requires to lock the control socket. */ |
516 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
173af62f DG |
517 | ret = relayd_send_close_stream(&relayd->control_sock, |
518 | stream->relayd_stream_id, | |
519 | stream->next_net_seq_num - 1); | |
3f8e211f | 520 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
173af62f | 521 | if (ret < 0) { |
a4b92340 DG |
522 | DBG("Unable to close stream on the relayd. Continuing"); |
523 | /* | |
524 | * Continue here. There is nothing we can do for the relayd. | |
525 | * Chances are that the relayd has closed the socket so we just | |
526 | * continue cleaning up. | |
527 | */ | |
173af62f DG |
528 | } |
529 | ||
530 | /* Both conditions are met, we destroy the relayd. */ | |
531 | if (uatomic_read(&relayd->refcount) == 0 && | |
532 | uatomic_read(&relayd->destroy_flag)) { | |
d09e1200 | 533 | destroy_relayd(relayd); |
00e2e675 | 534 | } |
00e2e675 | 535 | } |
b0b335c8 | 536 | rcu_read_unlock(); |
00e2e675 | 537 | |
f2ad556d | 538 | if (!uatomic_sub_return(&stream->chan->refcount, 1) |
ffe60014 | 539 | && !uatomic_read(&stream->chan->nb_init_stream_left)) { |
3bd1e081 | 540 | free_chan = stream->chan; |
00e2e675 DG |
541 | } |
542 | ||
3bd1e081 MD |
543 | end: |
544 | consumer_data.need_update = 1; | |
8994307f | 545 | pthread_mutex_unlock(&stream->lock); |
74251bb8 | 546 | pthread_mutex_unlock(&consumer_data.lock); |
3bd1e081 | 547 | |
c30aaa51 | 548 | if (free_chan) { |
3bd1e081 | 549 | consumer_del_channel(free_chan); |
c30aaa51 | 550 | } |
e316aad5 | 551 | |
ffe60014 DG |
552 | free_stream_rcu: |
553 | call_rcu(&stream->node.head, free_stream_rcu); | |
3bd1e081 MD |
554 | } |
555 | ||
d88aee68 DG |
556 | struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, |
557 | uint64_t stream_key, | |
3bd1e081 | 558 | enum lttng_consumer_stream_state state, |
ffe60014 | 559 | const char *channel_name, |
6df2e2c9 | 560 | uid_t uid, |
00e2e675 | 561 | gid_t gid, |
ffe60014 | 562 | int relayd_id, |
53632229 | 563 | uint64_t session_id, |
ffe60014 DG |
564 | int cpu, |
565 | int *alloc_ret, | |
566 | enum consumer_channel_type type) | |
3bd1e081 | 567 | { |
ffe60014 | 568 | int ret; |
3bd1e081 | 569 | struct lttng_consumer_stream *stream; |
3bd1e081 | 570 | |
effcf122 | 571 | stream = zmalloc(sizeof(*stream)); |
3bd1e081 | 572 | if (stream == NULL) { |
7a57cf92 | 573 | PERROR("malloc struct lttng_consumer_stream"); |
ffe60014 | 574 | ret = -ENOMEM; |
7a57cf92 | 575 | goto end; |
3bd1e081 | 576 | } |
7a57cf92 | 577 | |
d56db448 DG |
578 | rcu_read_lock(); |
579 | ||
3bd1e081 | 580 | stream->key = stream_key; |
3bd1e081 MD |
581 | stream->out_fd = -1; |
582 | stream->out_fd_offset = 0; | |
583 | stream->state = state; | |
6df2e2c9 MD |
584 | stream->uid = uid; |
585 | stream->gid = gid; | |
ffe60014 | 586 | stream->net_seq_idx = relayd_id; |
53632229 | 587 | stream->session_id = session_id; |
53632229 | 588 | pthread_mutex_init(&stream->lock, NULL); |
58b1f425 | 589 | |
ffe60014 DG |
590 | /* If channel is the metadata, flag this stream as metadata. */ |
591 | if (type == CONSUMER_CHANNEL_TYPE_METADATA) { | |
592 | stream->metadata_flag = 1; | |
593 | /* Metadata is flat out. */ | |
594 | strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name)); | |
58b1f425 | 595 | } else { |
ffe60014 DG |
596 | /* Format stream name to <channel_name>_<cpu_number> */ |
597 | ret = snprintf(stream->name, sizeof(stream->name), "%s_%d", | |
598 | channel_name, cpu); | |
599 | if (ret < 0) { | |
600 | PERROR("snprintf stream name"); | |
601 | goto error; | |
602 | } | |
58b1f425 | 603 | } |
c30aaa51 | 604 | |
ffe60014 | 605 | /* Key is always the wait_fd for streams. */ |
d88aee68 | 606 | lttng_ht_node_init_u64(&stream->node, stream->key); |
ffe60014 | 607 | |
d8ef542d MD |
608 | /* Init node per channel id key */ |
609 | lttng_ht_node_init_u64(&stream->node_channel_id, channel_key); | |
610 | ||
53632229 | 611 | /* Init session id node with the stream session id */ |
d88aee68 | 612 | lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id); |
53632229 | 613 | |
d8ef542d MD |
614 | DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64 " relayd_id %" PRIu64 ", session_id %" PRIu64, |
615 | stream->name, stream->key, channel_key, stream->net_seq_idx, stream->session_id); | |
d56db448 DG |
616 | |
617 | rcu_read_unlock(); | |
3bd1e081 | 618 | return stream; |
c80048c6 MD |
619 | |
620 | error: | |
d56db448 | 621 | rcu_read_unlock(); |
c80048c6 | 622 | free(stream); |
7a57cf92 | 623 | end: |
ffe60014 DG |
624 | if (alloc_ret) { |
625 | *alloc_ret = ret; | |
626 | } | |
c80048c6 | 627 | return NULL; |
3bd1e081 MD |
628 | } |
629 | ||
630 | /* | |
631 | * Add a stream to the global list protected by a mutex. | |
632 | */ | |
ffe60014 | 633 | static int add_stream(struct lttng_consumer_stream *stream, |
43c34bc3 | 634 | struct lttng_ht *ht) |
3bd1e081 MD |
635 | { |
636 | int ret = 0; | |
00e2e675 | 637 | struct consumer_relayd_sock_pair *relayd; |
3bd1e081 | 638 | |
e316aad5 | 639 | assert(stream); |
43c34bc3 | 640 | assert(ht); |
c77fc10a | 641 | |
d88aee68 | 642 | DBG3("Adding consumer stream %" PRIu64, stream->key); |
e316aad5 DG |
643 | |
644 | pthread_mutex_lock(&consumer_data.lock); | |
2e818a6a | 645 | pthread_mutex_lock(&stream->lock); |
b0b335c8 | 646 | rcu_read_lock(); |
e316aad5 | 647 | |
43c34bc3 | 648 | /* Steal stream identifier to avoid having streams with the same key */ |
ffe60014 | 649 | steal_stream_key(stream->key, ht); |
43c34bc3 | 650 | |
d88aee68 | 651 | lttng_ht_add_unique_u64(ht, &stream->node); |
00e2e675 | 652 | |
d8ef542d MD |
653 | lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, |
654 | &stream->node_channel_id); | |
655 | ||
ca22feea DG |
656 | /* |
657 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
658 | * the key since the HT does not use it and we allow to add redundant keys | |
659 | * into this table. | |
660 | */ | |
d88aee68 | 661 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 662 | |
00e2e675 DG |
663 | /* Check and cleanup relayd */ |
664 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
665 | if (relayd != NULL) { | |
b0b335c8 | 666 | uatomic_inc(&relayd->refcount); |
00e2e675 DG |
667 | } |
668 | ||
e316aad5 | 669 | /* |
ffe60014 DG |
670 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
671 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
672 | * causes the count to become 0 also causes a stream to be added. The |
673 | * channel deletion will thus be triggered by the following removal of this | |
674 | * stream. | |
675 | */ | |
ffe60014 | 676 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
677 | /* Increment refcount before decrementing nb_init_stream_left */ |
678 | cmm_smp_wmb(); | |
ffe60014 | 679 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
680 | } |
681 | ||
682 | /* Update consumer data once the node is inserted. */ | |
3bd1e081 MD |
683 | consumer_data.stream_count++; |
684 | consumer_data.need_update = 1; | |
685 | ||
e316aad5 | 686 | rcu_read_unlock(); |
2e818a6a | 687 | pthread_mutex_unlock(&stream->lock); |
3bd1e081 | 688 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 689 | |
3bd1e081 MD |
690 | return ret; |
691 | } | |
692 | ||
00e2e675 | 693 | /* |
3f8e211f DG |
694 | * Add relayd socket to global consumer data hashtable. RCU read side lock MUST |
695 | * be acquired before calling this. | |
00e2e675 | 696 | */ |
d09e1200 | 697 | static int add_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
698 | { |
699 | int ret = 0; | |
d88aee68 | 700 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
701 | struct lttng_ht_iter iter; |
702 | ||
ffe60014 | 703 | assert(relayd); |
00e2e675 | 704 | |
00e2e675 | 705 | lttng_ht_lookup(consumer_data.relayd_ht, |
d88aee68 DG |
706 | &relayd->net_seq_idx, &iter); |
707 | node = lttng_ht_iter_get_node_u64(&iter); | |
00e2e675 | 708 | if (node != NULL) { |
00e2e675 DG |
709 | goto end; |
710 | } | |
d88aee68 | 711 | lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node); |
00e2e675 | 712 | |
00e2e675 DG |
713 | end: |
714 | return ret; | |
715 | } | |
716 | ||
717 | /* | |
718 | * Allocate and return a consumer relayd socket. | |
719 | */ | |
720 | struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( | |
721 | int net_seq_idx) | |
722 | { | |
723 | struct consumer_relayd_sock_pair *obj = NULL; | |
724 | ||
725 | /* Negative net sequence index is a failure */ | |
726 | if (net_seq_idx < 0) { | |
727 | goto error; | |
728 | } | |
729 | ||
730 | obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); | |
731 | if (obj == NULL) { | |
732 | PERROR("zmalloc relayd sock"); | |
733 | goto error; | |
734 | } | |
735 | ||
736 | obj->net_seq_idx = net_seq_idx; | |
737 | obj->refcount = 0; | |
173af62f | 738 | obj->destroy_flag = 0; |
f96e4545 MD |
739 | obj->control_sock.sock.fd = -1; |
740 | obj->data_sock.sock.fd = -1; | |
d88aee68 | 741 | lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx); |
00e2e675 DG |
742 | pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); |
743 | ||
744 | error: | |
745 | return obj; | |
746 | } | |
747 | ||
748 | /* | |
749 | * Find a relayd socket pair in the global consumer data. | |
750 | * | |
751 | * Return the object if found else NULL. | |
b0b335c8 MD |
752 | * RCU read-side lock must be held across this call and while using the |
753 | * returned object. | |
00e2e675 | 754 | */ |
d88aee68 | 755 | struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key) |
00e2e675 DG |
756 | { |
757 | struct lttng_ht_iter iter; | |
d88aee68 | 758 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
759 | struct consumer_relayd_sock_pair *relayd = NULL; |
760 | ||
761 | /* Negative keys are lookup failures */ | |
d88aee68 | 762 | if (key == (uint64_t) -1ULL) { |
00e2e675 DG |
763 | goto error; |
764 | } | |
765 | ||
d88aee68 | 766 | lttng_ht_lookup(consumer_data.relayd_ht, &key, |
00e2e675 | 767 | &iter); |
d88aee68 | 768 | node = lttng_ht_iter_get_node_u64(&iter); |
00e2e675 DG |
769 | if (node != NULL) { |
770 | relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
771 | } | |
772 | ||
00e2e675 DG |
773 | error: |
774 | return relayd; | |
775 | } | |
776 | ||
777 | /* | |
778 | * Handle stream for relayd transmission if the stream applies for network | |
779 | * streaming where the net sequence index is set. | |
780 | * | |
781 | * Return destination file descriptor or negative value on error. | |
782 | */ | |
6197aea7 | 783 | static int write_relayd_stream_header(struct lttng_consumer_stream *stream, |
1d4dfdef DG |
784 | size_t data_size, unsigned long padding, |
785 | struct consumer_relayd_sock_pair *relayd) | |
00e2e675 DG |
786 | { |
787 | int outfd = -1, ret; | |
00e2e675 DG |
788 | struct lttcomm_relayd_data_hdr data_hdr; |
789 | ||
790 | /* Safety net */ | |
791 | assert(stream); | |
6197aea7 | 792 | assert(relayd); |
00e2e675 DG |
793 | |
794 | /* Reset data header */ | |
795 | memset(&data_hdr, 0, sizeof(data_hdr)); | |
796 | ||
00e2e675 DG |
797 | if (stream->metadata_flag) { |
798 | /* Caller MUST acquire the relayd control socket lock */ | |
799 | ret = relayd_send_metadata(&relayd->control_sock, data_size); | |
800 | if (ret < 0) { | |
801 | goto error; | |
802 | } | |
803 | ||
804 | /* Metadata are always sent on the control socket. */ | |
6151a90f | 805 | outfd = relayd->control_sock.sock.fd; |
00e2e675 DG |
806 | } else { |
807 | /* Set header with stream information */ | |
808 | data_hdr.stream_id = htobe64(stream->relayd_stream_id); | |
809 | data_hdr.data_size = htobe32(data_size); | |
1d4dfdef | 810 | data_hdr.padding_size = htobe32(padding); |
39df6d9f DG |
811 | /* |
812 | * Note that net_seq_num below is assigned with the *current* value of | |
813 | * next_net_seq_num and only after that the next_net_seq_num will be | |
814 | * increment. This is why when issuing a command on the relayd using | |
815 | * this next value, 1 should always be substracted in order to compare | |
816 | * the last seen sequence number on the relayd side to the last sent. | |
817 | */ | |
3604f373 | 818 | data_hdr.net_seq_num = htobe64(stream->next_net_seq_num); |
00e2e675 DG |
819 | /* Other fields are zeroed previously */ |
820 | ||
821 | ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, | |
822 | sizeof(data_hdr)); | |
823 | if (ret < 0) { | |
824 | goto error; | |
825 | } | |
826 | ||
3604f373 DG |
827 | ++stream->next_net_seq_num; |
828 | ||
00e2e675 | 829 | /* Set to go on data socket */ |
6151a90f | 830 | outfd = relayd->data_sock.sock.fd; |
00e2e675 DG |
831 | } |
832 | ||
833 | error: | |
834 | return outfd; | |
835 | } | |
836 | ||
3bd1e081 | 837 | /* |
ffe60014 DG |
838 | * Allocate and return a new lttng_consumer_channel object using the given key |
839 | * to initialize the hash table node. | |
840 | * | |
841 | * On error, return NULL. | |
3bd1e081 | 842 | */ |
886224ff | 843 | struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, |
ffe60014 DG |
844 | uint64_t session_id, |
845 | const char *pathname, | |
846 | const char *name, | |
847 | uid_t uid, | |
848 | gid_t gid, | |
849 | int relayd_id, | |
1624d5b7 JD |
850 | enum lttng_event_output output, |
851 | uint64_t tracefile_size, | |
852 | uint64_t tracefile_count) | |
3bd1e081 MD |
853 | { |
854 | struct lttng_consumer_channel *channel; | |
3bd1e081 | 855 | |
276b26d1 | 856 | channel = zmalloc(sizeof(*channel)); |
3bd1e081 | 857 | if (channel == NULL) { |
7a57cf92 | 858 | PERROR("malloc struct lttng_consumer_channel"); |
3bd1e081 MD |
859 | goto end; |
860 | } | |
ffe60014 DG |
861 | |
862 | channel->key = key; | |
3bd1e081 | 863 | channel->refcount = 0; |
ffe60014 DG |
864 | channel->session_id = session_id; |
865 | channel->uid = uid; | |
866 | channel->gid = gid; | |
867 | channel->relayd_id = relayd_id; | |
868 | channel->output = output; | |
1624d5b7 JD |
869 | channel->tracefile_size = tracefile_size; |
870 | channel->tracefile_count = tracefile_count; | |
