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> | |
f263b7fd | 37 | #include <common/compat/endian.h> |
309167d2 | 38 | #include <common/index/index.h> |
10a8a223 | 39 | #include <common/kernel-ctl/kernel-ctl.h> |
00e2e675 | 40 | #include <common/sessiond-comm/relayd.h> |
10a8a223 DG |
41 | #include <common/sessiond-comm/sessiond-comm.h> |
42 | #include <common/kernel-consumer/kernel-consumer.h> | |
00e2e675 | 43 | #include <common/relayd/relayd.h> |
10a8a223 | 44 | #include <common/ust-consumer/ust-consumer.h> |
d3e2ba59 | 45 | #include <common/consumer-timer.h> |
10a8a223 DG |
46 | |
47 | #include "consumer.h" | |
1d1a276c | 48 | #include "consumer-stream.h" |
2d57de81 | 49 | #include "consumer-testpoint.h" |
135347bd | 50 | #include "align.h" |
3bd1e081 MD |
51 | |
52 | struct lttng_consumer_global_data consumer_data = { | |
3bd1e081 MD |
53 | .stream_count = 0, |
54 | .need_update = 1, | |
55 | .type = LTTNG_CONSUMER_UNKNOWN, | |
56 | }; | |
57 | ||
d8ef542d MD |
58 | enum consumer_channel_action { |
59 | CONSUMER_CHANNEL_ADD, | |
a0cbdd2e | 60 | CONSUMER_CHANNEL_DEL, |
d8ef542d MD |
61 | CONSUMER_CHANNEL_QUIT, |
62 | }; | |
63 | ||
64 | struct consumer_channel_msg { | |
65 | enum consumer_channel_action action; | |
a0cbdd2e MD |
66 | struct lttng_consumer_channel *chan; /* add */ |
67 | uint64_t key; /* del */ | |
d8ef542d MD |
68 | }; |
69 | ||
3bd1e081 MD |
70 | /* |
71 | * Flag to inform the polling thread to quit when all fd hung up. Updated by | |
72 | * the consumer_thread_receive_fds when it notices that all fds has hung up. | |
73 | * Also updated by the signal handler (consumer_should_exit()). Read by the | |
74 | * polling threads. | |
75 | */ | |
a98dae5f | 76 | volatile int consumer_quit; |
3bd1e081 | 77 | |
43c34bc3 | 78 | /* |
43c34bc3 DG |
79 | * Global hash table containing respectively metadata and data streams. The |
80 | * stream element in this ht should only be updated by the metadata poll thread | |
81 | * for the metadata and the data poll thread for the data. | |
82 | */ | |
40dc48e0 DG |
83 | static struct lttng_ht *metadata_ht; |
84 | static struct lttng_ht *data_ht; | |
43c34bc3 | 85 | |
acdb9057 DG |
86 | /* |
87 | * Notify a thread lttng pipe to poll back again. This usually means that some | |
88 | * global state has changed so we just send back the thread in a poll wait | |
89 | * call. | |
90 | */ | |
91 | static void notify_thread_lttng_pipe(struct lttng_pipe *pipe) | |
92 | { | |
93 | struct lttng_consumer_stream *null_stream = NULL; | |
94 | ||
95 | assert(pipe); | |
96 | ||
97 | (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream)); | |
98 | } | |
99 | ||
5c635c72 MD |
100 | static void notify_health_quit_pipe(int *pipe) |
101 | { | |
6cd525e8 | 102 | ssize_t ret; |
5c635c72 | 103 | |
6cd525e8 MD |
104 | ret = lttng_write(pipe[1], "4", 1); |
105 | if (ret < 1) { | |
5c635c72 MD |
106 | PERROR("write consumer health quit"); |
107 | } | |
108 | } | |
109 | ||
d8ef542d MD |
110 | static void notify_channel_pipe(struct lttng_consumer_local_data *ctx, |
111 | struct lttng_consumer_channel *chan, | |
a0cbdd2e | 112 | uint64_t key, |
d8ef542d MD |
113 | enum consumer_channel_action action) |
114 | { | |
115 | struct consumer_channel_msg msg; | |
6cd525e8 | 116 | ssize_t ret; |
d8ef542d | 117 | |
e56251fc DG |
118 | memset(&msg, 0, sizeof(msg)); |
119 | ||
d8ef542d MD |
120 | msg.action = action; |
121 | msg.chan = chan; | |
f21dae48 | 122 | msg.key = key; |
6cd525e8 MD |
123 | ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg)); |
124 | if (ret < sizeof(msg)) { | |
125 | PERROR("notify_channel_pipe write error"); | |
126 | } | |
d8ef542d MD |
127 | } |
128 | ||
a0cbdd2e MD |
129 | void notify_thread_del_channel(struct lttng_consumer_local_data *ctx, |
130 | uint64_t key) | |
131 | { | |
132 | notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL); | |
133 | } | |
134 | ||
d8ef542d MD |
135 | static int read_channel_pipe(struct lttng_consumer_local_data *ctx, |
136 | struct lttng_consumer_channel **chan, | |
a0cbdd2e | 137 | uint64_t *key, |
d8ef542d MD |
138 | enum consumer_channel_action *action) |
139 | { | |
140 | struct consumer_channel_msg msg; | |
6cd525e8 | 141 | ssize_t ret; |
d8ef542d | 142 | |
6cd525e8 MD |
143 | ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg)); |
144 | if (ret < sizeof(msg)) { | |
145 | ret = -1; | |
146 | goto error; | |
d8ef542d | 147 | } |
6cd525e8 MD |
148 | *action = msg.action; |
149 | *chan = msg.chan; | |
150 | *key = msg.key; | |
151 | error: | |
152 | return (int) ret; | |
d8ef542d MD |
153 | } |
154 | ||
212d67a2 DG |
155 | /* |
156 | * Cleanup the stream list of a channel. Those streams are not yet globally | |
157 | * visible | |
158 | */ | |
159 | static void clean_channel_stream_list(struct lttng_consumer_channel *channel) | |
160 | { | |
161 | struct lttng_consumer_stream *stream, *stmp; | |
162 | ||
163 | assert(channel); | |
164 | ||
165 | /* Delete streams that might have been left in the stream list. */ | |
166 | cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head, | |
167 | send_node) { | |
168 | cds_list_del(&stream->send_node); | |
169 | /* | |
170 | * Once a stream is added to this list, the buffers were created so we | |
171 | * have a guarantee that this call will succeed. Setting the monitor | |
172 | * mode to 0 so we don't lock nor try to delete the stream from the | |
173 | * global hash table. | |
174 | */ | |
175 | stream->monitor = 0; | |
176 | consumer_stream_destroy(stream, NULL); | |
177 | } | |
178 | } | |
179 | ||
3bd1e081 MD |
180 | /* |
181 | * Find a stream. The consumer_data.lock must be locked during this | |
182 | * call. | |
183 | */ | |
d88aee68 | 184 | static struct lttng_consumer_stream *find_stream(uint64_t key, |
8389e4f8 | 185 | struct lttng_ht *ht) |
3bd1e081 | 186 | { |
e4421fec | 187 | struct lttng_ht_iter iter; |
d88aee68 | 188 | struct lttng_ht_node_u64 *node; |
e4421fec | 189 | struct lttng_consumer_stream *stream = NULL; |
3bd1e081 | 190 | |
8389e4f8 DG |
191 | assert(ht); |
192 | ||
d88aee68 DG |
193 | /* -1ULL keys are lookup failures */ |
194 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 195 | return NULL; |
7a57cf92 | 196 | } |
e4421fec | 197 | |
6065ceec DG |
198 | rcu_read_lock(); |
199 | ||
d88aee68 DG |
200 | lttng_ht_lookup(ht, &key, &iter); |
201 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
202 | if (node != NULL) { |
203 | stream = caa_container_of(node, struct lttng_consumer_stream, node); | |
3bd1e081 | 204 | } |
e4421fec | 205 | |
6065ceec DG |
206 | rcu_read_unlock(); |
207 | ||
e4421fec | 208 | return stream; |
3bd1e081 MD |
209 | } |
210 | ||
da009f2c | 211 | static void steal_stream_key(uint64_t key, struct lttng_ht *ht) |
7ad0a0cb MD |
212 | { |
213 | struct lttng_consumer_stream *stream; | |
214 | ||
04253271 | 215 | rcu_read_lock(); |
ffe60014 | 216 | stream = find_stream(key, ht); |
04253271 | 217 | if (stream) { |
da009f2c | 218 | stream->key = (uint64_t) -1ULL; |
04253271 MD |
219 | /* |
220 | * We don't want the lookup to match, but we still need | |
221 | * to iterate on this stream when iterating over the hash table. Just | |
222 | * change the node key. | |
223 | */ | |
da009f2c | 224 | stream->node.key = (uint64_t) -1ULL; |
04253271 MD |
225 | } |
226 | rcu_read_unlock(); | |
7ad0a0cb MD |
227 | } |
228 | ||
d56db448 DG |
229 | /* |
230 | * Return a channel object for the given key. | |
231 | * | |
232 | * RCU read side lock MUST be acquired before calling this function and | |
233 | * protects the channel ptr. | |
234 | */ | |
d88aee68 | 235 | struct lttng_consumer_channel *consumer_find_channel(uint64_t key) |
3bd1e081 | 236 | { |
e4421fec | 237 | struct lttng_ht_iter iter; |
d88aee68 | 238 | struct lttng_ht_node_u64 *node; |
e4421fec | 239 | struct lttng_consumer_channel *channel = NULL; |
3bd1e081 | 240 | |
d88aee68 DG |
241 | /* -1ULL keys are lookup failures */ |
242 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 243 | return NULL; |
7a57cf92 | 244 | } |
e4421fec | 245 | |
d88aee68 DG |
246 | lttng_ht_lookup(consumer_data.channel_ht, &key, &iter); |
247 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
248 | if (node != NULL) { |
249 | channel = caa_container_of(node, struct lttng_consumer_channel, node); | |
3bd1e081 | 250 | } |
e4421fec DG |
251 | |
252 | return channel; | |
3bd1e081 MD |
253 | } |
254 | ||
b5a6470f DG |
255 | /* |
256 | * There is a possibility that the consumer does not have enough time between | |
257 | * the close of the channel on the session daemon and the cleanup in here thus | |
258 | * once we have a channel add with an existing key, we know for sure that this | |
259 | * channel will eventually get cleaned up by all streams being closed. | |
260 | * | |
261 | * This function just nullifies the already existing channel key. | |
262 | */ | |
263 | static void steal_channel_key(uint64_t key) | |
264 | { | |
265 | struct lttng_consumer_channel *channel; | |
266 | ||
267 | rcu_read_lock(); | |
268 | channel = consumer_find_channel(key); | |
269 | if (channel) { | |
270 | channel->key = (uint64_t) -1ULL; | |
271 | /* | |
272 | * We don't want the lookup to match, but we still need to iterate on | |
273 | * this channel when iterating over the hash table. Just change the | |
274 | * node key. | |
275 | */ | |
276 | channel->node.key = (uint64_t) -1ULL; | |
277 | } | |
278 | rcu_read_unlock(); | |
279 | } | |
280 | ||
ffe60014 | 281 | static void free_channel_rcu(struct rcu_head *head) |
702b1ea4 | 282 | { |
d88aee68 DG |
283 | struct lttng_ht_node_u64 *node = |
284 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
