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 | |
51a9e1c7 | 33 | #include <bin/lttng-consumerd/health-consumerd.h> |
990570ed | 34 | #include <common/common.h> |
fb3a43a9 DG |
35 | #include <common/utils.h> |
36 | #include <common/compat/poll.h> | |
309167d2 | 37 | #include <common/index/index.h> |
10a8a223 | 38 | #include <common/kernel-ctl/kernel-ctl.h> |
00e2e675 | 39 | #include <common/sessiond-comm/relayd.h> |
10a8a223 DG |
40 | #include <common/sessiond-comm/sessiond-comm.h> |
41 | #include <common/kernel-consumer/kernel-consumer.h> | |
00e2e675 | 42 | #include <common/relayd/relayd.h> |
10a8a223 | 43 | #include <common/ust-consumer/ust-consumer.h> |
d3e2ba59 | 44 | #include <common/consumer-timer.h> |
10a8a223 DG |
45 | |
46 | #include "consumer.h" | |
1d1a276c | 47 | #include "consumer-stream.h" |
ede905e6 | 48 | #include "consumer-testpoint.h" |
3bd1e081 MD |
49 | |
50 | struct lttng_consumer_global_data consumer_data = { | |
3bd1e081 MD |
51 | .stream_count = 0, |
52 | .need_update = 1, | |
53 | .type = LTTNG_CONSUMER_UNKNOWN, | |
54 | }; | |
55 | ||
d8ef542d MD |
56 | enum consumer_channel_action { |
57 | CONSUMER_CHANNEL_ADD, | |
a0cbdd2e | 58 | CONSUMER_CHANNEL_DEL, |
d8ef542d MD |
59 | CONSUMER_CHANNEL_QUIT, |
60 | }; | |
61 | ||
62 | struct consumer_channel_msg { | |
63 | enum consumer_channel_action action; | |
a0cbdd2e MD |
64 | struct lttng_consumer_channel *chan; /* add */ |
65 | uint64_t key; /* del */ | |
d8ef542d MD |
66 | }; |
67 | ||
3bd1e081 MD |
68 | /* |
69 | * Flag to inform the polling thread to quit when all fd hung up. Updated by | |
70 | * the consumer_thread_receive_fds when it notices that all fds has hung up. | |
71 | * Also updated by the signal handler (consumer_should_exit()). Read by the | |
72 | * polling threads. | |
73 | */ | |
a98dae5f | 74 | volatile int consumer_quit; |
3bd1e081 | 75 | |
43c34bc3 | 76 | /* |
43c34bc3 DG |
77 | * Global hash table containing respectively metadata and data streams. The |
78 | * stream element in this ht should only be updated by the metadata poll thread | |
79 | * for the metadata and the data poll thread for the data. | |
80 | */ | |
40dc48e0 DG |
81 | static struct lttng_ht *metadata_ht; |
82 | static struct lttng_ht *data_ht; | |
43c34bc3 | 83 | |
acdb9057 DG |
84 | /* |
85 | * Notify a thread lttng pipe to poll back again. This usually means that some | |
86 | * global state has changed so we just send back the thread in a poll wait | |
87 | * call. | |
88 | */ | |
89 | static void notify_thread_lttng_pipe(struct lttng_pipe *pipe) | |
90 | { | |
91 | struct lttng_consumer_stream *null_stream = NULL; | |
92 | ||
93 | assert(pipe); | |
94 | ||
95 | (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream)); | |
96 | } | |
97 | ||
5c635c72 MD |
98 | static void notify_health_quit_pipe(int *pipe) |
99 | { | |
6cd525e8 | 100 | ssize_t ret; |
5c635c72 | 101 | |
6cd525e8 MD |
102 | ret = lttng_write(pipe[1], "4", 1); |
103 | if (ret < 1) { | |
5c635c72 MD |
104 | PERROR("write consumer health quit"); |
105 | } | |
106 | } | |
107 | ||
d8ef542d MD |
108 | static void notify_channel_pipe(struct lttng_consumer_local_data *ctx, |
109 | struct lttng_consumer_channel *chan, | |
a0cbdd2e | 110 | uint64_t key, |
d8ef542d MD |
111 | enum consumer_channel_action action) |
112 | { | |
113 | struct consumer_channel_msg msg; | |
6cd525e8 | 114 | ssize_t ret; |
d8ef542d | 115 | |
e56251fc DG |
116 | memset(&msg, 0, sizeof(msg)); |
117 | ||
d8ef542d MD |
118 | msg.action = action; |
119 | msg.chan = chan; | |
f21dae48 | 120 | msg.key = key; |
6cd525e8 MD |
121 | ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg)); |
122 | if (ret < sizeof(msg)) { | |
123 | PERROR("notify_channel_pipe write error"); | |
124 | } | |
d8ef542d MD |
125 | } |
126 | ||
a0cbdd2e MD |
127 | void notify_thread_del_channel(struct lttng_consumer_local_data *ctx, |
128 | uint64_t key) | |
129 | { | |
130 | notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL); | |
131 | } | |
132 | ||
d8ef542d MD |
133 | static int read_channel_pipe(struct lttng_consumer_local_data *ctx, |
134 | struct lttng_consumer_channel **chan, | |
a0cbdd2e | 135 | uint64_t *key, |
d8ef542d MD |
136 | enum consumer_channel_action *action) |
137 | { | |
138 | struct consumer_channel_msg msg; | |
6cd525e8 | 139 | ssize_t ret; |
d8ef542d | 140 | |
6cd525e8 MD |
141 | ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg)); |
142 | if (ret < sizeof(msg)) { | |
143 | ret = -1; | |
144 | goto error; | |
d8ef542d | 145 | } |
6cd525e8 MD |
146 | *action = msg.action; |
147 | *chan = msg.chan; | |
148 | *key = msg.key; | |
149 | error: | |
150 | return (int) ret; | |
d8ef542d MD |
151 | } |
152 | ||
6217f8c7 DG |
153 | /* |
154 | * Cleanup the stream list of a channel. Those streams are not yet globally | |
155 | * visible | |
156 | */ | |
157 | static void clean_channel_stream_list(struct lttng_consumer_channel *channel) | |
158 | { | |
159 | struct lttng_consumer_stream *stream, *stmp; | |
160 | ||
161 | assert(channel); | |
162 | ||
163 | /* Delete streams that might have been left in the stream list. */ | |
164 | cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head, | |
165 | send_node) { | |
166 | cds_list_del(&stream->send_node); | |
167 | /* | |
168 | * Once a stream is added to this list, the buffers were created so we | |
169 | * have a guarantee that this call will succeed. Setting the monitor | |
170 | * mode to 0 so we don't lock nor try to delete the stream from the | |
171 | * global hash table. | |
172 | */ | |
173 | stream->monitor = 0; | |
174 | consumer_stream_destroy(stream, NULL); | |
175 | } | |
176 | } | |
177 | ||
3bd1e081 MD |
178 | /* |
179 | * Find a stream. The consumer_data.lock must be locked during this | |
180 | * call. | |
181 | */ | |
d88aee68 | 182 | static struct lttng_consumer_stream *find_stream(uint64_t key, |
8389e4f8 | 183 | struct lttng_ht *ht) |
3bd1e081 | 184 | { |
e4421fec | 185 | struct lttng_ht_iter iter; |
d88aee68 | 186 | struct lttng_ht_node_u64 *node; |
e4421fec | 187 | struct lttng_consumer_stream *stream = NULL; |
3bd1e081 | 188 | |
8389e4f8 DG |
189 | assert(ht); |
190 | ||
d88aee68 DG |
191 | /* -1ULL keys are lookup failures */ |
192 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 193 | return NULL; |
7a57cf92 | 194 | } |
e4421fec | 195 | |
6065ceec DG |
196 | rcu_read_lock(); |
197 | ||
d88aee68 DG |
198 | lttng_ht_lookup(ht, &key, &iter); |
199 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
200 | if (node != NULL) { |
201 | stream = caa_container_of(node, struct lttng_consumer_stream, node); | |
3bd1e081 | 202 | } |
e4421fec | 203 | |
6065ceec DG |
204 | rcu_read_unlock(); |
205 | ||
e4421fec | 206 | return stream; |
3bd1e081 MD |
207 | } |
208 | ||
da009f2c | 209 | static void steal_stream_key(uint64_t key, struct lttng_ht *ht) |
7ad0a0cb MD |
210 | { |
211 | struct lttng_consumer_stream *stream; | |
212 | ||
04253271 | 213 | rcu_read_lock(); |
ffe60014 | 214 | stream = find_stream(key, ht); |
04253271 | 215 | if (stream) { |
da009f2c | 216 | stream->key = (uint64_t) -1ULL; |
04253271 MD |
217 | /* |
218 | * We don't want the lookup to match, but we still need | |
219 | * to iterate on this stream when iterating over the hash table. Just | |
220 | * change the node key. | |
221 | */ | |
da009f2c | 222 | stream->node.key = (uint64_t) -1ULL; |
04253271 MD |
223 | } |
224 | rcu_read_unlock(); | |
7ad0a0cb MD |
225 | } |
226 | ||
d56db448 DG |
227 | /* |
228 | * Return a channel object for the given key. | |
229 | * | |
230 | * RCU read side lock MUST be acquired before calling this function and | |
231 | * protects the channel ptr. | |
232 | */ | |
d88aee68 | 233 | struct lttng_consumer_channel *consumer_find_channel(uint64_t key) |
3bd1e081 | 234 | { |
e4421fec | 235 | struct lttng_ht_iter iter; |
d88aee68 | 236 | struct lttng_ht_node_u64 *node; |
e4421fec | 237 | struct lttng_consumer_channel *channel = NULL; |
3bd1e081 | 238 | |
d88aee68 DG |
239 | /* -1ULL keys are lookup failures */ |
240 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 241 | return NULL; |
7a57cf92 | 242 | } |
e4421fec | 243 | |
d88aee68 DG |
244 | lttng_ht_lookup(consumer_data.channel_ht, &key, &iter); |
245 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
246 | if (node != NULL) { |
247 | channel = caa_container_of(node, struct lttng_consumer_channel, node); | |
3bd1e081 | 248 | } |
e4421fec DG |
249 | |
250 | return channel; | |
3bd1e081 MD |
251 | } |
252 | ||
ffe60014 | 253 | static void free_channel_rcu(struct rcu_head *head) |
702b1ea4 | 254 | { |
d88aee68 DG |
255 | struct lttng_ht_node_u64 *node = |
256 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
ffe60014 DG |
257 | struct lttng_consumer_channel *channel = |
258 | caa_container_of(node, struct lttng_consumer_channel, node); | |
702b1ea4 | 259 | |
ffe60014 | 260 | free(channel); |
702b1ea4 MD |
261 | } |
262 | ||
00e2e675 DG |
263 | /* |
264 | * RCU protected relayd socket pair free. | |
265 | */ | |
ffe60014 | 266 | static void free_relayd_rcu(struct rcu_head *head) |
00e2e675 | 267 | { |
d88aee68 DG |
268 | struct lttng_ht_node_u64 *node = |
269 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
00e2e675 DG |
270 | struct consumer_relayd_sock_pair *relayd = |
271 | caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
272 | ||
8994307f DG |
273 | /* |
274 | * Close all sockets. This is done in the call RCU since we don't want the | |
275 | * socket fds to be reassigned thus potentially creating bad state of the | |
276 | * relayd object. | |
277 | * | |
278 | * We do not have to lock the control socket mutex here since at this stage | |
279 | * there is no one referencing to this relayd object. | |
280 | */ | |
281 | (void) relayd_close(&relayd->control_sock); | |
282 | (void) relayd_close(&relayd->data_sock); | |
283 | ||
00e2e675 DG |
284 | free(relayd); |
285 | } | |
286 | ||
287 | /* | |
288 | * Destroy and free relayd socket pair object. | |
00e2e675 | 289 | */ |
51230d70 | 290 | void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
291 | { |
292 | int ret; | |
293 | struct lttng_ht_iter iter; | |
294 | ||
173af62f DG |
295 | if (relayd == NULL) { |
296 | return; | |
297 | } | |
298 | ||
00e2e675 DG |
299 | DBG("Consumer destroy and close relayd socket pair"); |
300 | ||
301 | iter.iter.node = &relayd->node.node; | |
302 | ret = lttng_ht_del(consumer_data.relayd_ht, &iter); | |
173af62f | 303 | if (ret != 0) { |
8994307f | 304 | /* We assume the relayd is being or is destroyed */ |
173af62f DG |
305 | return; |
306 | } | |
00e2e675 | 307 | |
00e2e675 | 308 | /* RCU free() call */ |
ffe60014 DG |
309 | call_rcu(&relayd->node.head, free_relayd_rcu); |
310 | } | |
311 | ||
312 | /* | |
313 | * Remove a channel from the global list protected by a mutex. This function is | |
314 | * also responsible for freeing its data structures. | |
315 | */ | |
316 | void consumer_del_channel(struct lttng_consumer_channel *channel) | |
317 | { | |
318 | int ret; | |
319 | struct lttng_ht_iter iter; | |
320 | ||
d88aee68 | 321 | DBG("Consumer delete channel key %" PRIu64, channel->key); |
ffe60014 DG |
322 | |
323 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 324 | pthread_mutex_lock(&channel->lock); |
ffe60014 | 325 | |
6217f8c7 DG |
326 | /* Destroy streams that might have been left in the stream list. */ |
327 | clean_channel_stream_list(channel); | |
51e762e5 | 328 | |
d3e2ba59 JD |
329 | if (channel->live_timer_enabled == 1) { |
330 | consumer_timer_live_stop(channel); | |
331 | } | |
332 | ||
ffe60014 DG |
333 | switch (consumer_data.type) { |
334 | case LTTNG_CONSUMER_KERNEL: | |
335 | break; | |
336 | case LTTNG_CONSUMER32_UST: | |
337 | case LTTNG_CONSUMER64_UST: | |
338 | lttng_ustconsumer_del_channel(channel); | |
339 | break; | |
340 | default: | |
341 | ERR("Unknown consumer_data type"); | |
342 | assert(0); | |
343 | goto end; | |
344 | } | |
345 | ||
346 | rcu_read_lock(); | |
347 | iter.iter.node = &channel->node.node; | |
348 | ret = lttng_ht_del(consumer_data.channel_ht, &iter); | |
349 | assert(!ret); | |
350 | rcu_read_unlock(); | |
351 | ||
352 | call_rcu(&channel->node.head, free_channel_rcu); | |
353 | end: | |
a9838785 | 354 | pthread_mutex_unlock(&channel->lock); |
ffe60014 | 355 | pthread_mutex_unlock(&consumer_data.lock); |
00e2e675 DG |
356 | } |
357 | ||
228b5bf7 DG |
358 | /* |
359 | * Iterate over the relayd hash table and destroy each element. Finally, | |
360 | * destroy the whole hash table. | |
361 | */ | |
362 | static void cleanup_relayd_ht(void) | |
363 | { | |
364 | struct lttng_ht_iter iter; | |
365 | struct consumer_relayd_sock_pair *relayd; | |
366 | ||
367 | rcu_read_lock(); | |
368 | ||
369 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
370 | node.node) { | |
51230d70 | 371 | consumer_destroy_relayd(relayd); |
228b5bf7 DG |
372 | } |
373 | ||
228b5bf7 | 374 | rcu_read_unlock(); |
36b588ed MD |
375 | |
376 | lttng_ht_destroy(consumer_data.relayd_ht); | |
228b5bf7 DG |
377 | } |
378 | ||
8994307f DG |
379 | /* |
380 | * Update the end point status of all streams having the given network sequence | |
381 | * index (relayd index). | |
382 | * | |
383 | * It's atomically set without having the stream mutex locked which is fine | |
384 | * because we handle the write/read race with a pipe wakeup for each thread. | |
385 | */ | |
da009f2c | 386 | static void update_endpoint_status_by_netidx(uint64_t net_seq_idx, |
8994307f DG |
387 | enum consumer_endpoint_status status) |
388 | { | |
389 | struct lttng_ht_iter iter; | |
390 | struct lttng_consumer_stream *stream; | |
391 | ||
da009f2c | 392 | DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx); |
8994307f DG |
393 | |
394 | rcu_read_lock(); | |
395 | ||
396 | /* Let's begin with metadata */ | |
397 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
398 | if (stream->net_seq_idx == net_seq_idx) { | |
399 | uatomic_set(&stream->endpoint_status, status); | |
400 | DBG("Delete flag set to metadata stream %d", stream->wait_fd); | |
401 | } | |
402 | } | |
403 | ||
404 | /* Follow up by the data streams */ | |
405 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
406 | if (stream->net_seq_idx == net_seq_idx) { | |
407 | uatomic_set(&stream->endpoint_status, status); | |
408 | DBG("Delete flag set to data stream %d", stream->wait_fd); | |
409 | } | |
410 | } | |
411 | rcu_read_unlock(); | |
412 | } | |
413 | ||
414 | /* | |
415 | * Cleanup a relayd object by flagging every associated streams for deletion, | |
416 | * destroying the object meaning removing it from the relayd hash table, | |
417 | * closing the sockets and freeing the memory in a RCU call. | |
418 | * | |
419 | * If a local data context is available, notify the threads that the streams' | |
420 | * state have changed. | |
421 | */ | |
422 | static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd, | |
423 | struct lttng_consumer_local_data *ctx) | |
424 | { | |
da009f2c | 425 | uint64_t netidx; |
8994307f DG |
426 | |
427 | assert(relayd); | |
428 | ||
9617607b DG |
429 | DBG("Cleaning up relayd sockets"); |
430 | ||
8994307f DG |
431 | /* Save the net sequence index before destroying the object */ |
432 | netidx = relayd->net_seq_idx; | |
433 | ||
434 | /* | |
435 | * Delete the relayd from the relayd hash table, close the sockets and free | |
436 | * the object in a RCU call. | |
437 | */ | |
51230d70 | 438 | consumer_destroy_relayd(relayd); |
8994307f DG |
439 | |
440 | /* Set inactive endpoint to all streams */ | |
441 | update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE); | |
442 | ||
443 | /* | |
444 | * With a local data context, notify the threads that the streams' state | |
445 | * have changed. The write() action on the pipe acts as an "implicit" | |
446 | * memory barrier ordering the updates of the end point status from the | |
447 | * read of this status which happens AFTER receiving this notify. | |
448 | */ | |
449 | if (ctx) { | |
acdb9057 | 450 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
13886d2d | 451 | notify_thread_lttng_pipe(ctx->consumer_metadata_pipe); |
8994307f DG |
452 | } |
453 | } | |
454 | ||
a6ba4fe1 DG |
455 | /* |
456 | * Flag a relayd socket pair for destruction. Destroy it if the refcount | |
457 | * reaches zero. | |
458 | * | |
459 | * RCU read side lock MUST be aquired before calling this function. | |
460 | */ | |
461 | void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) | |
462 | { | |
463 | assert(relayd); | |
464 | ||
465 | /* Set destroy flag for this object */ | |
466 | uatomic_set(&relayd->destroy_flag, 1); | |
467 | ||
468 | /* Destroy the relayd if refcount is 0 */ | |
469 | if (uatomic_read(&relayd->refcount) == 0) { | |
51230d70 | 470 | consumer_destroy_relayd(relayd); |
a6ba4fe1 DG |
471 | } |
472 | } | |
473 | ||
3bd1e081 | 474 | /* |
1d1a276c DG |
475 | * Completly destroy stream from every visiable data structure and the given |
476 | * hash table if one. | |
477 | * | |
478 | * One this call returns, the stream object is not longer usable nor visible. | |
3bd1e081 | 479 | */ |
e316aad5 DG |
480 | void consumer_del_stream(struct lttng_consumer_stream *stream, |
481 | struct lttng_ht *ht) | |
3bd1e081 | 482 | { |
1d1a276c | 483 | consumer_stream_destroy(stream, ht); |
3bd1e081 MD |
484 | } |
485 | ||
5ab66908 MD |
486 | /* |
487 | * XXX naming of del vs destroy is all mixed up. | |