ffe60014 DG |
871 | |
872 | strncpy(channel->pathname, pathname, sizeof(channel->pathname)); | |
873 | channel->pathname[sizeof(channel->pathname) - 1] = '\0'; | |
874 | ||
875 | strncpy(channel->name, name, sizeof(channel->name)); | |
876 | channel->name[sizeof(channel->name) - 1] = '\0'; | |
877 | ||
d88aee68 | 878 | lttng_ht_node_init_u64(&channel->node, channel->key); |
d8ef542d MD |
879 | |
880 | channel->wait_fd = -1; | |
881 | ||
ffe60014 DG |
882 | CDS_INIT_LIST_HEAD(&channel->streams.head); |
883 | ||
d88aee68 | 884 | DBG("Allocated channel (key %" PRIu64 ")", channel->key) |
3bd1e081 | 885 | |
3bd1e081 MD |
886 | end: |
887 | return channel; | |
888 | } | |
889 | ||
890 | /* | |
891 | * Add a channel to the global list protected by a mutex. | |
892 | */ | |
d8ef542d MD |
893 | int consumer_add_channel(struct lttng_consumer_channel *channel, |
894 | struct lttng_consumer_local_data *ctx) | |
3bd1e081 | 895 | { |
ffe60014 | 896 | int ret = 0; |
d88aee68 | 897 | struct lttng_ht_node_u64 *node; |
c77fc10a DG |
898 | struct lttng_ht_iter iter; |
899 | ||
3bd1e081 | 900 | pthread_mutex_lock(&consumer_data.lock); |
6065ceec | 901 | rcu_read_lock(); |
c77fc10a | 902 | |
7972aab2 | 903 | lttng_ht_lookup(consumer_data.channel_ht, &channel->key, &iter); |
d88aee68 | 904 | node = lttng_ht_iter_get_node_u64(&iter); |
c77fc10a DG |
905 | if (node != NULL) { |
906 | /* Channel already exist. Ignore the insertion */ | |
d88aee68 DG |
907 | ERR("Consumer add channel key %" PRIu64 " already exists!", |
908 | channel->key); | |
e43c41c5 | 909 | ret = LTTNG_ERR_KERN_CHAN_EXIST; |
c77fc10a DG |
910 | goto end; |
911 | } | |
912 | ||
d88aee68 | 913 | lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node); |
c77fc10a DG |
914 | |
915 | end: | |
6065ceec | 916 | rcu_read_unlock(); |
3bd1e081 | 917 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 918 | |
d8ef542d MD |
919 | if (!ret && channel->wait_fd != -1 && |
920 | channel->metadata_stream == NULL) { | |
a0cbdd2e | 921 | notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD); |
d8ef542d | 922 | } |
ffe60014 | 923 | return ret; |
3bd1e081 MD |
924 | } |
925 | ||
926 | /* | |
927 | * Allocate the pollfd structure and the local view of the out fds to avoid | |
928 | * doing a lookup in the linked list and concurrency issues when writing is | |
929 | * needed. Called with consumer_data.lock held. | |
930 | * | |
931 | * Returns the number of fds in the structures. | |
932 | */ | |
ffe60014 DG |
933 | static int update_poll_array(struct lttng_consumer_local_data *ctx, |
934 | struct pollfd **pollfd, struct lttng_consumer_stream **local_stream, | |
935 | struct lttng_ht *ht) | |
3bd1e081 | 936 | { |
3bd1e081 | 937 | int i = 0; |
e4421fec DG |
938 | struct lttng_ht_iter iter; |
939 | struct lttng_consumer_stream *stream; | |
3bd1e081 | 940 | |
ffe60014 DG |
941 | assert(ctx); |
942 | assert(ht); | |
943 | assert(pollfd); | |
944 | assert(local_stream); | |
945 | ||
3bd1e081 | 946 | DBG("Updating poll fd array"); |
481d6c57 | 947 | rcu_read_lock(); |
43c34bc3 | 948 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
8994307f DG |
949 | /* |
950 | * Only active streams with an active end point can be added to the | |
951 | * poll set and local stream storage of the thread. | |
952 | * | |
953 | * There is a potential race here for endpoint_status to be updated | |
954 | * just after the check. However, this is OK since the stream(s) will | |
955 | * be deleted once the thread is notified that the end point state has | |
956 | * changed where this function will be called back again. | |
957 | */ | |
958 | if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM || | |
79d4ffb7 | 959 | stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { |
3bd1e081 MD |
960 | continue; |
961 | } | |
7972aab2 DG |
962 | /* |
963 | * This clobbers way too much the debug output. Uncomment that if you | |
964 | * need it for debugging purposes. | |
965 | * | |
966 | * DBG("Active FD %d", stream->wait_fd); | |
967 | */ | |
e4421fec | 968 | (*pollfd)[i].fd = stream->wait_fd; |
3bd1e081 | 969 | (*pollfd)[i].events = POLLIN | POLLPRI; |
e4421fec | 970 | local_stream[i] = stream; |
3bd1e081 MD |
971 | i++; |
972 | } | |
481d6c57 | 973 | rcu_read_unlock(); |
3bd1e081 MD |
974 | |
975 | /* | |
50f8ae69 | 976 | * Insert the consumer_data_pipe at the end of the array and don't |
3bd1e081 MD |
977 | * increment i so nb_fd is the number of real FD. |
978 | */ | |
acdb9057 | 979 | (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe); |
509bb1cf | 980 | (*pollfd)[i].events = POLLIN | POLLPRI; |
3bd1e081 MD |
981 | return i; |
982 | } | |
983 | ||
984 | /* | |
985 | * Poll on the should_quit pipe and the command socket return -1 on error and | |
986 | * should exit, 0 if data is available on the command socket | |
987 | */ | |
988 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) | |
989 | { | |
990 | int num_rdy; | |
991 | ||
88f2b785 | 992 | restart: |
3bd1e081 MD |
993 | num_rdy = poll(consumer_sockpoll, 2, -1); |
994 | if (num_rdy == -1) { | |
88f2b785 MD |
995 | /* |
996 | * Restart interrupted system call. | |
997 | */ | |
998 | if (errno == EINTR) { | |
999 | goto restart; | |
1000 | } | |
7a57cf92 | 1001 | PERROR("Poll error"); |
3bd1e081 MD |
1002 | goto exit; |
1003 | } | |
509bb1cf | 1004 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
3bd1e081 MD |
1005 | DBG("consumer_should_quit wake up"); |
1006 | goto exit; | |
1007 | } | |
1008 | return 0; | |
1009 | ||
1010 | exit: | |
1011 | return -1; | |
1012 | } | |
1013 | ||
1014 | /* | |
1015 | * Set the error socket. | |
1016 | */ | |
ffe60014 DG |
1017 | void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx, |
1018 | int sock) | |
3bd1e081 MD |
1019 | { |
1020 | ctx->consumer_error_socket = sock; | |
1021 | } | |
1022 | ||
1023 | /* | |
1024 | * Set the command socket path. | |
1025 | */ | |
3bd1e081 MD |
1026 | void lttng_consumer_set_command_sock_path( |
1027 | struct lttng_consumer_local_data *ctx, char *sock) | |
1028 | { | |
1029 | ctx->consumer_command_sock_path = sock; | |
1030 | } | |
1031 | ||
1032 | /* | |
1033 | * Send return code to the session daemon. | |
1034 | * If the socket is not defined, we return 0, it is not a fatal error | |
1035 | */ | |
ffe60014 | 1036 | int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd) |
3bd1e081 MD |
1037 | { |
1038 | if (ctx->consumer_error_socket > 0) { | |
1039 | return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd, | |
1040 | sizeof(enum lttcomm_sessiond_command)); | |
1041 | } | |
1042 | ||
1043 | return 0; | |
1044 | } | |
1045 | ||
1046 | /* | |
228b5bf7 DG |
1047 | * Close all the tracefiles and stream fds and MUST be called when all |
1048 | * instances are destroyed i.e. when all threads were joined and are ended. | |
3bd1e081 MD |
1049 | */ |
1050 | void lttng_consumer_cleanup(void) | |
1051 | { | |
e4421fec | 1052 | struct lttng_ht_iter iter; |
ffe60014 | 1053 | struct lttng_consumer_channel *channel; |
6065ceec DG |
1054 | |
1055 | rcu_read_lock(); | |
3bd1e081 | 1056 | |
ffe60014 DG |
1057 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel, |
1058 | node.node) { | |
702b1ea4 | 1059 | consumer_del_channel(channel); |
3bd1e081 | 1060 | } |
6065ceec DG |
1061 | |
1062 | rcu_read_unlock(); | |
d6ce1df2 | 1063 | |
d6ce1df2 | 1064 | lttng_ht_destroy(consumer_data.channel_ht); |
228b5bf7 DG |
1065 | |
1066 | cleanup_relayd_ht(); | |
1067 | ||
d8ef542d MD |
1068 | lttng_ht_destroy(consumer_data.stream_per_chan_id_ht); |
1069 | ||
228b5bf7 DG |
1070 | /* |
1071 | * This HT contains streams that are freed by either the metadata thread or | |
1072 | * the data thread so we do *nothing* on the hash table and simply destroy | |
1073 | * it. | |
1074 | */ | |
1075 | lttng_ht_destroy(consumer_data.stream_list_ht); | |
3bd1e081 MD |
1076 | } |
1077 | ||
1078 | /* | |
1079 | * Called from signal handler. | |
1080 | */ | |
1081 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) | |
1082 | { | |
1083 | int ret; | |
1084 | consumer_quit = 1; | |
6f94560a MD |
1085 | do { |
1086 | ret = write(ctx->consumer_should_quit[1], "4", 1); | |
1087 | } while (ret < 0 && errno == EINTR); | |
4cec016f | 1088 | if (ret < 0 || ret != 1) { |
7a57cf92 | 1089 | PERROR("write consumer quit"); |
3bd1e081 | 1090 | } |
ab1027f4 DG |
1091 | |
1092 | DBG("Consumer flag that it should quit"); | |
3bd1e081 MD |
1093 | } |
1094 | ||
00e2e675 DG |
1095 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
1096 | off_t orig_offset) | |
3bd1e081 MD |
1097 | { |
1098 | int outfd = stream->out_fd; | |
1099 | ||
1100 | /* | |
1101 | * This does a blocking write-and-wait on any page that belongs to the | |
1102 | * subbuffer prior to the one we just wrote. | |
1103 | * Don't care about error values, as these are just hints and ways to | |
1104 | * limit the amount of page cache used. | |
1105 | */ | |
ffe60014 | 1106 | if (orig_offset < stream->max_sb_size) { |
3bd1e081 MD |
1107 | return; |
1108 | } | |
ffe60014 DG |
1109 | lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size, |
1110 | stream->max_sb_size, | |
3bd1e081 MD |
1111 | SYNC_FILE_RANGE_WAIT_BEFORE |
1112 | | SYNC_FILE_RANGE_WRITE | |
1113 | | SYNC_FILE_RANGE_WAIT_AFTER); | |
1114 | /* | |
1115 | * Give hints to the kernel about how we access the file: | |
1116 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after | |
1117 | * we write it. | |
1118 | * | |
1119 | * We need to call fadvise again after the file grows because the | |
1120 | * kernel does not seem to apply fadvise to non-existing parts of the | |
1121 | * file. | |
1122 | * | |
1123 | * Call fadvise _after_ having waited for the page writeback to | |
1124 | * complete because the dirty page writeback semantic is not well | |
1125 | * defined. So it can be expected to lead to lower throughput in | |
1126 | * streaming. | |
1127 | */ | |
ffe60014 DG |
1128 | posix_fadvise(outfd, orig_offset - stream->max_sb_size, |
1129 | stream->max_sb_size, POSIX_FADV_DONTNEED); | |
3bd1e081 MD |
1130 | } |
1131 | ||
1132 | /* | |
1133 | * Initialise the necessary environnement : | |
1134 | * - create a new context | |
1135 | * - create the poll_pipe | |
1136 | * - create the should_quit pipe (for signal handler) | |
1137 | * - create the thread pipe (for splice) | |
1138 | * | |
1139 | * Takes a function pointer as argument, this function is called when data is | |
1140 | * available on a buffer. This function is responsible to do the | |
1141 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the | |
1142 | * buffer configuration and then kernctl_put_next_subbuf at the end. | |
1143 | * | |
1144 | * Returns a pointer to the new context or NULL on error. | |
1145 | */ | |
1146 | struct lttng_consumer_local_data *lttng_consumer_create( | |
1147 | enum lttng_consumer_type type, | |
4078b776 | 1148 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
d41f73b7 | 1149 | struct lttng_consumer_local_data *ctx), |
3bd1e081 MD |
1150 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
1151 | int (*recv_stream)(struct lttng_consumer_stream *stream), | |
1152 | int (*update_stream)(int stream_key, uint32_t state)) | |
1153 | { | |
d8ef542d | 1154 | int ret; |
3bd1e081 MD |
1155 | struct lttng_consumer_local_data *ctx; |
1156 | ||
1157 | assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || | |
1158 | consumer_data.type == type); | |
1159 | consumer_data.type = type; | |
1160 | ||
effcf122 | 1161 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
3bd1e081 | 1162 | if (ctx == NULL) { |
7a57cf92 | 1163 | PERROR("allocating context"); |
3bd1e081 MD |
1164 | goto error; |
1165 | } | |
1166 | ||
1167 | ctx->consumer_error_socket = -1; | |
331744e3 | 1168 | ctx->consumer_metadata_socket = -1; |
3bd1e081 MD |
1169 | /* assign the callbacks */ |
1170 | ctx->on_buffer_ready = buffer_ready; | |
1171 | ctx->on_recv_channel = recv_channel; | |
1172 | ctx->on_recv_stream = recv_stream; | |
1173 | ctx->on_update_stream = update_stream; | |
1174 | ||
acdb9057 DG |
1175 | ctx->consumer_data_pipe = lttng_pipe_open(0); |
1176 | if (!ctx->consumer_data_pipe) { | |
3bd1e081 MD |
1177 | goto error_poll_pipe; |
1178 | } | |
1179 | ||
1180 | ret = pipe(ctx->consumer_should_quit); | |
1181 | if (ret < 0) { | |
7a57cf92 | 1182 | PERROR("Error creating recv pipe"); |
3bd1e081 MD |
1183 | goto error_quit_pipe; |
1184 | } | |
1185 | ||
1186 | ret = pipe(ctx->consumer_thread_pipe); | |
1187 | if (ret < 0) { | |
7a57cf92 | 1188 | PERROR("Error creating thread pipe"); |
3bd1e081 MD |
1189 | goto error_thread_pipe; |
1190 | } | |
1191 | ||
d8ef542d MD |
1192 | ret = pipe(ctx->consumer_channel_pipe); |
1193 | if (ret < 0) { | |
1194 | PERROR("Error creating channel pipe"); | |
1195 | goto error_channel_pipe; | |
1196 | } | |
1197 | ||
13886d2d DG |
1198 | ctx->consumer_metadata_pipe = lttng_pipe_open(0); |
1199 | if (!ctx->consumer_metadata_pipe) { | |
fb3a43a9 DG |
1200 | goto error_metadata_pipe; |
1201 | } | |
3bd1e081 | 1202 | |
fb3a43a9 DG |
1203 | ret = utils_create_pipe(ctx->consumer_splice_metadata_pipe); |
1204 | if (ret < 0) { | |
1205 | goto error_splice_pipe; | |
1206 | } | |
1207 | ||
1208 | return ctx; | |
3bd1e081 | 1209 | |
fb3a43a9 | 1210 | error_splice_pipe: |
13886d2d | 1211 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
fb3a43a9 | 1212 | error_metadata_pipe: |
d8ef542d MD |
1213 | utils_close_pipe(ctx->consumer_channel_pipe); |
1214 | error_channel_pipe: | |
fb3a43a9 | 1215 | utils_close_pipe(ctx->consumer_thread_pipe); |
3bd1e081 | 1216 | error_thread_pipe: |
d8ef542d | 1217 | utils_close_pipe(ctx->consumer_should_quit); |
3bd1e081 | 1218 | error_quit_pipe: |
acdb9057 | 1219 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
3bd1e081 MD |
1220 | error_poll_pipe: |
1221 | free(ctx); | |
1222 | error: | |
1223 | return NULL; | |
1224 | } | |
1225 | ||
1226 | /* | |
1227 | * Close all fds associated with the instance and free the context. | |