ffe60014 DG |
285 | struct lttng_consumer_channel *channel = |
286 | caa_container_of(node, struct lttng_consumer_channel, node); | |
702b1ea4 | 287 | |
4534c65b MD |
288 | switch (consumer_data.type) { |
289 | case LTTNG_CONSUMER_KERNEL: | |
290 | break; | |
291 | case LTTNG_CONSUMER32_UST: | |
292 | case LTTNG_CONSUMER64_UST: | |
293 | lttng_ustconsumer_free_channel(channel); | |
294 | break; | |
295 | default: | |
296 | ERR("Unknown consumer_data type"); | |
297 | abort(); | |
298 | } | |
ffe60014 | 299 | free(channel); |
702b1ea4 MD |
300 | } |
301 | ||
00e2e675 DG |
302 | /* |
303 | * RCU protected relayd socket pair free. | |
304 | */ | |
ffe60014 | 305 | static void free_relayd_rcu(struct rcu_head *head) |
00e2e675 | 306 | { |
d88aee68 DG |
307 | struct lttng_ht_node_u64 *node = |
308 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
00e2e675 DG |
309 | struct consumer_relayd_sock_pair *relayd = |
310 | caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
311 | ||
8994307f DG |
312 | /* |
313 | * Close all sockets. This is done in the call RCU since we don't want the | |
314 | * socket fds to be reassigned thus potentially creating bad state of the | |
315 | * relayd object. | |
316 | * | |
317 | * We do not have to lock the control socket mutex here since at this stage | |
318 | * there is no one referencing to this relayd object. | |
319 | */ | |
320 | (void) relayd_close(&relayd->control_sock); | |
321 | (void) relayd_close(&relayd->data_sock); | |
322 | ||
00e2e675 DG |
323 | free(relayd); |
324 | } | |
325 | ||
326 | /* | |
327 | * Destroy and free relayd socket pair object. | |
00e2e675 | 328 | */ |
51230d70 | 329 | void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
330 | { |
331 | int ret; | |
332 | struct lttng_ht_iter iter; | |
333 | ||
173af62f DG |
334 | if (relayd == NULL) { |
335 | return; | |
336 | } | |
337 | ||
00e2e675 DG |
338 | DBG("Consumer destroy and close relayd socket pair"); |
339 | ||
340 | iter.iter.node = &relayd->node.node; | |
341 | ret = lttng_ht_del(consumer_data.relayd_ht, &iter); | |
173af62f | 342 | if (ret != 0) { |
8994307f | 343 | /* We assume the relayd is being or is destroyed */ |
173af62f DG |
344 | return; |
345 | } | |
00e2e675 | 346 | |
00e2e675 | 347 | /* RCU free() call */ |
ffe60014 DG |
348 | call_rcu(&relayd->node.head, free_relayd_rcu); |
349 | } | |
350 | ||
351 | /* | |
352 | * Remove a channel from the global list protected by a mutex. This function is | |
353 | * also responsible for freeing its data structures. | |
354 | */ | |
355 | void consumer_del_channel(struct lttng_consumer_channel *channel) | |
356 | { | |
357 | int ret; | |
358 | struct lttng_ht_iter iter; | |
359 | ||
d88aee68 | 360 | DBG("Consumer delete channel key %" PRIu64, channel->key); |
ffe60014 DG |
361 | |
362 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 363 | pthread_mutex_lock(&channel->lock); |
ffe60014 | 364 | |
212d67a2 DG |
365 | /* Destroy streams that might have been left in the stream list. */ |
366 | clean_channel_stream_list(channel); | |
51e762e5 | 367 | |
d3e2ba59 JD |
368 | if (channel->live_timer_enabled == 1) { |
369 | consumer_timer_live_stop(channel); | |
370 | } | |
371 | ||
ffe60014 DG |
372 | switch (consumer_data.type) { |
373 | case LTTNG_CONSUMER_KERNEL: | |
374 | break; | |
375 | case LTTNG_CONSUMER32_UST: | |
376 | case LTTNG_CONSUMER64_UST: | |
377 | lttng_ustconsumer_del_channel(channel); | |
378 | break; | |
379 | default: | |
380 | ERR("Unknown consumer_data type"); | |
381 | assert(0); | |
382 | goto end; | |
383 | } | |
384 | ||
385 | rcu_read_lock(); | |
386 | iter.iter.node = &channel->node.node; | |
387 | ret = lttng_ht_del(consumer_data.channel_ht, &iter); | |
388 | assert(!ret); | |
389 | rcu_read_unlock(); | |
390 | ||
391 | call_rcu(&channel->node.head, free_channel_rcu); | |
392 | end: | |
a9838785 | 393 | pthread_mutex_unlock(&channel->lock); |
ffe60014 | 394 | pthread_mutex_unlock(&consumer_data.lock); |
00e2e675 DG |
395 | } |
396 | ||
228b5bf7 DG |
397 | /* |
398 | * Iterate over the relayd hash table and destroy each element. Finally, | |
399 | * destroy the whole hash table. | |
400 | */ | |
401 | static void cleanup_relayd_ht(void) | |
402 | { | |
403 | struct lttng_ht_iter iter; | |
404 | struct consumer_relayd_sock_pair *relayd; | |
405 | ||
406 | rcu_read_lock(); | |
407 | ||
408 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
409 | node.node) { | |
51230d70 | 410 | consumer_destroy_relayd(relayd); |
228b5bf7 DG |
411 | } |
412 | ||
228b5bf7 | 413 | rcu_read_unlock(); |
36b588ed MD |
414 | |
415 | lttng_ht_destroy(consumer_data.relayd_ht); | |
228b5bf7 DG |
416 | } |
417 | ||
8994307f DG |
418 | /* |
419 | * Update the end point status of all streams having the given network sequence | |
420 | * index (relayd index). | |
421 | * | |
422 | * It's atomically set without having the stream mutex locked which is fine | |
423 | * because we handle the write/read race with a pipe wakeup for each thread. | |
424 | */ | |
da009f2c | 425 | static void update_endpoint_status_by_netidx(uint64_t net_seq_idx, |
8994307f DG |
426 | enum consumer_endpoint_status status) |
427 | { | |
428 | struct lttng_ht_iter iter; | |
429 | struct lttng_consumer_stream *stream; | |
430 | ||
da009f2c | 431 | DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx); |
8994307f DG |
432 | |
433 | rcu_read_lock(); | |
434 | ||
435 | /* Let's begin with metadata */ | |
436 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
437 | if (stream->net_seq_idx == net_seq_idx) { | |
438 | uatomic_set(&stream->endpoint_status, status); | |
439 | DBG("Delete flag set to metadata stream %d", stream->wait_fd); | |
440 | } | |
441 | } | |
442 | ||
443 | /* Follow up by the data streams */ | |
444 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
445 | if (stream->net_seq_idx == net_seq_idx) { | |
446 | uatomic_set(&stream->endpoint_status, status); | |
447 | DBG("Delete flag set to data stream %d", stream->wait_fd); | |
448 | } | |
449 | } | |
450 | rcu_read_unlock(); | |
451 | } | |
452 | ||
453 | /* | |
454 | * Cleanup a relayd object by flagging every associated streams for deletion, | |
455 | * destroying the object meaning removing it from the relayd hash table, | |
456 | * closing the sockets and freeing the memory in a RCU call. | |
457 | * | |
458 | * If a local data context is available, notify the threads that the streams' | |
459 | * state have changed. | |
460 | */ | |
461 | static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd, | |
462 | struct lttng_consumer_local_data *ctx) | |
463 | { | |
da009f2c | 464 | uint64_t netidx; |
8994307f DG |
465 | |
466 | assert(relayd); | |
467 | ||
9617607b DG |
468 | DBG("Cleaning up relayd sockets"); |
469 | ||
8994307f DG |
470 | /* Save the net sequence index before destroying the object */ |
471 | netidx = relayd->net_seq_idx; | |
472 | ||
473 | /* | |
474 | * Delete the relayd from the relayd hash table, close the sockets and free | |
475 | * the object in a RCU call. | |
476 | */ | |
51230d70 | 477 | consumer_destroy_relayd(relayd); |
8994307f DG |
478 | |
479 | /* Set inactive endpoint to all streams */ | |
480 | update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE); | |
481 | ||
482 | /* | |
483 | * With a local data context, notify the threads that the streams' state | |
484 | * have changed. The write() action on the pipe acts as an "implicit" | |
485 | * memory barrier ordering the updates of the end point status from the | |
486 | * read of this status which happens AFTER receiving this notify. | |
487 | */ | |
488 | if (ctx) { | |
acdb9057 | 489 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
13886d2d | 490 | notify_thread_lttng_pipe(ctx->consumer_metadata_pipe); |
8994307f DG |
491 | } |
492 | } | |
493 | ||
a6ba4fe1 DG |
494 | /* |
495 | * Flag a relayd socket pair for destruction. Destroy it if the refcount | |
496 | * reaches zero. | |
497 | * | |
498 | * RCU read side lock MUST be aquired before calling this function. | |
499 | */ | |
500 | void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) | |
501 | { | |
502 | assert(relayd); | |
503 | ||
504 | /* Set destroy flag for this object */ | |
505 | uatomic_set(&relayd->destroy_flag, 1); | |
506 | ||
507 | /* Destroy the relayd if refcount is 0 */ | |
508 | if (uatomic_read(&relayd->refcount) == 0) { | |
51230d70 | 509 | consumer_destroy_relayd(relayd); |
a6ba4fe1 DG |
510 | } |
511 | } | |
512 | ||
3bd1e081 | 513 | /* |
1d1a276c DG |
514 | * Completly destroy stream from every visiable data structure and the given |
515 | * hash table if one. | |
516 | * | |
517 | * One this call returns, the stream object is not longer usable nor visible. | |
3bd1e081 | 518 | */ |
e316aad5 DG |
519 | void consumer_del_stream(struct lttng_consumer_stream *stream, |
520 | struct lttng_ht *ht) | |
3bd1e081 | 521 | { |
1d1a276c | 522 | consumer_stream_destroy(stream, ht); |
3bd1e081 MD |
523 | } |
524 | ||
5ab66908 MD |
525 | /* |
526 | * XXX naming of del vs destroy is all mixed up. | |
527 | */ | |
528 | void consumer_del_stream_for_data(struct lttng_consumer_stream *stream) | |
529 | { | |
530 | consumer_stream_destroy(stream, data_ht); | |
531 | } | |
532 | ||
533 | void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream) | |
534 | { | |
535 | consumer_stream_destroy(stream, metadata_ht); | |
536 | } | |
537 | ||
d88aee68 DG |
538 | struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, |
539 | uint64_t stream_key, | |
3bd1e081 | 540 | enum lttng_consumer_stream_state state, |
ffe60014 | 541 | const char *channel_name, |