488 | */ | |
489 | void consumer_del_stream_for_data(struct lttng_consumer_stream *stream) | |
490 | { | |
491 | consumer_stream_destroy(stream, data_ht); | |
492 | } | |
493 | ||
494 | void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream) | |
495 | { | |
496 | consumer_stream_destroy(stream, metadata_ht); | |
497 | } | |
498 | ||
d88aee68 DG |
499 | struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, |
500 | uint64_t stream_key, | |
3bd1e081 | 501 | enum lttng_consumer_stream_state state, |
ffe60014 | 502 | const char *channel_name, |
6df2e2c9 | 503 | uid_t uid, |
00e2e675 | 504 | gid_t gid, |
57a269f2 | 505 | uint64_t relayd_id, |
53632229 | 506 | uint64_t session_id, |
ffe60014 DG |
507 | int cpu, |
508 | int *alloc_ret, | |
4891ece8 DG |
509 | enum consumer_channel_type type, |
510 | unsigned int monitor) | |
3bd1e081 | 511 | { |
ffe60014 | 512 | int ret; |
3bd1e081 | 513 | struct lttng_consumer_stream *stream; |
3bd1e081 | 514 | |
effcf122 | 515 | stream = zmalloc(sizeof(*stream)); |
3bd1e081 | 516 | if (stream == NULL) { |
7a57cf92 | 517 | PERROR("malloc struct lttng_consumer_stream"); |
ffe60014 | 518 | ret = -ENOMEM; |
7a57cf92 | 519 | goto end; |
3bd1e081 | 520 | } |
7a57cf92 | 521 | |
d56db448 DG |
522 | rcu_read_lock(); |
523 | ||
3bd1e081 | 524 | stream->key = stream_key; |
3bd1e081 MD |
525 | stream->out_fd = -1; |
526 | stream->out_fd_offset = 0; | |
e5d1a9b3 | 527 | stream->output_written = 0; |
3bd1e081 | 528 | stream->state = state; |
6df2e2c9 MD |
529 | stream->uid = uid; |
530 | stream->gid = gid; | |
ffe60014 | 531 | stream->net_seq_idx = relayd_id; |
53632229 | 532 | stream->session_id = session_id; |
4891ece8 | 533 | stream->monitor = monitor; |
774d490c | 534 | stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE; |
309167d2 | 535 | stream->index_fd = -1; |
53632229 | 536 | pthread_mutex_init(&stream->lock, NULL); |
58b1f425 | 537 | |
ffe60014 DG |
538 | /* If channel is the metadata, flag this stream as metadata. */ |
539 | if (type == CONSUMER_CHANNEL_TYPE_METADATA) { | |
540 | stream->metadata_flag = 1; | |
541 | /* Metadata is flat out. */ | |
542 | strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name)); | |
94d49140 JD |
543 | /* Live rendez-vous point. */ |
544 | pthread_cond_init(&stream->metadata_rdv, NULL); | |
545 | pthread_mutex_init(&stream->metadata_rdv_lock, NULL); | |
58b1f425 | 546 | } else { |
ffe60014 DG |
547 | /* Format stream name to <channel_name>_<cpu_number> */ |
548 | ret = snprintf(stream->name, sizeof(stream->name), "%s_%d", | |
549 | channel_name, cpu); | |
550 | if (ret < 0) { | |
551 | PERROR("snprintf stream name"); | |
552 | goto error; | |
553 | } | |
58b1f425 | 554 | } |
c30aaa51 | 555 | |
ffe60014 | 556 | /* Key is always the wait_fd for streams. */ |
d88aee68 | 557 | lttng_ht_node_init_u64(&stream->node, stream->key); |
ffe60014 | 558 | |
d8ef542d MD |
559 | /* Init node per channel id key */ |
560 | lttng_ht_node_init_u64(&stream->node_channel_id, channel_key); | |
561 | ||
53632229 | 562 | /* Init session id node with the stream session id */ |
d88aee68 | 563 | lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id); |
53632229 | 564 | |
07b86b52 JD |
565 | DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64 |
566 | " relayd_id %" PRIu64 ", session_id %" PRIu64, | |
567 | stream->name, stream->key, channel_key, | |
568 | stream->net_seq_idx, stream->session_id); | |
d56db448 DG |
569 | |
570 | rcu_read_unlock(); | |
3bd1e081 | 571 | return stream; |
c80048c6 MD |
572 | |
573 | error: | |
d56db448 | 574 | rcu_read_unlock(); |
c80048c6 | 575 | free(stream); |
7a57cf92 | 576 | end: |
ffe60014 DG |
577 | if (alloc_ret) { |
578 | *alloc_ret = ret; | |
579 | } | |
c80048c6 | 580 | return NULL; |
3bd1e081 MD |
581 | } |
582 | ||
583 | /* | |
584 | * Add a stream to the global list protected by a mutex. | |
585 | */ | |
5ab66908 | 586 | int consumer_add_data_stream(struct lttng_consumer_stream *stream) |
3bd1e081 | 587 | { |
5ab66908 | 588 | struct lttng_ht *ht = data_ht; |
3bd1e081 MD |
589 | int ret = 0; |
590 | ||
e316aad5 | 591 | assert(stream); |
43c34bc3 | 592 | assert(ht); |
c77fc10a | 593 | |
d88aee68 | 594 | DBG3("Adding consumer stream %" PRIu64, stream->key); |
e316aad5 DG |
595 | |
596 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 597 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 598 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 599 | pthread_mutex_lock(&stream->lock); |
b0b335c8 | 600 | rcu_read_lock(); |
e316aad5 | 601 | |
43c34bc3 | 602 | /* Steal stream identifier to avoid having streams with the same key */ |
ffe60014 | 603 | steal_stream_key(stream->key, ht); |
43c34bc3 | 604 | |
d88aee68 | 605 | lttng_ht_add_unique_u64(ht, &stream->node); |
00e2e675 | 606 | |
d8ef542d MD |
607 | lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, |
608 | &stream->node_channel_id); | |
609 | ||
ca22feea DG |
610 | /* |
611 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
612 | * the key since the HT does not use it and we allow to add redundant keys | |
613 | * into this table. | |
614 | */ | |
d88aee68 | 615 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 616 | |
e316aad5 | 617 | /* |
ffe60014 DG |
618 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
619 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
620 | * causes the count to become 0 also causes a stream to be added. The |
621 | * channel deletion will thus be triggered by the following removal of this | |
622 | * stream. | |
623 | */ | |
ffe60014 | 624 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
625 | /* Increment refcount before decrementing nb_init_stream_left */ |
626 | cmm_smp_wmb(); | |
ffe60014 | 627 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
628 | } |
629 | ||
630 | /* Update consumer data once the node is inserted. */ | |
3bd1e081 MD |
631 | consumer_data.stream_count++; |
632 | consumer_data.need_update = 1; | |
633 | ||
e316aad5 | 634 | rcu_read_unlock(); |
2e818a6a | 635 | pthread_mutex_unlock(&stream->lock); |
ec6ea7d0 | 636 | pthread_mutex_unlock(&stream->chan->timer_lock); |
a9838785 | 637 | pthread_mutex_unlock(&stream->chan->lock); |
3bd1e081 | 638 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 639 | |
3bd1e081 MD |
640 | return ret; |
641 | } | |
642 | ||
5ab66908 MD |
643 | void consumer_del_data_stream(struct lttng_consumer_stream *stream) |
644 | { | |
645 | consumer_del_stream(stream, data_ht); | |
646 | } | |
647 | ||
00e2e675 | 648 | /* |
3f8e211f DG |
649 | * Add relayd socket to global consumer data hashtable. RCU read side lock MUST |
650 | * be acquired before calling this. | |
00e2e675 | 651 | */ |
d09e1200 | 652 | static int add_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
653 | { |
654 | int ret = 0; | |
d88aee68 | 655 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
656 | struct lttng_ht_iter iter; |
657 | ||
ffe60014 | 658 | assert(relayd); |
00e2e675 | 659 | |
00e2e675 | 660 | lttng_ht_lookup(consumer_data.relayd_ht, |
d88aee68 DG |
661 | &relayd->net_seq_idx, &iter); |
662 | node = lttng_ht_iter_get_node_u64(&iter); | |
00e2e675 | 663 | if (node != NULL) { |
00e2e675 DG |
664 | goto end; |
665 | } | |
d88aee68 | 666 | lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node); |
00e2e675 | 667 | |
00e2e675 DG |
668 | end: |
669 | return ret; | |
670 | } | |
671 | ||
672 | /* | |
673 | * Allocate and return a consumer relayd socket. | |
674 | */ | |
675 | struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( | |
da009f2c | 676 | uint64_t net_seq_idx) |
00e2e675 DG |
677 | { |
678 | struct consumer_relayd_sock_pair *obj = NULL; | |
679 | ||
da009f2c MD |
680 | /* net sequence index of -1 is a failure */ |
681 | if (net_seq_idx == (uint64_t) -1ULL) { | |
00e2e675 DG |
682 | goto error; |
683 | } | |
684 | ||
685 | obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); | |
686 | if (obj == NULL) { | |
687 | PERROR("zmalloc relayd sock"); | |
688 | goto error; | |
689 | } | |
690 | ||
691 | obj->net_seq_idx = net_seq_idx; | |
692 | obj->refcount = 0; | |
173af62f | 693 | obj->destroy_flag = 0; |
f96e4545 MD |
694 | obj->control_sock.sock.fd = -1; |
695 | obj->data_sock.sock.fd = -1; | |
d88aee68 | 696 | lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx); |
00e2e675 DG |
697 | pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); |
698 | ||
699 | error: | |
700 | return obj; | |
701 | } | |
702 | ||
703 | /* | |
704 | * Find a relayd socket pair in the global consumer data. | |
705 | * | |
706 | * Return the object if found else NULL. | |
b0b335c8 MD |
707 | * RCU read-side lock must be held across this call and while using the |
708 | * returned object. | |
00e2e675 | 709 | */ |
d88aee68 | 710 | struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key) |
00e2e675 DG |
711 | { |
712 | struct lttng_ht_iter iter; | |
d88aee68 | 713 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
714 | struct consumer_relayd_sock_pair *relayd = NULL; |
715 | ||
716 | /* Negative keys are lookup failures */ | |
d88aee68 | 717 | if (key == (uint64_t) -1ULL) { |
00e2e675 DG |
718 | goto error; |
719 | } | |
720 | ||
d88aee68 | 721 | lttng_ht_lookup(consumer_data.relayd_ht, &key, |
00e2e675 | 722 | &iter); |
d88aee68 | 723 | node = lttng_ht_iter_get_node_u64(&iter); |
00e2e675 DG |
724 | if (node != NULL) { |
725 | relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
726 | } | |
727 | ||
00e2e675 DG |
728 | error: |
729 | return relayd; | |
730 | } | |
731 | ||
10a50311 JD |
732 | /* |
733 | * Find a relayd and send the stream | |
734 | * | |
735 | * Returns 0 on success, < 0 on error | |
736 | */ | |
737 | int consumer_send_relayd_stream(struct lttng_consumer_stream *stream, | |
738 | char *path) | |
739 | { | |
740 | int ret = 0; | |
741 | struct consumer_relayd_sock_pair *relayd; | |
742 | ||
743 | assert(stream); | |
744 | assert(stream->net_seq_idx != -1ULL); | |
745 | assert(path); | |
746 | ||
747 | /* The stream is not metadata. Get relayd reference if exists. */ | |
748 | rcu_read_lock(); | |
749 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
750 | if (relayd != NULL) { | |
751 | /* Add stream on the relayd */ | |
752 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
753 | ret = relayd_add_stream(&relayd->control_sock, stream->name, | |
754 | path, &stream->relayd_stream_id, | |
755 | stream->chan->tracefile_size, stream->chan->tracefile_count); | |
756 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
757 | if (ret < 0) { | |
758 | goto end; | |
759 | } | |
1c20f0e2 | 760 | |
10a50311 | 761 | uatomic_inc(&relayd->refcount); |
d01178b6 | 762 | stream->sent_to_relayd = 1; |
10a50311 JD |
763 | } else { |
764 | ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.", | |
765 | stream->key, stream->net_seq_idx); | |
766 | ret = -1; | |
767 | goto end; | |
768 | } | |
769 | ||
770 | DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64, | |
771 | stream->name, stream->key, stream->net_seq_idx); | |
772 | ||
773 | end: | |
774 | rcu_read_unlock(); | |
775 | return ret; | |
776 | } | |
777 | ||
814fcae4 JD |
778 | /* |
779 | * Find a relayd and send the streams sent message | |
780 | * | |
781 | * Returns 0 on success, < 0 on error | |
782 | */ | |
783 | int consumer_send_relayd_streams_sent(uint64_t net_seq_idx) | |
784 | { | |
785 | int ret = 0; | |
786 | struct consumer_relayd_sock_pair *relayd; | |
787 | ||
788 | assert(net_seq_idx != -1ULL); | |
789 | ||
790 | /* The stream is not metadata. Get relayd reference if exists. */ | |
791 | rcu_read_lock(); | |
792 | relayd = consumer_find_relayd(net_seq_idx); | |
793 | if (relayd != NULL) { | |
794 | /* Add stream on the relayd */ | |
795 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
796 | ret = relayd_streams_sent(&relayd->control_sock); | |
797 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
798 | if (ret < 0) { | |
799 | goto end; | |
800 | } | |
801 | } else { | |
802 | ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.", | |
803 | net_seq_idx); | |
804 | ret = -1; | |
805 | goto end; | |
806 | } | |
807 | ||
808 | ret = 0; | |
809 | DBG("All streams sent relayd id %" PRIu64, net_seq_idx); | |
810 | ||
811 | end: | |
812 | rcu_read_unlock(); | |
813 | return ret; | |
814 | } | |
815 | ||
10a50311 JD |
816 | /* |
817 | * Find a relayd and close the stream | |
818 | */ | |
819 | void close_relayd_stream(struct lttng_consumer_stream *stream) | |
820 | { | |
821 | struct consumer_relayd_sock_pair *relayd; | |
822 | ||
823 | /* The stream is not metadata. Get relayd reference if exists. */ | |
824 | rcu_read_lock(); | |
825 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
826 | if (relayd) { | |
827 | consumer_stream_relayd_close(stream, relayd); | |
828 | } | |
829 | rcu_read_unlock(); | |
830 | } | |
831 | ||
00e2e675 DG |
832 | /* |
833 | * Handle stream for relayd transmission if the stream applies for network | |
834 | * streaming where the net sequence index is set. | |
835 | * | |
836 | * Return destination file descriptor or negative value on error. | |
837 | */ | |
6197aea7 | 838 | static int write_relayd_stream_header(struct lttng_consumer_stream *stream, |
1d4dfdef DG |
839 | size_t data_size, unsigned long padding, |
840 | struct consumer_relayd_sock_pair *relayd) | |
00e2e675 DG |
841 | { |
842 | int outfd = -1, ret; | |
00e2e675 DG |
843 | struct lttcomm_relayd_data_hdr data_hdr; |
844 | ||
845 | /* Safety net */ | |
846 | assert(stream); | |
6197aea7 | 847 | assert(relayd); |
00e2e675 DG |
848 | |
849 | /* Reset data header */ | |
850 | memset(&data_hdr, 0, sizeof(data_hdr)); | |
851 | ||
00e2e675 DG |
852 | if (stream->metadata_flag) { |
853 | /* Caller MUST acquire the relayd control socket lock */ | |
854 | ret = relayd_send_metadata(&relayd->control_sock, data_size); | |
855 | if (ret < 0) { | |
856 | goto error; | |
857 | } | |
858 | ||
859 | /* Metadata are always sent on the control socket. */ | |
6151a90f | 860 | outfd = relayd->control_sock.sock.fd; |
00e2e675 DG |
861 | } else { |
862 | /* Set header with stream information */ | |
863 | data_hdr.stream_id = htobe64(stream->relayd_stream_id); | |
864 | data_hdr.data_size = htobe32(data_size); | |
1d4dfdef | 865 | data_hdr.padding_size = htobe32(padding); |
39df6d9f DG |
866 | /* |
867 | * Note that net_seq_num below is assigned with the *current* value of | |
868 | * next_net_seq_num and only after that the next_net_seq_num will be | |
869 | * increment. This is why when issuing a command on the relayd using | |
870 | * this next value, 1 should always be substracted in order to compare | |
871 | * the last seen sequence number on the relayd side to the last sent. | |
872 | */ | |
3604f373 | 873 | data_hdr.net_seq_num = htobe64(stream->next_net_seq_num); |
00e2e675 DG |
874 | /* Other fields are zeroed previously */ |
875 | ||
876 | ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, | |
877 | sizeof(data_hdr)); | |
878 | if (ret < 0) { | |
879 | goto error; | |
880 | } | |
881 | ||
3604f373 DG |
882 | ++stream->next_net_seq_num; |
883 | ||
00e2e675 | 884 | /* Set to go on data socket */ |
6151a90f | 885 | outfd = relayd->data_sock.sock.fd; |
00e2e675 DG |
886 | } |
887 | ||
888 | error: | |
889 | return outfd; | |
890 | } | |
891 | ||
3bd1e081 | 892 | /* |
ffe60014 DG |
893 | * Allocate and return a new lttng_consumer_channel object using the given key |
894 | * to initialize the hash table node. | |
895 | * | |
896 | * On error, return NULL. | |
3bd1e081 | 897 | */ |
886224ff | 898 | struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, |
ffe60014 DG |
899 | uint64_t session_id, |
900 | const char *pathname, | |
901 | const char *name, | |
902 | uid_t uid, | |
903 | gid_t gid, | |
57a269f2 | 904 | uint64_t relayd_id, |
1624d5b7 JD |
905 | enum lttng_event_output output, |
906 | uint64_t tracefile_size, | |
2bba9e53 | 907 | uint64_t tracefile_count, |
1950109e | 908 | uint64_t session_id_per_pid, |
ecc48a90 JD |
909 | unsigned int monitor, |
910 | unsigned int live_timer_interval) | |
3bd1e081 MD |
911 | { |
912 | struct lttng_consumer_channel *channel; | |
3bd1e081 | 913 | |
276b26d1 | 914 | channel = zmalloc(sizeof(*channel)); |
3bd1e081 | 915 | if (channel == NULL) { |
7a57cf92 | 916 | PERROR("malloc struct lttng_consumer_channel"); |
3bd1e081 MD |
917 | goto end; |
918 | } | |
ffe60014 DG |
919 | |
920 | channel->key = key; | |
3bd1e081 | 921 | channel->refcount = 0; |
ffe60014 | 922 | channel->session_id = session_id; |
1950109e | 923 | channel->session_id_per_pid = session_id_per_pid; |
ffe60014 DG |
924 | channel->uid = uid; |
925 | channel->gid = gid; | |
926 | channel->relayd_id = relayd_id; | |
1624d5b7 JD |
927 | channel->tracefile_size = tracefile_size; |
928 | channel->tracefile_count = tracefile_count; | |
2bba9e53 | 929 | channel->monitor = monitor; |
ecc48a90 | 930 | channel->live_timer_interval = live_timer_interval; |
a9838785 | 931 | pthread_mutex_init(&channel->lock, NULL); |
ec6ea7d0 | 932 | pthread_mutex_init(&channel->timer_lock, NULL); |
ffe60014 | 933 | |
0c759fc9 DG |
934 | switch (output) { |
935 | case LTTNG_EVENT_SPLICE: | |
936 | channel->output = CONSUMER_CHANNEL_SPLICE; | |
937 | break; | |
938 | case LTTNG_EVENT_MMAP: | |
939 | channel->output = CONSUMER_CHANNEL_MMAP; | |
940 | break; | |
941 | default: | |
942 | assert(0); | |
943 | free(channel); | |