1228 | */ | |
1229 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) | |
1230 | { | |
4c462e79 MD |
1231 | int ret; |
1232 | ||
ab1027f4 DG |
1233 | DBG("Consumer destroying it. Closing everything."); |
1234 | ||
4c462e79 MD |
1235 | ret = close(ctx->consumer_error_socket); |
1236 | if (ret) { | |
1237 | PERROR("close"); | |
1238 | } | |
331744e3 JD |
1239 | ret = close(ctx->consumer_metadata_socket); |
1240 | if (ret) { | |
1241 | PERROR("close"); | |
1242 | } | |
d8ef542d MD |
1243 | utils_close_pipe(ctx->consumer_thread_pipe); |
1244 | utils_close_pipe(ctx->consumer_channel_pipe); | |
acdb9057 | 1245 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
13886d2d | 1246 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
d8ef542d | 1247 | utils_close_pipe(ctx->consumer_should_quit); |
fb3a43a9 DG |
1248 | utils_close_pipe(ctx->consumer_splice_metadata_pipe); |
1249 | ||
3bd1e081 MD |
1250 | unlink(ctx->consumer_command_sock_path); |
1251 | free(ctx); | |
1252 | } | |
1253 | ||
6197aea7 DG |
1254 | /* |
1255 | * Write the metadata stream id on the specified file descriptor. | |
1256 | */ | |
1257 | static int write_relayd_metadata_id(int fd, | |
1258 | struct lttng_consumer_stream *stream, | |
ffe60014 | 1259 | struct consumer_relayd_sock_pair *relayd, unsigned long padding) |
6197aea7 DG |
1260 | { |
1261 | int ret; | |
1d4dfdef | 1262 | struct lttcomm_relayd_metadata_payload hdr; |
6197aea7 | 1263 | |
1d4dfdef DG |
1264 | hdr.stream_id = htobe64(stream->relayd_stream_id); |
1265 | hdr.padding_size = htobe32(padding); | |
6197aea7 | 1266 | do { |
1d4dfdef | 1267 | ret = write(fd, (void *) &hdr, sizeof(hdr)); |
6197aea7 | 1268 | } while (ret < 0 && errno == EINTR); |
4cec016f | 1269 | if (ret < 0 || ret != sizeof(hdr)) { |
d7b75ec8 DG |
1270 | /* |
1271 | * This error means that the fd's end is closed so ignore the perror | |
1272 | * not to clubber the error output since this can happen in a normal | |
1273 | * code path. | |
1274 | */ | |
1275 | if (errno != EPIPE) { | |
1276 | PERROR("write metadata stream id"); | |
1277 | } | |
1278 | DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno); | |
534d2592 DG |
1279 | /* |
1280 | * Set ret to a negative value because if ret != sizeof(hdr), we don't | |
1281 | * handle writting the missing part so report that as an error and | |
1282 | * don't lie to the caller. | |
1283 | */ | |
1284 | ret = -1; | |
6197aea7 DG |
1285 | goto end; |
1286 | } | |
1d4dfdef DG |
1287 | DBG("Metadata stream id %" PRIu64 " with padding %lu written before data", |
1288 | stream->relayd_stream_id, padding); | |
6197aea7 DG |
1289 | |
1290 | end: | |
1291 | return ret; | |
1292 | } | |
1293 | ||
3bd1e081 | 1294 | /* |
09e26845 DG |
1295 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
1296 | * core function for writing trace buffers to either the local filesystem or | |
1297 | * the network. | |
1298 | * | |
79d4ffb7 DG |
1299 | * It must be called with the stream lock held. |
1300 | * | |
09e26845 | 1301 | * Careful review MUST be put if any changes occur! |
3bd1e081 MD |
1302 | * |
1303 | * Returns the number of bytes written | |
1304 | */ | |
4078b776 | 1305 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
3bd1e081 | 1306 | struct lttng_consumer_local_data *ctx, |
1d4dfdef DG |
1307 | struct lttng_consumer_stream *stream, unsigned long len, |
1308 | unsigned long padding) | |
3bd1e081 | 1309 | { |
f02e1e8a | 1310 | unsigned long mmap_offset; |
ffe60014 | 1311 | void *mmap_base; |
f02e1e8a DG |
1312 | ssize_t ret = 0, written = 0; |
1313 | off_t orig_offset = stream->out_fd_offset; | |
1314 | /* Default is on the disk */ | |
1315 | int outfd = stream->out_fd; | |
f02e1e8a | 1316 | struct consumer_relayd_sock_pair *relayd = NULL; |
8994307f | 1317 | unsigned int relayd_hang_up = 0; |
f02e1e8a DG |
1318 | |
1319 | /* RCU lock for the relayd pointer */ | |
1320 | rcu_read_lock(); | |
1321 | ||
1322 | /* Flag that the current stream if set for network streaming. */ | |
1323 | if (stream->net_seq_idx != -1) { | |
1324 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
1325 | if (relayd == NULL) { | |
1326 | goto end; | |
1327 | } | |
1328 | } | |
1329 | ||
1330 | /* get the offset inside the fd to mmap */ | |
3bd1e081 MD |
1331 | switch (consumer_data.type) { |
1332 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1333 | mmap_base = stream->mmap_base; |
f02e1e8a DG |
1334 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); |
1335 | break; | |
7753dea8 MD |
1336 | case LTTNG_CONSUMER32_UST: |
1337 | case LTTNG_CONSUMER64_UST: | |
ffe60014 DG |
1338 | mmap_base = lttng_ustctl_get_mmap_base(stream); |
1339 | if (!mmap_base) { | |
1340 | ERR("read mmap get mmap base for stream %s", stream->name); | |
1341 | written = -1; | |
1342 | goto end; | |
1343 | } | |
1344 | ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset); | |
331744e3 | 1345 | |
f02e1e8a | 1346 | break; |
3bd1e081 MD |
1347 | default: |
1348 | ERR("Unknown consumer_data type"); | |
1349 | assert(0); | |
1350 | } | |
f02e1e8a DG |
1351 | if (ret != 0) { |
1352 | errno = -ret; | |
1353 | PERROR("tracer ctl get_mmap_read_offset"); | |
1354 | written = ret; | |
1355 | goto end; | |
1356 | } | |
b9182dd9 | 1357 | |
f02e1e8a DG |
1358 | /* Handle stream on the relayd if the output is on the network */ |
1359 | if (relayd) { | |
1360 | unsigned long netlen = len; | |
1361 | ||
1362 | /* | |
1363 | * Lock the control socket for the complete duration of the function | |
1364 | * since from this point on we will use the socket. | |
1365 | */ | |
1366 | if (stream->metadata_flag) { | |
1367 | /* Metadata requires the control socket. */ | |
1368 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1d4dfdef | 1369 | netlen += sizeof(struct lttcomm_relayd_metadata_payload); |
f02e1e8a DG |
1370 | } |
1371 | ||
1d4dfdef | 1372 | ret = write_relayd_stream_header(stream, netlen, padding, relayd); |
f02e1e8a DG |
1373 | if (ret >= 0) { |
1374 | /* Use the returned socket. */ | |
1375 | outfd = ret; | |
1376 | ||
1377 | /* Write metadata stream id before payload */ | |
1378 | if (stream->metadata_flag) { | |
1d4dfdef | 1379 | ret = write_relayd_metadata_id(outfd, stream, relayd, padding); |
f02e1e8a | 1380 | if (ret < 0) { |
f02e1e8a | 1381 | written = ret; |
8994307f DG |
1382 | /* Socket operation failed. We consider the relayd dead */ |
1383 | if (ret == -EPIPE || ret == -EINVAL) { | |
1384 | relayd_hang_up = 1; | |
1385 | goto write_error; | |
1386 | } | |
f02e1e8a DG |
1387 | goto end; |
1388 | } | |
f02e1e8a | 1389 | } |
8994307f DG |
1390 | } else { |
1391 | /* Socket operation failed. We consider the relayd dead */ | |
1392 | if (ret == -EPIPE || ret == -EINVAL) { | |
1393 | relayd_hang_up = 1; | |
1394 | goto write_error; | |
1395 | } | |
1396 | /* Else, use the default set before which is the filesystem. */ | |
f02e1e8a | 1397 | } |
1d4dfdef DG |
1398 | } else { |
1399 | /* No streaming, we have to set the len with the full padding */ | |
1400 | len += padding; | |
1624d5b7 JD |
1401 | |
1402 | /* | |
1403 | * Check if we need to change the tracefile before writing the packet. | |
1404 | */ | |
1405 | if (stream->chan->tracefile_size > 0 && | |
1406 | (stream->tracefile_size_current + len) > | |
1407 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1408 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1409 | stream->name, stream->chan->tracefile_size, | |
1410 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
1411 | stream->out_fd, &(stream->tracefile_count_current)); | |
1624d5b7 JD |
1412 | if (ret < 0) { |
1413 | ERR("Rotating output file"); | |
1414 | goto end; | |
1415 | } | |
fe4477ee | 1416 | outfd = stream->out_fd = ret; |
a6976990 DG |
1417 | /* Reset current size because we just perform a rotation. */ |
1418 | stream->tracefile_size_current = 0; | |
1624d5b7 JD |
1419 | } |
1420 | stream->tracefile_size_current += len; | |
f02e1e8a DG |
1421 | } |
1422 | ||
1423 | while (len > 0) { | |
1424 | do { | |
ffe60014 | 1425 | ret = write(outfd, mmap_base + mmap_offset, len); |
f02e1e8a | 1426 | } while (ret < 0 && errno == EINTR); |
1d4dfdef | 1427 | DBG("Consumer mmap write() ret %zd (len %lu)", ret, len); |
f02e1e8a | 1428 | if (ret < 0) { |
c5c45efa DG |
1429 | /* |
1430 | * This is possible if the fd is closed on the other side (outfd) | |
1431 | * or any write problem. It can be verbose a bit for a normal | |
1432 | * execution if for instance the relayd is stopped abruptly. This | |
1433 | * can happen so set this to a DBG statement. | |
1434 | */ | |
1435 | DBG("Error in file write mmap"); | |
f02e1e8a DG |
1436 | if (written == 0) { |
1437 | written = ret; | |
1438 | } | |
8994307f DG |
1439 | /* Socket operation failed. We consider the relayd dead */ |
1440 | if (errno == EPIPE || errno == EINVAL) { | |
1441 | relayd_hang_up = 1; | |
1442 | goto write_error; | |
1443 | } | |
f02e1e8a DG |
1444 | goto end; |
1445 | } else if (ret > len) { | |
77c7c900 | 1446 | PERROR("Error in file write (ret %zd > len %lu)", ret, len); |
f02e1e8a DG |
1447 | written += ret; |
1448 | goto end; | |
1449 | } else { | |
1450 | len -= ret; | |
1451 | mmap_offset += ret; | |
1452 | } | |
f02e1e8a DG |
1453 | |
1454 | /* This call is useless on a socket so better save a syscall. */ | |
1455 | if (!relayd) { | |
1456 | /* This won't block, but will start writeout asynchronously */ | |
1457 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret, | |
1458 | SYNC_FILE_RANGE_WRITE); | |
1459 | stream->out_fd_offset += ret; | |
1460 | } | |
1461 | written += ret; | |
1462 | } | |
1463 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1464 | ||
8994307f DG |
1465 | write_error: |
1466 | /* | |
1467 | * This is a special case that the relayd has closed its socket. Let's | |
1468 | * cleanup the relayd object and all associated streams. | |
1469 | */ | |
1470 | if (relayd && relayd_hang_up) { | |
1471 | cleanup_relayd(relayd, ctx); | |
1472 | } | |
1473 | ||
f02e1e8a DG |
1474 | end: |
1475 | /* Unlock only if ctrl socket used */ | |
1476 | if (relayd && stream->metadata_flag) { | |
1477 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1478 | } | |
1479 | ||
1480 | rcu_read_unlock(); | |
1481 | return written; | |
3bd1e081 MD |
1482 | } |
1483 | ||
1484 | /* | |
1485 | * Splice the data from the ring buffer to the tracefile. | |
1486 | * | |
79d4ffb7 DG |
1487 | * It must be called with the stream lock held. |
1488 | * | |
3bd1e081 MD |
1489 | * Returns the number of bytes spliced. |
1490 | */ | |
4078b776 | 1491 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
3bd1e081 | 1492 | struct lttng_consumer_local_data *ctx, |
1d4dfdef DG |
1493 | struct lttng_consumer_stream *stream, unsigned long len, |
1494 | unsigned long padding) | |
3bd1e081 | 1495 | { |
f02e1e8a DG |
1496 | ssize_t ret = 0, written = 0, ret_splice = 0; |
1497 | loff_t offset = 0; | |
1498 | off_t orig_offset = stream->out_fd_offset; | |
1499 | int fd = stream->wait_fd; | |
1500 | /* Default is on the disk */ | |
1501 | int outfd = stream->out_fd; | |
f02e1e8a | 1502 | struct consumer_relayd_sock_pair *relayd = NULL; |
fb3a43a9 | 1503 | int *splice_pipe; |
8994307f | 1504 | unsigned int relayd_hang_up = 0; |
f02e1e8a | 1505 | |
3bd1e081 MD |
1506 | switch (consumer_data.type) { |
1507 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a | 1508 | break; |
7753dea8 MD |
1509 | case LTTNG_CONSUMER32_UST: |
1510 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a | 1511 | /* Not supported for user space tracing */ |
3bd1e081 MD |
1512 | return -ENOSYS; |
1513 | default: | |
1514 | ERR("Unknown consumer_data type"); | |
1515 | assert(0); | |
3bd1e081 MD |
1516 | } |
1517 | ||
f02e1e8a DG |
1518 | /* RCU lock for the relayd pointer */ |
1519 | rcu_read_lock(); | |
1520 | ||
1521 | /* Flag that the current stream if set for network streaming. */ | |
1522 | if (stream->net_seq_idx != -1) { | |
1523 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
1524 | if (relayd == NULL) { | |
1525 | goto end; | |
1526 | } | |
1527 | } | |
1528 | ||
fb3a43a9 DG |
1529 | /* |
1530 | * Choose right pipe for splice. Metadata and trace data are handled by | |
1531 | * different threads hence the use of two pipes in order not to race or | |
1532 | * corrupt the written data. | |
1533 | */ | |
1534 | if (stream->metadata_flag) { | |
1535 | splice_pipe = ctx->consumer_splice_metadata_pipe; | |
1536 | } else { | |
1537 | splice_pipe = ctx->consumer_thread_pipe; | |
1538 | } | |
1539 | ||
f02e1e8a | 1540 | /* Write metadata stream id before payload */ |
1d4dfdef DG |
1541 | if (relayd) { |
1542 | int total_len = len; | |
f02e1e8a | 1543 | |
1d4dfdef DG |
1544 | if (stream->metadata_flag) { |
1545 | /* | |
1546 | * Lock the control socket for the complete duration of the function | |
1547 | * since from this point on we will use the socket. | |
1548 | */ | |
1549 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1550 | ||
1551 | ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd, | |
1552 | padding); | |
1553 | if (ret < 0) { | |
1554 | written = ret; | |
8994307f DG |
1555 | /* Socket operation failed. We consider the relayd dead */ |
1556 | if (ret == -EBADF) { | |
1557 | WARN("Remote relayd disconnected. Stopping"); | |
1558 | relayd_hang_up = 1; | |
1559 | goto write_error; | |
1560 | } | |
1d4dfdef DG |
1561 | goto end; |
1562 | } | |
1563 | ||
1564 | total_len += sizeof(struct lttcomm_relayd_metadata_payload); | |
1565 | } | |
1566 | ||
1567 | ret = write_relayd_stream_header(stream, total_len, padding, relayd); | |
1568 | if (ret >= 0) { | |
1569 | /* Use the returned socket. */ | |
1570 | outfd = ret; | |
1571 | } else { | |
8994307f DG |
1572 | /* Socket operation failed. We consider the relayd dead */ |
1573 | if (ret == -EBADF) { | |
1574 | WARN("Remote relayd disconnected. Stopping"); | |
1575 | relayd_hang_up = 1; | |
1576 | goto write_error; | |