6df2e2c9 | 542 | uid_t uid, |
00e2e675 | 543 | gid_t gid, |
57a269f2 | 544 | uint64_t relayd_id, |
53632229 | 545 | uint64_t session_id, |
ffe60014 DG |
546 | int cpu, |
547 | int *alloc_ret, | |
4891ece8 DG |
548 | enum consumer_channel_type type, |
549 | unsigned int monitor) | |
3bd1e081 | 550 | { |
ffe60014 | 551 | int ret; |
3bd1e081 | 552 | struct lttng_consumer_stream *stream; |
3bd1e081 | 553 | |
effcf122 | 554 | stream = zmalloc(sizeof(*stream)); |
3bd1e081 | 555 | if (stream == NULL) { |
7a57cf92 | 556 | PERROR("malloc struct lttng_consumer_stream"); |
ffe60014 | 557 | ret = -ENOMEM; |
7a57cf92 | 558 | goto end; |
3bd1e081 | 559 | } |
7a57cf92 | 560 | |
d56db448 DG |
561 | rcu_read_lock(); |
562 | ||
3bd1e081 | 563 | stream->key = stream_key; |
3bd1e081 MD |
564 | stream->out_fd = -1; |
565 | stream->out_fd_offset = 0; | |
e5d1a9b3 | 566 | stream->output_written = 0; |
3bd1e081 | 567 | stream->state = state; |
6df2e2c9 MD |
568 | stream->uid = uid; |
569 | stream->gid = gid; | |
ffe60014 | 570 | stream->net_seq_idx = relayd_id; |
53632229 | 571 | stream->session_id = session_id; |
4891ece8 | 572 | stream->monitor = monitor; |
774d490c | 573 | stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE; |
309167d2 | 574 | stream->index_fd = -1; |
53632229 | 575 | pthread_mutex_init(&stream->lock, NULL); |
5cc6b3e4 | 576 | pthread_mutex_init(&stream->metadata_timer_lock, NULL); |
58b1f425 | 577 | |
ffe60014 DG |
578 | /* If channel is the metadata, flag this stream as metadata. */ |
579 | if (type == CONSUMER_CHANNEL_TYPE_METADATA) { | |
580 | stream->metadata_flag = 1; | |
581 | /* Metadata is flat out. */ | |
582 | strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name)); | |
94d49140 JD |
583 | /* Live rendez-vous point. */ |
584 | pthread_cond_init(&stream->metadata_rdv, NULL); | |
585 | pthread_mutex_init(&stream->metadata_rdv_lock, NULL); | |
58b1f425 | 586 | } else { |
ffe60014 DG |
587 | /* Format stream name to <channel_name>_<cpu_number> */ |
588 | ret = snprintf(stream->name, sizeof(stream->name), "%s_%d", | |
589 | channel_name, cpu); | |
590 | if (ret < 0) { | |
591 | PERROR("snprintf stream name"); | |
592 | goto error; | |
593 | } | |
58b1f425 | 594 | } |
c30aaa51 | 595 | |
ffe60014 | 596 | /* Key is always the wait_fd for streams. */ |
d88aee68 | 597 | lttng_ht_node_init_u64(&stream->node, stream->key); |
ffe60014 | 598 | |
d8ef542d MD |
599 | /* Init node per channel id key */ |
600 | lttng_ht_node_init_u64(&stream->node_channel_id, channel_key); | |
601 | ||
53632229 | 602 | /* Init session id node with the stream session id */ |
d88aee68 | 603 | lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id); |
53632229 | 604 | |
07b86b52 JD |
605 | DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64 |
606 | " relayd_id %" PRIu64 ", session_id %" PRIu64, | |
607 | stream->name, stream->key, channel_key, | |
608 | stream->net_seq_idx, stream->session_id); | |
d56db448 DG |
609 | |
610 | rcu_read_unlock(); | |
3bd1e081 | 611 | return stream; |
c80048c6 MD |
612 | |
613 | error: | |
d56db448 | 614 | rcu_read_unlock(); |
c80048c6 | 615 | free(stream); |
7a57cf92 | 616 | end: |
ffe60014 DG |
617 | if (alloc_ret) { |
618 | *alloc_ret = ret; | |
619 | } | |
c80048c6 | 620 | return NULL; |
3bd1e081 MD |
621 | } |
622 | ||
623 | /* | |
624 | * Add a stream to the global list protected by a mutex. | |
625 | */ | |
5ab66908 | 626 | int consumer_add_data_stream(struct lttng_consumer_stream *stream) |
3bd1e081 | 627 | { |
5ab66908 | 628 | struct lttng_ht *ht = data_ht; |
3bd1e081 MD |
629 | int ret = 0; |
630 | ||
e316aad5 | 631 | assert(stream); |
43c34bc3 | 632 | assert(ht); |
c77fc10a | 633 | |
d88aee68 | 634 | DBG3("Adding consumer stream %" PRIu64, stream->key); |
e316aad5 DG |
635 | |
636 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 637 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 638 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 639 | pthread_mutex_lock(&stream->lock); |
b0b335c8 | 640 | rcu_read_lock(); |
e316aad5 | 641 | |
43c34bc3 | 642 | /* Steal stream identifier to avoid having streams with the same key */ |
ffe60014 | 643 | steal_stream_key(stream->key, ht); |
43c34bc3 | 644 | |
d88aee68 | 645 | lttng_ht_add_unique_u64(ht, &stream->node); |
00e2e675 | 646 | |
d8ef542d MD |
647 | lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, |
648 | &stream->node_channel_id); | |
649 | ||
ca22feea DG |
650 | /* |
651 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
652 | * the key since the HT does not use it and we allow to add redundant keys | |
653 | * into this table. | |
654 | */ | |
d88aee68 | 655 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 656 | |
e316aad5 | 657 | /* |
ffe60014 DG |
658 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
659 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
660 | * causes the count to become 0 also causes a stream to be added. The |
661 | * channel deletion will thus be triggered by the following removal of this | |
662 | * stream. | |
663 | */ | |
ffe60014 | 664 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
665 | /* Increment refcount before decrementing nb_init_stream_left */ |
666 | cmm_smp_wmb(); | |
ffe60014 | 667 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
668 | } |
669 | ||
670 | /* Update consumer data once the node is inserted. */ | |
3bd1e081 MD |
671 | consumer_data.stream_count++; |
672 | consumer_data.need_update = 1; | |
673 | ||
e316aad5 | 674 | rcu_read_unlock(); |
2e818a6a | 675 | pthread_mutex_unlock(&stream->lock); |
ec6ea7d0 | 676 | pthread_mutex_unlock(&stream->chan->timer_lock); |
a9838785 | 677 | pthread_mutex_unlock(&stream->chan->lock); |
3bd1e081 | 678 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 679 | |
3bd1e081 MD |
680 | return ret; |
681 | } | |
682 | ||
5ab66908 MD |
683 | void consumer_del_data_stream(struct lttng_consumer_stream *stream) |
684 | { | |
685 | consumer_del_stream(stream, data_ht); | |
686 | } | |
687 | ||
00e2e675 | 688 | /* |
3f8e211f DG |
689 | * Add relayd socket to global consumer data hashtable. RCU read side lock MUST |
690 | * be acquired before calling this. | |
00e2e675 | 691 | */ |
d09e1200 | 692 | static int add_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
693 | { |
694 | int ret = 0; | |
d88aee68 | 695 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
696 | struct lttng_ht_iter iter; |
697 | ||
ffe60014 | 698 | assert(relayd); |
00e2e675 | 699 | |
00e2e675 | 700 | lttng_ht_lookup(consumer_data.relayd_ht, |
d88aee68 DG |
701 | &relayd->net_seq_idx, &iter); |
702 | node = lttng_ht_iter_get_node_u64(&iter); | |
00e2e675 | 703 | if (node != NULL) { |
00e2e675 DG |
704 | goto end; |
705 | } | |
d88aee68 | 706 | lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node); |
00e2e675 | 707 | |
00e2e675 DG |
708 | end: |
709 | return ret; | |
710 | } | |
711 | ||
712 | /* | |
713 | * Allocate and return a consumer relayd socket. | |
714 | */ | |
715 | struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( | |
da009f2c | 716 | uint64_t net_seq_idx) |
00e2e675 DG |
717 | { |
718 | struct consumer_relayd_sock_pair *obj = NULL; | |
719 | ||
da009f2c MD |
720 | /* net sequence index of -1 is a failure */ |
721 | if (net_seq_idx == (uint64_t) -1ULL) { | |
00e2e675 DG |
722 | goto error; |
723 | } | |
724 | ||
725 | obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); | |
726 | if (obj == NULL) { | |
727 | PERROR("zmalloc relayd sock"); | |
728 | goto error; | |
729 | } | |
730 | ||
731 | obj->net_seq_idx = net_seq_idx; | |
732 | obj->refcount = 0; | |
173af62f | 733 | obj->destroy_flag = 0; |
f96e4545 MD |
734 | obj->control_sock.sock.fd = -1; |
735 | obj->data_sock.sock.fd = -1; | |
d88aee68 | 736 | lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx); |
00e2e675 DG |
737 | pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); |
738 | ||
739 | error: | |
740 | return obj; | |
741 | } | |
742 | ||
743 | /* | |
744 | * Find a relayd socket pair in the global consumer data. | |
745 | * | |
746 | * Return the object if found else NULL. | |
b0b335c8 MD |
747 | * RCU read-side lock must be held across this call and while using the |
748 | * returned object. | |
00e2e675 | 749 | */ |
d88aee68 | 750 | struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key) |
00e2e675 DG |
751 | { |
752 | struct lttng_ht_iter iter; | |
d88aee68 | 753 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
754 | struct consumer_relayd_sock_pair *relayd = NULL; |
755 | ||
756 | /* Negative keys are lookup failures */ | |
d88aee68 | 757 | if (key == (uint64_t) -1ULL) { |
00e2e675 DG |
758 | goto error; |
759 | } | |
760 | ||
d88aee68 | 761 | lttng_ht_lookup(consumer_data.relayd_ht, &key, |
00e2e675 | 762 | &iter); |
d88aee68 | 763 | node = lttng_ht_iter_get_node_u64(&iter); |
00e2e675 DG |
764 | if (node != NULL) { |
765 | relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
766 | } | |
767 | ||
00e2e675 DG |
768 | error: |
769 | return relayd; | |
770 | } | |
771 | ||
10a50311 JD |
772 | /* |
773 | * Find a relayd and send the stream | |
774 | * | |
775 | * Returns 0 on success, < 0 on error | |
776 | */ | |
777 | int consumer_send_relayd_stream(struct lttng_consumer_stream *stream, | |
778 | char *path) | |
779 | { | |
780 | int ret = 0; | |
781 | struct consumer_relayd_sock_pair *relayd; | |
782 | ||
783 | assert(stream); | |
784 | assert(stream->net_seq_idx != -1ULL); | |
785 | assert(path); | |
786 | ||
787 | /* The stream is not metadata. Get relayd reference if exists. */ | |
788 | rcu_read_lock(); | |