944 | channel = NULL; | |
945 | goto end; | |
946 | } | |
947 | ||
07b86b52 JD |
948 | /* |
949 | * In monitor mode, the streams associated with the channel will be put in | |
950 | * a special list ONLY owned by this channel. So, the refcount is set to 1 | |
951 | * here meaning that the channel itself has streams that are referenced. | |
952 | * | |
953 | * On a channel deletion, once the channel is no longer visible, the | |
954 | * refcount is decremented and checked for a zero value to delete it. With | |
955 | * streams in no monitor mode, it will now be safe to destroy the channel. | |
956 | */ | |
957 | if (!channel->monitor) { | |
958 | channel->refcount = 1; | |
959 | } | |
960 | ||
ffe60014 DG |
961 | strncpy(channel->pathname, pathname, sizeof(channel->pathname)); |
962 | channel->pathname[sizeof(channel->pathname) - 1] = '\0'; | |
963 | ||
964 | strncpy(channel->name, name, sizeof(channel->name)); | |
965 | channel->name[sizeof(channel->name) - 1] = '\0'; | |
966 | ||
d88aee68 | 967 | lttng_ht_node_init_u64(&channel->node, channel->key); |
d8ef542d MD |
968 | |
969 | channel->wait_fd = -1; | |
970 | ||
ffe60014 DG |
971 | CDS_INIT_LIST_HEAD(&channel->streams.head); |
972 | ||
d88aee68 | 973 | DBG("Allocated channel (key %" PRIu64 ")", channel->key) |
3bd1e081 | 974 | |
3bd1e081 MD |
975 | end: |
976 | return channel; | |
977 | } | |
978 | ||
979 | /* | |
980 | * Add a channel to the global list protected by a mutex. | |
821fffb2 DG |
981 | * |
982 | * On success 0 is returned else a negative value. | |
3bd1e081 | 983 | */ |
d8ef542d MD |
984 | int consumer_add_channel(struct lttng_consumer_channel *channel, |
985 | struct lttng_consumer_local_data *ctx) | |
3bd1e081 | 986 | { |
ffe60014 | 987 | int ret = 0; |
d88aee68 | 988 | struct lttng_ht_node_u64 *node; |
c77fc10a DG |
989 | struct lttng_ht_iter iter; |
990 | ||
3bd1e081 | 991 | pthread_mutex_lock(&consumer_data.lock); |
a9838785 | 992 | pthread_mutex_lock(&channel->lock); |
ec6ea7d0 | 993 | pthread_mutex_lock(&channel->timer_lock); |
6065ceec | 994 | rcu_read_lock(); |
c77fc10a | 995 | |
7972aab2 | 996 | lttng_ht_lookup(consumer_data.channel_ht, &channel->key, &iter); |
d88aee68 | 997 | node = lttng_ht_iter_get_node_u64(&iter); |
c77fc10a DG |
998 | if (node != NULL) { |
999 | /* Channel already exist. Ignore the insertion */ | |
d88aee68 DG |
1000 | ERR("Consumer add channel key %" PRIu64 " already exists!", |
1001 | channel->key); | |
821fffb2 | 1002 | ret = -EEXIST; |
c77fc10a DG |
1003 | goto end; |
1004 | } | |
1005 | ||
d88aee68 | 1006 | lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node); |
c77fc10a DG |
1007 | |
1008 | end: | |
6065ceec | 1009 | rcu_read_unlock(); |
ec6ea7d0 | 1010 | pthread_mutex_unlock(&channel->timer_lock); |
a9838785 | 1011 | pthread_mutex_unlock(&channel->lock); |
3bd1e081 | 1012 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 1013 | |
d8ef542d | 1014 | if (!ret && channel->wait_fd != -1 && |
10a50311 | 1015 | channel->type == CONSUMER_CHANNEL_TYPE_DATA) { |
a0cbdd2e | 1016 | notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD); |
d8ef542d | 1017 | } |
ffe60014 | 1018 | return ret; |
3bd1e081 MD |
1019 | } |
1020 | ||
1021 | /* | |
1022 | * Allocate the pollfd structure and the local view of the out fds to avoid | |
1023 | * doing a lookup in the linked list and concurrency issues when writing is | |
1024 | * needed. Called with consumer_data.lock held. | |
1025 | * | |
1026 | * Returns the number of fds in the structures. | |
1027 | */ | |
ffe60014 DG |
1028 | static int update_poll_array(struct lttng_consumer_local_data *ctx, |
1029 | struct pollfd **pollfd, struct lttng_consumer_stream **local_stream, | |
1030 | struct lttng_ht *ht) | |
3bd1e081 | 1031 | { |
3bd1e081 | 1032 | int i = 0; |
e4421fec DG |
1033 | struct lttng_ht_iter iter; |
1034 | struct lttng_consumer_stream *stream; | |
3bd1e081 | 1035 | |
ffe60014 DG |
1036 | assert(ctx); |
1037 | assert(ht); | |
1038 | assert(pollfd); | |
1039 | assert(local_stream); | |
1040 | ||
3bd1e081 | 1041 | DBG("Updating poll fd array"); |
481d6c57 | 1042 | rcu_read_lock(); |
43c34bc3 | 1043 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
8994307f DG |
1044 | /* |
1045 | * Only active streams with an active end point can be added to the | |
1046 | * poll set and local stream storage of the thread. | |
1047 | * | |
1048 | * There is a potential race here for endpoint_status to be updated | |
1049 | * just after the check. However, this is OK since the stream(s) will | |
1050 | * be deleted once the thread is notified that the end point state has | |
1051 | * changed where this function will be called back again. | |
1052 | */ | |
1053 | if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM || | |
79d4ffb7 | 1054 | stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { |
3bd1e081 MD |
1055 | continue; |
1056 | } | |
7972aab2 DG |
1057 | /* |
1058 | * This clobbers way too much the debug output. Uncomment that if you | |
1059 | * need it for debugging purposes. | |
1060 | * | |
1061 | * DBG("Active FD %d", stream->wait_fd); | |
1062 | */ | |
e4421fec | 1063 | (*pollfd)[i].fd = stream->wait_fd; |
3bd1e081 | 1064 | (*pollfd)[i].events = POLLIN | POLLPRI; |
e4421fec | 1065 | local_stream[i] = stream; |
3bd1e081 MD |
1066 | i++; |
1067 | } | |
481d6c57 | 1068 | rcu_read_unlock(); |
3bd1e081 MD |
1069 | |
1070 | /* | |
50f8ae69 | 1071 | * Insert the consumer_data_pipe at the end of the array and don't |
3bd1e081 MD |
1072 | * increment i so nb_fd is the number of real FD. |
1073 | */ | |
acdb9057 | 1074 | (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe); |
509bb1cf | 1075 | (*pollfd)[i].events = POLLIN | POLLPRI; |
3bd1e081 MD |
1076 | return i; |
1077 | } | |
1078 | ||
1079 | /* | |
1080 | * Poll on the should_quit pipe and the command socket return -1 on error and | |
1081 | * should exit, 0 if data is available on the command socket | |
1082 | */ | |
1083 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) | |
1084 | { | |
1085 | int num_rdy; | |
1086 | ||
88f2b785 | 1087 | restart: |
3bd1e081 MD |
1088 | num_rdy = poll(consumer_sockpoll, 2, -1); |
1089 | if (num_rdy == -1) { | |
88f2b785 MD |
1090 | /* |
1091 | * Restart interrupted system call. | |
1092 | */ | |
1093 | if (errno == EINTR) { | |
1094 | goto restart; | |
1095 | } | |
7a57cf92 | 1096 | PERROR("Poll error"); |
3bd1e081 MD |
1097 | goto exit; |
1098 | } | |
509bb1cf | 1099 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
3bd1e081 MD |
1100 | DBG("consumer_should_quit wake up"); |
1101 | goto exit; | |
1102 | } | |
1103 | return 0; | |
1104 | ||
1105 | exit: | |
1106 | return -1; | |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * Set the error socket. | |
1111 | */ | |
ffe60014 DG |
1112 | void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx, |
1113 | int sock) | |
3bd1e081 MD |
1114 | { |
1115 | ctx->consumer_error_socket = sock; | |
1116 | } | |
1117 | ||
1118 | /* | |
1119 | * Set the command socket path. | |
1120 | */ | |
3bd1e081 MD |
1121 | void lttng_consumer_set_command_sock_path( |
1122 | struct lttng_consumer_local_data *ctx, char *sock) | |
1123 | { | |
1124 | ctx->consumer_command_sock_path = sock; | |
1125 | } | |
1126 | ||
1127 | /* | |
1128 | * Send return code to the session daemon. | |
1129 | * If the socket is not defined, we return 0, it is not a fatal error | |
1130 | */ | |
ffe60014 | 1131 | int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd) |
3bd1e081 MD |
1132 | { |
1133 | if (ctx->consumer_error_socket > 0) { | |
1134 | return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd, | |
1135 | sizeof(enum lttcomm_sessiond_command)); | |
1136 | } | |
1137 | ||
1138 | return 0; | |
1139 | } | |
1140 | ||
1141 | /* | |
228b5bf7 DG |
1142 | * Close all the tracefiles and stream fds and MUST be called when all |
1143 | * instances are destroyed i.e. when all threads were joined and are ended. | |
3bd1e081 MD |
1144 | */ |
1145 | void lttng_consumer_cleanup(void) | |
1146 | { | |
e4421fec | 1147 | struct lttng_ht_iter iter; |
ffe60014 | 1148 | struct lttng_consumer_channel *channel; |
6065ceec DG |
1149 | |
1150 | rcu_read_lock(); | |
3bd1e081 | 1151 | |
ffe60014 DG |
1152 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel, |
1153 | node.node) { | |
702b1ea4 | 1154 | consumer_del_channel(channel); |
3bd1e081 | 1155 | } |
6065ceec DG |
1156 | |
1157 | rcu_read_unlock(); | |
d6ce1df2 | 1158 | |
d6ce1df2 | 1159 | lttng_ht_destroy(consumer_data.channel_ht); |
228b5bf7 DG |
1160 | |
1161 | cleanup_relayd_ht(); | |
1162 | ||
d8ef542d MD |
1163 | lttng_ht_destroy(consumer_data.stream_per_chan_id_ht); |
1164 | ||
228b5bf7 DG |
1165 | /* |
1166 | * This HT contains streams that are freed by either the metadata thread or | |
1167 | * the data thread so we do *nothing* on the hash table and simply destroy | |
1168 | * it. | |
1169 | */ | |
1170 | lttng_ht_destroy(consumer_data.stream_list_ht); | |
3bd1e081 MD |
1171 | } |
1172 | ||
1173 | /* | |
1174 | * Called from signal handler. | |
1175 | */ | |
1176 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) | |
1177 | { | |
6cd525e8 MD |
1178 | ssize_t ret; |
1179 | ||
3bd1e081 | 1180 | consumer_quit = 1; |
6cd525e8 MD |
1181 | ret = lttng_write(ctx->consumer_should_quit[1], "4", 1); |
1182 | if (ret < 1) { | |
7a57cf92 | 1183 | PERROR("write consumer quit"); |
3bd1e081 | 1184 | } |
ab1027f4 DG |
1185 | |
1186 | DBG("Consumer flag that it should quit"); | |
3bd1e081 MD |
1187 | } |
1188 | ||
00e2e675 DG |
1189 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
1190 | off_t orig_offset) | |
3bd1e081 MD |
1191 | { |
1192 | int outfd = stream->out_fd; | |
1193 | ||
1194 | /* | |
1195 | * This does a blocking write-and-wait on any page that belongs to the | |
1196 | * subbuffer prior to the one we just wrote. | |
1197 | * Don't care about error values, as these are just hints and ways to | |
1198 | * limit the amount of page cache used. | |
1199 | */ | |
ffe60014 | 1200 | if (orig_offset < stream->max_sb_size) { |
3bd1e081 MD |
1201 | return; |
1202 | } | |
ffe60014 DG |
1203 | lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size, |
1204 | stream->max_sb_size, | |
3bd1e081 MD |
1205 | SYNC_FILE_RANGE_WAIT_BEFORE |
1206 | | SYNC_FILE_RANGE_WRITE | |
1207 | | SYNC_FILE_RANGE_WAIT_AFTER); | |
1208 | /* | |
1209 | * Give hints to the kernel about how we access the file: | |
1210 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after | |
1211 | * we write it. | |
1212 | * | |
1213 | * We need to call fadvise again after the file grows because the | |
1214 | * kernel does not seem to apply fadvise to non-existing parts of the | |
1215 | * file. | |
1216 | * | |
1217 | * Call fadvise _after_ having waited for the page writeback to | |
1218 | * complete because the dirty page writeback semantic is not well | |
1219 | * defined. So it can be expected to lead to lower throughput in | |
1220 | * streaming. | |
1221 | */ | |
ffe60014 DG |
1222 | posix_fadvise(outfd, orig_offset - stream->max_sb_size, |
1223 | stream->max_sb_size, POSIX_FADV_DONTNEED); | |
3bd1e081 MD |
1224 | } |
1225 | ||
1226 | /* | |
1227 | * Initialise the necessary environnement : | |
1228 | * - create a new context | |
1229 | * - create the poll_pipe | |
1230 | * - create the should_quit pipe (for signal handler) | |
1231 | * - create the thread pipe (for splice) | |
1232 | * | |
1233 | * Takes a function pointer as argument, this function is called when data is | |
1234 | * available on a buffer. This function is responsible to do the | |
1235 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the | |
1236 | * buffer configuration and then kernctl_put_next_subbuf at the end. | |
1237 | * | |
1238 | * Returns a pointer to the new context or NULL on error. | |
1239 | */ | |
1240 | struct lttng_consumer_local_data *lttng_consumer_create( | |
1241 | enum lttng_consumer_type type, | |
4078b776 | 1242 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
d41f73b7 | 1243 | struct lttng_consumer_local_data *ctx), |
3bd1e081 MD |
1244 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
1245 | int (*recv_stream)(struct lttng_consumer_stream *stream), | |
30319bcb | 1246 | int (*update_stream)(uint64_t stream_key, uint32_t state)) |
3bd1e081 | 1247 | { |
d8ef542d | 1248 | int ret; |
3bd1e081 MD |
1249 | struct lttng_consumer_local_data *ctx; |
1250 | ||
1251 | assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || | |
1252 | consumer_data.type == type); | |
1253 | consumer_data.type = type; | |
1254 | ||
effcf122 | 1255 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
3bd1e081 | 1256 | if (ctx == NULL) { |
7a57cf92 | 1257 | PERROR("allocating context"); |
3bd1e081 MD |
1258 | goto error; |
1259 | } | |
1260 | ||
1261 | ctx->consumer_error_socket = -1; | |
331744e3 | 1262 | ctx->consumer_metadata_socket = -1; |
75d83e50 | 1263 | pthread_mutex_init(&ctx->metadata_socket_lock, NULL); |
3bd1e081 MD |
1264 | /* assign the callbacks */ |
1265 | ctx->on_buffer_ready = buffer_ready; | |
1266 | ctx->on_recv_channel = recv_channel; | |
1267 | ctx->on_recv_stream = recv_stream; | |
1268 | ctx->on_update_stream = update_stream; | |
1269 | ||
acdb9057 DG |
1270 | ctx->consumer_data_pipe = lttng_pipe_open(0); |
1271 | if (!ctx->consumer_data_pipe) { | |
3bd1e081 MD |
1272 | goto error_poll_pipe; |
1273 | } | |
1274 | ||
1275 | ret = pipe(ctx->consumer_should_quit); | |
1276 | if (ret < 0) { | |
7a57cf92 | 1277 | PERROR("Error creating recv pipe"); |
3bd1e081 MD |
1278 | goto error_quit_pipe; |
1279 | } | |
1280 | ||
1281 | ret = pipe(ctx->consumer_thread_pipe); | |
1282 | if (ret < 0) { | |
7a57cf92 | 1283 | PERROR("Error creating thread pipe"); |
3bd1e081 MD |
1284 | goto error_thread_pipe; |
1285 | } | |
1286 | ||
d8ef542d MD |
1287 | ret = pipe(ctx->consumer_channel_pipe); |
1288 | if (ret < 0) { | |
1289 | PERROR("Error creating channel pipe"); | |
1290 | goto error_channel_pipe; | |
1291 | } | |
1292 | ||
13886d2d DG |
1293 | ctx->consumer_metadata_pipe = lttng_pipe_open(0); |
1294 | if (!ctx->consumer_metadata_pipe) { | |
fb3a43a9 DG |
1295 | goto error_metadata_pipe; |
1296 | } | |
3bd1e081 | 1297 | |
fb3a43a9 DG |
1298 | ret = utils_create_pipe(ctx->consumer_splice_metadata_pipe); |
1299 | if (ret < 0) { | |
1300 | goto error_splice_pipe; | |
1301 | } | |
1302 | ||
1303 | return ctx; | |
3bd1e081 | 1304 | |
fb3a43a9 | 1305 | error_splice_pipe: |
13886d2d | 1306 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
fb3a43a9 | 1307 | error_metadata_pipe: |
d8ef542d MD |
1308 | utils_close_pipe(ctx->consumer_channel_pipe); |
1309 | error_channel_pipe: | |
fb3a43a9 | 1310 | utils_close_pipe(ctx->consumer_thread_pipe); |
3bd1e081 | 1311 | error_thread_pipe: |
d8ef542d | 1312 | utils_close_pipe(ctx->consumer_should_quit); |
3bd1e081 | 1313 | error_quit_pipe: |
acdb9057 | 1314 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
3bd1e081 MD |
1315 | error_poll_pipe: |
1316 | free(ctx); | |
1317 | error: | |
1318 | return NULL; | |
1319 | } | |
1320 | ||
3b6b25ef MD |
1321 | /* |
1322 | * Iterate over all streams of the hashtable and free them properly. | |
1323 | */ | |
1324 | static void destroy_data_stream_ht(struct lttng_ht *ht) | |
1325 | { | |
1326 | struct lttng_ht_iter iter; | |
1327 | struct lttng_consumer_stream *stream; | |
1328 | ||
1329 | if (ht == NULL) { | |
1330 | return; | |
1331 | } | |
1332 | ||
1333 | rcu_read_lock(); | |
1334 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1335 | /* | |
1336 | * Ignore return value since we are currently cleaning up so any error | |
1337 | * can't be handled. | |
1338 | */ | |
1339 | (void) consumer_del_stream(stream, ht); | |
1340 | } | |
1341 | rcu_read_unlock(); | |
1342 | ||
1343 | lttng_ht_destroy(ht); | |
1344 | } | |
1345 | ||
1346 | /* | |
1347 | * Iterate over all streams of the metadata hashtable and free them | |
1348 | * properly. | |
1349 | */ | |
1350 | static void destroy_metadata_stream_ht(struct lttng_ht *ht) | |
1351 | { | |
1352 | struct lttng_ht_iter iter; | |
1353 | struct lttng_consumer_stream *stream; | |
1354 | ||
1355 | if (ht == NULL) { | |
1356 | return; | |
1357 | } | |
1358 | ||
1359 | rcu_read_lock(); | |
1360 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1361 | /* | |
1362 | * Ignore return value since we are currently cleaning up so any error | |
1363 | * can't be handled. | |
1364 | */ | |
1365 | (void) consumer_del_metadata_stream(stream, ht); | |
1366 | } | |
1367 | rcu_read_unlock(); | |
1368 | ||
1369 | lttng_ht_destroy(ht); | |
1370 | } | |
1371 | ||
3bd1e081 MD |
1372 | /* |
1373 | * Close all fds associated with the instance and free the context. | |
1374 | */ | |
1375 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) | |
1376 | { | |
4c462e79 MD |
1377 | int ret; |
1378 | ||
ab1027f4 DG |
1379 | DBG("Consumer destroying it. Closing everything."); |
1380 | ||
3b6b25ef MD |
1381 | destroy_data_stream_ht(data_ht); |
1382 | destroy_metadata_stream_ht(metadata_ht); | |
1383 | ||
4c462e79 MD |
1384 | ret = close(ctx->consumer_error_socket); |
1385 | if (ret) { | |
1386 | PERROR("close"); | |
1387 | } | |
331744e3 JD |
1388 | ret = close(ctx->consumer_metadata_socket); |
1389 | if (ret) { | |
1390 | PERROR("close"); | |
1391 | } | |
d8ef542d MD |
1392 | utils_close_pipe(ctx->consumer_thread_pipe); |
1393 | utils_close_pipe(ctx->consumer_channel_pipe); | |