1577 | } | |
f02e1e8a DG |
1578 | goto end; |
1579 | } | |
1d4dfdef DG |
1580 | } else { |
1581 | /* No streaming, we have to set the len with the full padding */ | |
1582 | len += padding; | |
1624d5b7 JD |
1583 | |
1584 | /* | |
1585 | * Check if we need to change the tracefile before writing the packet. | |
1586 | */ | |
1587 | if (stream->chan->tracefile_size > 0 && | |
1588 | (stream->tracefile_size_current + len) > | |
1589 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1590 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1591 | stream->name, stream->chan->tracefile_size, | |
1592 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
1593 | stream->out_fd, &(stream->tracefile_count_current)); | |
1624d5b7 JD |
1594 | if (ret < 0) { |
1595 | ERR("Rotating output file"); | |
1596 | goto end; | |
1597 | } | |
fe4477ee | 1598 | outfd = stream->out_fd = ret; |
a6976990 DG |
1599 | /* Reset current size because we just perform a rotation. */ |
1600 | stream->tracefile_size_current = 0; | |
1624d5b7 JD |
1601 | } |
1602 | stream->tracefile_size_current += len; | |
f02e1e8a DG |
1603 | } |
1604 | ||
1605 | while (len > 0) { | |
1d4dfdef DG |
1606 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)", |
1607 | (unsigned long)offset, len, fd, splice_pipe[1]); | |
fb3a43a9 | 1608 | ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len, |
f02e1e8a DG |
1609 | SPLICE_F_MOVE | SPLICE_F_MORE); |
1610 | DBG("splice chan to pipe, ret %zd", ret_splice); | |
1611 | if (ret_splice < 0) { | |
1612 | PERROR("Error in relay splice"); | |
1613 | if (written == 0) { | |
1614 | written = ret_splice; | |
1615 | } | |
1616 | ret = errno; | |
1617 | goto splice_error; | |
1618 | } | |
1619 | ||
1620 | /* Handle stream on the relayd if the output is on the network */ | |
1621 | if (relayd) { | |
1622 | if (stream->metadata_flag) { | |
1d4dfdef DG |
1623 | size_t metadata_payload_size = |
1624 | sizeof(struct lttcomm_relayd_metadata_payload); | |
1625 | ||
f02e1e8a | 1626 | /* Update counter to fit the spliced data */ |
1d4dfdef DG |
1627 | ret_splice += metadata_payload_size; |
1628 | len += metadata_payload_size; | |
f02e1e8a DG |
1629 | /* |
1630 | * We do this so the return value can match the len passed as | |
1631 | * argument to this function. | |
1632 | */ | |
1d4dfdef | 1633 | written -= metadata_payload_size; |
f02e1e8a DG |
1634 | } |
1635 | } | |
1636 | ||
1637 | /* Splice data out */ | |
fb3a43a9 | 1638 | ret_splice = splice(splice_pipe[0], NULL, outfd, NULL, |
f02e1e8a | 1639 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); |
1d4dfdef | 1640 | DBG("Consumer splice pipe to file, ret %zd", ret_splice); |
f02e1e8a DG |
1641 | if (ret_splice < 0) { |
1642 | PERROR("Error in file splice"); | |
1643 | if (written == 0) { | |
1644 | written = ret_splice; | |
1645 | } | |
8994307f | 1646 | /* Socket operation failed. We consider the relayd dead */ |
00c8752b | 1647 | if (errno == EBADF || errno == EPIPE) { |
8994307f DG |
1648 | WARN("Remote relayd disconnected. Stopping"); |
1649 | relayd_hang_up = 1; | |
1650 | goto write_error; | |
1651 | } | |
f02e1e8a DG |
1652 | ret = errno; |
1653 | goto splice_error; | |
1654 | } else if (ret_splice > len) { | |
1655 | errno = EINVAL; | |
1656 | PERROR("Wrote more data than requested %zd (len: %lu)", | |
1657 | ret_splice, len); | |
1658 | written += ret_splice; | |
1659 | ret = errno; | |
1660 | goto splice_error; | |
1661 | } | |
1662 | len -= ret_splice; | |
1663 | ||
1664 | /* This call is useless on a socket so better save a syscall. */ | |
1665 | if (!relayd) { | |
1666 | /* This won't block, but will start writeout asynchronously */ | |
1667 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, | |
1668 | SYNC_FILE_RANGE_WRITE); | |
1669 | stream->out_fd_offset += ret_splice; | |
1670 | } | |
1671 | written += ret_splice; | |
1672 | } | |
1673 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1674 | ||
1675 | ret = ret_splice; | |
1676 | ||
1677 | goto end; | |
1678 | ||
8994307f DG |
1679 | write_error: |
1680 | /* | |
1681 | * This is a special case that the relayd has closed its socket. Let's | |
1682 | * cleanup the relayd object and all associated streams. | |
1683 | */ | |
1684 | if (relayd && relayd_hang_up) { | |
1685 | cleanup_relayd(relayd, ctx); | |
1686 | /* Skip splice error so the consumer does not fail */ | |
1687 | goto end; | |
1688 | } | |
1689 | ||
f02e1e8a DG |
1690 | splice_error: |
1691 | /* send the appropriate error description to sessiond */ | |
1692 | switch (ret) { | |
f02e1e8a | 1693 | case EINVAL: |
f73fabfd | 1694 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); |
f02e1e8a DG |
1695 | break; |
1696 | case ENOMEM: | |
f73fabfd | 1697 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); |
f02e1e8a DG |
1698 | break; |
1699 | case ESPIPE: | |
f73fabfd | 1700 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); |
f02e1e8a DG |
1701 | break; |
1702 | } | |
1703 | ||
1704 | end: | |
1705 | if (relayd && stream->metadata_flag) { | |
1706 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1707 | } | |
1708 | ||
1709 | rcu_read_unlock(); | |
1710 | return written; | |
3bd1e081 MD |
1711 | } |
1712 | ||
1713 | /* | |
1714 | * Take a snapshot for a specific fd | |
1715 | * | |
1716 | * Returns 0 on success, < 0 on error | |
1717 | */ | |
ffe60014 | 1718 | int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) |
3bd1e081 MD |
1719 | { |
1720 | switch (consumer_data.type) { | |
1721 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1722 | return lttng_kconsumer_take_snapshot(stream); |
7753dea8 MD |
1723 | case LTTNG_CONSUMER32_UST: |
1724 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 1725 | return lttng_ustconsumer_take_snapshot(stream); |
3bd1e081 MD |
1726 | default: |
1727 | ERR("Unknown consumer_data type"); | |
1728 | assert(0); | |
1729 | return -ENOSYS; | |
1730 | } | |
3bd1e081 MD |
1731 | } |
1732 | ||
1733 | /* | |
1734 | * Get the produced position | |
1735 | * | |
1736 | * Returns 0 on success, < 0 on error | |
1737 | */ | |
ffe60014 | 1738 | int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, |
3bd1e081 MD |
1739 | unsigned long *pos) |
1740 | { | |
1741 | switch (consumer_data.type) { | |
1742 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1743 | return lttng_kconsumer_get_produced_snapshot(stream, pos); |
7753dea8 MD |
1744 | case LTTNG_CONSUMER32_UST: |
1745 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 1746 | return lttng_ustconsumer_get_produced_snapshot(stream, pos); |
3bd1e081 MD |
1747 | default: |
1748 | ERR("Unknown consumer_data type"); | |
1749 | assert(0); | |
1750 | return -ENOSYS; | |
1751 | } | |
1752 | } | |
1753 | ||
1754 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, | |
1755 | int sock, struct pollfd *consumer_sockpoll) | |
1756 | { | |
1757 | switch (consumer_data.type) { | |
1758 | case LTTNG_CONSUMER_KERNEL: | |
1759 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
7753dea8 MD |
1760 | case LTTNG_CONSUMER32_UST: |
1761 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1762 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
1763 | default: | |
1764 | ERR("Unknown consumer_data type"); | |
1765 | assert(0); | |
1766 | return -ENOSYS; | |
1767 | } | |
1768 | } | |
1769 | ||
43c34bc3 DG |
1770 | /* |
1771 | * Iterate over all streams of the hashtable and free them properly. | |
1772 | * | |
1773 | * WARNING: *MUST* be used with data stream only. | |
1774 | */ | |
1775 | static void destroy_data_stream_ht(struct lttng_ht *ht) | |
1776 | { | |
43c34bc3 DG |
1777 | struct lttng_ht_iter iter; |
1778 | struct lttng_consumer_stream *stream; | |
1779 | ||
1780 | if (ht == NULL) { | |
1781 | return; | |
1782 | } | |
1783 | ||
1784 | rcu_read_lock(); | |
1785 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
5c540210 DG |
1786 | /* |
1787 | * Ignore return value since we are currently cleaning up so any error | |
1788 | * can't be handled. | |
1789 | */ | |
1790 | (void) consumer_del_stream(stream, ht); | |
43c34bc3 DG |
1791 | } |
1792 | rcu_read_unlock(); | |
1793 | ||
1794 | lttng_ht_destroy(ht); | |
1795 | } | |
1796 | ||
fb3a43a9 | 1797 | /* |
f724d81e | 1798 | * Iterate over all streams of the hashtable and free them properly. |
e316aad5 DG |
1799 | * |
1800 | * XXX: Should not be only for metadata stream or else use an other name. | |
fb3a43a9 DG |
1801 | */ |
1802 | static void destroy_stream_ht(struct lttng_ht *ht) | |
1803 | { | |
fb3a43a9 DG |
1804 | struct lttng_ht_iter iter; |
1805 | struct lttng_consumer_stream *stream; | |
1806 | ||
1807 | if (ht == NULL) { | |
1808 | return; | |
1809 | } | |
1810 | ||
d09e1200 | 1811 | rcu_read_lock(); |
58b1f425 | 1812 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
5c540210 DG |
1813 | /* |
1814 | * Ignore return value since we are currently cleaning up so any error | |
1815 | * can't be handled. | |
1816 | */ | |
1817 | (void) consumer_del_metadata_stream(stream, ht); | |
fb3a43a9 | 1818 | } |
d09e1200 | 1819 | rcu_read_unlock(); |
fb3a43a9 DG |
1820 | |
1821 | lttng_ht_destroy(ht); | |
1822 | } | |
1823 | ||
d88aee68 DG |
1824 | void lttng_consumer_close_metadata(void) |
1825 | { | |
1826 | switch (consumer_data.type) { | |
1827 | case LTTNG_CONSUMER_KERNEL: | |
1828 | /* | |
1829 | * The Kernel consumer has a different metadata scheme so we don't | |
1830 | * close anything because the stream will be closed by the session | |
1831 | * daemon. | |
1832 | */ | |
1833 | break; | |
1834 | case LTTNG_CONSUMER32_UST: | |
1835 | case LTTNG_CONSUMER64_UST: | |
1836 | /* | |
1837 | * Close all metadata streams. The metadata hash table is passed and | |
1838 | * this call iterates over it by closing all wakeup fd. This is safe | |
1839 | * because at this point we are sure that the metadata producer is | |
1840 | * either dead or blocked. | |
1841 | */ | |
1842 | lttng_ustconsumer_close_metadata(metadata_ht); | |
1843 | break; | |
1844 | default: | |
1845 | ERR("Unknown consumer_data type"); | |
1846 | assert(0); | |
1847 | } | |
1848 | } | |
1849 | ||
fb3a43a9 DG |
1850 | /* |
1851 | * Clean up a metadata stream and free its memory. | |
1852 | */ | |
e316aad5 DG |
1853 | void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, |
1854 | struct lttng_ht *ht) | |
fb3a43a9 DG |
1855 | { |
1856 | int ret; | |
e316aad5 DG |
1857 | struct lttng_ht_iter iter; |
1858 | struct lttng_consumer_channel *free_chan = NULL; | |
fb3a43a9 DG |
1859 | struct consumer_relayd_sock_pair *relayd; |
1860 | ||
1861 | assert(stream); | |
1862 | /* | |
1863 | * This call should NEVER receive regular stream. It must always be | |
1864 | * metadata stream and this is crucial for data structure synchronization. | |
1865 | */ | |
1866 | assert(stream->metadata_flag); | |
1867 | ||
e316aad5 DG |
1868 | DBG3("Consumer delete metadata stream %d", stream->wait_fd); |
1869 | ||
1870 | if (ht == NULL) { | |
1871 | /* Means the stream was allocated but not successfully added */ | |
ffe60014 | 1872 | goto free_stream_rcu; |
e316aad5 DG |
1873 | } |
1874 | ||
74251bb8 | 1875 | pthread_mutex_lock(&consumer_data.lock); |
8994307f DG |
1876 | pthread_mutex_lock(&stream->lock); |
1877 | ||
fb3a43a9 DG |
1878 | switch (consumer_data.type) { |
1879 | case LTTNG_CONSUMER_KERNEL: | |
1880 | if (stream->mmap_base != NULL) { | |
1881 | ret = munmap(stream->mmap_base, stream->mmap_len); | |
1882 | if (ret != 0) { | |
1883 | PERROR("munmap metadata stream"); | |
1884 | } | |
1885 | } | |
4c95e622 JD |
1886 | |
1887 | if (stream->wait_fd >= 0) { | |
1888 | ret = close(stream->wait_fd); | |
1889 | if (ret < 0) { | |
1890 | PERROR("close kernel metadata wait_fd"); | |
1891 | } | |
1892 | } | |
fb3a43a9 DG |
1893 | break; |
1894 | case LTTNG_CONSUMER32_UST: | |
1895 | case LTTNG_CONSUMER64_UST: | |
1896 | lttng_ustconsumer_del_stream(stream); | |
1897 | break; | |
1898 | default: | |
1899 | ERR("Unknown consumer_data type"); | |
1900 | assert(0); | |
e316aad5 | 1901 | goto end; |
fb3a43a9 | 1902 | } |
fb3a43a9 | 1903 | |
c869f647 | 1904 | rcu_read_lock(); |
58b1f425 | 1905 | iter.iter.node = &stream->node.node; |
c869f647 DG |
1906 | ret = lttng_ht_del(ht, &iter); |
1907 | assert(!ret); | |
ca22feea | 1908 | |
d8ef542d MD |
1909 | iter.iter.node = &stream->node_channel_id.node; |
1910 | ret = lttng_ht_del(consumer_data.stream_per_chan_id_ht, &iter); | |
1911 | assert(!ret); | |
1912 | ||
ca22feea DG |
1913 | iter.iter.node = &stream->node_session_id.node; |
1914 | ret = lttng_ht_del(consumer_data.stream_list_ht, &iter); | |
1915 | assert(!ret); | |
c869f647 DG |
1916 | rcu_read_unlock(); |
1917 | ||
fb3a43a9 DG |
1918 | if (stream->out_fd >= 0) { |
1919 | ret = close(stream->out_fd); | |
1920 | if (ret) { | |
1921 | PERROR("close"); | |
1922 | } | |
1923 | } | |
1924 | ||
fb3a43a9 DG |
1925 | /* Check and cleanup relayd */ |
1926 | rcu_read_lock(); | |
1927 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
1928 | if (relayd != NULL) { | |
1929 | uatomic_dec(&relayd->refcount); | |
1930 | assert(uatomic_read(&relayd->refcount) >= 0); | |
1931 | ||
1932 | /* Closing streams requires to lock the control socket. */ | |
1933 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1934 | ret = relayd_send_close_stream(&relayd->control_sock, | |
1935 | stream->relayd_stream_id, stream->next_net_seq_num - 1); | |
1936 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1937 | if (ret < 0) { | |
1938 | DBG("Unable to close stream on the relayd. Continuing"); | |
1939 | /* | |
1940 | * Continue here. There is nothing we can do for the relayd. | |
1941 | * Chances are that the relayd has closed the socket so we just | |
1942 | * continue cleaning up. | |
1943 | */ | |
1944 | } | |
1945 | ||
1946 | /* Both conditions are met, we destroy the relayd. */ | |
1947 | if (uatomic_read(&relayd->refcount) == 0 && | |
1948 | uatomic_read(&relayd->destroy_flag)) { | |
d09e1200 | 1949 | destroy_relayd(relayd); |