789 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
790 | if (relayd != NULL) { | |
791 | /* Add stream on the relayd */ | |
792 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
793 | ret = relayd_add_stream(&relayd->control_sock, stream->name, | |
794 | path, &stream->relayd_stream_id, | |
795 | stream->chan->tracefile_size, stream->chan->tracefile_count); | |
796 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
797 | if (ret < 0) { | |
798 | goto end; | |
799 | } | |
1c20f0e2 | 800 | |
10a50311 | 801 | uatomic_inc(&relayd->refcount); |
d01178b6 | 802 | stream->sent_to_relayd = 1; |
10a50311 JD |
803 | } else { |
804 | ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.", | |
805 | stream->key, stream->net_seq_idx); | |
806 | ret = -1; | |
807 | goto end; | |
808 | } | |
809 | ||
810 | DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64, | |
811 | stream->name, stream->key, stream->net_seq_idx); | |
812 | ||
813 | end: | |
814 | rcu_read_unlock(); | |
815 | return ret; | |
816 | } | |
817 | ||
a4baae1b JD |
818 | /* |
819 | * Find a relayd and send the streams sent message | |
820 | * | |
821 | * Returns 0 on success, < 0 on error | |
822 | */ | |
823 | int consumer_send_relayd_streams_sent(uint64_t net_seq_idx) | |
824 | { | |
825 | int ret = 0; | |
826 | struct consumer_relayd_sock_pair *relayd; | |
827 | ||
828 | assert(net_seq_idx != -1ULL); | |
829 | ||
830 | /* The stream is not metadata. Get relayd reference if exists. */ | |
831 | rcu_read_lock(); | |
832 | relayd = consumer_find_relayd(net_seq_idx); | |
833 | if (relayd != NULL) { | |
834 | /* Add stream on the relayd */ | |
835 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
836 | ret = relayd_streams_sent(&relayd->control_sock); | |
837 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
838 | if (ret < 0) { | |
839 | goto end; | |
840 | } | |
841 | } else { | |
842 | ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.", | |
843 | net_seq_idx); | |
844 | ret = -1; | |
845 | goto end; | |
846 | } | |
847 | ||
848 | ret = 0; | |
849 | DBG("All streams sent relayd id %" PRIu64, net_seq_idx); | |
850 | ||
851 | end: | |
852 | rcu_read_unlock(); | |
853 | return ret; | |
854 | } | |
855 | ||
10a50311 JD |
856 | /* |
857 | * Find a relayd and close the stream | |
858 | */ | |
859 | void close_relayd_stream(struct lttng_consumer_stream *stream) | |
860 | { | |
861 | struct consumer_relayd_sock_pair *relayd; | |
862 | ||
863 | /* The stream is not metadata. Get relayd reference if exists. */ | |
864 | rcu_read_lock(); | |
865 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
866 | if (relayd) { | |
867 | consumer_stream_relayd_close(stream, relayd); | |
868 | } | |
869 | rcu_read_unlock(); | |
870 | } | |
871 | ||
00e2e675 DG |
872 | /* |
873 | * Handle stream for relayd transmission if the stream applies for network | |
874 | * streaming where the net sequence index is set. | |
875 | * | |
876 | * Return destination file descriptor or negative value on error. | |
877 | */ | |
6197aea7 | 878 | static int write_relayd_stream_header(struct lttng_consumer_stream *stream, |
1d4dfdef DG |
879 | size_t data_size, unsigned long padding, |
880 | struct consumer_relayd_sock_pair *relayd) | |
00e2e675 DG |
881 | { |
882 | int outfd = -1, ret; | |
00e2e675 DG |
883 | struct lttcomm_relayd_data_hdr data_hdr; |
884 | ||
885 | /* Safety net */ | |
886 | assert(stream); | |
6197aea7 | 887 | assert(relayd); |
00e2e675 DG |
888 | |
889 | /* Reset data header */ | |
890 | memset(&data_hdr, 0, sizeof(data_hdr)); | |
891 | ||
00e2e675 DG |
892 | if (stream->metadata_flag) { |
893 | /* Caller MUST acquire the relayd control socket lock */ | |
894 | ret = relayd_send_metadata(&relayd->control_sock, data_size); | |
895 | if (ret < 0) { | |
896 | goto error; | |
897 | } | |
898 | ||
899 | /* Metadata are always sent on the control socket. */ | |
6151a90f | 900 | outfd = relayd->control_sock.sock.fd; |
00e2e675 DG |
901 | } else { |
902 | /* Set header with stream information */ | |
903 | data_hdr.stream_id = htobe64(stream->relayd_stream_id); | |
904 | data_hdr.data_size = htobe32(data_size); | |
1d4dfdef | 905 | data_hdr.padding_size = htobe32(padding); |
39df6d9f DG |
906 | /* |
907 | * Note that net_seq_num below is assigned with the *current* value of | |
908 | * next_net_seq_num and only after that the next_net_seq_num will be | |
909 | * increment. This is why when issuing a command on the relayd using | |
910 | * this next value, 1 should always be substracted in order to compare | |
911 | * the last seen sequence number on the relayd side to the last sent. | |
912 | */ | |
3604f373 | 913 | data_hdr.net_seq_num = htobe64(stream->next_net_seq_num); |
00e2e675 DG |
914 | /* Other fields are zeroed previously */ |
915 | ||
916 | ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, | |
917 | sizeof(data_hdr)); | |
918 | if (ret < 0) { | |
919 | goto error; | |
920 | } | |
921 | ||
3604f373 DG |
922 | ++stream->next_net_seq_num; |
923 | ||
00e2e675 | 924 | /* Set to go on data socket */ |
6151a90f | 925 | outfd = relayd->data_sock.sock.fd; |
00e2e675 DG |
926 | } |
927 | ||
928 | error: | |
929 | return outfd; | |
930 | } | |
931 | ||
3bd1e081 | 932 | /* |
ffe60014 DG |
933 | * Allocate and return a new lttng_consumer_channel object using the given key |
934 | * to initialize the hash table node. | |
935 | * | |
936 | * On error, return NULL. | |
3bd1e081 | 937 | */ |
886224ff | 938 | struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, |
ffe60014 DG |
939 | uint64_t session_id, |
940 | const char *pathname, | |
941 | const char *name, | |
942 | uid_t uid, | |
943 | gid_t gid, | |
57a269f2 | 944 | uint64_t relayd_id, |
1624d5b7 JD |
945 | enum lttng_event_output output, |
946 | uint64_t tracefile_size, | |
2bba9e53 | 947 | uint64_t tracefile_count, |
1950109e | 948 | uint64_t session_id_per_pid, |
ecc48a90 JD |
949 | unsigned int monitor, |
950 | unsigned int live_timer_interval) | |
3bd1e081 MD |
951 | { |
952 | struct lttng_consumer_channel *channel; | |
3bd1e081 | 953 | |
276b26d1 | 954 | channel = zmalloc(sizeof(*channel)); |
3bd1e081 | 955 | if (channel == NULL) { |
7a57cf92 | 956 | PERROR("malloc struct lttng_consumer_channel"); |
3bd1e081 MD |
957 | goto end; |
958 | } | |
ffe60014 DG |
959 | |
960 | channel->key = key; | |
3bd1e081 | 961 | channel->refcount = 0; |
ffe60014 | 962 | channel->session_id = session_id; |
1950109e | 963 | channel->session_id_per_pid = session_id_per_pid; |
ffe60014 DG |
964 | channel->uid = uid; |
965 | channel->gid = gid; | |
966 | channel->relayd_id = relayd_id; | |
1624d5b7 JD |
967 | channel->tracefile_size = tracefile_size; |
968 | channel->tracefile_count = tracefile_count; | |
2bba9e53 | 969 | channel->monitor = monitor; |
ecc48a90 | 970 | channel->live_timer_interval = live_timer_interval; |
a9838785 | 971 | pthread_mutex_init(&channel->lock, NULL); |
ec6ea7d0 | 972 | pthread_mutex_init(&channel->timer_lock, NULL); |
ffe60014 | 973 | |
0c759fc9 DG |
974 | switch (output) { |
975 | case LTTNG_EVENT_SPLICE: | |
976 | channel->output = CONSUMER_CHANNEL_SPLICE; | |
977 | break; | |
978 | case LTTNG_EVENT_MMAP: | |
979 | channel->output = CONSUMER_CHANNEL_MMAP; | |
980 | break; | |
981 | default: | |
982 | assert(0); | |
983 | free(channel); | |
984 | channel = NULL; | |
985 | goto end; | |
986 | } | |
987 | ||
07b86b52 JD |
988 | /* |
989 | * In monitor mode, the streams associated with the channel will be put in | |
990 | * a special list ONLY owned by this channel. So, the refcount is set to 1 | |
991 | * here meaning that the channel itself has streams that are referenced. | |
992 | * | |
993 | * On a channel deletion, once the channel is no longer visible, the | |
994 | * refcount is decremented and checked for a zero value to delete it. With | |
995 | * streams in no monitor mode, it will now be safe to destroy the channel. | |
996 | */ | |
997 | if (!channel->monitor) { | |
998 | channel->refcount = 1; | |
999 | } | |
1000 | ||
ffe60014 DG |
1001 | strncpy(channel->pathname, pathname, sizeof(channel->pathname)); |
1002 | channel->pathname[sizeof(channel->pathname) - 1] = '\0'; | |
1003 | ||
1004 | strncpy(channel->name, name, sizeof(channel->name)); | |
1005 | channel->name[sizeof(channel->name) - 1] = '\0'; | |
1006 | ||
d88aee68 | 1007 | lttng_ht_node_init_u64(&channel->node, channel->key); |
d8ef542d MD |
1008 | |
1009 | channel->wait_fd = -1; | |
1010 | ||
ffe60014 DG |
1011 | CDS_INIT_LIST_HEAD(&channel->streams.head); |
1012 | ||
d88aee68 | 1013 | DBG("Allocated channel (key %" PRIu64 ")", channel->key) |
3bd1e081 | 1014 | |
3bd1e081 MD |
1015 | end: |
1016 | return channel; | |
1017 | } | |
1018 | ||
1019 | /* | |
1020 | * Add a channel to the global list protected by a mutex. | |
821fffb2 | 1021 | * |
b5a6470f | 1022 | * Always return 0 indicating success. |
3bd1e081 | 1023 | */ |
d8ef542d MD |
1024 | int consumer_add_channel(struct lttng_consumer_channel *channel, |
1025 | struct lttng_consumer_local_data *ctx) | |
3bd1e081 | 1026 | { |
3bd1e081 | 1027 | pthread_mutex_lock(&consumer_data.lock); |
a9838785 | 1028 | pthread_mutex_lock(&channel->lock); |
ec6ea7d0 | 1029 | pthread_mutex_lock(&channel->timer_lock); |
c77fc10a | 1030 | |
b5a6470f DG |
1031 | /* |
1032 | * This gives us a guarantee that the channel we are about to add to the | |
1033 | * channel hash table will be unique. See this function comment on the why | |
1034 | * we need to steel the channel key at this stage. | |
1035 | */ | |
1036 | steal_channel_key(channel->key); | |
c77fc10a | 1037 | |
b5a6470f | 1038 | rcu_read_lock(); |