acdb9057 | 1394 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
13886d2d | 1395 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
d8ef542d | 1396 | utils_close_pipe(ctx->consumer_should_quit); |
fb3a43a9 DG |
1397 | utils_close_pipe(ctx->consumer_splice_metadata_pipe); |
1398 | ||
3bd1e081 MD |
1399 | unlink(ctx->consumer_command_sock_path); |
1400 | free(ctx); | |
1401 | } | |
1402 | ||
6197aea7 DG |
1403 | /* |
1404 | * Write the metadata stream id on the specified file descriptor. | |
1405 | */ | |
1406 | static int write_relayd_metadata_id(int fd, | |
1407 | struct lttng_consumer_stream *stream, | |
ffe60014 | 1408 | struct consumer_relayd_sock_pair *relayd, unsigned long padding) |
6197aea7 | 1409 | { |
6cd525e8 | 1410 | ssize_t ret; |
1d4dfdef | 1411 | struct lttcomm_relayd_metadata_payload hdr; |
6197aea7 | 1412 | |
1d4dfdef DG |
1413 | hdr.stream_id = htobe64(stream->relayd_stream_id); |
1414 | hdr.padding_size = htobe32(padding); | |
6cd525e8 MD |
1415 | ret = lttng_write(fd, (void *) &hdr, sizeof(hdr)); |
1416 | if (ret < sizeof(hdr)) { | |
d7b75ec8 DG |
1417 | /* |
1418 | * This error means that the fd's end is closed so ignore the perror | |
1419 | * not to clubber the error output since this can happen in a normal | |
1420 | * code path. | |
1421 | */ | |
1422 | if (errno != EPIPE) { | |
1423 | PERROR("write metadata stream id"); | |
1424 | } | |
1425 | DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno); | |
534d2592 DG |
1426 | /* |
1427 | * Set ret to a negative value because if ret != sizeof(hdr), we don't | |
1428 | * handle writting the missing part so report that as an error and | |
1429 | * don't lie to the caller. | |
1430 | */ | |
1431 | ret = -1; | |
6197aea7 DG |
1432 | goto end; |
1433 | } | |
1d4dfdef DG |
1434 | DBG("Metadata stream id %" PRIu64 " with padding %lu written before data", |
1435 | stream->relayd_stream_id, padding); | |
6197aea7 DG |
1436 | |
1437 | end: | |
6cd525e8 | 1438 | return (int) ret; |
6197aea7 DG |
1439 | } |
1440 | ||
3bd1e081 | 1441 | /* |
09e26845 DG |
1442 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
1443 | * core function for writing trace buffers to either the local filesystem or | |
1444 | * the network. | |
1445 | * | |
79d4ffb7 DG |
1446 | * It must be called with the stream lock held. |
1447 | * | |
09e26845 | 1448 | * Careful review MUST be put if any changes occur! |
3bd1e081 MD |
1449 | * |
1450 | * Returns the number of bytes written | |
1451 | */ | |
4078b776 | 1452 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
3bd1e081 | 1453 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1454 | struct lttng_consumer_stream *stream, unsigned long len, |
309167d2 | 1455 | unsigned long padding, |
50adc264 | 1456 | struct ctf_packet_index *index) |
3bd1e081 | 1457 | { |
f02e1e8a | 1458 | unsigned long mmap_offset; |
ffe60014 | 1459 | void *mmap_base; |
1b6f00fa | 1460 | ssize_t ret = 0; |
f02e1e8a DG |
1461 | off_t orig_offset = stream->out_fd_offset; |
1462 | /* Default is on the disk */ | |
1463 | int outfd = stream->out_fd; | |
f02e1e8a | 1464 | struct consumer_relayd_sock_pair *relayd = NULL; |
8994307f | 1465 | unsigned int relayd_hang_up = 0; |
f02e1e8a DG |
1466 | |
1467 | /* RCU lock for the relayd pointer */ | |
1468 | rcu_read_lock(); | |
1469 | ||
1470 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1471 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1472 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1473 | if (relayd == NULL) { | |
56591bac | 1474 | ret = -EPIPE; |
f02e1e8a DG |
1475 | goto end; |
1476 | } | |
1477 | } | |
1478 | ||
1479 | /* get the offset inside the fd to mmap */ | |
3bd1e081 MD |
1480 | switch (consumer_data.type) { |
1481 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1482 | mmap_base = stream->mmap_base; |
f02e1e8a | 1483 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); |
1b6f00fa DG |
1484 | if (ret < 0) { |
1485 | ret = -errno; | |
56591bac | 1486 | PERROR("tracer ctl get_mmap_read_offset"); |
56591bac MD |
1487 | goto end; |
1488 | } | |
f02e1e8a | 1489 | break; |
7753dea8 MD |
1490 | case LTTNG_CONSUMER32_UST: |
1491 | case LTTNG_CONSUMER64_UST: | |
ffe60014 DG |
1492 | mmap_base = lttng_ustctl_get_mmap_base(stream); |
1493 | if (!mmap_base) { | |
1494 | ERR("read mmap get mmap base for stream %s", stream->name); | |
1b6f00fa | 1495 | ret = -EPERM; |
ffe60014 DG |
1496 | goto end; |
1497 | } | |
1498 | ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset); | |
56591bac MD |
1499 | if (ret != 0) { |
1500 | PERROR("tracer ctl get_mmap_read_offset"); | |
1b6f00fa | 1501 | ret = -EINVAL; |
56591bac MD |
1502 | goto end; |
1503 | } | |
f02e1e8a | 1504 | break; |
3bd1e081 MD |
1505 | default: |
1506 | ERR("Unknown consumer_data type"); | |
1507 | assert(0); | |
1508 | } | |
b9182dd9 | 1509 | |
f02e1e8a DG |
1510 | /* Handle stream on the relayd if the output is on the network */ |
1511 | if (relayd) { | |
1512 | unsigned long netlen = len; | |
1513 | ||
1514 | /* | |
1515 | * Lock the control socket for the complete duration of the function | |
1516 | * since from this point on we will use the socket. | |
1517 | */ | |
1518 | if (stream->metadata_flag) { | |
1519 | /* Metadata requires the control socket. */ | |
1520 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1d4dfdef | 1521 | netlen += sizeof(struct lttcomm_relayd_metadata_payload); |
f02e1e8a DG |
1522 | } |
1523 | ||
1d4dfdef | 1524 | ret = write_relayd_stream_header(stream, netlen, padding, relayd); |
1b6f00fa DG |
1525 | if (ret < 0) { |
1526 | relayd_hang_up = 1; | |
1527 | goto write_error; | |
1528 | } | |
1529 | /* Use the returned socket. */ | |
1530 | outfd = ret; | |
f02e1e8a | 1531 | |
1b6f00fa DG |
1532 | /* Write metadata stream id before payload */ |
1533 | if (stream->metadata_flag) { | |
1534 | ret = write_relayd_metadata_id(outfd, stream, relayd, padding); | |
1535 | if (ret < 0) { | |
8994307f DG |
1536 | relayd_hang_up = 1; |
1537 | goto write_error; | |
1538 | } | |
f02e1e8a | 1539 | } |
1d4dfdef DG |
1540 | } else { |
1541 | /* No streaming, we have to set the len with the full padding */ | |
1542 | len += padding; | |
1624d5b7 JD |
1543 | |
1544 | /* | |
1545 | * Check if we need to change the tracefile before writing the packet. | |
1546 | */ | |
1547 | if (stream->chan->tracefile_size > 0 && | |
1548 | (stream->tracefile_size_current + len) > | |
1549 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1550 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1551 | stream->name, stream->chan->tracefile_size, | |
1552 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
309167d2 JD |
1553 | stream->out_fd, &(stream->tracefile_count_current), |
1554 | &stream->out_fd); | |
1624d5b7 JD |
1555 | if (ret < 0) { |
1556 | ERR("Rotating output file"); | |
1557 | goto end; | |
1558 | } | |
309167d2 JD |
1559 | outfd = stream->out_fd; |
1560 | ||
1561 | if (stream->index_fd >= 0) { | |
1562 | ret = index_create_file(stream->chan->pathname, | |
1563 | stream->name, stream->uid, stream->gid, | |
1564 | stream->chan->tracefile_size, | |
1565 | stream->tracefile_count_current); | |
1566 | if (ret < 0) { | |
1567 | goto end; | |
1568 | } | |
1569 | stream->index_fd = ret; | |
1570 | } | |
1571 | ||
a6976990 DG |
1572 | /* Reset current size because we just perform a rotation. */ |
1573 | stream->tracefile_size_current = 0; | |
a1ae300f JD |
1574 | stream->out_fd_offset = 0; |
1575 | orig_offset = 0; | |
1624d5b7 JD |
1576 | } |
1577 | stream->tracefile_size_current += len; | |
309167d2 JD |
1578 | if (index) { |
1579 | index->offset = htobe64(stream->out_fd_offset); | |
1580 | } | |
f02e1e8a DG |
1581 | } |
1582 | ||
e90d31be DG |
1583 | /* |
1584 | * This call guarantee that len or less is returned. It's impossible to | |
1585 | * receive a ret value that is bigger than len. | |
1586 | */ | |
1587 | ret = lttng_write(outfd, mmap_base + mmap_offset, len); | |
1588 | DBG("Consumer mmap write() ret %zd (len %lu)", ret, len); | |
1589 | if (ret < 0 || ((size_t) ret != len)) { | |
1590 | /* | |
1591 | * Report error to caller if nothing was written else at least send the | |
1592 | * amount written. | |
1593 | */ | |
1594 | if (ret < 0) { | |
1b6f00fa | 1595 | ret = -errno; |
f02e1e8a | 1596 | } |
1b6f00fa | 1597 | relayd_hang_up = 1; |
f02e1e8a | 1598 | |
e90d31be | 1599 | /* Socket operation failed. We consider the relayd dead */ |
1b6f00fa | 1600 | if (errno == EPIPE || errno == EINVAL || errno == EBADF) { |
e90d31be DG |
1601 | /* |
1602 | * This is possible if the fd is closed on the other side | |
1603 | * (outfd) or any write problem. It can be verbose a bit for a | |
1604 | * normal execution if for instance the relayd is stopped | |
1605 | * abruptly. This can happen so set this to a DBG statement. | |
1606 | */ | |
1607 | DBG("Consumer mmap write detected relayd hang up"); | |
1b6f00fa DG |
1608 | } else { |
1609 | /* Unhandled error, print it and stop function right now. */ | |
1610 | PERROR("Error in write mmap (ret %zd != len %lu)", ret, len); | |
f02e1e8a | 1611 | } |
1b6f00fa | 1612 | goto write_error; |
e90d31be DG |
1613 | } |
1614 | stream->output_written += ret; | |
e90d31be DG |
1615 | |
1616 | /* This call is useless on a socket so better save a syscall. */ | |
1617 | if (!relayd) { | |
1618 | /* This won't block, but will start writeout asynchronously */ | |
1619 | lttng_sync_file_range(outfd, stream->out_fd_offset, len, | |
1620 | SYNC_FILE_RANGE_WRITE); | |
1621 | stream->out_fd_offset += len; | |
f02e1e8a DG |
1622 | } |
1623 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1624 | ||
8994307f DG |
1625 | write_error: |
1626 | /* | |
1627 | * This is a special case that the relayd has closed its socket. Let's | |
1628 | * cleanup the relayd object and all associated streams. | |
1629 | */ | |
1630 | if (relayd && relayd_hang_up) { | |
1631 | cleanup_relayd(relayd, ctx); | |
1632 | } | |
1633 | ||
f02e1e8a DG |
1634 | end: |
1635 | /* Unlock only if ctrl socket used */ | |
1636 | if (relayd && stream->metadata_flag) { | |
1637 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1638 | } | |
1639 | ||
1640 | rcu_read_unlock(); | |
1b6f00fa | 1641 | return ret; |
3bd1e081 MD |
1642 | } |
1643 | ||
1644 | /* | |
1645 | * Splice the data from the ring buffer to the tracefile. | |
1646 | * | |
79d4ffb7 DG |
1647 | * It must be called with the stream lock held. |
1648 | * | |
3bd1e081 MD |
1649 | * Returns the number of bytes spliced. |
1650 | */ | |
4078b776 | 1651 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
3bd1e081 | 1652 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1653 | struct lttng_consumer_stream *stream, unsigned long len, |
309167d2 | 1654 | unsigned long padding, |
50adc264 | 1655 | struct ctf_packet_index *index) |
3bd1e081 | 1656 | { |
f02e1e8a DG |
1657 | ssize_t ret = 0, written = 0, ret_splice = 0; |
1658 | loff_t offset = 0; | |
1659 | off_t orig_offset = stream->out_fd_offset; | |
1660 | int fd = stream->wait_fd; | |
1661 | /* Default is on the disk */ | |
1662 | int outfd = stream->out_fd; | |
f02e1e8a | 1663 | struct consumer_relayd_sock_pair *relayd = NULL; |
fb3a43a9 | 1664 | int *splice_pipe; |
8994307f | 1665 | unsigned int relayd_hang_up = 0; |
f02e1e8a | 1666 | |
3bd1e081 MD |
1667 | switch (consumer_data.type) { |
1668 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a | 1669 | break; |
7753dea8 MD |
1670 | case LTTNG_CONSUMER32_UST: |
1671 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a | 1672 | /* Not supported for user space tracing */ |
3bd1e081 MD |
1673 | return -ENOSYS; |
1674 | default: | |
1675 | ERR("Unknown consumer_data type"); | |
1676 | assert(0); | |
3bd1e081 MD |
1677 | } |
1678 | ||
f02e1e8a DG |
1679 | /* RCU lock for the relayd pointer */ |
1680 | rcu_read_lock(); | |
1681 | ||
1682 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1683 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1684 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1685 | if (relayd == NULL) { | |
c258c659 | 1686 | written = -ret; |
f02e1e8a DG |
1687 | goto end; |
1688 | } | |
1689 | } | |
1690 | ||
fb3a43a9 DG |
1691 | /* |
1692 | * Choose right pipe for splice. Metadata and trace data are handled by | |
1693 | * different threads hence the use of two pipes in order not to race or | |
1694 | * corrupt the written data. | |
1695 | */ | |
1696 | if (stream->metadata_flag) { | |
1697 | splice_pipe = ctx->consumer_splice_metadata_pipe; | |
1698 | } else { | |
1699 | splice_pipe = ctx->consumer_thread_pipe; | |
1700 | } | |
1701 | ||
f02e1e8a | 1702 | /* Write metadata stream id before payload */ |
1d4dfdef | 1703 | if (relayd) { |
c258c659 | 1704 | unsigned long total_len = len; |
f02e1e8a | 1705 | |
1d4dfdef DG |
1706 | if (stream->metadata_flag) { |
1707 | /* | |
1708 | * Lock the control socket for the complete duration of the function | |
1709 | * since from this point on we will use the socket. | |
1710 | */ | |
1711 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1712 | ||
1713 | ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd, | |
1714 | padding); | |
1715 | if (ret < 0) { | |
1716 | written = ret; | |
c258c659 DG |
1717 | relayd_hang_up = 1; |
1718 | goto write_error; | |
1d4dfdef DG |
1719 | } |
1720 | ||
1721 | total_len += sizeof(struct lttcomm_relayd_metadata_payload); | |
1722 | } | |
1723 | ||
1724 | ret = write_relayd_stream_header(stream, total_len, padding, relayd); | |
c258c659 DG |
1725 | if (ret < 0) { |
1726 | written = ret; | |
1727 | relayd_hang_up = 1; | |
1728 | goto write_error; | |
f02e1e8a | 1729 | } |
c258c659 DG |
1730 | /* Use the returned socket. */ |
1731 | outfd = ret; | |
1d4dfdef DG |
1732 | } else { |
1733 | /* No streaming, we have to set the len with the full padding */ | |
1734 | len += padding; | |
1624d5b7 JD |
1735 | |
1736 | /* | |
1737 | * Check if we need to change the tracefile before writing the packet. | |
1738 | */ | |
1739 | if (stream->chan->tracefile_size > 0 && | |
1740 | (stream->tracefile_size_current + len) > | |
1741 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1742 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1743 | stream->name, stream->chan->tracefile_size, | |
1744 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
309167d2 JD |
1745 | stream->out_fd, &(stream->tracefile_count_current), |
1746 | &stream->out_fd); | |
1624d5b7 | 1747 | if (ret < 0) { |
c258c659 | 1748 | written = ret; |
1624d5b7 JD |
1749 | ERR("Rotating output file"); |
1750 | goto end; | |
1751 | } | |
309167d2 JD |
1752 | outfd = stream->out_fd; |
1753 | ||
1754 | if (stream->index_fd >= 0) { | |
1755 | ret = index_create_file(stream->chan->pathname, | |
1756 | stream->name, stream->uid, stream->gid, | |
1757 | stream->chan->tracefile_size, | |
1758 | stream->tracefile_count_current); | |
1759 | if (ret < 0) { | |
c258c659 | 1760 | written = ret; |
309167d2 JD |
1761 | goto end; |
1762 | } | |
1763 | stream->index_fd = ret; | |
1764 | } | |
1765 | ||
a6976990 DG |
1766 | /* Reset current size because we just perform a rotation. */ |
1767 | stream->tracefile_size_current = 0; | |
a1ae300f JD |
1768 | stream->out_fd_offset = 0; |
1769 | orig_offset = 0; | |
1624d5b7 JD |
1770 | } |
1771 | stream->tracefile_size_current += len; | |
309167d2 | 1772 | index->offset = htobe64(stream->out_fd_offset); |
f02e1e8a DG |
1773 | } |
1774 | ||
1775 | while (len > 0) { | |
1d4dfdef DG |
1776 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)", |
1777 | (unsigned long)offset, len, fd, splice_pipe[1]); | |
fb3a43a9 | 1778 | ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len, |
f02e1e8a DG |
1779 | SPLICE_F_MOVE | SPLICE_F_MORE); |
1780 | DBG("splice chan to pipe, ret %zd", ret_splice); | |
1781 | if (ret_splice < 0) { | |
e90d31be | 1782 | ret = errno; |
c258c659 | 1783 | written = -ret; |
e90d31be | 1784 | PERROR("Error in relay splice"); |
f02e1e8a DG |
1785 | goto splice_error; |
1786 | } | |
1787 | ||
1788 | /* Handle stream on the relayd if the output is on the network */ | |
c258c659 DG |
1789 | if (relayd && stream->metadata_flag) { |
1790 | size_t metadata_payload_size = | |
1791 | sizeof(struct lttcomm_relayd_metadata_payload); | |
1792 | ||
1793 | /* Update counter to fit the spliced data */ | |
1794 | ret_splice += metadata_payload_size; | |
1795 | len += metadata_payload_size; | |
1796 | /* | |
1797 | * We do this so the return value can match the len passed as | |
1798 | * argument to this function. | |
1799 | */ | |
1800 | written -= metadata_payload_size; | |
f02e1e8a DG |
1801 | } |
1802 | ||
1803 | /* Splice data out */ | |
fb3a43a9 | 1804 | ret_splice = splice(splice_pipe[0], NULL, outfd, NULL, |
f02e1e8a | 1805 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); |
1d4dfdef | 1806 | DBG("Consumer splice pipe to file, ret %zd", ret_splice); |
f02e1e8a | 1807 | if (ret_splice < 0) { |
e90d31be | 1808 | ret = errno; |
c258c659 DG |
1809 | written = -ret; |
1810 | relayd_hang_up = 1; | |
1811 | goto write_error; | |
f02e1e8a | 1812 | } else if (ret_splice > len) { |
e90d31be DG |
1813 | /* |
1814 | * We don't expect this code path to be executed but you never know | |
1815 | * so this is an extra protection agains a buggy splice(). | |
1816 | */ | |
f02e1e8a | 1817 | ret = errno; |
c258c659 | 1818 | written += ret_splice; |
e90d31be DG |
1819 | PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice, |