fb3a43a9 DG |
1950 | } |
1951 | } | |
1952 | rcu_read_unlock(); | |
1953 | ||
1954 | /* Atomically decrement channel refcount since other threads can use it. */ | |
f2ad556d | 1955 | if (!uatomic_sub_return(&stream->chan->refcount, 1) |
ffe60014 | 1956 | && !uatomic_read(&stream->chan->nb_init_stream_left)) { |
c30aaa51 | 1957 | /* Go for channel deletion! */ |
e316aad5 | 1958 | free_chan = stream->chan; |
fb3a43a9 DG |
1959 | } |
1960 | ||
e316aad5 | 1961 | end: |
8994307f | 1962 | pthread_mutex_unlock(&stream->lock); |
74251bb8 | 1963 | pthread_mutex_unlock(&consumer_data.lock); |
e316aad5 DG |
1964 | |
1965 | if (free_chan) { | |
1966 | consumer_del_channel(free_chan); | |
1967 | } | |
1968 | ||
ffe60014 DG |
1969 | free_stream_rcu: |
1970 | call_rcu(&stream->node.head, free_stream_rcu); | |
fb3a43a9 DG |
1971 | } |
1972 | ||
1973 | /* | |
1974 | * Action done with the metadata stream when adding it to the consumer internal | |
1975 | * data structures to handle it. | |
1976 | */ | |
ffe60014 | 1977 | static int add_metadata_stream(struct lttng_consumer_stream *stream, |
e316aad5 | 1978 | struct lttng_ht *ht) |
fb3a43a9 | 1979 | { |
e316aad5 | 1980 | int ret = 0; |
fb3a43a9 | 1981 | struct consumer_relayd_sock_pair *relayd; |
76082088 | 1982 | struct lttng_ht_iter iter; |
d88aee68 | 1983 | struct lttng_ht_node_u64 *node; |
fb3a43a9 | 1984 | |
e316aad5 DG |
1985 | assert(stream); |
1986 | assert(ht); | |
1987 | ||
d88aee68 | 1988 | DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key); |
e316aad5 DG |
1989 | |
1990 | pthread_mutex_lock(&consumer_data.lock); | |
2e818a6a | 1991 | pthread_mutex_lock(&stream->lock); |
e316aad5 | 1992 | |
e316aad5 DG |
1993 | /* |
1994 | * From here, refcounts are updated so be _careful_ when returning an error | |
1995 | * after this point. | |
1996 | */ | |
1997 | ||
fb3a43a9 | 1998 | rcu_read_lock(); |
76082088 DG |
1999 | |
2000 | /* | |
2001 | * Lookup the stream just to make sure it does not exist in our internal | |
2002 | * state. This should NEVER happen. | |
2003 | */ | |
d88aee68 DG |
2004 | lttng_ht_lookup(ht, &stream->key, &iter); |
2005 | node = lttng_ht_iter_get_node_u64(&iter); | |
76082088 DG |
2006 | assert(!node); |
2007 | ||
e316aad5 | 2008 | /* Find relayd and, if one is found, increment refcount. */ |
fb3a43a9 DG |
2009 | relayd = consumer_find_relayd(stream->net_seq_idx); |
2010 | if (relayd != NULL) { | |
2011 | uatomic_inc(&relayd->refcount); | |
2012 | } | |
e316aad5 DG |
2013 | |
2014 | /* Update channel refcount once added without error(s). */ | |
2015 | uatomic_inc(&stream->chan->refcount); | |
2016 | ||
2017 | /* | |
ffe60014 DG |
2018 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
2019 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
2020 | * causes the count to become 0 also causes a stream to be added. The |
2021 | * channel deletion will thus be triggered by the following removal of this | |
2022 | * stream. | |
2023 | */ | |
ffe60014 | 2024 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
2025 | /* Increment refcount before decrementing nb_init_stream_left */ |
2026 | cmm_smp_wmb(); | |
ffe60014 | 2027 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
2028 | } |
2029 | ||
d88aee68 | 2030 | lttng_ht_add_unique_u64(ht, &stream->node); |
ca22feea | 2031 | |
d8ef542d MD |
2032 | lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht, |
2033 | &stream->node_channel_id); | |
2034 | ||
ca22feea DG |
2035 | /* |
2036 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
2037 | * the key since the HT does not use it and we allow to add redundant keys | |
2038 | * into this table. | |
2039 | */ | |
d88aee68 | 2040 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 2041 | |
fb3a43a9 | 2042 | rcu_read_unlock(); |
e316aad5 | 2043 | |
2e818a6a | 2044 | pthread_mutex_unlock(&stream->lock); |
e316aad5 DG |
2045 | pthread_mutex_unlock(&consumer_data.lock); |
2046 | return ret; | |
fb3a43a9 DG |
2047 | } |
2048 | ||
8994307f DG |
2049 | /* |
2050 | * Delete data stream that are flagged for deletion (endpoint_status). | |
2051 | */ | |
2052 | static void validate_endpoint_status_data_stream(void) | |
2053 | { | |
2054 | struct lttng_ht_iter iter; | |
2055 | struct lttng_consumer_stream *stream; | |
2056 | ||
2057 | DBG("Consumer delete flagged data stream"); | |
2058 | ||
2059 | rcu_read_lock(); | |
2060 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
2061 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2062 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2063 | continue; |
2064 | } | |
2065 | /* Delete it right now */ | |
2066 | consumer_del_stream(stream, data_ht); | |
2067 | } | |
2068 | rcu_read_unlock(); | |
2069 | } | |
2070 | ||
2071 | /* | |
2072 | * Delete metadata stream that are flagged for deletion (endpoint_status). | |
2073 | */ | |
2074 | static void validate_endpoint_status_metadata_stream( | |
2075 | struct lttng_poll_event *pollset) | |
2076 | { | |
2077 | struct lttng_ht_iter iter; | |
2078 | struct lttng_consumer_stream *stream; | |
2079 | ||
2080 | DBG("Consumer delete flagged metadata stream"); | |
2081 | ||
2082 | assert(pollset); | |
2083 | ||
2084 | rcu_read_lock(); | |
2085 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
2086 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2087 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2088 | continue; |
2089 | } | |
2090 | /* | |
2091 | * Remove from pollset so the metadata thread can continue without | |
2092 | * blocking on a deleted stream. | |
2093 | */ | |
2094 | lttng_poll_del(pollset, stream->wait_fd); | |
2095 | ||
2096 | /* Delete it right now */ | |
2097 | consumer_del_metadata_stream(stream, metadata_ht); | |
2098 | } | |
2099 | rcu_read_unlock(); | |
2100 | } | |
2101 | ||
fb3a43a9 DG |
2102 | /* |
2103 | * Thread polls on metadata file descriptor and write them on disk or on the | |
2104 | * network. | |
2105 | */ | |
7d980def | 2106 | void *consumer_thread_metadata_poll(void *data) |
fb3a43a9 DG |
2107 | { |
2108 | int ret, i, pollfd; | |
2109 | uint32_t revents, nb_fd; | |
e316aad5 | 2110 | struct lttng_consumer_stream *stream = NULL; |
fb3a43a9 | 2111 | struct lttng_ht_iter iter; |
d88aee68 | 2112 | struct lttng_ht_node_u64 *node; |
fb3a43a9 DG |
2113 | struct lttng_poll_event events; |
2114 | struct lttng_consumer_local_data *ctx = data; | |
2115 | ssize_t len; | |
2116 | ||
2117 | rcu_register_thread(); | |
2118 | ||
d88aee68 | 2119 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
04bb2b64 DG |
2120 | if (!metadata_ht) { |
2121 | /* ENOMEM at this point. Better to bail out. */ | |
d8ef542d | 2122 | goto end_ht; |
04bb2b64 DG |
2123 | } |
2124 | ||
fb3a43a9 DG |
2125 | DBG("Thread metadata poll started"); |
2126 | ||
fb3a43a9 DG |
2127 | /* Size is set to 1 for the consumer_metadata pipe */ |
2128 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2129 | if (ret < 0) { | |
2130 | ERR("Poll set creation failed"); | |
d8ef542d | 2131 | goto end_poll; |
fb3a43a9 DG |
2132 | } |
2133 | ||
13886d2d DG |
2134 | ret = lttng_poll_add(&events, |
2135 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN); | |
fb3a43a9 DG |
2136 | if (ret < 0) { |
2137 | goto end; | |
2138 | } | |
2139 | ||
2140 | /* Main loop */ | |
2141 | DBG("Metadata main loop started"); | |
2142 | ||
2143 | while (1) { | |
fb3a43a9 | 2144 | /* Only the metadata pipe is set */ |
d21b0d71 | 2145 | if (LTTNG_POLL_GETNB(&events) == 0 && consumer_quit == 1) { |
fb3a43a9 DG |
2146 | goto end; |
2147 | } | |
2148 | ||
2149 | restart: | |
d21b0d71 | 2150 | DBG("Metadata poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events)); |
fb3a43a9 DG |
2151 | ret = lttng_poll_wait(&events, -1); |
2152 | DBG("Metadata event catched in thread"); | |
2153 | if (ret < 0) { | |
2154 | if (errno == EINTR) { | |
e316aad5 | 2155 | ERR("Poll EINTR catched"); |
fb3a43a9 DG |
2156 | goto restart; |
2157 | } | |
2158 | goto error; | |
2159 | } | |
2160 | ||
0d9c5d77 DG |
2161 | nb_fd = ret; |
2162 | ||
e316aad5 | 2163 | /* From here, the event is a metadata wait fd */ |
fb3a43a9 DG |
2164 | for (i = 0; i < nb_fd; i++) { |
2165 | revents = LTTNG_POLL_GETEV(&events, i); | |
2166 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2167 | ||
e316aad5 DG |
2168 | /* Just don't waste time if no returned events for the fd */ |
2169 | if (!revents) { | |
2170 | continue; | |
2171 | } | |
2172 | ||
13886d2d | 2173 | if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) { |
4adabd61 | 2174 | if (revents & (LPOLLERR | LPOLLHUP )) { |
fb3a43a9 DG |
2175 | DBG("Metadata thread pipe hung up"); |
2176 | /* | |
2177 | * Remove the pipe from the poll set and continue the loop | |
2178 | * since their might be data to consume. | |
2179 | */ | |
13886d2d DG |
2180 | lttng_poll_del(&events, |
2181 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2182 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
fb3a43a9 DG |
2183 | continue; |
2184 | } else if (revents & LPOLLIN) { | |
13886d2d DG |
2185 | ssize_t pipe_len; |
2186 | ||
2187 | pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe, | |
2188 | &stream, sizeof(stream)); | |
2189 | if (pipe_len < 0) { | |
2190 | ERR("read metadata stream, ret: %ld", pipe_len); | |
fb3a43a9 | 2191 | /* |
13886d2d | 2192 | * Continue here to handle the rest of the streams. |
fb3a43a9 DG |
2193 | */ |
2194 | continue; | |
2195 | } | |
2196 | ||
8994307f DG |
2197 | /* A NULL stream means that the state has changed. */ |
2198 | if (stream == NULL) { | |
2199 | /* Check for deleted streams. */ | |
2200 | validate_endpoint_status_metadata_stream(&events); | |
3714380f | 2201 | goto restart; |
8994307f DG |
2202 | } |
2203 | ||
fb3a43a9 DG |
2204 | DBG("Adding metadata stream %d to poll set", |
2205 | stream->wait_fd); | |
2206 | ||
ffe60014 | 2207 | ret = add_metadata_stream(stream, metadata_ht); |
e316aad5 DG |
2208 | if (ret) { |
2209 | ERR("Unable to add metadata stream"); | |
2210 | /* Stream was not setup properly. Continuing. */ | |
2211 | consumer_del_metadata_stream(stream, NULL); | |
2212 | continue; | |
2213 | } | |
fb3a43a9 DG |
2214 | |
2215 | /* Add metadata stream to the global poll events list */ | |
2216 | lttng_poll_add(&events, stream->wait_fd, | |
2217 | LPOLLIN | LPOLLPRI); | |
fb3a43a9 DG |
2218 | } |
2219 | ||
e316aad5 | 2220 | /* Handle other stream */ |
fb3a43a9 DG |
2221 | continue; |
2222 | } | |
2223 | ||
d09e1200 | 2224 | rcu_read_lock(); |
d88aee68 DG |
2225 | { |
2226 | uint64_t tmp_id = (uint64_t) pollfd; | |
2227 | ||
2228 | lttng_ht_lookup(metadata_ht, &tmp_id, &iter); | |
2229 | } | |
2230 | node = lttng_ht_iter_get_node_u64(&iter); | |
e316aad5 | 2231 | assert(node); |
fb3a43a9 DG |
2232 | |
2233 | stream = caa_container_of(node, struct lttng_consumer_stream, | |
58b1f425 | 2234 | node); |
fb3a43a9 | 2235 | |
e316aad5 | 2236 | /* Check for error event */ |
4adabd61 | 2237 | if (revents & (LPOLLERR | LPOLLHUP)) { |
e316aad5 | 2238 | DBG("Metadata fd %d is hup|err.", pollfd); |
fb3a43a9 DG |
2239 | if (!stream->hangup_flush_done |
2240 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2241 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2242 | DBG("Attempting to flush and consume the UST buffers"); | |
2243 | lttng_ustconsumer_on_stream_hangup(stream); | |
2244 | ||
2245 | /* We just flushed the stream now read it. */ | |
4bb94b75 DG |
2246 | do { |
2247 | len = ctx->on_buffer_ready(stream, ctx); | |
2248 | /* | |
2249 | * We don't check the return value here since if we get | |
2250 | * a negative len, it means an error occured thus we | |
2251 | * simply remove it from the poll set and free the | |
2252 | * stream. | |
2253 | */ | |
2254 | } while (len > 0); | |
fb3a43a9 DG |
2255 | } |
2256 | ||
fb3a43a9 | 2257 | lttng_poll_del(&events, stream->wait_fd); |
e316aad5 DG |
2258 | /* |
2259 | * This call update the channel states, closes file descriptors | |
2260 | * and securely free the stream. | |
2261 | */ | |
2262 | consumer_del_metadata_stream(stream, metadata_ht); | |
2263 | } else if (revents & (LPOLLIN | LPOLLPRI)) { | |
2264 | /* Get the data out of the metadata file descriptor */ | |
2265 | DBG("Metadata available on fd %d", pollfd); | |
2266 | assert(stream->wait_fd == pollfd); | |
2267 | ||
2268 | len = ctx->on_buffer_ready(stream, ctx); | |
2269 | /* It's ok to have an unavailable sub-buffer */ | |
b64403e3 | 2270 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2271 | /* Clean up stream from consumer and free it. */ |
2272 | lttng_poll_del(&events, stream->wait_fd); | |
2273 | consumer_del_metadata_stream(stream, metadata_ht); | |
e316aad5 DG |
2274 | } else if (len > 0) { |
2275 | stream->data_read = 1; | |
2276 | } | |
fb3a43a9 | 2277 | } |
e316aad5 DG |
2278 | |
2279 | /* Release RCU lock for the stream looked up */ | |
d09e1200 | 2280 | rcu_read_unlock(); |
fb3a43a9 DG |
2281 | } |
2282 | } | |
2283 | ||
2284 | error: | |
2285 | end: | |
2286 | DBG("Metadata poll thread exiting"); | |
fb3a43a9 | 2287 | |
d8ef542d MD |
2288 | lttng_poll_clean(&events); |
2289 | end_poll: | |
04bb2b64 | 2290 | destroy_stream_ht(metadata_ht); |
d8ef542d | 2291 | end_ht: |
fb3a43a9 DG |
2292 | rcu_unregister_thread(); |
2293 | return NULL; | |
2294 | } | |
2295 | ||
3bd1e081 | 2296 | /* |
e4421fec | 2297 | * This thread polls the fds in the set to consume the data and write |
3bd1e081 MD |
2298 | * it to tracefile if necessary. |
2299 | */ | |
7d980def | 2300 | void *consumer_thread_data_poll(void *data) |
3bd1e081 MD |
2301 | { |
2302 | int num_rdy, num_hup, high_prio, ret, i; | |
2303 | struct pollfd *pollfd = NULL; | |
2304 | /* local view of the streams */ | |
c869f647 | 2305 | struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; |
3bd1e081 MD |
2306 | /* local view of consumer_data.fds_count */ |
2307 | int nb_fd = 0; | |
3bd1e081 | 2308 | struct lttng_consumer_local_data *ctx = data; |
00e2e675 | 2309 | ssize_t len; |
3bd1e081 | 2310 | |