d88aee68 | 1039 | lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node); |
6065ceec | 1040 | rcu_read_unlock(); |
b5a6470f | 1041 | |
ec6ea7d0 | 1042 | pthread_mutex_unlock(&channel->timer_lock); |
a9838785 | 1043 | pthread_mutex_unlock(&channel->lock); |
3bd1e081 | 1044 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 1045 | |
b5a6470f | 1046 | if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) { |
a0cbdd2e | 1047 | notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD); |
d8ef542d | 1048 | } |
b5a6470f DG |
1049 | |
1050 | return 0; | |
3bd1e081 MD |
1051 | } |
1052 | ||
1053 | /* | |
1054 | * Allocate the pollfd structure and the local view of the out fds to avoid | |
1055 | * doing a lookup in the linked list and concurrency issues when writing is | |
1056 | * needed. Called with consumer_data.lock held. | |
1057 | * | |
1058 | * Returns the number of fds in the structures. | |
1059 | */ | |
ffe60014 DG |
1060 | static int update_poll_array(struct lttng_consumer_local_data *ctx, |
1061 | struct pollfd **pollfd, struct lttng_consumer_stream **local_stream, | |
1062 | struct lttng_ht *ht) | |
3bd1e081 | 1063 | { |
3bd1e081 | 1064 | int i = 0; |
e4421fec DG |
1065 | struct lttng_ht_iter iter; |
1066 | struct lttng_consumer_stream *stream; | |
3bd1e081 | 1067 | |
ffe60014 DG |
1068 | assert(ctx); |
1069 | assert(ht); | |
1070 | assert(pollfd); | |
1071 | assert(local_stream); | |
1072 | ||
3bd1e081 | 1073 | DBG("Updating poll fd array"); |
481d6c57 | 1074 | rcu_read_lock(); |
43c34bc3 | 1075 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
8994307f DG |
1076 | /* |
1077 | * Only active streams with an active end point can be added to the | |
1078 | * poll set and local stream storage of the thread. | |
1079 | * | |
1080 | * There is a potential race here for endpoint_status to be updated | |
1081 | * just after the check. However, this is OK since the stream(s) will | |
1082 | * be deleted once the thread is notified that the end point state has | |
1083 | * changed where this function will be called back again. | |
1084 | */ | |
1085 | if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM || | |
79d4ffb7 | 1086 | stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { |
3bd1e081 MD |
1087 | continue; |
1088 | } | |
7972aab2 DG |
1089 | /* |
1090 | * This clobbers way too much the debug output. Uncomment that if you | |
1091 | * need it for debugging purposes. | |
1092 | * | |
1093 | * DBG("Active FD %d", stream->wait_fd); | |
1094 | */ | |
e4421fec | 1095 | (*pollfd)[i].fd = stream->wait_fd; |
3bd1e081 | 1096 | (*pollfd)[i].events = POLLIN | POLLPRI; |
e4421fec | 1097 | local_stream[i] = stream; |
3bd1e081 MD |
1098 | i++; |
1099 | } | |
481d6c57 | 1100 | rcu_read_unlock(); |
3bd1e081 MD |
1101 | |
1102 | /* | |
50f8ae69 | 1103 | * Insert the consumer_data_pipe at the end of the array and don't |
3bd1e081 MD |
1104 | * increment i so nb_fd is the number of real FD. |
1105 | */ | |
acdb9057 | 1106 | (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe); |
509bb1cf | 1107 | (*pollfd)[i].events = POLLIN | POLLPRI; |
02b3d176 DG |
1108 | |
1109 | (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe); | |
1110 | (*pollfd)[i + 1].events = POLLIN | POLLPRI; | |
3bd1e081 MD |
1111 | return i; |
1112 | } | |
1113 | ||
1114 | /* | |
84382d49 MD |
1115 | * Poll on the should_quit pipe and the command socket return -1 on |
1116 | * error, 1 if should exit, 0 if data is available on the command socket | |
3bd1e081 MD |
1117 | */ |
1118 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) | |
1119 | { | |
1120 | int num_rdy; | |
1121 | ||
88f2b785 | 1122 | restart: |
3bd1e081 MD |
1123 | num_rdy = poll(consumer_sockpoll, 2, -1); |
1124 | if (num_rdy == -1) { | |
88f2b785 MD |
1125 | /* |
1126 | * Restart interrupted system call. | |
1127 | */ | |
1128 | if (errno == EINTR) { | |
1129 | goto restart; | |
1130 | } | |
7a57cf92 | 1131 | PERROR("Poll error"); |
84382d49 | 1132 | return -1; |
3bd1e081 | 1133 | } |
509bb1cf | 1134 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
3bd1e081 | 1135 | DBG("consumer_should_quit wake up"); |
84382d49 | 1136 | return 1; |
3bd1e081 MD |
1137 | } |
1138 | return 0; | |
3bd1e081 MD |
1139 | } |
1140 | ||
1141 | /* | |
1142 | * Set the error socket. | |
1143 | */ | |
ffe60014 DG |
1144 | void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx, |
1145 | int sock) | |
3bd1e081 MD |
1146 | { |
1147 | ctx->consumer_error_socket = sock; | |
1148 | } | |
1149 | ||
1150 | /* | |
1151 | * Set the command socket path. | |
1152 | */ | |
3bd1e081 MD |
1153 | void lttng_consumer_set_command_sock_path( |
1154 | struct lttng_consumer_local_data *ctx, char *sock) | |
1155 | { | |
1156 | ctx->consumer_command_sock_path = sock; | |
1157 | } | |
1158 | ||
1159 | /* | |
1160 | * Send return code to the session daemon. | |
1161 | * If the socket is not defined, we return 0, it is not a fatal error | |
1162 | */ | |
ffe60014 | 1163 | int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd) |
3bd1e081 MD |
1164 | { |
1165 | if (ctx->consumer_error_socket > 0) { | |
1166 | return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd, | |
1167 | sizeof(enum lttcomm_sessiond_command)); | |
1168 | } | |
1169 | ||
1170 | return 0; | |
1171 | } | |
1172 | ||
1173 | /* | |
228b5bf7 DG |
1174 | * Close all the tracefiles and stream fds and MUST be called when all |
1175 | * instances are destroyed i.e. when all threads were joined and are ended. | |
3bd1e081 MD |
1176 | */ |
1177 | void lttng_consumer_cleanup(void) | |
1178 | { | |
e4421fec | 1179 | struct lttng_ht_iter iter; |
ffe60014 | 1180 | struct lttng_consumer_channel *channel; |
6065ceec DG |
1181 | |
1182 | rcu_read_lock(); | |
3bd1e081 | 1183 | |
ffe60014 DG |
1184 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel, |
1185 | node.node) { | |
702b1ea4 | 1186 | consumer_del_channel(channel); |
3bd1e081 | 1187 | } |
6065ceec DG |
1188 | |
1189 | rcu_read_unlock(); | |
d6ce1df2 | 1190 | |
d6ce1df2 | 1191 | lttng_ht_destroy(consumer_data.channel_ht); |
228b5bf7 DG |
1192 | |
1193 | cleanup_relayd_ht(); | |
1194 | ||
d8ef542d MD |
1195 | lttng_ht_destroy(consumer_data.stream_per_chan_id_ht); |
1196 | ||
228b5bf7 DG |
1197 | /* |
1198 | * This HT contains streams that are freed by either the metadata thread or | |
1199 | * the data thread so we do *nothing* on the hash table and simply destroy | |
1200 | * it. | |
1201 | */ | |
1202 | lttng_ht_destroy(consumer_data.stream_list_ht); | |
3bd1e081 MD |
1203 | } |
1204 | ||
1205 | /* | |
1206 | * Called from signal handler. | |
1207 | */ | |
1208 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) | |
1209 | { | |
6cd525e8 MD |
1210 | ssize_t ret; |
1211 | ||
3bd1e081 | 1212 | consumer_quit = 1; |
6cd525e8 MD |
1213 | ret = lttng_write(ctx->consumer_should_quit[1], "4", 1); |
1214 | if (ret < 1) { | |
7a57cf92 | 1215 | PERROR("write consumer quit"); |
3bd1e081 | 1216 | } |
ab1027f4 DG |
1217 | |
1218 | DBG("Consumer flag that it should quit"); | |
3bd1e081 MD |
1219 | } |
1220 | ||
00e2e675 DG |
1221 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
1222 | off_t orig_offset) | |
3bd1e081 | 1223 | { |
19b049ec | 1224 | int ret; |
3bd1e081 MD |
1225 | int outfd = stream->out_fd; |
1226 | ||
1227 | /* | |
1228 | * This does a blocking write-and-wait on any page that belongs to the | |
1229 | * subbuffer prior to the one we just wrote. | |
1230 | * Don't care about error values, as these are just hints and ways to | |
1231 | * limit the amount of page cache used. | |
1232 | */ | |
ffe60014 | 1233 | if (orig_offset < stream->max_sb_size) { |
3bd1e081 MD |
1234 | return; |
1235 | } | |
ffe60014 DG |
1236 | lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size, |
1237 | stream->max_sb_size, | |
3bd1e081 MD |
1238 | SYNC_FILE_RANGE_WAIT_BEFORE |
1239 | | SYNC_FILE_RANGE_WRITE | |
1240 | | SYNC_FILE_RANGE_WAIT_AFTER); | |
1241 | /* | |
1242 | * Give hints to the kernel about how we access the file: | |
1243 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after | |
1244 | * we write it. | |
1245 | * | |
1246 | * We need to call fadvise again after the file grows because the | |
1247 | * kernel does not seem to apply fadvise to non-existing parts of the | |
1248 | * file. | |
1249 | * | |
1250 | * Call fadvise _after_ having waited for the page writeback to | |
1251 | * complete because the dirty page writeback semantic is not well | |
1252 | * defined. So it can be expected to lead to lower throughput in | |
1253 | * streaming. | |
1254 | */ | |
19b049ec | 1255 | ret = posix_fadvise(outfd, orig_offset - stream->max_sb_size, |
ffe60014 | 1256 | stream->max_sb_size, POSIX_FADV_DONTNEED); |
6a381a6e JG |
1257 | if (ret && ret != -ENOSYS) { |
1258 | errno = -ret; | |
1259 | PERROR("posix_fadvise"); | |
19b049ec | 1260 | } |
3bd1e081 MD |
1261 | } |
1262 | ||
1263 | /* | |
1264 | * Initialise the necessary environnement : | |
1265 | * - create a new context | |
1266 | * - create the poll_pipe | |
1267 | * - create the should_quit pipe (for signal handler) | |
1268 | * - create the thread pipe (for splice) | |
1269 | * | |
1270 | * Takes a function pointer as argument, this function is called when data is | |
1271 | * available on a buffer. This function is responsible to do the | |
1272 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the | |
1273 | * buffer configuration and then kernctl_put_next_subbuf at the end. | |
1274 | * | |
1275 | * Returns a pointer to the new context or NULL on error. | |
1276 | */ | |
1277 | struct lttng_consumer_local_data *lttng_consumer_create( | |
1278 | enum lttng_consumer_type type, | |