1820 | len); | |
f02e1e8a | 1821 | goto splice_error; |
e90d31be DG |
1822 | } else { |
1823 | /* All good, update current len and continue. */ | |
1824 | len -= ret_splice; | |
f02e1e8a | 1825 | } |
f02e1e8a DG |
1826 | |
1827 | /* This call is useless on a socket so better save a syscall. */ | |
1828 | if (!relayd) { | |
1829 | /* This won't block, but will start writeout asynchronously */ | |
1830 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, | |
1831 | SYNC_FILE_RANGE_WRITE); | |
1832 | stream->out_fd_offset += ret_splice; | |
1833 | } | |
e5d1a9b3 | 1834 | stream->output_written += ret_splice; |
f02e1e8a DG |
1835 | written += ret_splice; |
1836 | } | |
1837 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
f02e1e8a DG |
1838 | goto end; |
1839 | ||
8994307f DG |
1840 | write_error: |
1841 | /* | |
1842 | * This is a special case that the relayd has closed its socket. Let's | |
1843 | * cleanup the relayd object and all associated streams. | |
1844 | */ | |
1845 | if (relayd && relayd_hang_up) { | |
1846 | cleanup_relayd(relayd, ctx); | |
1847 | /* Skip splice error so the consumer does not fail */ | |
1848 | goto end; | |
1849 | } | |
1850 | ||
f02e1e8a DG |
1851 | splice_error: |
1852 | /* send the appropriate error description to sessiond */ | |
1853 | switch (ret) { | |
f02e1e8a | 1854 | case EINVAL: |
f73fabfd | 1855 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); |
f02e1e8a DG |
1856 | break; |
1857 | case ENOMEM: | |
f73fabfd | 1858 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); |
f02e1e8a DG |
1859 | break; |
1860 | case ESPIPE: | |
f73fabfd | 1861 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); |
f02e1e8a DG |
1862 | break; |
1863 | } | |
1864 | ||
1865 | end: | |
1866 | if (relayd && stream->metadata_flag) { | |
1867 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1868 | } | |
1869 | ||
1870 | rcu_read_unlock(); | |
1871 | return written; | |
3bd1e081 MD |
1872 | } |
1873 | ||
1874 | /* | |
1875 | * Take a snapshot for a specific fd | |
1876 | * | |
1877 | * Returns 0 on success, < 0 on error | |
1878 | */ | |
ffe60014 | 1879 | int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) |
3bd1e081 MD |
1880 | { |
1881 | switch (consumer_data.type) { | |
1882 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1883 | return lttng_kconsumer_take_snapshot(stream); |
7753dea8 MD |
1884 | case LTTNG_CONSUMER32_UST: |
1885 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 1886 | return lttng_ustconsumer_take_snapshot(stream); |
3bd1e081 MD |
1887 | default: |
1888 | ERR("Unknown consumer_data type"); | |
1889 | assert(0); | |
1890 | return -ENOSYS; | |
1891 | } | |
3bd1e081 MD |
1892 | } |
1893 | ||
1894 | /* | |
1895 | * Get the produced position | |
1896 | * | |
1897 | * Returns 0 on success, < 0 on error | |
1898 | */ | |
ffe60014 | 1899 | int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, |
3bd1e081 MD |
1900 | unsigned long *pos) |
1901 | { | |
1902 | switch (consumer_data.type) { | |
1903 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1904 | return lttng_kconsumer_get_produced_snapshot(stream, pos); |
7753dea8 MD |
1905 | case LTTNG_CONSUMER32_UST: |
1906 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 1907 | return lttng_ustconsumer_get_produced_snapshot(stream, pos); |
3bd1e081 MD |
1908 | default: |
1909 | ERR("Unknown consumer_data type"); | |
1910 | assert(0); | |
1911 | return -ENOSYS; | |
1912 | } | |
1913 | } | |
1914 | ||
1915 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, | |
1916 | int sock, struct pollfd *consumer_sockpoll) | |
1917 | { | |
1918 | switch (consumer_data.type) { | |
1919 | case LTTNG_CONSUMER_KERNEL: | |
1920 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
7753dea8 MD |
1921 | case LTTNG_CONSUMER32_UST: |
1922 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1923 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
1924 | default: | |
1925 | ERR("Unknown consumer_data type"); | |
1926 | assert(0); | |
1927 | return -ENOSYS; | |
1928 | } | |
1929 | } | |
1930 | ||
59a6e1d8 | 1931 | void lttng_consumer_close_all_metadata(void) |
d88aee68 DG |
1932 | { |
1933 | switch (consumer_data.type) { | |
1934 | case LTTNG_CONSUMER_KERNEL: | |
1935 | /* | |
1936 | * The Kernel consumer has a different metadata scheme so we don't | |
1937 | * close anything because the stream will be closed by the session | |
1938 | * daemon. | |
1939 | */ | |
1940 | break; | |
1941 | case LTTNG_CONSUMER32_UST: | |
1942 | case LTTNG_CONSUMER64_UST: | |
1943 | /* | |
1944 | * Close all metadata streams. The metadata hash table is passed and | |
1945 | * this call iterates over it by closing all wakeup fd. This is safe | |
1946 | * because at this point we are sure that the metadata producer is | |
1947 | * either dead or blocked. | |
1948 | */ | |
59a6e1d8 | 1949 | lttng_ustconsumer_close_all_metadata(metadata_ht); |
d88aee68 DG |
1950 | break; |
1951 | default: | |
1952 | ERR("Unknown consumer_data type"); | |
1953 | assert(0); | |
1954 | } | |
1955 | } | |
1956 | ||
fb3a43a9 DG |
1957 | /* |
1958 | * Clean up a metadata stream and free its memory. | |
1959 | */ | |
e316aad5 DG |
1960 | void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, |
1961 | struct lttng_ht *ht) | |
fb3a43a9 | 1962 | { |
e316aad5 | 1963 | struct lttng_consumer_channel *free_chan = NULL; |
fb3a43a9 DG |
1964 | |
1965 | assert(stream); | |
1966 | /* | |
1967 | * This call should NEVER receive regular stream. It must always be | |
1968 | * metadata stream and this is crucial for data structure synchronization. | |
1969 | */ | |
1970 | assert(stream->metadata_flag); | |
1971 | ||
e316aad5 DG |
1972 | DBG3("Consumer delete metadata stream %d", stream->wait_fd); |
1973 | ||
74251bb8 | 1974 | pthread_mutex_lock(&consumer_data.lock); |
a9838785 | 1975 | pthread_mutex_lock(&stream->chan->lock); |
8994307f DG |
1976 | pthread_mutex_lock(&stream->lock); |
1977 | ||
59a6e1d8 DG |
1978 | /* Remove any reference to that stream. */ |
1979 | consumer_stream_delete(stream, ht); | |
ca22feea | 1980 | |
59a6e1d8 DG |
1981 | /* Close down everything including the relayd if one. */ |
1982 | consumer_stream_close(stream); | |
1983 | /* Destroy tracer buffers of the stream. */ | |
1984 | consumer_stream_destroy_buffers(stream); | |
fb3a43a9 DG |
1985 | |
1986 | /* Atomically decrement channel refcount since other threads can use it. */ | |
f2ad556d | 1987 | if (!uatomic_sub_return(&stream->chan->refcount, 1) |
ffe60014 | 1988 | && !uatomic_read(&stream->chan->nb_init_stream_left)) { |
c30aaa51 | 1989 | /* Go for channel deletion! */ |
e316aad5 | 1990 | free_chan = stream->chan; |
fb3a43a9 DG |
1991 | } |
1992 | ||
73811ecc DG |
1993 | /* |
1994 | * Nullify the stream reference so it is not used after deletion. The | |
59a6e1d8 DG |
1995 | * channel lock MUST be acquired before being able to check for a NULL |
1996 | * pointer value. | |
73811ecc DG |
1997 | */ |
1998 | stream->chan->metadata_stream = NULL; | |
1999 | ||
8994307f | 2000 | pthread_mutex_unlock(&stream->lock); |
a9838785 | 2001 | pthread_mutex_unlock(&stream->chan->lock); |
74251bb8 | 2002 | pthread_mutex_unlock(&consumer_data.lock); |
e316aad5 DG |
2003 | |
2004 | if (free_chan) { | |
2005 | consumer_del_channel(free_chan); | |
2006 | } | |
2007 | ||
59a6e1d8 | 2008 | consumer_stream_free(stream); |
fb3a43a9 DG |
2009 | } |
2010 | ||
2011 | /* | |
2012 | * Action done with the metadata stream when adding it to the consumer internal | |
2013 | * data structures to handle it. | |
2014 | */ | |
5ab66908 | 2015 | int consumer_add_metadata_stream(struct lttng_consumer_stream *stream) |
fb3a43a9 | 2016 | { |
5ab66908 | 2017 | struct lttng_ht *ht = metadata_ht; |
e316aad5 | 2018 | int ret = 0; |
76082088 | 2019 | struct lttng_ht_iter iter; |
d88aee68 | 2020 | struct lttng_ht_node_u64 *node; |
fb3a43a9 | 2021 | |
e316aad5 DG |
2022 | assert(stream); |
2023 | assert(ht); | |
2024 | ||
d88aee68 | 2025 | DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key); |
e316aad5 DG |
2026 | |
2027 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 2028 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 2029 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 2030 | pthread_mutex_lock(&stream->lock); |
e316aad5 | 2031 | |
e316aad5 DG |
2032 | /* |
2033 | * From here, refcounts are updated so be _careful_ when returning an error | |
2034 | * after this point. | |
2035 | */ | |
2036 | ||
fb3a43a9 | 2037 | rcu_read_lock(); |
76082088 DG |
2038 | |
2039 | /* | |
2040 | * Lookup the stream just to make sure it does not exist in our internal | |
2041 | * state. This should NEVER happen. | |
2042 | */ | |
d88aee68 DG |
2043 | lttng_ht_lookup(ht, &stream->key, &iter); |
2044 | node = lttng_ht_iter_get_node_u64(&iter); | |
76082088 DG |
2045 | assert(!node); |
2046 | ||
e316aad5 | 2047 | /* |
ffe60014 DG |
2048 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
2049 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
2050 | * causes the count to become 0 also causes a stream to be added. The |
2051 | * channel deletion will thus be triggered by the following removal of this | |
2052 | * stream. | |
2053 | */ | |
ffe60014 | 2054 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
2055 | /* Increment refcount before decrementing nb_init_stream_left */ |
2056 | cmm_smp_wmb(); | |
ffe60014 | 2057 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
2058 | } |
2059 | ||
d88aee68 | 2060 | lttng_ht_add_unique_u64(ht, &stream->node); |
ca22feea | 2061 | |
d8ef542d MD |
2062 | lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht, |
2063 | &stream->node_channel_id); | |
2064 | ||
ca22feea DG |
2065 | /* |
2066 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
2067 | * the key since the HT does not use it and we allow to add redundant keys | |
2068 | * into this table. | |
2069 | */ | |
d88aee68 | 2070 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 2071 | |
fb3a43a9 | 2072 | rcu_read_unlock(); |
e316aad5 | 2073 | |
2e818a6a | 2074 | pthread_mutex_unlock(&stream->lock); |
a9838785 | 2075 | pthread_mutex_unlock(&stream->chan->lock); |
ec6ea7d0 | 2076 | pthread_mutex_unlock(&stream->chan->timer_lock); |
e316aad5 DG |
2077 | pthread_mutex_unlock(&consumer_data.lock); |
2078 | return ret; | |
fb3a43a9 DG |
2079 | } |
2080 | ||
8994307f DG |
2081 | /* |
2082 | * Delete data stream that are flagged for deletion (endpoint_status). | |
2083 | */ | |
2084 | static void validate_endpoint_status_data_stream(void) | |
2085 | { | |
2086 | struct lttng_ht_iter iter; | |
2087 | struct lttng_consumer_stream *stream; | |
2088 | ||
2089 | DBG("Consumer delete flagged data stream"); | |
2090 | ||
2091 | rcu_read_lock(); | |
2092 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
2093 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2094 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2095 | continue; |
2096 | } | |
2097 | /* Delete it right now */ | |
2098 | consumer_del_stream(stream, data_ht); | |
2099 | } | |
2100 | rcu_read_unlock(); | |
2101 | } | |
2102 | ||
2103 | /* | |
2104 | * Delete metadata stream that are flagged for deletion (endpoint_status). | |
2105 | */ | |
2106 | static void validate_endpoint_status_metadata_stream( | |
2107 | struct lttng_poll_event *pollset) | |
2108 | { | |
2109 | struct lttng_ht_iter iter; | |
2110 | struct lttng_consumer_stream *stream; | |
2111 | ||
2112 | DBG("Consumer delete flagged metadata stream"); | |
2113 | ||
2114 | assert(pollset); | |
2115 | ||
2116 | rcu_read_lock(); | |
2117 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
2118 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2119 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2120 | continue; |
2121 | } | |
2122 | /* | |
2123 | * Remove from pollset so the metadata thread can continue without | |
2124 | * blocking on a deleted stream. | |
2125 | */ | |
2126 | lttng_poll_del(pollset, stream->wait_fd); | |
2127 | ||
2128 | /* Delete it right now */ | |
2129 | consumer_del_metadata_stream(stream, metadata_ht); | |
2130 | } | |
2131 | rcu_read_unlock(); | |
2132 | } | |
2133 | ||
fb3a43a9 DG |
2134 | /* |
2135 | * Thread polls on metadata file descriptor and write them on disk or on the | |
2136 | * network. | |
2137 | */ | |
7d980def | 2138 | void *consumer_thread_metadata_poll(void *data) |
fb3a43a9 | 2139 | { |
1fc79fb4 | 2140 | int ret, i, pollfd, err = -1; |
fb3a43a9 | 2141 | uint32_t revents, nb_fd; |
e316aad5 | 2142 | struct lttng_consumer_stream *stream = NULL; |
fb3a43a9 | 2143 | struct lttng_ht_iter iter; |
d88aee68 | 2144 | struct lttng_ht_node_u64 *node; |
fb3a43a9 DG |
2145 | struct lttng_poll_event events; |
2146 | struct lttng_consumer_local_data *ctx = data; | |
2147 | ssize_t len; | |
2148 | ||
2149 | rcu_register_thread(); | |
2150 | ||
1fc79fb4 MD |
2151 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA); |
2152 | ||
ede905e6 MD |
2153 | if (testpoint(consumerd_thread_metadata)) { |
2154 | goto error_testpoint; | |
2155 | } | |
2156 | ||
9ce5646a MD |
2157 | health_code_update(); |
2158 | ||
fb3a43a9 DG |
2159 | DBG("Thread metadata poll started"); |
2160 | ||
fb3a43a9 DG |
2161 | /* Size is set to 1 for the consumer_metadata pipe */ |
2162 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2163 | if (ret < 0) { | |
2164 | ERR("Poll set creation failed"); | |
d8ef542d | 2165 | goto end_poll; |
fb3a43a9 DG |
2166 | } |
2167 | ||
13886d2d DG |
2168 | ret = lttng_poll_add(&events, |
2169 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN); | |
fb3a43a9 DG |
2170 | if (ret < 0) { |
2171 | goto end; | |
2172 | } | |
2173 | ||
2174 | /* Main loop */ | |
2175 | DBG("Metadata main loop started"); | |
2176 | ||
2177 | while (1) { | |
9ce5646a MD |
2178 | health_code_update(); |
2179 | ||
fb3a43a9 | 2180 | /* Only the metadata pipe is set */ |
d21b0d71 | 2181 | if (LTTNG_POLL_GETNB(&events) == 0 && consumer_quit == 1) { |
1fc79fb4 | 2182 | err = 0; /* All is OK */ |
fb3a43a9 DG |
2183 | goto end; |
2184 | } | |
2185 | ||
2186 | restart: | |
d21b0d71 | 2187 | DBG("Metadata poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events)); |
9ce5646a | 2188 | health_poll_entry(); |
fb3a43a9 | 2189 | ret = lttng_poll_wait(&events, -1); |
9ce5646a | 2190 | health_poll_exit(); |
fb3a43a9 DG |
2191 | DBG("Metadata event catched in thread"); |
2192 | if (ret < 0) { | |
2193 | if (errno == EINTR) { | |
e316aad5 | 2194 | ERR("Poll EINTR catched"); |
fb3a43a9 DG |
2195 | goto restart; |
2196 | } | |
2197 | goto error; | |
2198 | } | |
2199 | ||
0d9c5d77 DG |
2200 | nb_fd = ret; |
2201 | ||
e316aad5 | 2202 | /* From here, the event is a metadata wait fd */ |
fb3a43a9 | 2203 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2204 | health_code_update(); |
2205 | ||
fb3a43a9 DG |
2206 | revents = LTTNG_POLL_GETEV(&events, i); |
2207 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2208 | ||
13886d2d | 2209 | if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) { |
4adabd61 | 2210 | if (revents & (LPOLLERR | LPOLLHUP )) { |
fb3a43a9 DG |
2211 | DBG("Metadata thread pipe hung up"); |
2212 | /* | |
2213 | * Remove the pipe from the poll set and continue the loop | |
2214 | * since their might be data to consume. | |
2215 | */ | |
13886d2d DG |
2216 | lttng_poll_del(&events, |
2217 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2218 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
fb3a43a9 DG |
2219 | continue; |
2220 | } else if (revents & LPOLLIN) { | |
13886d2d DG |
2221 | ssize_t pipe_len; |
2222 | ||
2223 | pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe, | |
2224 | &stream, sizeof(stream)); | |
6cd525e8 MD |
2225 | if (pipe_len < sizeof(stream)) { |
2226 | PERROR("read metadata stream"); | |
fb3a43a9 | 2227 | /* |
13886d2d | 2228 | * Continue here to handle the rest of the streams. |
fb3a43a9 DG |
2229 | */ |
2230 | continue; | |
2231 | } | |
2232 | ||
8994307f DG |
2233 | /* A NULL stream means that the state has changed. */ |
2234 | if (stream == NULL) { | |
2235 | /* Check for deleted streams. */ | |
2236 | validate_endpoint_status_metadata_stream(&events); | |
3714380f | 2237 | goto restart; |
8994307f DG |
2238 | } |
2239 | ||
fb3a43a9 DG |
2240 | DBG("Adding metadata stream %d to poll set", |
2241 | stream->wait_fd); | |
2242 | ||
fb3a43a9 DG |
2243 | /* Add metadata stream to the global poll events list */ |
2244 | lttng_poll_add(&events, stream->wait_fd, | |
59a6e1d8 | 2245 | LPOLLIN | LPOLLPRI | LPOLLHUP); |
fb3a43a9 DG |
2246 | } |
2247 | ||
e316aad5 | 2248 | /* Handle other stream */ |
fb3a43a9 DG |
2249 | continue; |
2250 | } | |
2251 | ||
d09e1200 | 2252 | rcu_read_lock(); |
d88aee68 DG |
2253 | { |
2254 | uint64_t tmp_id = (uint64_t) pollfd; | |
2255 | ||
2256 | lttng_ht_lookup(metadata_ht, &tmp_id, &iter); | |
2257 | } | |
2258 | node = lttng_ht_iter_get_node_u64(&iter); | |
e316aad5 | 2259 | assert(node); |
fb3a43a9 DG |
2260 | |
2261 | stream = caa_container_of(node, struct lttng_consumer_stream, | |
58b1f425 | 2262 | node); |
fb3a43a9 | 2263 | |
e316aad5 | 2264 | /* Check for error event */ |
4adabd61 | 2265 | if (revents & (LPOLLERR | LPOLLHUP)) { |
e316aad5 | 2266 | DBG("Metadata fd %d is hup|err.", pollfd); |
fb3a43a9 DG |
2267 | if (!stream->hangup_flush_done |
2268 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2269 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2270 | DBG("Attempting to flush and consume the UST buffers"); | |