e7b994a3 DG |
2311 | rcu_register_thread(); |
2312 | ||
d88aee68 | 2313 | data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
43c34bc3 | 2314 | if (data_ht == NULL) { |
04bb2b64 | 2315 | /* ENOMEM at this point. Better to bail out. */ |
43c34bc3 DG |
2316 | goto end; |
2317 | } | |
2318 | ||
effcf122 | 2319 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream)); |
3bd1e081 MD |
2320 | |
2321 | while (1) { | |
2322 | high_prio = 0; | |
2323 | num_hup = 0; | |
2324 | ||
2325 | /* | |
e4421fec | 2326 | * the fds set has been updated, we need to update our |
3bd1e081 MD |
2327 | * local array as well |
2328 | */ | |
2329 | pthread_mutex_lock(&consumer_data.lock); | |
2330 | if (consumer_data.need_update) { | |
0e428499 DG |
2331 | free(pollfd); |
2332 | pollfd = NULL; | |
2333 | ||
2334 | free(local_stream); | |
2335 | local_stream = NULL; | |
3bd1e081 | 2336 | |
50f8ae69 | 2337 | /* allocate for all fds + 1 for the consumer_data_pipe */ |
effcf122 | 2338 | pollfd = zmalloc((consumer_data.stream_count + 1) * sizeof(struct pollfd)); |
3bd1e081 | 2339 | if (pollfd == NULL) { |
7a57cf92 | 2340 | PERROR("pollfd malloc"); |
3bd1e081 MD |
2341 | pthread_mutex_unlock(&consumer_data.lock); |
2342 | goto end; | |
2343 | } | |
2344 | ||
50f8ae69 | 2345 | /* allocate for all fds + 1 for the consumer_data_pipe */ |
effcf122 | 2346 | local_stream = zmalloc((consumer_data.stream_count + 1) * |
3bd1e081 MD |
2347 | sizeof(struct lttng_consumer_stream)); |
2348 | if (local_stream == NULL) { | |
7a57cf92 | 2349 | PERROR("local_stream malloc"); |
3bd1e081 MD |
2350 | pthread_mutex_unlock(&consumer_data.lock); |
2351 | goto end; | |
2352 | } | |
ffe60014 | 2353 | ret = update_poll_array(ctx, &pollfd, local_stream, |
43c34bc3 | 2354 | data_ht); |
3bd1e081 MD |
2355 | if (ret < 0) { |
2356 | ERR("Error in allocating pollfd or local_outfds"); | |
f73fabfd | 2357 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2358 | pthread_mutex_unlock(&consumer_data.lock); |
2359 | goto end; | |
2360 | } | |
2361 | nb_fd = ret; | |
2362 | consumer_data.need_update = 0; | |
2363 | } | |
2364 | pthread_mutex_unlock(&consumer_data.lock); | |
2365 | ||
4078b776 MD |
2366 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
2367 | if (nb_fd == 0 && consumer_quit == 1) { | |
2368 | goto end; | |
2369 | } | |
3bd1e081 | 2370 | /* poll on the array of fds */ |
88f2b785 | 2371 | restart: |
3bd1e081 | 2372 | DBG("polling on %d fd", nb_fd + 1); |
cb365c03 | 2373 | num_rdy = poll(pollfd, nb_fd + 1, -1); |
3bd1e081 MD |
2374 | DBG("poll num_rdy : %d", num_rdy); |
2375 | if (num_rdy == -1) { | |
88f2b785 MD |
2376 | /* |
2377 | * Restart interrupted system call. | |
2378 | */ | |
2379 | if (errno == EINTR) { | |
2380 | goto restart; | |
2381 | } | |
7a57cf92 | 2382 | PERROR("Poll error"); |
f73fabfd | 2383 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2384 | goto end; |
2385 | } else if (num_rdy == 0) { | |
2386 | DBG("Polling thread timed out"); | |
2387 | goto end; | |
2388 | } | |
2389 | ||
3bd1e081 | 2390 | /* |
50f8ae69 | 2391 | * If the consumer_data_pipe triggered poll go directly to the |
00e2e675 DG |
2392 | * beginning of the loop to update the array. We want to prioritize |
2393 | * array update over low-priority reads. | |
3bd1e081 | 2394 | */ |
509bb1cf | 2395 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
ab30f567 | 2396 | ssize_t pipe_readlen; |
04fdd819 | 2397 | |
50f8ae69 | 2398 | DBG("consumer_data_pipe wake up"); |
acdb9057 DG |
2399 | pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe, |
2400 | &new_stream, sizeof(new_stream)); | |
23f5f35d | 2401 | if (pipe_readlen < 0) { |
acdb9057 | 2402 | ERR("Consumer data pipe ret %ld", pipe_readlen); |
23f5f35d DG |
2403 | /* Continue so we can at least handle the current stream(s). */ |
2404 | continue; | |
2405 | } | |
c869f647 DG |
2406 | |
2407 | /* | |
2408 | * If the stream is NULL, just ignore it. It's also possible that | |
2409 | * the sessiond poll thread changed the consumer_quit state and is | |
2410 | * waking us up to test it. | |
2411 | */ | |
2412 | if (new_stream == NULL) { | |
8994307f | 2413 | validate_endpoint_status_data_stream(); |
c869f647 DG |
2414 | continue; |
2415 | } | |
2416 | ||
ffe60014 | 2417 | ret = add_stream(new_stream, data_ht); |
c869f647 | 2418 | if (ret) { |
d88aee68 | 2419 | ERR("Consumer add stream %" PRIu64 " failed. Continuing", |
c869f647 DG |
2420 | new_stream->key); |
2421 | /* | |
2422 | * At this point, if the add_stream fails, it is not in the | |
2423 | * hash table thus passing the NULL value here. | |
2424 | */ | |
2425 | consumer_del_stream(new_stream, NULL); | |
2426 | } | |
2427 | ||
2428 | /* Continue to update the local streams and handle prio ones */ | |
3bd1e081 MD |
2429 | continue; |
2430 | } | |
2431 | ||
2432 | /* Take care of high priority channels first. */ | |
2433 | for (i = 0; i < nb_fd; i++) { | |
9617607b DG |
2434 | if (local_stream[i] == NULL) { |
2435 | continue; | |
2436 | } | |
fb3a43a9 | 2437 | if (pollfd[i].revents & POLLPRI) { |
d41f73b7 MD |
2438 | DBG("Urgent read on fd %d", pollfd[i].fd); |
2439 | high_prio = 1; | |
4078b776 | 2440 | len = ctx->on_buffer_ready(local_stream[i], ctx); |
d41f73b7 | 2441 | /* it's ok to have an unavailable sub-buffer */ |
b64403e3 | 2442 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2443 | /* Clean the stream and free it. */ |
2444 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2445 | local_stream[i] = NULL; |
4078b776 MD |
2446 | } else if (len > 0) { |
2447 | local_stream[i]->data_read = 1; | |
d41f73b7 | 2448 | } |
3bd1e081 MD |
2449 | } |
2450 | } | |
2451 | ||
4078b776 MD |
2452 | /* |
2453 | * If we read high prio channel in this loop, try again | |
2454 | * for more high prio data. | |
2455 | */ | |
2456 | if (high_prio) { | |
3bd1e081 MD |
2457 | continue; |
2458 | } | |
2459 | ||
2460 | /* Take care of low priority channels. */ | |
4078b776 | 2461 | for (i = 0; i < nb_fd; i++) { |
9617607b DG |
2462 | if (local_stream[i] == NULL) { |
2463 | continue; | |
2464 | } | |
4078b776 MD |
2465 | if ((pollfd[i].revents & POLLIN) || |
2466 | local_stream[i]->hangup_flush_done) { | |
4078b776 MD |
2467 | DBG("Normal read on fd %d", pollfd[i].fd); |
2468 | len = ctx->on_buffer_ready(local_stream[i], ctx); | |
2469 | /* it's ok to have an unavailable sub-buffer */ | |
b64403e3 | 2470 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2471 | /* Clean the stream and free it. */ |
2472 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2473 | local_stream[i] = NULL; |
4078b776 MD |
2474 | } else if (len > 0) { |
2475 | local_stream[i]->data_read = 1; | |
2476 | } | |
2477 | } | |
2478 | } | |
2479 | ||
2480 | /* Handle hangup and errors */ | |
2481 | for (i = 0; i < nb_fd; i++) { | |
9617607b DG |
2482 | if (local_stream[i] == NULL) { |
2483 | continue; | |
2484 | } | |
4078b776 MD |
2485 | if (!local_stream[i]->hangup_flush_done |
2486 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) | |
2487 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2488 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2489 | DBG("fd %d is hup|err|nval. Attempting flush and read.", | |
9617607b | 2490 | pollfd[i].fd); |
4078b776 MD |
2491 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); |
2492 | /* Attempt read again, for the data we just flushed. */ | |
2493 | local_stream[i]->data_read = 1; | |
2494 | } | |
2495 | /* | |
2496 | * If the poll flag is HUP/ERR/NVAL and we have | |
2497 | * read no data in this pass, we can remove the | |
2498 | * stream from its hash table. | |
2499 | */ | |
2500 | if ((pollfd[i].revents & POLLHUP)) { | |
2501 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); | |
2502 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2503 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2504 | local_stream[i] = NULL; |
4078b776 MD |
2505 | num_hup++; |
2506 | } | |
2507 | } else if (pollfd[i].revents & POLLERR) { | |
2508 | ERR("Error returned in polling fd %d.", pollfd[i].fd); | |
2509 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2510 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2511 | local_stream[i] = NULL; |
4078b776 MD |
2512 | num_hup++; |
2513 | } | |
2514 | } else if (pollfd[i].revents & POLLNVAL) { | |
2515 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); | |
2516 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2517 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2518 | local_stream[i] = NULL; |
4078b776 | 2519 | num_hup++; |
3bd1e081 MD |
2520 | } |
2521 | } | |
9617607b DG |
2522 | if (local_stream[i] != NULL) { |
2523 | local_stream[i]->data_read = 0; | |
2524 | } | |
3bd1e081 MD |
2525 | } |
2526 | } | |
2527 | end: | |
2528 | DBG("polling thread exiting"); | |
0e428499 DG |
2529 | free(pollfd); |
2530 | free(local_stream); | |
fb3a43a9 DG |
2531 | |
2532 | /* | |
2533 | * Close the write side of the pipe so epoll_wait() in | |
7d980def DG |
2534 | * consumer_thread_metadata_poll can catch it. The thread is monitoring the |
2535 | * read side of the pipe. If we close them both, epoll_wait strangely does | |
2536 | * not return and could create a endless wait period if the pipe is the | |
2537 | * only tracked fd in the poll set. The thread will take care of closing | |
2538 | * the read side. | |
fb3a43a9 | 2539 | */ |
13886d2d | 2540 | (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe); |
fb3a43a9 | 2541 | |
04bb2b64 | 2542 | destroy_data_stream_ht(data_ht); |
43c34bc3 | 2543 | |
e7b994a3 | 2544 | rcu_unregister_thread(); |
3bd1e081 MD |
2545 | return NULL; |
2546 | } | |
2547 | ||
d8ef542d MD |
2548 | /* |
2549 | * Close wake-up end of each stream belonging to the channel. This will | |
2550 | * allow the poll() on the stream read-side to detect when the | |
2551 | * write-side (application) finally closes them. | |
2552 | */ | |
2553 | static | |
2554 | void consumer_close_channel_streams(struct lttng_consumer_channel *channel) | |
2555 | { | |
2556 | struct lttng_ht *ht; | |
2557 | struct lttng_consumer_stream *stream; | |
2558 | struct lttng_ht_iter iter; | |
2559 | ||
2560 | ht = consumer_data.stream_per_chan_id_ht; | |
2561 | ||
2562 | rcu_read_lock(); | |
2563 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
2564 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
2565 | ht->match_fct, &channel->key, | |
2566 | &iter.iter, stream, node_channel_id.node) { | |
f2ad556d MD |
2567 | /* |
2568 | * Protect against teardown with mutex. | |
2569 | */ | |
2570 | pthread_mutex_lock(&stream->lock); | |
2571 | if (cds_lfht_is_node_deleted(&stream->node.node)) { | |
2572 | goto next; | |
2573 | } | |
d8ef542d MD |
2574 | switch (consumer_data.type) { |
2575 | case LTTNG_CONSUMER_KERNEL: | |
2576 | break; | |
2577 | case LTTNG_CONSUMER32_UST: | |
2578 | case LTTNG_CONSUMER64_UST: | |
2579 | /* | |
2580 | * Note: a mutex is taken internally within | |
2581 | * liblttng-ust-ctl to protect timer wakeup_fd | |
2582 | * use from concurrent close. | |
2583 | */ | |
2584 | lttng_ustconsumer_close_stream_wakeup(stream); | |
2585 | break; | |
2586 | default: | |
2587 | ERR("Unknown consumer_data type"); | |
2588 | assert(0); | |
2589 | } | |
f2ad556d MD |
2590 | next: |
2591 | pthread_mutex_unlock(&stream->lock); | |
d8ef542d MD |
2592 | } |
2593 | rcu_read_unlock(); | |
2594 | } | |
2595 | ||
2596 | static void destroy_channel_ht(struct lttng_ht *ht) | |
2597 | { | |
2598 | struct lttng_ht_iter iter; | |
2599 | struct lttng_consumer_channel *channel; | |
2600 | int ret; | |
2601 | ||
2602 | if (ht == NULL) { | |
2603 | return; | |
2604 | } | |
2605 | ||
2606 | rcu_read_lock(); | |
2607 | cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) { | |
2608 | ret = lttng_ht_del(ht, &iter); | |
2609 | assert(ret != 0); | |
2610 | } | |
2611 | rcu_read_unlock(); | |
2612 | ||
2613 | lttng_ht_destroy(ht); | |
2614 | } | |
2615 | ||
2616 | /* | |
2617 | * This thread polls the channel fds to detect when they are being | |
2618 | * closed. It closes all related streams if the channel is detected as | |
2619 | * closed. It is currently only used as a shim layer for UST because the | |
2620 | * consumerd needs to keep the per-stream wakeup end of pipes open for | |
2621 | * periodical flush. | |
2622 | */ | |
2623 | void *consumer_thread_channel_poll(void *data) | |
2624 | { | |
2625 | int ret, i, pollfd; | |
2626 | uint32_t revents, nb_fd; | |
2627 | struct lttng_consumer_channel *chan = NULL; | |
2628 | struct lttng_ht_iter iter; | |
2629 | struct lttng_ht_node_u64 *node; | |
2630 | struct lttng_poll_event events; | |
2631 | struct lttng_consumer_local_data *ctx = data; | |
2632 | struct lttng_ht *channel_ht; | |
2633 | ||
2634 | rcu_register_thread(); | |
2635 | ||
2636 | channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
2637 | if (!channel_ht) { | |
2638 | /* ENOMEM at this point. Better to bail out. */ | |
2639 | goto end_ht; | |
2640 | } | |
2641 | ||
2642 | DBG("Thread channel poll started"); | |
2643 | ||
2644 | /* Size is set to 1 for the consumer_channel pipe */ | |
2645 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2646 | if (ret < 0) { | |
2647 | ERR("Poll set creation failed"); | |
2648 | goto end_poll; | |
2649 | } | |
2650 | ||
2651 | ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN); | |
2652 | if (ret < 0) { | |
2653 | goto end; | |
2654 | } | |
2655 | ||
2656 | /* Main loop */ | |
2657 | DBG("Channel main loop started"); | |
2658 | ||
2659 | while (1) { | |
2660 | /* Only the channel pipe is set */ | |
2661 | if (LTTNG_POLL_GETNB(&events) == 0 && consumer_quit == 1) { | |
2662 | goto end; | |
2663 | } | |
2664 | ||
2665 | restart: | |
2666 | DBG("Channel poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events)); | |
2667 | ret = lttng_poll_wait(&events, -1); | |
2668 | DBG("Channel event catched in thread"); | |
2669 | if (ret < 0) { | |
2670 | if (errno == EINTR) { | |