4078b776 | 1279 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
d41f73b7 | 1280 | struct lttng_consumer_local_data *ctx), |
3bd1e081 MD |
1281 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
1282 | int (*recv_stream)(struct lttng_consumer_stream *stream), | |
30319bcb | 1283 | int (*update_stream)(uint64_t stream_key, uint32_t state)) |
3bd1e081 | 1284 | { |
d8ef542d | 1285 | int ret; |
3bd1e081 MD |
1286 | struct lttng_consumer_local_data *ctx; |
1287 | ||
1288 | assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || | |
1289 | consumer_data.type == type); | |
1290 | consumer_data.type = type; | |
1291 | ||
effcf122 | 1292 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
3bd1e081 | 1293 | if (ctx == NULL) { |
7a57cf92 | 1294 | PERROR("allocating context"); |
3bd1e081 MD |
1295 | goto error; |
1296 | } | |
1297 | ||
1298 | ctx->consumer_error_socket = -1; | |
331744e3 | 1299 | ctx->consumer_metadata_socket = -1; |
75d83e50 | 1300 | pthread_mutex_init(&ctx->metadata_socket_lock, NULL); |
3bd1e081 MD |
1301 | /* assign the callbacks */ |
1302 | ctx->on_buffer_ready = buffer_ready; | |
1303 | ctx->on_recv_channel = recv_channel; | |
1304 | ctx->on_recv_stream = recv_stream; | |
1305 | ctx->on_update_stream = update_stream; | |
1306 | ||
acdb9057 DG |
1307 | ctx->consumer_data_pipe = lttng_pipe_open(0); |
1308 | if (!ctx->consumer_data_pipe) { | |
3bd1e081 MD |
1309 | goto error_poll_pipe; |
1310 | } | |
1311 | ||
02b3d176 DG |
1312 | ctx->consumer_wakeup_pipe = lttng_pipe_open(0); |
1313 | if (!ctx->consumer_wakeup_pipe) { | |
1314 | goto error_wakeup_pipe; | |
1315 | } | |
1316 | ||
3bd1e081 MD |
1317 | ret = pipe(ctx->consumer_should_quit); |
1318 | if (ret < 0) { | |
7a57cf92 | 1319 | PERROR("Error creating recv pipe"); |
3bd1e081 MD |
1320 | goto error_quit_pipe; |
1321 | } | |
1322 | ||
d8ef542d MD |
1323 | ret = pipe(ctx->consumer_channel_pipe); |
1324 | if (ret < 0) { | |
1325 | PERROR("Error creating channel pipe"); | |
1326 | goto error_channel_pipe; | |
1327 | } | |
1328 | ||
13886d2d DG |
1329 | ctx->consumer_metadata_pipe = lttng_pipe_open(0); |
1330 | if (!ctx->consumer_metadata_pipe) { | |
fb3a43a9 DG |
1331 | goto error_metadata_pipe; |
1332 | } | |
3bd1e081 | 1333 | |
fb3a43a9 | 1334 | return ctx; |
3bd1e081 | 1335 | |
fb3a43a9 | 1336 | error_metadata_pipe: |
d8ef542d MD |
1337 | utils_close_pipe(ctx->consumer_channel_pipe); |
1338 | error_channel_pipe: | |
d8ef542d | 1339 | utils_close_pipe(ctx->consumer_should_quit); |
3bd1e081 | 1340 | error_quit_pipe: |
02b3d176 DG |
1341 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
1342 | error_wakeup_pipe: | |
acdb9057 | 1343 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
3bd1e081 MD |
1344 | error_poll_pipe: |
1345 | free(ctx); | |
1346 | error: | |
1347 | return NULL; | |
1348 | } | |
1349 | ||
282dadbc MD |
1350 | /* |
1351 | * Iterate over all streams of the hashtable and free them properly. | |
1352 | */ | |
1353 | static void destroy_data_stream_ht(struct lttng_ht *ht) | |
1354 | { | |
1355 | struct lttng_ht_iter iter; | |
1356 | struct lttng_consumer_stream *stream; | |
1357 | ||
1358 | if (ht == NULL) { | |
1359 | return; | |
1360 | } | |
1361 | ||
1362 | rcu_read_lock(); | |
1363 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1364 | /* | |
1365 | * Ignore return value since we are currently cleaning up so any error | |
1366 | * can't be handled. | |
1367 | */ | |
1368 | (void) consumer_del_stream(stream, ht); | |
1369 | } | |
1370 | rcu_read_unlock(); | |
1371 | ||
1372 | lttng_ht_destroy(ht); | |
1373 | } | |
1374 | ||
1375 | /* | |
1376 | * Iterate over all streams of the metadata hashtable and free them | |
1377 | * properly. | |
1378 | */ | |
1379 | static void destroy_metadata_stream_ht(struct lttng_ht *ht) | |
1380 | { | |
1381 | struct lttng_ht_iter iter; | |
1382 | struct lttng_consumer_stream *stream; | |
1383 | ||
1384 | if (ht == NULL) { | |
1385 | return; | |
1386 | } | |
1387 | ||
1388 | rcu_read_lock(); | |
1389 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1390 | /* | |
1391 | * Ignore return value since we are currently cleaning up so any error | |
1392 | * can't be handled. | |
1393 | */ | |
1394 | (void) consumer_del_metadata_stream(stream, ht); | |
1395 | } | |
1396 | rcu_read_unlock(); | |
1397 | ||
1398 | lttng_ht_destroy(ht); | |
1399 | } | |
1400 | ||
3bd1e081 MD |
1401 | /* |
1402 | * Close all fds associated with the instance and free the context. | |
1403 | */ | |
1404 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) | |
1405 | { | |
4c462e79 MD |
1406 | int ret; |
1407 | ||
ab1027f4 DG |
1408 | DBG("Consumer destroying it. Closing everything."); |
1409 | ||
4f2e75b9 DG |
1410 | if (!ctx) { |
1411 | return; | |
1412 | } | |
1413 | ||
282dadbc MD |
1414 | destroy_data_stream_ht(data_ht); |
1415 | destroy_metadata_stream_ht(metadata_ht); | |
1416 | ||
4c462e79 MD |
1417 | ret = close(ctx->consumer_error_socket); |
1418 | if (ret) { | |
1419 | PERROR("close"); | |
1420 | } | |
331744e3 JD |
1421 | ret = close(ctx->consumer_metadata_socket); |
1422 | if (ret) { | |
1423 | PERROR("close"); | |
1424 | } | |
d8ef542d | 1425 | utils_close_pipe(ctx->consumer_channel_pipe); |
acdb9057 | 1426 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
13886d2d | 1427 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
02b3d176 | 1428 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
d8ef542d | 1429 | utils_close_pipe(ctx->consumer_should_quit); |
fb3a43a9 | 1430 | |
3bd1e081 MD |
1431 | unlink(ctx->consumer_command_sock_path); |
1432 | free(ctx); | |
1433 | } | |
1434 | ||
6197aea7 DG |
1435 | /* |
1436 | * Write the metadata stream id on the specified file descriptor. | |
1437 | */ | |
1438 | static int write_relayd_metadata_id(int fd, | |
1439 | struct lttng_consumer_stream *stream, | |
ffe60014 | 1440 | struct consumer_relayd_sock_pair *relayd, unsigned long padding) |
6197aea7 | 1441 | { |
6cd525e8 | 1442 | ssize_t ret; |
1d4dfdef | 1443 | struct lttcomm_relayd_metadata_payload hdr; |
6197aea7 | 1444 | |
1d4dfdef DG |
1445 | hdr.stream_id = htobe64(stream->relayd_stream_id); |
1446 | hdr.padding_size = htobe32(padding); | |
6cd525e8 MD |
1447 | ret = lttng_write(fd, (void *) &hdr, sizeof(hdr)); |
1448 | if (ret < sizeof(hdr)) { | |
d7b75ec8 DG |
1449 | /* |
1450 | * This error means that the fd's end is closed so ignore the perror | |
1451 | * not to clubber the error output since this can happen in a normal | |
1452 | * code path. | |
1453 | */ | |
1454 | if (errno != EPIPE) { | |
1455 | PERROR("write metadata stream id"); | |
1456 | } | |
1457 | DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno); | |
534d2592 DG |
1458 | /* |
1459 | * Set ret to a negative value because if ret != sizeof(hdr), we don't | |
1460 | * handle writting the missing part so report that as an error and | |
1461 | * don't lie to the caller. | |
1462 | */ | |
1463 | ret = -1; | |
6197aea7 DG |
1464 | goto end; |
1465 | } | |
1d4dfdef DG |
1466 | DBG("Metadata stream id %" PRIu64 " with padding %lu written before data", |
1467 | stream->relayd_stream_id, padding); | |
6197aea7 DG |
1468 | |
1469 | end: | |
6cd525e8 | 1470 | return (int) ret; |
6197aea7 DG |
1471 | } |
1472 | ||
3bd1e081 | 1473 | /* |
09e26845 DG |
1474 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
1475 | * core function for writing trace buffers to either the local filesystem or | |
1476 | * the network. | |
1477 | * | |
79d4ffb7 DG |
1478 | * It must be called with the stream lock held. |
1479 | * | |
09e26845 | 1480 | * Careful review MUST be put if any changes occur! |
3bd1e081 MD |
1481 | * |
1482 | * Returns the number of bytes written | |
1483 | */ | |
4078b776 | 1484 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
3bd1e081 | 1485 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1486 | struct lttng_consumer_stream *stream, unsigned long len, |
309167d2 | 1487 | unsigned long padding, |
50adc264 | 1488 | struct ctf_packet_index *index) |
3bd1e081 | 1489 | { |
f02e1e8a | 1490 | unsigned long mmap_offset; |
ffe60014 | 1491 | void *mmap_base; |
994ab360 | 1492 | ssize_t ret = 0; |
f02e1e8a DG |
1493 | off_t orig_offset = stream->out_fd_offset; |
1494 | /* Default is on the disk */ | |
1495 | int outfd = stream->out_fd; | |
f02e1e8a | 1496 | struct consumer_relayd_sock_pair *relayd = NULL; |
8994307f | 1497 | unsigned int relayd_hang_up = 0; |
f02e1e8a DG |
1498 | |
1499 | /* RCU lock for the relayd pointer */ | |
1500 | rcu_read_lock(); | |
1501 | ||
1502 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1503 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1504 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1505 | if (relayd == NULL) { | |
56591bac | 1506 | ret = -EPIPE; |
f02e1e8a DG |
1507 | goto end; |
1508 | } | |
1509 | } | |
1510 | ||
1511 | /* get the offset inside the fd to mmap */ | |
3bd1e081 MD |
1512 | switch (consumer_data.type) { |
1513 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1514 | mmap_base = stream->mmap_base; |
f02e1e8a | 1515 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); |
994ab360 DG |
1516 | if (ret < 0) { |
1517 | ret = -errno; | |
56591bac | 1518 | PERROR("tracer ctl get_mmap_read_offset"); |
56591bac MD |
1519 | goto end; |
1520 | } | |
f02e1e8a | 1521 | break; |
7753dea8 MD |
1522 | case LTTNG_CONSUMER32_UST: |
1523 | case LTTNG_CONSUMER64_UST: | |
ffe60014 DG |
1524 | mmap_base = lttng_ustctl_get_mmap_base(stream); |
1525 | if (!mmap_base) { | |
1526 | ERR("read mmap get mmap base for stream %s", stream->name); | |
994ab360 | 1527 | ret = -EPERM; |
ffe60014 DG |