2271 | lttng_ustconsumer_on_stream_hangup(stream); | |
2272 | ||
2273 | /* We just flushed the stream now read it. */ | |
4bb94b75 | 2274 | do { |
9ce5646a MD |
2275 | health_code_update(); |
2276 | ||
4bb94b75 DG |
2277 | len = ctx->on_buffer_ready(stream, ctx); |
2278 | /* | |
2279 | * We don't check the return value here since if we get | |
2280 | * a negative len, it means an error occured thus we | |
2281 | * simply remove it from the poll set and free the | |
2282 | * stream. | |
2283 | */ | |
2284 | } while (len > 0); | |
fb3a43a9 DG |
2285 | } |
2286 | ||
fb3a43a9 | 2287 | lttng_poll_del(&events, stream->wait_fd); |
e316aad5 DG |
2288 | /* |
2289 | * This call update the channel states, closes file descriptors | |
2290 | * and securely free the stream. | |
2291 | */ | |
2292 | consumer_del_metadata_stream(stream, metadata_ht); | |
2293 | } else if (revents & (LPOLLIN | LPOLLPRI)) { | |
2294 | /* Get the data out of the metadata file descriptor */ | |
2295 | DBG("Metadata available on fd %d", pollfd); | |
2296 | assert(stream->wait_fd == pollfd); | |
2297 | ||
04ef1097 | 2298 | do { |
9ce5646a MD |
2299 | health_code_update(); |
2300 | ||
04ef1097 MD |
2301 | len = ctx->on_buffer_ready(stream, ctx); |
2302 | /* | |
2303 | * We don't check the return value here since if we get | |
2304 | * a negative len, it means an error occured thus we | |
2305 | * simply remove it from the poll set and free the | |
2306 | * stream. | |
2307 | */ | |
2308 | } while (len > 0); | |
2309 | ||
e316aad5 | 2310 | /* It's ok to have an unavailable sub-buffer */ |
b64403e3 | 2311 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2312 | /* Clean up stream from consumer and free it. */ |
2313 | lttng_poll_del(&events, stream->wait_fd); | |
2314 | consumer_del_metadata_stream(stream, metadata_ht); | |
e316aad5 | 2315 | } |
fb3a43a9 | 2316 | } |
e316aad5 DG |
2317 | |
2318 | /* Release RCU lock for the stream looked up */ | |
d09e1200 | 2319 | rcu_read_unlock(); |
fb3a43a9 DG |
2320 | } |
2321 | } | |
2322 | ||
1fc79fb4 MD |
2323 | /* All is OK */ |
2324 | err = 0; | |
fb3a43a9 DG |
2325 | error: |
2326 | end: | |
2327 | DBG("Metadata poll thread exiting"); | |
fb3a43a9 | 2328 | |
d8ef542d MD |
2329 | lttng_poll_clean(&events); |
2330 | end_poll: | |
ede905e6 | 2331 | error_testpoint: |
1fc79fb4 MD |
2332 | if (err) { |
2333 | health_error(); | |
2334 | ERR("Health error occurred in %s", __func__); | |
2335 | } | |
2336 | health_unregister(health_consumerd); | |
fb3a43a9 DG |
2337 | rcu_unregister_thread(); |
2338 | return NULL; | |
2339 | } | |
2340 | ||
3bd1e081 | 2341 | /* |
e4421fec | 2342 | * This thread polls the fds in the set to consume the data and write |
3bd1e081 MD |
2343 | * it to tracefile if necessary. |
2344 | */ | |
7d980def | 2345 | void *consumer_thread_data_poll(void *data) |
3bd1e081 | 2346 | { |
1fc79fb4 | 2347 | int num_rdy, num_hup, high_prio, ret, i, err = -1; |
3bd1e081 MD |
2348 | struct pollfd *pollfd = NULL; |
2349 | /* local view of the streams */ | |
c869f647 | 2350 | struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; |
3bd1e081 MD |
2351 | /* local view of consumer_data.fds_count */ |
2352 | int nb_fd = 0; | |
3bd1e081 | 2353 | struct lttng_consumer_local_data *ctx = data; |
00e2e675 | 2354 | ssize_t len; |
3bd1e081 | 2355 | |
e7b994a3 DG |
2356 | rcu_register_thread(); |
2357 | ||
1fc79fb4 MD |
2358 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA); |
2359 | ||
ede905e6 MD |
2360 | if (testpoint(consumerd_thread_data)) { |
2361 | goto error_testpoint; | |
2362 | } | |
2363 | ||
9ce5646a MD |
2364 | health_code_update(); |
2365 | ||
4df6c8cb MD |
2366 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream *)); |
2367 | if (local_stream == NULL) { | |
2368 | PERROR("local_stream malloc"); | |
2369 | goto end; | |
2370 | } | |
3bd1e081 MD |
2371 | |
2372 | while (1) { | |
9ce5646a MD |
2373 | health_code_update(); |
2374 | ||
3bd1e081 MD |
2375 | high_prio = 0; |
2376 | num_hup = 0; | |
2377 | ||
2378 | /* | |
e4421fec | 2379 | * the fds set has been updated, we need to update our |
3bd1e081 MD |
2380 | * local array as well |
2381 | */ | |
2382 | pthread_mutex_lock(&consumer_data.lock); | |
2383 | if (consumer_data.need_update) { | |
0e428499 DG |
2384 | free(pollfd); |
2385 | pollfd = NULL; | |
2386 | ||
2387 | free(local_stream); | |
2388 | local_stream = NULL; | |
3bd1e081 | 2389 | |
50f8ae69 | 2390 | /* allocate for all fds + 1 for the consumer_data_pipe */ |
effcf122 | 2391 | pollfd = zmalloc((consumer_data.stream_count + 1) * sizeof(struct pollfd)); |
3bd1e081 | 2392 | if (pollfd == NULL) { |
7a57cf92 | 2393 | PERROR("pollfd malloc"); |
3bd1e081 MD |
2394 | pthread_mutex_unlock(&consumer_data.lock); |
2395 | goto end; | |
2396 | } | |
2397 | ||
50f8ae69 | 2398 | /* allocate for all fds + 1 for the consumer_data_pipe */ |
effcf122 | 2399 | local_stream = zmalloc((consumer_data.stream_count + 1) * |
747f8642 | 2400 | sizeof(struct lttng_consumer_stream *)); |
3bd1e081 | 2401 | if (local_stream == NULL) { |
7a57cf92 | 2402 | PERROR("local_stream malloc"); |
3bd1e081 MD |
2403 | pthread_mutex_unlock(&consumer_data.lock); |
2404 | goto end; | |
2405 | } | |
ffe60014 | 2406 | ret = update_poll_array(ctx, &pollfd, local_stream, |
43c34bc3 | 2407 | data_ht); |
3bd1e081 MD |
2408 | if (ret < 0) { |
2409 | ERR("Error in allocating pollfd or local_outfds"); | |
f73fabfd | 2410 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2411 | pthread_mutex_unlock(&consumer_data.lock); |
2412 | goto end; | |
2413 | } | |
2414 | nb_fd = ret; | |
2415 | consumer_data.need_update = 0; | |
2416 | } | |
2417 | pthread_mutex_unlock(&consumer_data.lock); | |
2418 | ||
4078b776 MD |
2419 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
2420 | if (nb_fd == 0 && consumer_quit == 1) { | |
1fc79fb4 | 2421 | err = 0; /* All is OK */ |
4078b776 MD |
2422 | goto end; |
2423 | } | |
3bd1e081 | 2424 | /* poll on the array of fds */ |
88f2b785 | 2425 | restart: |
3bd1e081 | 2426 | DBG("polling on %d fd", nb_fd + 1); |
9ce5646a | 2427 | health_poll_entry(); |
cb365c03 | 2428 | num_rdy = poll(pollfd, nb_fd + 1, -1); |
9ce5646a | 2429 | health_poll_exit(); |
3bd1e081 MD |
2430 | DBG("poll num_rdy : %d", num_rdy); |
2431 | if (num_rdy == -1) { | |
88f2b785 MD |
2432 | /* |
2433 | * Restart interrupted system call. | |
2434 | */ | |
2435 | if (errno == EINTR) { | |
2436 | goto restart; | |
2437 | } | |
7a57cf92 | 2438 | PERROR("Poll error"); |
f73fabfd | 2439 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2440 | goto end; |
2441 | } else if (num_rdy == 0) { | |
2442 | DBG("Polling thread timed out"); | |
2443 | goto end; | |
2444 | } | |
2445 | ||
3bd1e081 | 2446 | /* |
50f8ae69 | 2447 | * If the consumer_data_pipe triggered poll go directly to the |
00e2e675 DG |
2448 | * beginning of the loop to update the array. We want to prioritize |
2449 | * array update over low-priority reads. | |
3bd1e081 | 2450 | */ |
509bb1cf | 2451 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
ab30f567 | 2452 | ssize_t pipe_readlen; |
04fdd819 | 2453 | |
50f8ae69 | 2454 | DBG("consumer_data_pipe wake up"); |
acdb9057 DG |
2455 | pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe, |
2456 | &new_stream, sizeof(new_stream)); | |
6cd525e8 MD |
2457 | if (pipe_readlen < sizeof(new_stream)) { |
2458 | PERROR("Consumer data pipe"); | |
23f5f35d DG |
2459 | /* Continue so we can at least handle the current stream(s). */ |
2460 | continue; | |
2461 | } | |
c869f647 DG |
2462 | |
2463 | /* | |
2464 | * If the stream is NULL, just ignore it. It's also possible that | |
2465 | * the sessiond poll thread changed the consumer_quit state and is | |
2466 | * waking us up to test it. | |
2467 | */ | |
2468 | if (new_stream == NULL) { | |
8994307f | 2469 | validate_endpoint_status_data_stream(); |
c869f647 DG |
2470 | continue; |
2471 | } | |
2472 | ||
c869f647 | 2473 | /* Continue to update the local streams and handle prio ones */ |
3bd1e081 MD |
2474 | continue; |
2475 | } | |
2476 | ||
2477 | /* Take care of high priority channels first. */ | |
2478 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2479 | health_code_update(); |
2480 | ||
9617607b DG |
2481 | if (local_stream[i] == NULL) { |
2482 | continue; | |
2483 | } | |
fb3a43a9 | 2484 | if (pollfd[i].revents & POLLPRI) { |
d41f73b7 MD |
2485 | DBG("Urgent read on fd %d", pollfd[i].fd); |
2486 | high_prio = 1; | |
4078b776 | 2487 | len = ctx->on_buffer_ready(local_stream[i], ctx); |
d41f73b7 | 2488 | /* it's ok to have an unavailable sub-buffer */ |
b64403e3 | 2489 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2490 | /* Clean the stream and free it. */ |
2491 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2492 | local_stream[i] = NULL; |
4078b776 MD |
2493 | } else if (len > 0) { |
2494 | local_stream[i]->data_read = 1; | |
d41f73b7 | 2495 | } |
3bd1e081 MD |
2496 | } |
2497 | } | |
2498 | ||
4078b776 MD |
2499 | /* |
2500 | * If we read high prio channel in this loop, try again | |
2501 | * for more high prio data. | |
2502 | */ | |
2503 | if (high_prio) { | |
3bd1e081 MD |
2504 | continue; |
2505 | } | |
2506 | ||
2507 | /* Take care of low priority channels. */ | |
4078b776 | 2508 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2509 | health_code_update(); |
2510 | ||
9617607b DG |
2511 | if (local_stream[i] == NULL) { |
2512 | continue; | |
2513 | } | |
4078b776 MD |
2514 | if ((pollfd[i].revents & POLLIN) || |
2515 | local_stream[i]->hangup_flush_done) { | |
4078b776 MD |
2516 | DBG("Normal read on fd %d", pollfd[i].fd); |
2517 | len = ctx->on_buffer_ready(local_stream[i], ctx); | |
2518 | /* it's ok to have an unavailable sub-buffer */ | |
b64403e3 | 2519 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2520 | /* Clean the stream and free it. */ |
2521 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2522 | local_stream[i] = NULL; |
4078b776 MD |
2523 | } else if (len > 0) { |
2524 | local_stream[i]->data_read = 1; | |
2525 | } | |
2526 | } | |
2527 | } | |
2528 | ||
2529 | /* Handle hangup and errors */ | |
2530 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2531 | health_code_update(); |
2532 | ||
9617607b DG |
2533 | if (local_stream[i] == NULL) { |
2534 | continue; | |
2535 | } | |
4078b776 MD |
2536 | if (!local_stream[i]->hangup_flush_done |
2537 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) | |
2538 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2539 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2540 | DBG("fd %d is hup|err|nval. Attempting flush and read.", | |
9617607b | 2541 | pollfd[i].fd); |
4078b776 MD |
2542 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); |
2543 | /* Attempt read again, for the data we just flushed. */ | |
2544 | local_stream[i]->data_read = 1; | |
2545 | } | |
2546 | /* | |
2547 | * If the poll flag is HUP/ERR/NVAL and we have | |
2548 | * read no data in this pass, we can remove the | |
2549 | * stream from its hash table. | |
2550 | */ | |
2551 | if ((pollfd[i].revents & POLLHUP)) { | |
2552 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); | |
2553 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2554 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2555 | local_stream[i] = NULL; |
4078b776 MD |
2556 | num_hup++; |
2557 | } | |
2558 | } else if (pollfd[i].revents & POLLERR) { | |
2559 | ERR("Error returned in polling fd %d.", pollfd[i].fd); | |
2560 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2561 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2562 | local_stream[i] = NULL; |
4078b776 MD |
2563 | num_hup++; |
2564 | } | |
2565 | } else if (pollfd[i].revents & POLLNVAL) { | |
2566 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); | |
2567 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2568 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2569 | local_stream[i] = NULL; |
4078b776 | 2570 | num_hup++; |
3bd1e081 MD |
2571 | } |
2572 | } | |
9617607b DG |
2573 | if (local_stream[i] != NULL) { |
2574 | local_stream[i]->data_read = 0; | |
2575 | } | |
3bd1e081 MD |
2576 | } |
2577 | } | |
1fc79fb4 MD |
2578 | /* All is OK */ |
2579 | err = 0; | |
3bd1e081 MD |
2580 | end: |
2581 | DBG("polling thread exiting"); | |
0e428499 DG |
2582 | free(pollfd); |
2583 | free(local_stream); | |
fb3a43a9 DG |
2584 | |
2585 | /* | |
2586 | * Close the write side of the pipe so epoll_wait() in | |
7d980def DG |
2587 | * consumer_thread_metadata_poll can catch it. The thread is monitoring the |
2588 | * read side of the pipe. If we close them both, epoll_wait strangely does | |
2589 | * not return and could create a endless wait period if the pipe is the | |
2590 | * only tracked fd in the poll set. The thread will take care of closing | |
2591 | * the read side. | |
fb3a43a9 | 2592 | */ |
13886d2d | 2593 | (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe); |
fb3a43a9 | 2594 | |
ede905e6 | 2595 | error_testpoint: |
1fc79fb4 MD |
2596 | if (err) { |
2597 | health_error(); | |
2598 | ERR("Health error occurred in %s", __func__); | |
2599 | } | |
2600 | health_unregister(health_consumerd); | |
2601 | ||
e7b994a3 | 2602 | rcu_unregister_thread(); |
3bd1e081 MD |
2603 | return NULL; |
2604 | } | |
2605 | ||
d8ef542d MD |
2606 | /* |
2607 | * Close wake-up end of each stream belonging to the channel. This will | |
2608 | * allow the poll() on the stream read-side to detect when the | |
2609 | * write-side (application) finally closes them. | |
2610 | */ | |
2611 | static | |
2612 | void consumer_close_channel_streams(struct lttng_consumer_channel *channel) | |
2613 | { | |
2614 | struct lttng_ht *ht; | |
2615 | struct lttng_consumer_stream *stream; | |
2616 | struct lttng_ht_iter iter; | |
2617 | ||
2618 | ht = consumer_data.stream_per_chan_id_ht; | |
2619 | ||
2620 | rcu_read_lock(); | |
2621 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
2622 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
2623 | ht->match_fct, &channel->key, | |
2624 | &iter.iter, stream, node_channel_id.node) { | |
f2ad556d MD |
2625 | /* |
2626 | * Protect against teardown with mutex. | |
2627 | */ | |
2628 | pthread_mutex_lock(&stream->lock); | |
2629 | if (cds_lfht_is_node_deleted(&stream->node.node)) { | |
2630 | goto next; | |
2631 | } | |
d8ef542d MD |
2632 | switch (consumer_data.type) { |
2633 | case LTTNG_CONSUMER_KERNEL: | |
2634 | break; | |
2635 | case LTTNG_CONSUMER32_UST: | |
2636 | case LTTNG_CONSUMER64_UST: | |
509dbbbe DG |
2637 | if (stream->metadata_flag) { |
2638 | /* Safe and protected by the stream lock. */ | |
2639 | lttng_ustconsumer_close_metadata(stream->chan); | |
2640 | } else { | |
2641 | /* | |
2642 | * Note: a mutex is taken internally within | |
2643 | * liblttng-ust-ctl to protect timer wakeup_fd | |
2644 | * use from concurrent close. | |
2645 | */ | |
2646 | lttng_ustconsumer_close_stream_wakeup(stream); | |
2647 | } | |
d8ef542d MD |
2648 | break; |
2649 | default: | |
2650 | ERR("Unknown consumer_data type"); | |
2651 | assert(0); | |
2652 | } | |
f2ad556d MD |
2653 | next: |
2654 | pthread_mutex_unlock(&stream->lock); | |
d8ef542d MD |
2655 | } |
2656 | rcu_read_unlock(); | |
2657 | } | |
2658 | ||
2659 | static void destroy_channel_ht(struct lttng_ht *ht) | |
2660 | { | |
2661 | struct lttng_ht_iter iter; | |
2662 | struct lttng_consumer_channel *channel; | |
2663 | int ret; | |
2664 | ||
2665 | if (ht == NULL) { | |
2666 | return; | |
2667 | } | |
2668 | ||
2669 | rcu_read_lock(); | |
2670 | cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) { | |
2671 | ret = lttng_ht_del(ht, &iter); | |
2672 | assert(ret != 0); | |
2673 | } | |
2674 | rcu_read_unlock(); | |
2675 | ||
2676 | lttng_ht_destroy(ht); | |
2677 | } | |
2678 | ||
2679 | /* | |
2680 | * This thread polls the channel fds to detect when they are being | |
2681 | * closed. It closes all related streams if the channel is detected as | |
2682 | * closed. It is currently only used as a shim layer for UST because the | |
2683 | * consumerd needs to keep the per-stream wakeup end of pipes open for | |
2684 | * periodical flush. | |
2685 | */ | |
2686 | void *consumer_thread_channel_poll(void *data) | |
2687 | { | |
1fc79fb4 | 2688 | int ret, i, pollfd, err = -1; |
d8ef542d MD |
2689 | uint32_t revents, nb_fd; |
2690 | struct lttng_consumer_channel *chan = NULL; | |
2691 | struct lttng_ht_iter iter; | |
2692 | struct lttng_ht_node_u64 *node; | |
2693 | struct lttng_poll_event events; | |
2694 | struct lttng_consumer_local_data *ctx = data; | |
2695 | struct lttng_ht *channel_ht; | |
2696 | ||
2697 | rcu_register_thread(); | |
2698 | ||
1fc79fb4 MD |
2699 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL); |
2700 | ||
ede905e6 MD |
2701 | if (testpoint(consumerd_thread_channel)) { |
2702 | goto error_testpoint; | |
2703 | } | |
2704 | ||
9ce5646a MD |
2705 | health_code_update(); |
2706 | ||
d8ef542d MD |
2707 | channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
2708 | if (!channel_ht) { | |
2709 | /* ENOMEM at this point. Better to bail out. */ | |
2710 | goto end_ht; | |
2711 | } | |
2712 | ||
2713 | DBG("Thread channel poll started"); | |
2714 | ||
2715 | /* Size is set to 1 for the consumer_channel pipe */ | |
2716 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2717 | if (ret < 0) { | |