2671 | ERR("Poll EINTR catched"); | |
2672 | goto restart; | |
2673 | } | |
2674 | goto end; | |
2675 | } | |
2676 | ||
2677 | nb_fd = ret; | |
2678 | ||
2679 | /* From here, the event is a channel wait fd */ | |
2680 | for (i = 0; i < nb_fd; i++) { | |
2681 | revents = LTTNG_POLL_GETEV(&events, i); | |
2682 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2683 | ||
2684 | /* Just don't waste time if no returned events for the fd */ | |
2685 | if (!revents) { | |
2686 | continue; | |
2687 | } | |
2688 | if (pollfd == ctx->consumer_channel_pipe[0]) { | |
2689 | if (revents & (LPOLLERR | LPOLLHUP)) { | |
2690 | DBG("Channel thread pipe hung up"); | |
2691 | /* | |
2692 | * Remove the pipe from the poll set and continue the loop | |
2693 | * since their might be data to consume. | |
2694 | */ | |
2695 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
2696 | continue; | |
2697 | } else if (revents & LPOLLIN) { | |
2698 | enum consumer_channel_action action; | |
a0cbdd2e | 2699 | uint64_t key; |
d8ef542d | 2700 | |
a0cbdd2e | 2701 | ret = read_channel_pipe(ctx, &chan, &key, &action); |
d8ef542d MD |
2702 | if (ret <= 0) { |
2703 | ERR("Error reading channel pipe"); | |
2704 | continue; | |
2705 | } | |
2706 | ||
2707 | switch (action) { | |
2708 | case CONSUMER_CHANNEL_ADD: | |
2709 | DBG("Adding channel %d to poll set", | |
2710 | chan->wait_fd); | |
2711 | ||
2712 | lttng_ht_node_init_u64(&chan->wait_fd_node, | |
2713 | chan->wait_fd); | |
2714 | lttng_ht_add_unique_u64(channel_ht, | |
2715 | &chan->wait_fd_node); | |
2716 | /* Add channel to the global poll events list */ | |
2717 | lttng_poll_add(&events, chan->wait_fd, | |
2718 | LPOLLIN | LPOLLPRI); | |
2719 | break; | |
a0cbdd2e MD |
2720 | case CONSUMER_CHANNEL_DEL: |
2721 | { | |
2722 | chan = consumer_find_channel(key); | |
2723 | if (!chan) { | |
2724 | ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key); | |
2725 | break; | |
2726 | } | |
2727 | lttng_poll_del(&events, chan->wait_fd); | |
2728 | ret = lttng_ht_del(channel_ht, &iter); | |
2729 | assert(ret == 0); | |
2730 | consumer_close_channel_streams(chan); | |
2731 | ||
2732 | /* | |
2733 | * Release our own refcount. Force channel deletion even if | |
2734 | * streams were not initialized. | |
2735 | */ | |
2736 | if (!uatomic_sub_return(&chan->refcount, 1)) { | |
2737 | consumer_del_channel(chan); | |
2738 | } | |
2739 | goto restart; | |
2740 | } | |
d8ef542d MD |
2741 | case CONSUMER_CHANNEL_QUIT: |
2742 | /* | |
2743 | * Remove the pipe from the poll set and continue the loop | |
2744 | * since their might be data to consume. | |
2745 | */ | |
2746 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
2747 | continue; | |
2748 | default: | |
2749 | ERR("Unknown action"); | |
2750 | break; | |
2751 | } | |
2752 | } | |
2753 | ||
2754 | /* Handle other stream */ | |
2755 | continue; | |
2756 | } | |
2757 | ||
2758 | rcu_read_lock(); | |
2759 | { | |
2760 | uint64_t tmp_id = (uint64_t) pollfd; | |
2761 | ||
2762 | lttng_ht_lookup(channel_ht, &tmp_id, &iter); | |
2763 | } | |
2764 | node = lttng_ht_iter_get_node_u64(&iter); | |
2765 | assert(node); | |
2766 | ||
2767 | chan = caa_container_of(node, struct lttng_consumer_channel, | |
2768 | wait_fd_node); | |
2769 | ||
2770 | /* Check for error event */ | |
2771 | if (revents & (LPOLLERR | LPOLLHUP)) { | |
2772 | DBG("Channel fd %d is hup|err.", pollfd); | |
2773 | ||
2774 | lttng_poll_del(&events, chan->wait_fd); | |
2775 | ret = lttng_ht_del(channel_ht, &iter); | |
2776 | assert(ret == 0); | |
2777 | consumer_close_channel_streams(chan); | |
f2ad556d MD |
2778 | |
2779 | /* Release our own refcount */ | |
2780 | if (!uatomic_sub_return(&chan->refcount, 1) | |
2781 | && !uatomic_read(&chan->nb_init_stream_left)) { | |
2782 | consumer_del_channel(chan); | |
2783 | } | |
d8ef542d MD |
2784 | } |
2785 | ||
2786 | /* Release RCU lock for the channel looked up */ | |
2787 | rcu_read_unlock(); | |
2788 | } | |
2789 | } | |
2790 | ||
2791 | end: | |
2792 | lttng_poll_clean(&events); | |
2793 | end_poll: | |
2794 | destroy_channel_ht(channel_ht); | |
2795 | end_ht: | |
2796 | DBG("Channel poll thread exiting"); | |
2797 | rcu_unregister_thread(); | |
2798 | return NULL; | |
2799 | } | |
2800 | ||
331744e3 JD |
2801 | static int set_metadata_socket(struct lttng_consumer_local_data *ctx, |
2802 | struct pollfd *sockpoll, int client_socket) | |
2803 | { | |
2804 | int ret; | |
2805 | ||
2806 | assert(ctx); | |
2807 | assert(sockpoll); | |
2808 | ||
2809 | if (lttng_consumer_poll_socket(sockpoll) < 0) { | |
2810 | ret = -1; | |
2811 | goto error; | |
2812 | } | |
2813 | DBG("Metadata connection on client_socket"); | |
2814 | ||
2815 | /* Blocking call, waiting for transmission */ | |
2816 | ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket); | |
2817 | if (ctx->consumer_metadata_socket < 0) { | |
2818 | WARN("On accept metadata"); | |
2819 | ret = -1; | |
2820 | goto error; | |
2821 | } | |
2822 | ret = 0; | |
2823 | ||
2824 | error: | |
2825 | return ret; | |
2826 | } | |
2827 | ||
3bd1e081 MD |
2828 | /* |
2829 | * This thread listens on the consumerd socket and receives the file | |
2830 | * descriptors from the session daemon. | |
2831 | */ | |
7d980def | 2832 | void *consumer_thread_sessiond_poll(void *data) |
3bd1e081 | 2833 | { |
d96f09c6 | 2834 | int sock = -1, client_socket, ret; |
3bd1e081 MD |
2835 | /* |
2836 | * structure to poll for incoming data on communication socket avoids | |
2837 | * making blocking sockets. | |
2838 | */ | |
2839 | struct pollfd consumer_sockpoll[2]; | |
2840 | struct lttng_consumer_local_data *ctx = data; | |
2841 | ||
e7b994a3 DG |
2842 | rcu_register_thread(); |
2843 | ||
3bd1e081 MD |
2844 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
2845 | unlink(ctx->consumer_command_sock_path); | |
2846 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); | |
2847 | if (client_socket < 0) { | |
2848 | ERR("Cannot create command socket"); | |
2849 | goto end; | |
2850 | } | |
2851 | ||
2852 | ret = lttcomm_listen_unix_sock(client_socket); | |
2853 | if (ret < 0) { | |
2854 | goto end; | |
2855 | } | |
2856 | ||
32258573 | 2857 | DBG("Sending ready command to lttng-sessiond"); |
f73fabfd | 2858 | ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); |
3bd1e081 MD |
2859 | /* return < 0 on error, but == 0 is not fatal */ |
2860 | if (ret < 0) { | |
32258573 | 2861 | ERR("Error sending ready command to lttng-sessiond"); |
3bd1e081 MD |
2862 | goto end; |
2863 | } | |
2864 | ||
2865 | ret = fcntl(client_socket, F_SETFL, O_NONBLOCK); | |
2866 | if (ret < 0) { | |
7a57cf92 | 2867 | PERROR("fcntl O_NONBLOCK"); |
3bd1e081 MD |
2868 | goto end; |
2869 | } | |
2870 | ||
2871 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ | |
2872 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; | |
2873 | consumer_sockpoll[0].events = POLLIN | POLLPRI; | |
2874 | consumer_sockpoll[1].fd = client_socket; | |
2875 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
2876 | ||
2877 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
2878 | goto end; | |
2879 | } | |
2880 | DBG("Connection on client_socket"); | |
2881 | ||
2882 | /* Blocking call, waiting for transmission */ | |
2883 | sock = lttcomm_accept_unix_sock(client_socket); | |
534d2592 | 2884 | if (sock < 0) { |
3bd1e081 MD |
2885 | WARN("On accept"); |
2886 | goto end; | |
2887 | } | |
2888 | ret = fcntl(sock, F_SETFL, O_NONBLOCK); | |
2889 | if (ret < 0) { | |
7a57cf92 | 2890 | PERROR("fcntl O_NONBLOCK"); |
3bd1e081 MD |
2891 | goto end; |
2892 | } | |
2893 | ||
331744e3 JD |
2894 | /* |
2895 | * Setup metadata socket which is the second socket connection on the | |
2896 | * command unix socket. | |
2897 | */ | |
2898 | ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket); | |
2899 | if (ret < 0) { | |
2900 | goto end; | |
2901 | } | |
2902 | ||
d96f09c6 DG |
2903 | /* This socket is not useful anymore. */ |
2904 | ret = close(client_socket); | |
2905 | if (ret < 0) { | |
2906 | PERROR("close client_socket"); | |
2907 | } | |
2908 | client_socket = -1; | |
2909 | ||
3bd1e081 MD |
2910 | /* update the polling structure to poll on the established socket */ |
2911 | consumer_sockpoll[1].fd = sock; | |
2912 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
2913 | ||
2914 | while (1) { | |
2915 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
2916 | goto end; | |
2917 | } | |
2918 | DBG("Incoming command on sock"); | |
2919 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
2920 | if (ret == -ENOENT) { | |
2921 | DBG("Received STOP command"); | |
2922 | goto end; | |
2923 | } | |
4cbc1a04 DG |
2924 | if (ret <= 0) { |
2925 | /* | |
2926 | * This could simply be a session daemon quitting. Don't output | |
2927 | * ERR() here. | |
2928 | */ | |
2929 | DBG("Communication interrupted on command socket"); | |
3bd1e081 MD |
2930 | goto end; |
2931 | } | |
2932 | if (consumer_quit) { | |
2933 | DBG("consumer_thread_receive_fds received quit from signal"); | |
2934 | goto end; | |
2935 | } | |
ffe60014 | 2936 | DBG("received command on sock"); |
3bd1e081 MD |
2937 | } |
2938 | end: | |
ffe60014 | 2939 | DBG("Consumer thread sessiond poll exiting"); |
3bd1e081 | 2940 | |
d88aee68 DG |
2941 | /* |
2942 | * Close metadata streams since the producer is the session daemon which | |
2943 | * just died. | |
2944 | * | |
2945 | * NOTE: for now, this only applies to the UST tracer. | |
2946 | */ | |
2947 | lttng_consumer_close_metadata(); | |
2948 | ||
3bd1e081 MD |
2949 | /* |
2950 | * when all fds have hung up, the polling thread | |
2951 | * can exit cleanly | |
2952 | */ | |
2953 | consumer_quit = 1; | |
2954 | ||
04fdd819 | 2955 | /* |
c869f647 | 2956 | * Notify the data poll thread to poll back again and test the |
8994307f | 2957 | * consumer_quit state that we just set so to quit gracefully. |
04fdd819 | 2958 | */ |
acdb9057 | 2959 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
c869f647 | 2960 | |
a0cbdd2e | 2961 | notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT); |
d8ef542d | 2962 | |
d96f09c6 DG |
2963 | /* Cleaning up possibly open sockets. */ |
2964 | if (sock >= 0) { | |
2965 | ret = close(sock); | |
2966 | if (ret < 0) { | |
2967 | PERROR("close sock sessiond poll"); | |
2968 | } | |
2969 | } | |
2970 | if (client_socket >= 0) { | |
2971 | ret = close(sock); | |
2972 | if (ret < 0) { | |
2973 | PERROR("close client_socket sessiond poll"); | |
2974 | } | |
2975 | } | |
2976 | ||
e7b994a3 | 2977 | rcu_unregister_thread(); |
3bd1e081 MD |
2978 | return NULL; |
2979 | } | |
d41f73b7 | 2980 | |
4078b776 | 2981 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
d41f73b7 MD |
2982 | struct lttng_consumer_local_data *ctx) |
2983 | { | |
74251bb8 DG |
2984 | ssize_t ret; |
2985 | ||
2986 | pthread_mutex_lock(&stream->lock); | |
2987 | ||
d41f73b7 MD |
2988 | switch (consumer_data.type) { |
2989 | case LTTNG_CONSUMER_KERNEL: | |
74251bb8 DG |
2990 | ret = lttng_kconsumer_read_subbuffer(stream, ctx); |
2991 | break; | |
7753dea8 MD |
2992 | case LTTNG_CONSUMER32_UST: |
2993 | case LTTNG_CONSUMER64_UST: | |
74251bb8 DG |
2994 | ret = lttng_ustconsumer_read_subbuffer(stream, ctx); |
2995 | break; | |
d41f73b7 MD |
2996 | default: |
2997 | ERR("Unknown consumer_data type"); | |
2998 | assert(0); | |
74251bb8 DG |
2999 | ret = -ENOSYS; |
3000 | break; | |
d41f73b7 | 3001 | } |
74251bb8 DG |
3002 | |
3003 | pthread_mutex_unlock(&stream->lock); | |
3004 | return ret; | |
d41f73b7 MD |
3005 | } |
3006 | ||
3007 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
3008 | { | |
3009 | switch (consumer_data.type) { | |
3010 | case LTTNG_CONSUMER_KERNEL: | |
3011 | return lttng_kconsumer_on_recv_stream(stream); | |
7753dea8 MD |
3012 | case LTTNG_CONSUMER32_UST: |
3013 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
3014 | return lttng_ustconsumer_on_recv_stream(stream); |
3015 | default: | |
3016 | ERR("Unknown consumer_data type"); | |
3017 | assert(0); | |
3018 | return -ENOSYS; | |
3019 | } | |
3020 | } | |
e4421fec DG |
3021 | |
3022 | /* | |
3023 | * Allocate and set consumer data hash tables. | |
3024 | */ | |
3025 | void lttng_consumer_init(void) | |
3026 | { | |
d88aee68 DG |
3027 | consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
3028 | consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3029 | consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
d8ef542d | 3030 | consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
e4421fec | 3031 | } |
7735ef9e DG |
3032 | |
3033 | /* | |
3034 | * Process the ADD_RELAYD command receive by a consumer. | |
3035 | * | |
3036 | * This will create a relayd socket pair and add it to the relayd hash table. | |
3037 | * The caller MUST acquire a RCU read side lock before calling it. | |
3038 | */ | |
3039 | int consumer_add_relayd_socket(int net_seq_idx, int sock_type, | |
3040 | struct lttng_consumer_local_data *ctx, int sock, | |
6151a90f JD |
3041 | struct pollfd *consumer_sockpoll, |
3042 | struct lttcomm_relayd_sock *relayd_sock, unsigned int sessiond_id) | |
7735ef9e | 3043 | { |
cd2b09ed | 3044 | int fd = -1, ret = -1, relayd_created = 0; |
f50f23d9 | 3045 | enum lttng_error_code ret_code = LTTNG_OK; |
d4298c99 | 3046 | struct consumer_relayd_sock_pair *relayd = NULL; |
7735ef9e | 3047 | |
6151a90f JD |
3048 | assert(ctx); |
3049 | assert(relayd_sock); | |
3050 | ||
7735ef9e DG |
3051 | DBG("Consumer adding relayd socket (idx: %d)", net_seq_idx); |
3052 | ||
f50f23d9 DG |
3053 | /* First send a status message before receiving the fds. */ |
3054 | ret = consumer_send_status_msg(sock, ret_code); | |
3055 | if (ret < 0) { | |
3056 | /* Somehow, the session daemon is not responding anymore. */ | |
3057 | goto error; | |
3058 | } | |
3059 | ||
7735ef9e DG |
3060 | /* Get relayd reference if exists. */ |
3061 | relayd = consumer_find_relayd(net_seq_idx); | |
3062 | if (relayd == NULL) { | |
3063 | /* Not found. Allocate one. */ | |