1528 | goto end; |
1529 | } | |
1530 | ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset); | |
56591bac MD |
1531 | if (ret != 0) { |
1532 | PERROR("tracer ctl get_mmap_read_offset"); | |
994ab360 | 1533 | ret = -EINVAL; |
56591bac MD |
1534 | goto end; |
1535 | } | |
f02e1e8a | 1536 | break; |
3bd1e081 MD |
1537 | default: |
1538 | ERR("Unknown consumer_data type"); | |
1539 | assert(0); | |
1540 | } | |
b9182dd9 | 1541 | |
f02e1e8a DG |
1542 | /* Handle stream on the relayd if the output is on the network */ |
1543 | if (relayd) { | |
1544 | unsigned long netlen = len; | |
1545 | ||
1546 | /* | |
1547 | * Lock the control socket for the complete duration of the function | |
1548 | * since from this point on we will use the socket. | |
1549 | */ | |
1550 | if (stream->metadata_flag) { | |
1551 | /* Metadata requires the control socket. */ | |
1552 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1d4dfdef | 1553 | netlen += sizeof(struct lttcomm_relayd_metadata_payload); |
f02e1e8a DG |
1554 | } |
1555 | ||
1d4dfdef | 1556 | ret = write_relayd_stream_header(stream, netlen, padding, relayd); |
994ab360 DG |
1557 | if (ret < 0) { |
1558 | relayd_hang_up = 1; | |
1559 | goto write_error; | |
1560 | } | |
1561 | /* Use the returned socket. */ | |
1562 | outfd = ret; | |
f02e1e8a | 1563 | |
994ab360 DG |
1564 | /* Write metadata stream id before payload */ |
1565 | if (stream->metadata_flag) { | |
1566 | ret = write_relayd_metadata_id(outfd, stream, relayd, padding); | |
1567 | if (ret < 0) { | |
8994307f DG |
1568 | relayd_hang_up = 1; |
1569 | goto write_error; | |
1570 | } | |
f02e1e8a | 1571 | } |
1d4dfdef DG |
1572 | } else { |
1573 | /* No streaming, we have to set the len with the full padding */ | |
1574 | len += padding; | |
1624d5b7 JD |
1575 | |
1576 | /* | |
1577 | * Check if we need to change the tracefile before writing the packet. | |
1578 | */ | |
1579 | if (stream->chan->tracefile_size > 0 && | |
1580 | (stream->tracefile_size_current + len) > | |
1581 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1582 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1583 | stream->name, stream->chan->tracefile_size, | |
1584 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
309167d2 JD |
1585 | stream->out_fd, &(stream->tracefile_count_current), |
1586 | &stream->out_fd); | |
1624d5b7 JD |
1587 | if (ret < 0) { |
1588 | ERR("Rotating output file"); | |
1589 | goto end; | |
1590 | } | |
309167d2 JD |
1591 | outfd = stream->out_fd; |
1592 | ||
1593 | if (stream->index_fd >= 0) { | |
f5aa4ae8 JG |
1594 | ret = close(stream->index_fd); |
1595 | if (ret < 0) { | |
1596 | PERROR("Closing index"); | |
1597 | goto end; | |
1598 | } | |
1599 | stream->index_fd = -1; | |
309167d2 JD |
1600 | ret = index_create_file(stream->chan->pathname, |
1601 | stream->name, stream->uid, stream->gid, | |
1602 | stream->chan->tracefile_size, | |
1603 | stream->tracefile_count_current); | |
1604 | if (ret < 0) { | |
1605 | goto end; | |
1606 | } | |
1607 | stream->index_fd = ret; | |
1608 | } | |
1609 | ||
a6976990 DG |
1610 | /* Reset current size because we just perform a rotation. */ |
1611 | stream->tracefile_size_current = 0; | |
a1ae300f JD |
1612 | stream->out_fd_offset = 0; |
1613 | orig_offset = 0; | |
1624d5b7 JD |
1614 | } |
1615 | stream->tracefile_size_current += len; | |
309167d2 JD |
1616 | if (index) { |
1617 | index->offset = htobe64(stream->out_fd_offset); | |
1618 | } | |
f02e1e8a DG |
1619 | } |
1620 | ||
d02b8372 DG |
1621 | /* |
1622 | * This call guarantee that len or less is returned. It's impossible to | |
1623 | * receive a ret value that is bigger than len. | |
1624 | */ | |
1625 | ret = lttng_write(outfd, mmap_base + mmap_offset, len); | |
1626 | DBG("Consumer mmap write() ret %zd (len %lu)", ret, len); | |
1627 | if (ret < 0 || ((size_t) ret != len)) { | |
1628 | /* | |
1629 | * Report error to caller if nothing was written else at least send the | |
1630 | * amount written. | |
1631 | */ | |
1632 | if (ret < 0) { | |
994ab360 | 1633 | ret = -errno; |
f02e1e8a | 1634 | } |
994ab360 | 1635 | relayd_hang_up = 1; |
f02e1e8a | 1636 | |
d02b8372 | 1637 | /* Socket operation failed. We consider the relayd dead */ |
994ab360 | 1638 | if (errno == EPIPE || errno == EINVAL || errno == EBADF) { |
d02b8372 DG |
1639 | /* |
1640 | * This is possible if the fd is closed on the other side | |
1641 | * (outfd) or any write problem. It can be verbose a bit for a | |
1642 | * normal execution if for instance the relayd is stopped | |
1643 | * abruptly. This can happen so set this to a DBG statement. | |
1644 | */ | |
1645 | DBG("Consumer mmap write detected relayd hang up"); | |
994ab360 DG |
1646 | } else { |
1647 | /* Unhandled error, print it and stop function right now. */ | |
1648 | PERROR("Error in write mmap (ret %zd != len %lu)", ret, len); | |
f02e1e8a | 1649 | } |
994ab360 | 1650 | goto write_error; |
d02b8372 DG |
1651 | } |
1652 | stream->output_written += ret; | |
d02b8372 DG |
1653 | |
1654 | /* This call is useless on a socket so better save a syscall. */ | |
1655 | if (!relayd) { | |
1656 | /* This won't block, but will start writeout asynchronously */ | |
1657 | lttng_sync_file_range(outfd, stream->out_fd_offset, len, | |
1658 | SYNC_FILE_RANGE_WRITE); | |
1659 | stream->out_fd_offset += len; | |
f02e1e8a DG |
1660 | } |
1661 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1662 | ||
8994307f DG |
1663 | write_error: |
1664 | /* | |
1665 | * This is a special case that the relayd has closed its socket. Let's | |
1666 | * cleanup the relayd object and all associated streams. | |
1667 | */ | |
1668 | if (relayd && relayd_hang_up) { | |
1669 | cleanup_relayd(relayd, ctx); | |
1670 | } | |
1671 | ||
f02e1e8a DG |
1672 | end: |
1673 | /* Unlock only if ctrl socket used */ | |
1674 | if (relayd && stream->metadata_flag) { | |
1675 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1676 | } | |
1677 | ||
1678 | rcu_read_unlock(); | |
994ab360 | 1679 | return ret; |
3bd1e081 MD |
1680 | } |
1681 | ||
1682 | /* | |
1683 | * Splice the data from the ring buffer to the tracefile. | |
1684 | * | |
79d4ffb7 DG |
1685 | * It must be called with the stream lock held. |
1686 | * | |
3bd1e081 MD |
1687 | * Returns the number of bytes spliced. |
1688 | */ | |
4078b776 | 1689 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
3bd1e081 | 1690 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1691 | struct lttng_consumer_stream *stream, unsigned long len, |
309167d2 | 1692 | unsigned long padding, |
50adc264 | 1693 | struct ctf_packet_index *index) |
3bd1e081 | 1694 | { |
f02e1e8a DG |
1695 | ssize_t ret = 0, written = 0, ret_splice = 0; |
1696 | loff_t offset = 0; | |
1697 | off_t orig_offset = stream->out_fd_offset; | |
1698 | int fd = stream->wait_fd; | |
1699 | /* Default is on the disk */ | |
1700 | int outfd = stream->out_fd; | |
f02e1e8a | 1701 | struct consumer_relayd_sock_pair *relayd = NULL; |
fb3a43a9 | 1702 | int *splice_pipe; |
8994307f | 1703 | unsigned int relayd_hang_up = 0; |
f02e1e8a | 1704 | |
3bd1e081 MD |
1705 | switch (consumer_data.type) { |
1706 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a | 1707 | break; |
7753dea8 MD |
1708 | case LTTNG_CONSUMER32_UST: |
1709 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a | 1710 | /* Not supported for user space tracing */ |
3bd1e081 MD |
1711 | return -ENOSYS; |
1712 | default: | |
1713 | ERR("Unknown consumer_data type"); | |
1714 | assert(0); | |
3bd1e081 MD |
1715 | } |
1716 | ||
f02e1e8a DG |
1717 | /* RCU lock for the relayd pointer */ |
1718 | rcu_read_lock(); | |
1719 | ||
1720 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1721 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1722 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1723 | if (relayd == NULL) { | |
ad0b0d23 | 1724 | written = -ret; |
f02e1e8a DG |
1725 | goto end; |
1726 | } | |
1727 | } | |
6f2b2a45 | 1728 | splice_pipe = stream->splice_pipe; |
fb3a43a9 | 1729 | |
f02e1e8a | 1730 | /* Write metadata stream id before payload */ |
1d4dfdef | 1731 | if (relayd) { |
ad0b0d23 | 1732 | unsigned long total_len = len; |
f02e1e8a | 1733 | |
1d4dfdef DG |
1734 | if (stream->metadata_flag) { |
1735 | /* | |
1736 | * Lock the control socket for the complete duration of the function | |
1737 | * since from this point on we will use the socket. | |
1738 | */ | |
1739 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1740 | ||
1741 | ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd, | |
1742 | padding); | |
1743 | if (ret < 0) { | |
1744 | written = ret; | |
ad0b0d23 DG |
1745 | relayd_hang_up = 1; |
1746 | goto write_error; | |
1d4dfdef DG |
1747 | } |
1748 | ||
1749 | total_len += sizeof(struct lttcomm_relayd_metadata_payload); | |
1750 | } | |
1751 | ||
1752 | ret = write_relayd_stream_header(stream, total_len, padding, relayd); | |
ad0b0d23 DG |
1753 | if (ret < 0) { |
1754 | written = ret; | |
1755 | relayd_hang_up = 1; | |
1756 | goto write_error; | |
f02e1e8a | 1757 | } |
ad0b0d23 DG |
1758 | /* Use the returned socket. */ |
1759 | outfd = ret; | |
1d4dfdef DG |
1760 | } else { |
1761 | /* No streaming, we have to set the len with the full padding */ | |
1762 | len += padding; | |
1624d5b7 JD |
1763 | |
1764 | /* | |
1765 | * Check if we need to change the tracefile before writing the packet. | |
1766 | */ | |
1767 | if (stream->chan->tracefile_size > 0 && | |
1768 | (stream->tracefile_size_current + len) > | |
1769 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1770 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1771 | stream->name, stream->chan->tracefile_size, | |