2718 | ERR("Poll set creation failed"); | |
2719 | goto end_poll; | |
2720 | } | |
2721 | ||
2722 | ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN); | |
2723 | if (ret < 0) { | |
2724 | goto end; | |
2725 | } | |
2726 | ||
2727 | /* Main loop */ | |
2728 | DBG("Channel main loop started"); | |
2729 | ||
2730 | while (1) { | |
9ce5646a MD |
2731 | health_code_update(); |
2732 | ||
d8ef542d MD |
2733 | /* Only the channel pipe is set */ |
2734 | if (LTTNG_POLL_GETNB(&events) == 0 && consumer_quit == 1) { | |
1fc79fb4 | 2735 | err = 0; /* All is OK */ |
d8ef542d MD |
2736 | goto end; |
2737 | } | |
2738 | ||
2739 | restart: | |
2740 | DBG("Channel poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2741 | health_poll_entry(); |
d8ef542d | 2742 | ret = lttng_poll_wait(&events, -1); |
9ce5646a | 2743 | health_poll_exit(); |
d8ef542d MD |
2744 | DBG("Channel event catched in thread"); |
2745 | if (ret < 0) { | |
2746 | if (errno == EINTR) { | |
2747 | ERR("Poll EINTR catched"); | |
2748 | goto restart; | |
2749 | } | |
2750 | goto end; | |
2751 | } | |
2752 | ||
2753 | nb_fd = ret; | |
2754 | ||
2755 | /* From here, the event is a channel wait fd */ | |
2756 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2757 | health_code_update(); |
2758 | ||
d8ef542d MD |
2759 | revents = LTTNG_POLL_GETEV(&events, i); |
2760 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2761 | ||
2762 | /* Just don't waste time if no returned events for the fd */ | |
2763 | if (!revents) { | |
2764 | continue; | |
2765 | } | |
2766 | if (pollfd == ctx->consumer_channel_pipe[0]) { | |
2767 | if (revents & (LPOLLERR | LPOLLHUP)) { | |
2768 | DBG("Channel thread pipe hung up"); | |
2769 | /* | |
2770 | * Remove the pipe from the poll set and continue the loop | |
2771 | * since their might be data to consume. | |
2772 | */ | |
2773 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
2774 | continue; | |
2775 | } else if (revents & LPOLLIN) { | |
2776 | enum consumer_channel_action action; | |
a0cbdd2e | 2777 | uint64_t key; |
d8ef542d | 2778 | |
a0cbdd2e | 2779 | ret = read_channel_pipe(ctx, &chan, &key, &action); |
d8ef542d MD |
2780 | if (ret <= 0) { |
2781 | ERR("Error reading channel pipe"); | |
2782 | continue; | |
2783 | } | |
2784 | ||
2785 | switch (action) { | |
2786 | case CONSUMER_CHANNEL_ADD: | |
2787 | DBG("Adding channel %d to poll set", | |
2788 | chan->wait_fd); | |
2789 | ||
2790 | lttng_ht_node_init_u64(&chan->wait_fd_node, | |
2791 | chan->wait_fd); | |
c7260a81 | 2792 | rcu_read_lock(); |
d8ef542d MD |
2793 | lttng_ht_add_unique_u64(channel_ht, |
2794 | &chan->wait_fd_node); | |
c7260a81 | 2795 | rcu_read_unlock(); |
d8ef542d MD |
2796 | /* Add channel to the global poll events list */ |
2797 | lttng_poll_add(&events, chan->wait_fd, | |
2798 | LPOLLIN | LPOLLPRI); | |
2799 | break; | |
a0cbdd2e MD |
2800 | case CONSUMER_CHANNEL_DEL: |
2801 | { | |
509dbbbe DG |
2802 | /* |
2803 | * This command should never be called if the channel | |
2804 | * has streams monitored by either the data or metadata | |
2805 | * thread. The consumer only notify this thread with a | |
2806 | * channel del. command if it receives a destroy | |
2807 | * channel command from the session daemon that send it | |
2808 | * if a command prior to the GET_CHANNEL failed. | |
2809 | */ | |
2810 | ||
c7260a81 | 2811 | rcu_read_lock(); |
a0cbdd2e MD |
2812 | chan = consumer_find_channel(key); |
2813 | if (!chan) { | |
c7260a81 | 2814 | rcu_read_unlock(); |
a0cbdd2e MD |
2815 | ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key); |
2816 | break; | |
2817 | } | |
2818 | lttng_poll_del(&events, chan->wait_fd); | |
f623cc0b | 2819 | iter.iter.node = &chan->wait_fd_node.node; |
a0cbdd2e MD |
2820 | ret = lttng_ht_del(channel_ht, &iter); |
2821 | assert(ret == 0); | |
a0cbdd2e | 2822 | |
f2a444f1 DG |
2823 | switch (consumer_data.type) { |
2824 | case LTTNG_CONSUMER_KERNEL: | |
2825 | break; | |
2826 | case LTTNG_CONSUMER32_UST: | |
2827 | case LTTNG_CONSUMER64_UST: | |
6217f8c7 DG |
2828 | health_code_update(); |
2829 | /* Destroy streams that might have been left in the stream list. */ | |
2830 | clean_channel_stream_list(chan); | |
f2a444f1 DG |
2831 | break; |
2832 | default: | |
2833 | ERR("Unknown consumer_data type"); | |
2834 | assert(0); | |
2835 | } | |
2836 | ||
a0cbdd2e MD |
2837 | /* |
2838 | * Release our own refcount. Force channel deletion even if | |
2839 | * streams were not initialized. | |
2840 | */ | |
2841 | if (!uatomic_sub_return(&chan->refcount, 1)) { | |
2842 | consumer_del_channel(chan); | |
2843 | } | |
c7260a81 | 2844 | rcu_read_unlock(); |
a0cbdd2e MD |
2845 | goto restart; |
2846 | } | |
d8ef542d MD |
2847 | case CONSUMER_CHANNEL_QUIT: |
2848 | /* | |
2849 | * Remove the pipe from the poll set and continue the loop | |
2850 | * since their might be data to consume. | |
2851 | */ | |
2852 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
2853 | continue; | |
2854 | default: | |
2855 | ERR("Unknown action"); | |
2856 | break; | |
2857 | } | |
2858 | } | |
2859 | ||
2860 | /* Handle other stream */ | |
2861 | continue; | |
2862 | } | |
2863 | ||
2864 | rcu_read_lock(); | |
2865 | { | |
2866 | uint64_t tmp_id = (uint64_t) pollfd; | |
2867 | ||
2868 | lttng_ht_lookup(channel_ht, &tmp_id, &iter); | |
2869 | } | |
2870 | node = lttng_ht_iter_get_node_u64(&iter); | |
2871 | assert(node); | |
2872 | ||
2873 | chan = caa_container_of(node, struct lttng_consumer_channel, | |
2874 | wait_fd_node); | |
2875 | ||
2876 | /* Check for error event */ | |
2877 | if (revents & (LPOLLERR | LPOLLHUP)) { | |
2878 | DBG("Channel fd %d is hup|err.", pollfd); | |
2879 | ||
2880 | lttng_poll_del(&events, chan->wait_fd); | |
2881 | ret = lttng_ht_del(channel_ht, &iter); | |
2882 | assert(ret == 0); | |
509dbbbe DG |
2883 | |
2884 | /* | |
2885 | * This will close the wait fd for each stream associated to | |
2886 | * this channel AND monitored by the data/metadata thread thus | |
2887 | * will be clean by the right thread. | |
2888 | */ | |
d8ef542d | 2889 | consumer_close_channel_streams(chan); |
f2ad556d MD |
2890 | |
2891 | /* Release our own refcount */ | |
2892 | if (!uatomic_sub_return(&chan->refcount, 1) | |
2893 | && !uatomic_read(&chan->nb_init_stream_left)) { | |
2894 | consumer_del_channel(chan); | |
2895 | } | |
d8ef542d MD |
2896 | } |
2897 | ||
2898 | /* Release RCU lock for the channel looked up */ | |
2899 | rcu_read_unlock(); | |
2900 | } | |
2901 | } | |
2902 | ||
1fc79fb4 MD |
2903 | /* All is OK */ |
2904 | err = 0; | |
d8ef542d MD |
2905 | end: |
2906 | lttng_poll_clean(&events); | |
2907 | end_poll: | |
2908 | destroy_channel_ht(channel_ht); | |
2909 | end_ht: | |
ede905e6 | 2910 | error_testpoint: |
d8ef542d | 2911 | DBG("Channel poll thread exiting"); |
1fc79fb4 MD |
2912 | if (err) { |
2913 | health_error(); | |
2914 | ERR("Health error occurred in %s", __func__); | |
2915 | } | |
2916 | health_unregister(health_consumerd); | |
d8ef542d MD |
2917 | rcu_unregister_thread(); |
2918 | return NULL; | |
2919 | } | |
2920 | ||
331744e3 JD |
2921 | static int set_metadata_socket(struct lttng_consumer_local_data *ctx, |
2922 | struct pollfd *sockpoll, int client_socket) | |
2923 | { | |
2924 | int ret; | |
2925 | ||
2926 | assert(ctx); | |
2927 | assert(sockpoll); | |
2928 | ||
2929 | if (lttng_consumer_poll_socket(sockpoll) < 0) { | |
2930 | ret = -1; | |
2931 | goto error; | |
2932 | } | |
2933 | DBG("Metadata connection on client_socket"); | |
2934 | ||
2935 | /* Blocking call, waiting for transmission */ | |
2936 | ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket); | |
2937 | if (ctx->consumer_metadata_socket < 0) { | |
2938 | WARN("On accept metadata"); | |
2939 | ret = -1; | |
2940 | goto error; | |
2941 | } | |
2942 | ret = 0; | |
2943 | ||
2944 | error: | |
2945 | return ret; | |
2946 | } | |
2947 | ||
3bd1e081 MD |
2948 | /* |
2949 | * This thread listens on the consumerd socket and receives the file | |
2950 | * descriptors from the session daemon. | |
2951 | */ | |
7d980def | 2952 | void *consumer_thread_sessiond_poll(void *data) |
3bd1e081 | 2953 | { |
1fc79fb4 | 2954 | int sock = -1, client_socket, ret, err = -1; |
3bd1e081 MD |
2955 | /* |
2956 | * structure to poll for incoming data on communication socket avoids | |
2957 | * making blocking sockets. | |
2958 | */ | |
2959 | struct pollfd consumer_sockpoll[2]; | |
2960 | struct lttng_consumer_local_data *ctx = data; | |
2961 | ||
e7b994a3 DG |
2962 | rcu_register_thread(); |
2963 | ||
1fc79fb4 MD |
2964 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND); |
2965 | ||
ede905e6 MD |
2966 | if (testpoint(consumerd_thread_sessiond)) { |
2967 | goto error_testpoint; | |
2968 | } | |
2969 | ||
9ce5646a MD |
2970 | health_code_update(); |
2971 | ||
3bd1e081 MD |
2972 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
2973 | unlink(ctx->consumer_command_sock_path); | |
2974 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); | |
2975 | if (client_socket < 0) { | |
2976 | ERR("Cannot create command socket"); | |
2977 | goto end; | |
2978 | } | |
2979 | ||
2980 | ret = lttcomm_listen_unix_sock(client_socket); | |
2981 | if (ret < 0) { | |
2982 | goto end; | |
2983 | } | |
2984 | ||
32258573 | 2985 | DBG("Sending ready command to lttng-sessiond"); |
f73fabfd | 2986 | ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); |
3bd1e081 MD |
2987 | /* return < 0 on error, but == 0 is not fatal */ |
2988 | if (ret < 0) { | |
32258573 | 2989 | ERR("Error sending ready command to lttng-sessiond"); |
3bd1e081 MD |
2990 | goto end; |
2991 | } | |
2992 | ||
3bd1e081 MD |
2993 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ |
2994 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; | |
2995 | consumer_sockpoll[0].events = POLLIN | POLLPRI; | |
2996 | consumer_sockpoll[1].fd = client_socket; | |
2997 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
2998 | ||
2999 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
3000 | goto end; | |
3001 | } | |
3002 | DBG("Connection on client_socket"); | |
3003 | ||
3004 | /* Blocking call, waiting for transmission */ | |
3005 | sock = lttcomm_accept_unix_sock(client_socket); | |
534d2592 | 3006 | if (sock < 0) { |
3bd1e081 MD |
3007 | WARN("On accept"); |
3008 | goto end; | |
3009 | } | |
3bd1e081 | 3010 | |
331744e3 JD |
3011 | /* |
3012 | * Setup metadata socket which is the second socket connection on the | |
3013 | * command unix socket. | |
3014 | */ | |
3015 | ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket); | |
3016 | if (ret < 0) { | |
3017 | goto end; | |
3018 | } | |
3019 | ||
d96f09c6 DG |
3020 | /* This socket is not useful anymore. */ |
3021 | ret = close(client_socket); | |
3022 | if (ret < 0) { | |
3023 | PERROR("close client_socket"); | |
3024 | } | |
3025 | client_socket = -1; | |
3026 | ||
3bd1e081 MD |
3027 | /* update the polling structure to poll on the established socket */ |
3028 | consumer_sockpoll[1].fd = sock; | |
3029 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3030 | ||
3031 | while (1) { | |
9ce5646a MD |
3032 | health_code_update(); |
3033 | ||
3034 | health_poll_entry(); | |
3035 | ret = lttng_consumer_poll_socket(consumer_sockpoll); | |
3036 | health_poll_exit(); | |
3037 | if (ret < 0) { | |
3bd1e081 MD |
3038 | goto end; |
3039 | } | |
3040 | DBG("Incoming command on sock"); | |
3041 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
3042 | if (ret == -ENOENT) { | |
3043 | DBG("Received STOP command"); | |
3044 | goto end; | |
3045 | } | |
4cbc1a04 DG |
3046 | if (ret <= 0) { |
3047 | /* | |
3048 | * This could simply be a session daemon quitting. Don't output | |
3049 | * ERR() here. | |
3050 | */ | |
3051 | DBG("Communication interrupted on command socket"); | |
41ba6035 | 3052 | err = 0; |
3bd1e081 MD |
3053 | goto end; |
3054 | } | |
3055 | if (consumer_quit) { | |
3056 | DBG("consumer_thread_receive_fds received quit from signal"); | |
1fc79fb4 | 3057 | err = 0; /* All is OK */ |
3bd1e081 MD |
3058 | goto end; |
3059 | } | |
ffe60014 | 3060 | DBG("received command on sock"); |
3bd1e081 | 3061 | } |
1fc79fb4 MD |
3062 | /* All is OK */ |
3063 | err = 0; | |
3064 | ||
3bd1e081 | 3065 | end: |
ffe60014 | 3066 | DBG("Consumer thread sessiond poll exiting"); |
3bd1e081 | 3067 | |
d88aee68 DG |
3068 | /* |
3069 | * Close metadata streams since the producer is the session daemon which | |
3070 | * just died. | |
3071 | * | |
3072 | * NOTE: for now, this only applies to the UST tracer. | |
3073 | */ | |
59a6e1d8 | 3074 | lttng_consumer_close_all_metadata(); |
d88aee68 | 3075 | |
3bd1e081 MD |
3076 | /* |
3077 | * when all fds have hung up, the polling thread | |
3078 | * can exit cleanly | |
3079 | */ | |
3080 | consumer_quit = 1; | |
3081 | ||
04fdd819 | 3082 | /* |
c869f647 | 3083 | * Notify the data poll thread to poll back again and test the |
8994307f | 3084 | * consumer_quit state that we just set so to quit gracefully. |
04fdd819 | 3085 | */ |
acdb9057 | 3086 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
c869f647 | 3087 | |
a0cbdd2e | 3088 | notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT); |
d8ef542d | 3089 | |
5c635c72 MD |
3090 | notify_health_quit_pipe(health_quit_pipe); |
3091 | ||
d96f09c6 DG |
3092 | /* Cleaning up possibly open sockets. */ |
3093 | if (sock >= 0) { | |
3094 | ret = close(sock); | |
3095 | if (ret < 0) { | |
3096 | PERROR("close sock sessiond poll"); | |
3097 | } | |
3098 | } | |
3099 | if (client_socket >= 0) { | |
38476d24 | 3100 | ret = close(client_socket); |
d96f09c6 DG |
3101 | if (ret < 0) { |
3102 | PERROR("close client_socket sessiond poll"); | |
3103 | } | |
3104 | } | |
3105 | ||
ede905e6 | 3106 | error_testpoint: |
1fc79fb4 MD |
3107 | if (err) { |
3108 | health_error(); | |
3109 | ERR("Health error occurred in %s", __func__); | |
3110 | } | |
3111 | health_unregister(health_consumerd); | |
3112 | ||
e7b994a3 | 3113 | rcu_unregister_thread(); |
3bd1e081 MD |
3114 | return NULL; |
3115 | } | |
d41f73b7 | 3116 | |
4078b776 | 3117 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
d41f73b7 MD |
3118 | struct lttng_consumer_local_data *ctx) |
3119 | { | |
74251bb8 DG |
3120 | ssize_t ret; |
3121 | ||
3122 | pthread_mutex_lock(&stream->lock); | |
94d49140 JD |
3123 | if (stream->metadata_flag) { |
3124 | pthread_mutex_lock(&stream->metadata_rdv_lock); | |
3125 | } | |
74251bb8 | 3126 | |
d41f73b7 MD |
3127 | switch (consumer_data.type) { |
3128 | case LTTNG_CONSUMER_KERNEL: | |
74251bb8 DG |
3129 | ret = lttng_kconsumer_read_subbuffer(stream, ctx); |
3130 | break; | |
7753dea8 MD |
3131 | case LTTNG_CONSUMER32_UST: |
3132 | case LTTNG_CONSUMER64_UST: | |
74251bb8 DG |
3133 | ret = lttng_ustconsumer_read_subbuffer(stream, ctx); |
3134 | break; | |
d41f73b7 MD |
3135 | default: |
3136 | ERR("Unknown consumer_data type"); | |
3137 | assert(0); | |
74251bb8 DG |
3138 | ret = -ENOSYS; |
3139 | break; | |
d41f73b7 | 3140 | } |
74251bb8 | 3141 | |
94d49140 JD |
3142 | if (stream->metadata_flag) { |
3143 | pthread_cond_broadcast(&stream->metadata_rdv); | |
3144 | pthread_mutex_unlock(&stream->metadata_rdv_lock); | |
3145 | } | |
74251bb8 DG |
3146 | pthread_mutex_unlock(&stream->lock); |
3147 | return ret; | |
d41f73b7 MD |
3148 | } |
3149 | ||
3150 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
3151 | { | |
3152 | switch (consumer_data.type) { | |
3153 | case LTTNG_CONSUMER_KERNEL: | |
3154 | return lttng_kconsumer_on_recv_stream(stream); | |
7753dea8 MD |
3155 | case LTTNG_CONSUMER32_UST: |
3156 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
3157 | return lttng_ustconsumer_on_recv_stream(stream); |
3158 | default: | |
3159 | ERR("Unknown consumer_data type"); | |
3160 | assert(0); | |
3161 | return -ENOSYS; | |
3162 | } | |
3163 | } | |
e4421fec DG |
3164 | |
3165 | /* | |
3166 | * Allocate and set consumer data hash tables. | |
3167 | */ | |
3b6b25ef | 3168 | int lttng_consumer_init(void) |
e4421fec | 3169 | { |
d88aee68 | 3170 | consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
3b6b25ef MD |
3171 | if (!consumer_data.channel_ht) { |
3172 | goto error; | |
3173 | } | |
3174 | ||
d88aee68 | 3175 | consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
3b6b25ef MD |
3176 | if (!consumer_data.relayd_ht) { |
3177 | goto error; | |
3178 | } | |
3179 | ||
d88aee68 | 3180 | consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
3b6b25ef MD |
3181 | if (!consumer_data.stream_list_ht) { |
3182 | goto error; | |
3183 | } | |
3184 | ||
d8ef542d | 3185 | consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
3b6b25ef MD |
3186 | if (!consumer_data.stream_per_chan_id_ht) { |
3187 | goto error; | |
3188 | } | |
3189 | ||
3190 | data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3191 | if (!data_ht) { | |
3192 | goto error; | |
3193 | } | |
3194 | ||
3195 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3196 | if (!metadata_ht) { | |
3197 | goto error; | |
3198 | } | |
3199 | ||
3200 | return 0; | |
3201 | ||
3202 | error: | |
3203 | return -1; | |
e4421fec | 3204 | } |
7735ef9e DG |
3205 | |
3206 | /* | |
3207 | * Process the ADD_RELAYD command receive by a consumer. | |
3208 | * | |
3209 | * This will create a relayd socket pair and add it to the relayd hash table. | |
3210 | * The caller MUST acquire a RCU read side lock before calling it. | |
3211 | */ | |
da009f2c | 3212 | int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type, |