3064 | relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); | |
3065 | if (relayd == NULL) { | |
f73fabfd | 3066 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR); |
59e71485 | 3067 | ret = -1; |
7735ef9e DG |
3068 | goto error; |
3069 | } | |
f7079f67 | 3070 | relayd->sessiond_session_id = (uint64_t) sessiond_id; |
cd2b09ed | 3071 | relayd_created = 1; |
7735ef9e DG |
3072 | } |
3073 | ||
3074 | /* Poll on consumer socket. */ | |
3075 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
3076 | ret = -EINTR; | |
3077 | goto error; | |
3078 | } | |
3079 | ||
3080 | /* Get relayd socket from session daemon */ | |
3081 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); | |
3082 | if (ret != sizeof(fd)) { | |
f73fabfd | 3083 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); |
7735ef9e | 3084 | ret = -1; |
4028eeb9 | 3085 | fd = -1; /* Just in case it gets set with an invalid value. */ |
ffe60014 | 3086 | goto error_close; |
7735ef9e DG |
3087 | } |
3088 | ||
f50f23d9 DG |
3089 | /* We have the fds without error. Send status back. */ |
3090 | ret = consumer_send_status_msg(sock, ret_code); | |
3091 | if (ret < 0) { | |
3092 | /* Somehow, the session daemon is not responding anymore. */ | |
3093 | goto error; | |
3094 | } | |
3095 | ||
7735ef9e DG |
3096 | /* Copy socket information and received FD */ |
3097 | switch (sock_type) { | |
3098 | case LTTNG_STREAM_CONTROL: | |
3099 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3100 | lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock); |
3101 | ret = lttcomm_create_sock(&relayd->control_sock.sock); | |
4028eeb9 | 3102 | /* Immediately try to close the created socket if valid. */ |
6151a90f JD |
3103 | if (relayd->control_sock.sock.fd >= 0) { |
3104 | if (close(relayd->control_sock.sock.fd)) { | |
4028eeb9 DG |
3105 | PERROR("close relayd control socket"); |
3106 | } | |
7735ef9e | 3107 | } |
4028eeb9 | 3108 | /* Handle create_sock error. */ |
f66c074c | 3109 | if (ret < 0) { |
4028eeb9 | 3110 | goto error; |
f66c074c | 3111 | } |
7735ef9e DG |
3112 | |
3113 | /* Assign new file descriptor */ | |
6151a90f JD |
3114 | relayd->control_sock.sock.fd = fd; |
3115 | /* Assign version values. */ | |
3116 | relayd->control_sock.major = relayd_sock->major; | |
3117 | relayd->control_sock.minor = relayd_sock->minor; | |
c5b6f4f0 DG |
3118 | |
3119 | /* | |
59e71485 DG |
3120 | * Create a session on the relayd and store the returned id. Lock the |
3121 | * control socket mutex if the relayd was NOT created before. | |
c5b6f4f0 | 3122 | */ |
59e71485 DG |
3123 | if (!relayd_created) { |
3124 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3125 | } | |
c5b6f4f0 | 3126 | ret = relayd_create_session(&relayd->control_sock, |
f7079f67 | 3127 | &relayd->relayd_session_id); |
59e71485 DG |
3128 | if (!relayd_created) { |
3129 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3130 | } | |
c5b6f4f0 | 3131 | if (ret < 0) { |
ffe60014 DG |
3132 | /* |
3133 | * Close all sockets of a relayd object. It will be freed if it was | |
3134 | * created at the error code path or else it will be garbage | |
3135 | * collect. | |
3136 | */ | |
3137 | (void) relayd_close(&relayd->control_sock); | |
3138 | (void) relayd_close(&relayd->data_sock); | |
c5b6f4f0 DG |
3139 | goto error; |
3140 | } | |
3141 | ||
7735ef9e DG |
3142 | break; |
3143 | case LTTNG_STREAM_DATA: | |
3144 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3145 | lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock); |
3146 | ret = lttcomm_create_sock(&relayd->data_sock.sock); | |
4028eeb9 | 3147 | /* Immediately try to close the created socket if valid. */ |
6151a90f JD |
3148 | if (relayd->data_sock.sock.fd >= 0) { |
3149 | if (close(relayd->data_sock.sock.fd)) { | |
4028eeb9 DG |
3150 | PERROR("close relayd data socket"); |
3151 | } | |
7735ef9e | 3152 | } |
4028eeb9 | 3153 | /* Handle create_sock error. */ |
f66c074c | 3154 | if (ret < 0) { |
4028eeb9 | 3155 | goto error; |
f66c074c | 3156 | } |
7735ef9e DG |
3157 | |
3158 | /* Assign new file descriptor */ | |
6151a90f JD |
3159 | relayd->data_sock.sock.fd = fd; |
3160 | /* Assign version values. */ | |
3161 | relayd->data_sock.major = relayd_sock->major; | |
3162 | relayd->data_sock.minor = relayd_sock->minor; | |
7735ef9e DG |
3163 | break; |
3164 | default: | |
3165 | ERR("Unknown relayd socket type (%d)", sock_type); | |
59e71485 | 3166 | ret = -1; |
7735ef9e DG |
3167 | goto error; |
3168 | } | |
3169 | ||
d88aee68 | 3170 | DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)", |
7735ef9e DG |
3171 | sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", |
3172 | relayd->net_seq_idx, fd); | |
3173 | ||
3174 | /* | |
3175 | * Add relayd socket pair to consumer data hashtable. If object already | |
3176 | * exists or on error, the function gracefully returns. | |
3177 | */ | |
d09e1200 | 3178 | add_relayd(relayd); |
7735ef9e DG |
3179 | |
3180 | /* All good! */ | |
4028eeb9 | 3181 | return 0; |
7735ef9e DG |
3182 | |
3183 | error: | |
4028eeb9 DG |
3184 | /* Close received socket if valid. */ |
3185 | if (fd >= 0) { | |
3186 | if (close(fd)) { | |
3187 | PERROR("close received socket"); | |
3188 | } | |
3189 | } | |
cd2b09ed | 3190 | |
ffe60014 | 3191 | error_close: |
cd2b09ed | 3192 | if (relayd_created) { |
cd2b09ed DG |
3193 | free(relayd); |
3194 | } | |
3195 | ||
7735ef9e DG |
3196 | return ret; |
3197 | } | |
ca22feea | 3198 | |
4e9a4686 DG |
3199 | /* |
3200 | * Try to lock the stream mutex. | |
3201 | * | |
3202 | * On success, 1 is returned else 0 indicating that the mutex is NOT lock. | |
3203 | */ | |
3204 | static int stream_try_lock(struct lttng_consumer_stream *stream) | |
3205 | { | |
3206 | int ret; | |
3207 | ||
3208 | assert(stream); | |
3209 | ||
3210 | /* | |
3211 | * Try to lock the stream mutex. On failure, we know that the stream is | |
3212 | * being used else where hence there is data still being extracted. | |
3213 | */ | |
3214 | ret = pthread_mutex_trylock(&stream->lock); | |
3215 | if (ret) { | |
3216 | /* For both EBUSY and EINVAL error, the mutex is NOT locked. */ | |
3217 | ret = 0; | |
3218 | goto end; | |
3219 | } | |
3220 | ||
3221 | ret = 1; | |
3222 | ||
3223 | end: | |
3224 | return ret; | |
3225 | } | |
3226 | ||
f7079f67 DG |
3227 | /* |
3228 | * Search for a relayd associated to the session id and return the reference. | |
3229 | * | |
3230 | * A rcu read side lock MUST be acquire before calling this function and locked | |
3231 | * until the relayd object is no longer necessary. | |
3232 | */ | |
3233 | static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id) | |
3234 | { | |
3235 | struct lttng_ht_iter iter; | |
f7079f67 | 3236 | struct consumer_relayd_sock_pair *relayd = NULL; |
f7079f67 DG |
3237 | |
3238 | /* Iterate over all relayd since they are indexed by net_seq_idx. */ | |
3239 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
3240 | node.node) { | |
18261bd1 DG |
3241 | /* |
3242 | * Check by sessiond id which is unique here where the relayd session | |
3243 | * id might not be when having multiple relayd. | |
3244 | */ | |
3245 | if (relayd->sessiond_session_id == id) { | |
f7079f67 | 3246 | /* Found the relayd. There can be only one per id. */ |
18261bd1 | 3247 | goto found; |
f7079f67 DG |
3248 | } |
3249 | } | |
3250 | ||
18261bd1 DG |
3251 | return NULL; |
3252 | ||
3253 | found: | |
f7079f67 DG |
3254 | return relayd; |
3255 | } | |
3256 | ||
ca22feea DG |
3257 | /* |
3258 | * Check if for a given session id there is still data needed to be extract | |
3259 | * from the buffers. | |
3260 | * | |
6d805429 | 3261 | * Return 1 if data is pending or else 0 meaning ready to be read. |
ca22feea | 3262 | */ |
6d805429 | 3263 | int consumer_data_pending(uint64_t id) |
ca22feea DG |
3264 | { |
3265 | int ret; | |
3266 | struct lttng_ht_iter iter; | |
3267 | struct lttng_ht *ht; | |
3268 | struct lttng_consumer_stream *stream; | |
f7079f67 | 3269 | struct consumer_relayd_sock_pair *relayd = NULL; |
6d805429 | 3270 | int (*data_pending)(struct lttng_consumer_stream *); |
ca22feea | 3271 | |
6d805429 | 3272 | DBG("Consumer data pending command on session id %" PRIu64, id); |
ca22feea | 3273 | |
6f6eda74 | 3274 | rcu_read_lock(); |
ca22feea DG |
3275 | pthread_mutex_lock(&consumer_data.lock); |
3276 | ||
3277 | switch (consumer_data.type) { | |
3278 | case LTTNG_CONSUMER_KERNEL: | |
6d805429 | 3279 | data_pending = lttng_kconsumer_data_pending; |
ca22feea DG |
3280 | break; |
3281 | case LTTNG_CONSUMER32_UST: | |
3282 | case LTTNG_CONSUMER64_UST: | |
6d805429 | 3283 | data_pending = lttng_ustconsumer_data_pending; |
ca22feea DG |
3284 | break; |
3285 | default: | |
3286 | ERR("Unknown consumer data type"); | |
3287 | assert(0); | |
3288 | } | |
3289 | ||
3290 | /* Ease our life a bit */ | |
3291 | ht = consumer_data.stream_list_ht; | |
3292 | ||
f7079f67 DG |
3293 | relayd = find_relayd_by_session_id(id); |
3294 | if (relayd) { | |
3295 | /* Send init command for data pending. */ | |
3296 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3297 | ret = relayd_begin_data_pending(&relayd->control_sock, | |
3298 | relayd->relayd_session_id); | |
3299 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3300 | if (ret < 0) { | |
3301 | /* Communication error thus the relayd so no data pending. */ | |
3302 | goto data_not_pending; | |
3303 | } | |
3304 | } | |
3305 | ||
c8f59ee5 | 3306 | cds_lfht_for_each_entry_duplicate(ht->ht, |
d88aee68 DG |
3307 | ht->hash_fct(&id, lttng_ht_seed), |
3308 | ht->match_fct, &id, | |
ca22feea | 3309 | &iter.iter, stream, node_session_id.node) { |
4e9a4686 DG |
3310 | /* If this call fails, the stream is being used hence data pending. */ |
3311 | ret = stream_try_lock(stream); | |
3312 | if (!ret) { | |
f7079f67 | 3313 | goto data_pending; |
ca22feea | 3314 | } |
ca22feea | 3315 | |
4e9a4686 DG |
3316 | /* |
3317 | * A removed node from the hash table indicates that the stream has | |
3318 | * been deleted thus having a guarantee that the buffers are closed | |
3319 | * on the consumer side. However, data can still be transmitted | |
3320 | * over the network so don't skip the relayd check. | |
3321 | */ | |
3322 | ret = cds_lfht_is_node_deleted(&stream->node.node); | |
3323 | if (!ret) { | |
3324 | /* Check the stream if there is data in the buffers. */ | |
6d805429 DG |
3325 | ret = data_pending(stream); |
3326 | if (ret == 1) { | |
4e9a4686 | 3327 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3328 | goto data_pending; |
4e9a4686 DG |
3329 | } |
3330 | } | |
3331 | ||
3332 | /* Relayd check */ | |
f7079f67 | 3333 | if (relayd) { |
c8f59ee5 DG |
3334 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
3335 | if (stream->metadata_flag) { | |
ad7051c0 DG |
3336 | ret = relayd_quiescent_control(&relayd->control_sock, |
3337 | stream->relayd_stream_id); | |
c8f59ee5 | 3338 | } else { |
6d805429 | 3339 | ret = relayd_data_pending(&relayd->control_sock, |
39df6d9f DG |
3340 | stream->relayd_stream_id, |
3341 | stream->next_net_seq_num - 1); | |
c8f59ee5 DG |
3342 | } |
3343 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
6d805429 | 3344 | if (ret == 1) { |
4e9a4686 | 3345 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3346 | goto data_pending; |
c8f59ee5 DG |
3347 | } |
3348 | } | |
4e9a4686 | 3349 | pthread_mutex_unlock(&stream->lock); |
c8f59ee5 | 3350 | } |
ca22feea | 3351 | |
f7079f67 DG |
3352 | if (relayd) { |
3353 | unsigned int is_data_inflight = 0; | |
3354 | ||
3355 | /* Send init command for data pending. */ | |
3356 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3357 | ret = relayd_end_data_pending(&relayd->control_sock, | |
3358 | relayd->relayd_session_id, &is_data_inflight); | |
3359 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
bdd88757 | 3360 | if (ret < 0) { |
f7079f67 DG |
3361 | goto data_not_pending; |
3362 | } | |
bdd88757 DG |
3363 | if (is_data_inflight) { |
3364 | goto data_pending; | |
3365 | } | |
f7079f67 DG |
3366 | } |
3367 | ||
ca22feea | 3368 | /* |
f7079f67 DG |
3369 | * Finding _no_ node in the hash table and no inflight data means that the |
3370 | * stream(s) have been removed thus data is guaranteed to be available for | |
3371 | * analysis from the trace files. | |
ca22feea DG |
3372 | */ |
3373 | ||
f7079f67 | 3374 | data_not_pending: |
ca22feea DG |
3375 | /* Data is available to be read by a viewer. */ |
3376 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3377 | rcu_read_unlock(); |
6d805429 | 3378 | return 0; |
ca22feea | 3379 | |
f7079f67 | 3380 | data_pending: |
ca22feea DG |
3381 | /* Data is still being extracted from buffers. */ |
3382 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3383 | rcu_read_unlock(); |
6d805429 | 3384 | return 1; |
ca22feea | 3385 | } |
f50f23d9 DG |
3386 | |
3387 | /* | |
3388 | * Send a ret code status message to the sessiond daemon. | |
3389 | * | |
3390 | * Return the sendmsg() return value. | |
3391 | */ | |
3392 | int consumer_send_status_msg(int sock, int ret_code) | |
3393 | { | |
3394 | struct lttcomm_consumer_status_msg msg; | |
3395 | ||
3396 | msg.ret_code = ret_code; | |
3397 | ||
3398 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3399 | } | |
ffe60014 DG |
3400 | |
3401 | /* | |
3402 | * Send a channel status message to the sessiond daemon. | |
3403 | * | |
3404 | * Return the sendmsg() return value. | |
3405 | */ | |
3406 | int consumer_send_status_channel(int sock, | |
3407 | struct lttng_consumer_channel *channel) | |
3408 | { | |
3409 | struct lttcomm_consumer_status_channel msg; | |
3410 | ||
3411 | assert(sock >= 0); | |
3412 | ||
3413 | if (!channel) { | |
3414 | msg.ret_code = -LTTNG_ERR_UST_CHAN_FAIL; | |
3415 | } else { | |
3416 | msg.ret_code = LTTNG_OK; | |
3417 | msg.key = channel->key; | |
3418 | msg.stream_count = channel->streams.count; | |
3419 | } | |
3420 | ||
3421 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3422 | } |