7735ef9e | 3213 | struct lttng_consumer_local_data *ctx, int sock, |
6151a90f | 3214 | struct pollfd *consumer_sockpoll, |
d3e2ba59 JD |
3215 | struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id, |
3216 | uint64_t relayd_session_id) | |
7735ef9e | 3217 | { |
cd2b09ed | 3218 | int fd = -1, ret = -1, relayd_created = 0; |
0c759fc9 | 3219 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
d4298c99 | 3220 | struct consumer_relayd_sock_pair *relayd = NULL; |
7735ef9e | 3221 | |
6151a90f JD |
3222 | assert(ctx); |
3223 | assert(relayd_sock); | |
3224 | ||
da009f2c | 3225 | DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx); |
7735ef9e DG |
3226 | |
3227 | /* Get relayd reference if exists. */ | |
3228 | relayd = consumer_find_relayd(net_seq_idx); | |
3229 | if (relayd == NULL) { | |
da009f2c | 3230 | assert(sock_type == LTTNG_STREAM_CONTROL); |
7735ef9e DG |
3231 | /* Not found. Allocate one. */ |
3232 | relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); | |
3233 | if (relayd == NULL) { | |
0d08d75e | 3234 | ret = -ENOMEM; |
618a6a28 MD |
3235 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
3236 | goto error; | |
0d08d75e | 3237 | } else { |
30319bcb | 3238 | relayd->sessiond_session_id = sessiond_id; |
0d08d75e | 3239 | relayd_created = 1; |
7735ef9e | 3240 | } |
0d08d75e DG |
3241 | |
3242 | /* | |
3243 | * This code path MUST continue to the consumer send status message to | |
3244 | * we can notify the session daemon and continue our work without | |
3245 | * killing everything. | |
3246 | */ | |
da009f2c MD |
3247 | } else { |
3248 | /* | |
3249 | * relayd key should never be found for control socket. | |
3250 | */ | |
3251 | assert(sock_type != LTTNG_STREAM_CONTROL); | |
0d08d75e DG |
3252 | } |
3253 | ||
3254 | /* First send a status message before receiving the fds. */ | |
0c759fc9 | 3255 | ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS); |
618a6a28 | 3256 | if (ret < 0) { |
0d08d75e | 3257 | /* Somehow, the session daemon is not responding anymore. */ |
618a6a28 MD |
3258 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); |
3259 | goto error_nosignal; | |
7735ef9e DG |
3260 | } |
3261 | ||
3262 | /* Poll on consumer socket. */ | |
3263 | if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) { | |
0d08d75e | 3264 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
7735ef9e | 3265 | ret = -EINTR; |
618a6a28 | 3266 | goto error_nosignal; |
7735ef9e DG |
3267 | } |
3268 | ||
3269 | /* Get relayd socket from session daemon */ | |
3270 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); | |
3271 | if (ret != sizeof(fd)) { | |
7735ef9e | 3272 | ret = -1; |
4028eeb9 | 3273 | fd = -1; /* Just in case it gets set with an invalid value. */ |
0d08d75e DG |
3274 | |
3275 | /* | |
3276 | * Failing to receive FDs might indicate a major problem such as | |
3277 | * reaching a fd limit during the receive where the kernel returns a | |
3278 | * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we | |
3279 | * don't take any chances and stop everything. | |
3280 | * | |
3281 | * XXX: Feature request #558 will fix that and avoid this possible | |
3282 | * issue when reaching the fd limit. | |
3283 | */ | |
3284 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); | |
618a6a28 | 3285 | ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD; |
f50f23d9 DG |
3286 | goto error; |
3287 | } | |
3288 | ||
7735ef9e DG |
3289 | /* Copy socket information and received FD */ |
3290 | switch (sock_type) { | |
3291 | case LTTNG_STREAM_CONTROL: | |
3292 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3293 | lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock); |
3294 | ret = lttcomm_create_sock(&relayd->control_sock.sock); | |
4028eeb9 | 3295 | /* Handle create_sock error. */ |
f66c074c | 3296 | if (ret < 0) { |
618a6a28 | 3297 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3298 | goto error; |
f66c074c | 3299 | } |
da009f2c MD |
3300 | /* |
3301 | * Close the socket created internally by | |
3302 | * lttcomm_create_sock, so we can replace it by the one | |
3303 | * received from sessiond. | |
3304 | */ | |
3305 | if (close(relayd->control_sock.sock.fd)) { | |
3306 | PERROR("close"); | |
3307 | } | |
7735ef9e DG |
3308 | |
3309 | /* Assign new file descriptor */ | |
6151a90f | 3310 | relayd->control_sock.sock.fd = fd; |
4b29f1ce | 3311 | fd = -1; /* For error path */ |
6151a90f JD |
3312 | /* Assign version values. */ |
3313 | relayd->control_sock.major = relayd_sock->major; | |
3314 | relayd->control_sock.minor = relayd_sock->minor; | |
c5b6f4f0 | 3315 | |
d3e2ba59 | 3316 | relayd->relayd_session_id = relayd_session_id; |
c5b6f4f0 | 3317 | |
7735ef9e DG |
3318 | break; |
3319 | case LTTNG_STREAM_DATA: | |
3320 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3321 | lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock); |
3322 | ret = lttcomm_create_sock(&relayd->data_sock.sock); | |
4028eeb9 | 3323 | /* Handle create_sock error. */ |
f66c074c | 3324 | if (ret < 0) { |
618a6a28 | 3325 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3326 | goto error; |
f66c074c | 3327 | } |
da009f2c MD |
3328 | /* |
3329 | * Close the socket created internally by | |
3330 | * lttcomm_create_sock, so we can replace it by the one | |
3331 | * received from sessiond. | |
3332 | */ | |
3333 | if (close(relayd->data_sock.sock.fd)) { | |
3334 | PERROR("close"); | |
3335 | } | |
7735ef9e DG |
3336 | |
3337 | /* Assign new file descriptor */ | |
6151a90f | 3338 | relayd->data_sock.sock.fd = fd; |
4b29f1ce | 3339 | fd = -1; /* for eventual error paths */ |
6151a90f JD |
3340 | /* Assign version values. */ |
3341 | relayd->data_sock.major = relayd_sock->major; | |
3342 | relayd->data_sock.minor = relayd_sock->minor; | |
7735ef9e DG |
3343 | break; |
3344 | default: | |
3345 | ERR("Unknown relayd socket type (%d)", sock_type); | |
59e71485 | 3346 | ret = -1; |
618a6a28 | 3347 | ret_code = LTTCOMM_CONSUMERD_FATAL; |
7735ef9e DG |
3348 | goto error; |
3349 | } | |
3350 | ||
d88aee68 | 3351 | DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)", |
7735ef9e DG |
3352 | sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", |
3353 | relayd->net_seq_idx, fd); | |
3354 | ||
618a6a28 MD |
3355 | /* We successfully added the socket. Send status back. */ |
3356 | ret = consumer_send_status_msg(sock, ret_code); | |
3357 | if (ret < 0) { | |
3358 | /* Somehow, the session daemon is not responding anymore. */ | |
3359 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3360 | goto error_nosignal; | |
3361 | } | |
3362 | ||
7735ef9e DG |
3363 | /* |
3364 | * Add relayd socket pair to consumer data hashtable. If object already | |
3365 | * exists or on error, the function gracefully returns. | |
3366 | */ | |
d09e1200 | 3367 | add_relayd(relayd); |
7735ef9e DG |
3368 | |
3369 | /* All good! */ | |
4028eeb9 | 3370 | return 0; |
7735ef9e DG |
3371 | |
3372 | error: | |
618a6a28 MD |
3373 | if (consumer_send_status_msg(sock, ret_code) < 0) { |
3374 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3375 | } | |
3376 | ||
3377 | error_nosignal: | |
4028eeb9 DG |
3378 | /* Close received socket if valid. */ |
3379 | if (fd >= 0) { | |
3380 | if (close(fd)) { | |
3381 | PERROR("close received socket"); | |
3382 | } | |
3383 | } | |
cd2b09ed DG |
3384 | |
3385 | if (relayd_created) { | |
cd2b09ed DG |
3386 | free(relayd); |
3387 | } | |
3388 | ||
7735ef9e DG |
3389 | return ret; |
3390 | } | |
ca22feea | 3391 | |
4e9a4686 DG |
3392 | /* |
3393 | * Try to lock the stream mutex. | |
3394 | * | |
3395 | * On success, 1 is returned else 0 indicating that the mutex is NOT lock. | |
3396 | */ | |
3397 | static int stream_try_lock(struct lttng_consumer_stream *stream) | |
3398 | { | |
3399 | int ret; | |
3400 | ||
3401 | assert(stream); | |
3402 | ||
3403 | /* | |
3404 | * Try to lock the stream mutex. On failure, we know that the stream is | |
3405 | * being used else where hence there is data still being extracted. | |
3406 | */ | |
3407 | ret = pthread_mutex_trylock(&stream->lock); | |
3408 | if (ret) { | |
3409 | /* For both EBUSY and EINVAL error, the mutex is NOT locked. */ | |
3410 | ret = 0; | |
3411 | goto end; | |
3412 | } | |
3413 | ||
3414 | ret = 1; | |
3415 | ||
3416 | end: | |
3417 | return ret; | |
3418 | } | |
3419 | ||
f7079f67 DG |
3420 | /* |
3421 | * Search for a relayd associated to the session id and return the reference. | |
3422 | * | |
3423 | * A rcu read side lock MUST be acquire before calling this function and locked | |
3424 | * until the relayd object is no longer necessary. | |
3425 | */ | |
3426 | static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id) | |
3427 | { | |
3428 | struct lttng_ht_iter iter; | |
f7079f67 | 3429 | struct consumer_relayd_sock_pair *relayd = NULL; |
f7079f67 DG |
3430 | |
3431 | /* Iterate over all relayd since they are indexed by net_seq_idx. */ | |
3432 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
3433 | node.node) { | |
18261bd1 DG |
3434 | /* |
3435 | * Check by sessiond id which is unique here where the relayd session | |
3436 | * id might not be when having multiple relayd. | |
3437 | */ | |
3438 | if (relayd->sessiond_session_id == id) { | |
f7079f67 | 3439 | /* Found the relayd. There can be only one per id. */ |
18261bd1 | 3440 | goto found; |
f7079f67 DG |
3441 | } |
3442 | } | |
3443 | ||
18261bd1 DG |
3444 | return NULL; |
3445 | ||
3446 | found: | |
f7079f67 DG |
3447 | return relayd; |
3448 | } | |
3449 | ||
ca22feea DG |
3450 | /* |
3451 | * Check if for a given session id there is still data needed to be extract | |
3452 | * from the buffers. | |
3453 | * | |
6d805429 | 3454 | * Return 1 if data is pending or else 0 meaning ready to be read. |
ca22feea | 3455 | */ |
6d805429 | 3456 | int consumer_data_pending(uint64_t id) |
ca22feea DG |
3457 | { |
3458 | int ret; | |
3459 | struct lttng_ht_iter iter; | |
3460 | struct lttng_ht *ht; | |
3461 | struct lttng_consumer_stream *stream; | |
f7079f67 | 3462 | struct consumer_relayd_sock_pair *relayd = NULL; |
6d805429 | 3463 | int (*data_pending)(struct lttng_consumer_stream *); |
ca22feea | 3464 | |
6d805429 | 3465 | DBG("Consumer data pending command on session id %" PRIu64, id); |
ca22feea | 3466 | |
6f6eda74 | 3467 | rcu_read_lock(); |
ca22feea DG |
3468 | pthread_mutex_lock(&consumer_data.lock); |
3469 | ||
3470 | switch (consumer_data.type) { | |
3471 | case LTTNG_CONSUMER_KERNEL: | |
6d805429 | 3472 | data_pending = lttng_kconsumer_data_pending; |
ca22feea DG |
3473 | break; |
3474 | case LTTNG_CONSUMER32_UST: | |
3475 | case LTTNG_CONSUMER64_UST: | |
6d805429 | 3476 | data_pending = lttng_ustconsumer_data_pending; |
ca22feea DG |
3477 | break; |
3478 | default: | |
3479 | ERR("Unknown consumer data type"); | |
3480 | assert(0); | |
3481 | } | |
3482 | ||
3483 | /* Ease our life a bit */ | |
3484 | ht = consumer_data.stream_list_ht; | |
3485 | ||
f7079f67 DG |
3486 | relayd = find_relayd_by_session_id(id); |
3487 | if (relayd) { | |
3488 | /* Send init command for data pending. */ | |
3489 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3490 | ret = relayd_begin_data_pending(&relayd->control_sock, | |
3491 | relayd->relayd_session_id); | |
3492 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3493 | if (ret < 0) { | |
3494 | /* Communication error thus the relayd so no data pending. */ | |
3495 | goto data_not_pending; | |
3496 | } | |
3497 | } | |
3498 | ||
c8f59ee5 | 3499 | cds_lfht_for_each_entry_duplicate(ht->ht, |
d88aee68 DG |
3500 | ht->hash_fct(&id, lttng_ht_seed), |
3501 | ht->match_fct, &id, | |
ca22feea | 3502 | &iter.iter, stream, node_session_id.node) { |
4e9a4686 DG |
3503 | /* If this call fails, the stream is being used hence data pending. */ |
3504 | ret = stream_try_lock(stream); | |
3505 | if (!ret) { | |
f7079f67 | 3506 | goto data_pending; |
ca22feea | 3507 | } |
ca22feea | 3508 | |
4e9a4686 DG |
3509 | /* |
3510 | * A removed node from the hash table indicates that the stream has | |
3511 | * been deleted thus having a guarantee that the buffers are closed | |
3512 | * on the consumer side. However, data can still be transmitted | |
3513 | * over the network so don't skip the relayd check. | |
3514 | */ | |
3515 | ret = cds_lfht_is_node_deleted(&stream->node.node); | |
3516 | if (!ret) { | |
e5d1a9b3 MD |
3517 | /* |
3518 | * An empty output file is not valid. We need at least one packet | |
3519 | * generated per stream, even if it contains no event, so it | |
3520 | * contains at least one packet header. | |
3521 | */ | |
3522 | if (stream->output_written == 0) { | |
3523 | pthread_mutex_unlock(&stream->lock); | |
3524 | goto data_pending; | |
3525 | } | |
4e9a4686 | 3526 | /* Check the stream if there is data in the buffers. */ |
6d805429 DG |
3527 | ret = data_pending(stream); |
3528 | if (ret == 1) { | |
4e9a4686 | 3529 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3530 | goto data_pending; |
4e9a4686 DG |
3531 | } |
3532 | } | |
3533 | ||
3534 | /* Relayd check */ | |
f7079f67 | 3535 | if (relayd) { |
c8f59ee5 DG |
3536 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
3537 | if (stream->metadata_flag) { | |
ad7051c0 DG |
3538 | ret = relayd_quiescent_control(&relayd->control_sock, |
3539 | stream->relayd_stream_id); | |
c8f59ee5 | 3540 | } else { |
6d805429 | 3541 | ret = relayd_data_pending(&relayd->control_sock, |
39df6d9f DG |
3542 | stream->relayd_stream_id, |
3543 | stream->next_net_seq_num - 1); | |
c8f59ee5 DG |
3544 | } |
3545 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
6d805429 | 3546 | if (ret == 1) { |
4e9a4686 | 3547 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3548 | goto data_pending; |
c8f59ee5 DG |
3549 | } |
3550 | } | |
4e9a4686 | 3551 | pthread_mutex_unlock(&stream->lock); |
c8f59ee5 | 3552 | } |
ca22feea | 3553 | |
f7079f67 DG |
3554 | if (relayd) { |
3555 | unsigned int is_data_inflight = 0; | |
3556 | ||
3557 | /* Send init command for data pending. */ | |
3558 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3559 | ret = relayd_end_data_pending(&relayd->control_sock, | |
3560 | relayd->relayd_session_id, &is_data_inflight); | |
3561 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
bdd88757 | 3562 | if (ret < 0) { |
f7079f67 DG |
3563 | goto data_not_pending; |
3564 | } | |
bdd88757 DG |
3565 | if (is_data_inflight) { |
3566 | goto data_pending; | |
3567 | } | |
f7079f67 DG |
3568 | } |
3569 | ||
ca22feea | 3570 | /* |
f7079f67 DG |
3571 | * Finding _no_ node in the hash table and no inflight data means that the |
3572 | * stream(s) have been removed thus data is guaranteed to be available for | |
3573 | * analysis from the trace files. | |
ca22feea DG |
3574 | */ |
3575 | ||
f7079f67 | 3576 | data_not_pending: |
ca22feea DG |
3577 | /* Data is available to be read by a viewer. */ |
3578 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3579 | rcu_read_unlock(); |
6d805429 | 3580 | return 0; |
ca22feea | 3581 | |
f7079f67 | 3582 | data_pending: |
ca22feea DG |
3583 | /* Data is still being extracted from buffers. */ |
3584 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3585 | rcu_read_unlock(); |
6d805429 | 3586 | return 1; |
ca22feea | 3587 | } |
f50f23d9 DG |
3588 | |
3589 | /* | |
3590 | * Send a ret code status message to the sessiond daemon. | |
3591 | * | |
3592 | * Return the sendmsg() return value. | |
3593 | */ | |
3594 | int consumer_send_status_msg(int sock, int ret_code) | |
3595 | { | |
3596 | struct lttcomm_consumer_status_msg msg; | |
3597 | ||
3598 | msg.ret_code = ret_code; | |
3599 | ||
3600 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3601 | } | |
ffe60014 DG |
3602 | |
3603 | /* | |
3604 | * Send a channel status message to the sessiond daemon. | |
3605 | * | |
3606 | * Return the sendmsg() return value. | |
3607 | */ | |
3608 | int consumer_send_status_channel(int sock, | |
3609 | struct lttng_consumer_channel *channel) | |
3610 | { | |
3611 | struct lttcomm_consumer_status_channel msg; | |
3612 | ||
3613 | assert(sock >= 0); | |
3614 | ||
3615 | if (!channel) { | |
0c759fc9 | 3616 | msg.ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL; |
ffe60014 | 3617 | } else { |
0c759fc9 | 3618 | msg.ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
ffe60014 DG |
3619 | msg.key = channel->key; |
3620 | msg.stream_count = channel->streams.count; | |
3621 | } | |
3622 | ||
3623 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3624 | } | |
5c786ded JD |
3625 | |
3626 | /* | |
3627 | * Using a maximum stream size with the produced and consumed position of a | |
3628 | * stream, computes the new consumed position to be as close as possible to the | |
3629 | * maximum possible stream size. | |
3630 | * | |
3631 | * If maximum stream size is lower than the possible buffer size (produced - | |
3632 | * consumed), the consumed_pos given is returned untouched else the new value | |
3633 | * is returned. | |
3634 | */ | |
3635 | unsigned long consumer_get_consumed_maxsize(unsigned long consumed_pos, | |
3636 | unsigned long produced_pos, uint64_t max_stream_size) | |
3637 | { | |
3638 | if (max_stream_size && max_stream_size < (produced_pos - consumed_pos)) { | |
3639 | /* Offset from the produced position to get the latest buffers. */ | |
3640 | return produced_pos - max_stream_size; | |
3641 | } | |
3642 | ||
3643 | return consumed_pos; | |
3644 | } |