1772 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
309167d2 JD |
1773 | stream->out_fd, &(stream->tracefile_count_current), |
1774 | &stream->out_fd); | |
1624d5b7 | 1775 | if (ret < 0) { |
ad0b0d23 | 1776 | written = ret; |
1624d5b7 JD |
1777 | ERR("Rotating output file"); |
1778 | goto end; | |
1779 | } | |
309167d2 JD |
1780 | outfd = stream->out_fd; |
1781 | ||
1782 | if (stream->index_fd >= 0) { | |
79ead7f0 JD |
1783 | ret = close(stream->index_fd); |
1784 | if (ret < 0) { | |
1785 | PERROR("Closing index"); | |
1786 | goto end; | |
1787 | } | |
1788 | stream->index_fd = -1; | |
309167d2 JD |
1789 | ret = index_create_file(stream->chan->pathname, |
1790 | stream->name, stream->uid, stream->gid, | |
1791 | stream->chan->tracefile_size, | |
1792 | stream->tracefile_count_current); | |
1793 | if (ret < 0) { | |
ad0b0d23 | 1794 | written = ret; |
309167d2 JD |
1795 | goto end; |
1796 | } | |
1797 | stream->index_fd = ret; | |
1798 | } | |
1799 | ||
a6976990 DG |
1800 | /* Reset current size because we just perform a rotation. */ |
1801 | stream->tracefile_size_current = 0; | |
a1ae300f JD |
1802 | stream->out_fd_offset = 0; |
1803 | orig_offset = 0; | |
1624d5b7 JD |
1804 | } |
1805 | stream->tracefile_size_current += len; | |
309167d2 | 1806 | index->offset = htobe64(stream->out_fd_offset); |
f02e1e8a DG |
1807 | } |
1808 | ||
1809 | while (len > 0) { | |
1d4dfdef DG |
1810 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)", |
1811 | (unsigned long)offset, len, fd, splice_pipe[1]); | |
fb3a43a9 | 1812 | ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len, |
f02e1e8a DG |
1813 | SPLICE_F_MOVE | SPLICE_F_MORE); |
1814 | DBG("splice chan to pipe, ret %zd", ret_splice); | |
1815 | if (ret_splice < 0) { | |
d02b8372 | 1816 | ret = errno; |
ad0b0d23 | 1817 | written = -ret; |
d02b8372 | 1818 | PERROR("Error in relay splice"); |
f02e1e8a DG |
1819 | goto splice_error; |
1820 | } | |
1821 | ||
1822 | /* Handle stream on the relayd if the output is on the network */ | |
ad0b0d23 DG |
1823 | if (relayd && stream->metadata_flag) { |
1824 | size_t metadata_payload_size = | |
1825 | sizeof(struct lttcomm_relayd_metadata_payload); | |
1826 | ||
1827 | /* Update counter to fit the spliced data */ | |
1828 | ret_splice += metadata_payload_size; | |
1829 | len += metadata_payload_size; | |
1830 | /* | |
1831 | * We do this so the return value can match the len passed as | |
1832 | * argument to this function. | |
1833 | */ | |
1834 | written -= metadata_payload_size; | |
f02e1e8a DG |
1835 | } |
1836 | ||
1837 | /* Splice data out */ | |
fb3a43a9 | 1838 | ret_splice = splice(splice_pipe[0], NULL, outfd, NULL, |
f02e1e8a | 1839 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); |
6f2b2a45 JD |
1840 | DBG("Consumer splice pipe to file (out_fd: %d), ret %zd", |
1841 | outfd, ret_splice); | |
f02e1e8a | 1842 | if (ret_splice < 0) { |
d02b8372 | 1843 | ret = errno; |
ad0b0d23 DG |
1844 | written = -ret; |
1845 | relayd_hang_up = 1; | |
1846 | goto write_error; | |
f02e1e8a | 1847 | } else if (ret_splice > len) { |
d02b8372 DG |
1848 | /* |
1849 | * We don't expect this code path to be executed but you never know | |
1850 | * so this is an extra protection agains a buggy splice(). | |
1851 | */ | |
f02e1e8a | 1852 | ret = errno; |
ad0b0d23 | 1853 | written += ret_splice; |
d02b8372 DG |
1854 | PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice, |
1855 | len); | |
f02e1e8a | 1856 | goto splice_error; |
d02b8372 DG |
1857 | } else { |
1858 | /* All good, update current len and continue. */ | |
1859 | len -= ret_splice; | |
f02e1e8a | 1860 | } |
f02e1e8a DG |
1861 | |
1862 | /* This call is useless on a socket so better save a syscall. */ | |
1863 | if (!relayd) { | |
1864 | /* This won't block, but will start writeout asynchronously */ | |
1865 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, | |
1866 | SYNC_FILE_RANGE_WRITE); | |
1867 | stream->out_fd_offset += ret_splice; | |
1868 | } | |
e5d1a9b3 | 1869 | stream->output_written += ret_splice; |
f02e1e8a DG |
1870 | written += ret_splice; |
1871 | } | |
1872 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
f02e1e8a DG |
1873 | goto end; |
1874 | ||
8994307f DG |
1875 | write_error: |
1876 | /* | |
1877 | * This is a special case that the relayd has closed its socket. Let's | |
1878 | * cleanup the relayd object and all associated streams. | |
1879 | */ | |
1880 | if (relayd && relayd_hang_up) { | |
1881 | cleanup_relayd(relayd, ctx); | |
1882 | /* Skip splice error so the consumer does not fail */ | |
1883 | goto end; | |
1884 | } | |
1885 | ||
f02e1e8a DG |
1886 | splice_error: |
1887 | /* send the appropriate error description to sessiond */ | |
1888 | switch (ret) { | |
f02e1e8a | 1889 | case EINVAL: |
f73fabfd | 1890 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); |
f02e1e8a DG |
1891 | break; |
1892 | case ENOMEM: | |
f73fabfd | 1893 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); |
f02e1e8a DG |
1894 | break; |
1895 | case ESPIPE: | |
f73fabfd | 1896 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); |
f02e1e8a DG |
1897 | break; |
1898 | } | |
1899 | ||
1900 | end: | |
1901 | if (relayd && stream->metadata_flag) { | |
1902 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1903 | } | |
1904 | ||
1905 | rcu_read_unlock(); | |
1906 | return written; | |
3bd1e081 MD |
1907 | } |
1908 | ||
1909 | /* | |
1910 | * Take a snapshot for a specific fd | |
1911 | * | |
1912 | * Returns 0 on success, < 0 on error | |
1913 | */ | |
ffe60014 | 1914 | int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) |
3bd1e081 MD |
1915 | { |
1916 | switch (consumer_data.type) { | |
1917 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1918 | return lttng_kconsumer_take_snapshot(stream); |
7753dea8 MD |
1919 | case LTTNG_CONSUMER32_UST: |
1920 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 1921 | return lttng_ustconsumer_take_snapshot(stream); |
3bd1e081 MD |
1922 | default: |
1923 | ERR("Unknown consumer_data type"); | |
1924 | assert(0); | |
1925 | return -ENOSYS; | |
1926 | } | |
3bd1e081 MD |
1927 | } |
1928 | ||
1929 | /* | |
1930 | * Get the produced position | |
1931 | * | |
1932 | * Returns 0 on success, < 0 on error | |
1933 | */ | |
ffe60014 | 1934 | int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, |
3bd1e081 MD |
1935 | unsigned long *pos) |
1936 | { | |
1937 | switch (consumer_data.type) { | |
1938 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1939 | return lttng_kconsumer_get_produced_snapshot(stream, pos); |
7753dea8 MD |
1940 | case LTTNG_CONSUMER32_UST: |
1941 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 1942 | return lttng_ustconsumer_get_produced_snapshot(stream, pos); |
3bd1e081 MD |
1943 | default: |
1944 | ERR("Unknown consumer_data type"); | |
1945 | assert(0); | |
1946 | return -ENOSYS; | |
1947 | } | |
1948 | } | |
1949 | ||
1950 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, | |
1951 | int sock, struct pollfd *consumer_sockpoll) | |
1952 | { | |
1953 | switch (consumer_data.type) { | |
1954 | case LTTNG_CONSUMER_KERNEL: | |
1955 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
7753dea8 MD |
1956 | case LTTNG_CONSUMER32_UST: |
1957 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
1958 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
1959 | default: | |
1960 | ERR("Unknown consumer_data type"); | |
1961 | assert(0); | |
1962 | return -ENOSYS; | |
1963 | } | |
1964 | } | |
1965 | ||
6d574024 | 1966 | void lttng_consumer_close_all_metadata(void) |
d88aee68 DG |
1967 | { |
1968 | switch (consumer_data.type) { | |
1969 | case LTTNG_CONSUMER_KERNEL: | |
1970 | /* | |
1971 | * The Kernel consumer has a different metadata scheme so we don't | |
1972 | * close anything because the stream will be closed by the session | |
1973 | * daemon. | |
1974 | */ | |
1975 | break; | |
1976 | case LTTNG_CONSUMER32_UST: | |
1977 | case LTTNG_CONSUMER64_UST: | |
1978 | /* | |
1979 | * Close all metadata streams. The metadata hash table is passed and | |
1980 | * this call iterates over it by closing all wakeup fd. This is safe | |
1981 | * because at this point we are sure that the metadata producer is | |
1982 | * either dead or blocked. | |
1983 | */ | |
6d574024 | 1984 | lttng_ustconsumer_close_all_metadata(metadata_ht); |
d88aee68 DG |
1985 | break; |
1986 | default: | |
1987 | ERR("Unknown consumer_data type"); | |
1988 | assert(0); | |
1989 | } | |
1990 | } | |
1991 | ||
fb3a43a9 DG |
1992 | /* |
1993 | * Clean up a metadata stream and free its memory. | |
1994 | */ | |
e316aad5 DG |
1995 | void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, |
1996 | struct lttng_ht *ht) | |
fb3a43a9 | 1997 | { |
e316aad5 | 1998 | struct lttng_consumer_channel *free_chan = NULL; |
fb3a43a9 DG |
1999 | |
2000 | assert(stream); | |
2001 | /* | |
2002 | * This call should NEVER receive regular stream. It must always be | |
2003 | * metadata stream and this is crucial for data structure synchronization. | |
2004 | */ | |
2005 | assert(stream->metadata_flag); | |
2006 | ||
e316aad5 DG |
2007 | DBG3("Consumer delete metadata stream %d", stream->wait_fd); |
2008 | ||
74251bb8 | 2009 | pthread_mutex_lock(&consumer_data.lock); |
a9838785 | 2010 | pthread_mutex_lock(&stream->chan->lock); |
8994307f DG |
2011 | pthread_mutex_lock(&stream->lock); |
2012 | ||
6d574024 DG |
2013 | /* Remove any reference to that stream. */ |
2014 | consumer_stream_delete(stream, ht); | |
ca22feea | 2015 | |
6d574024 DG |
2016 | /* Close down everything including the relayd if one. */ |
2017 | consumer_stream_close(stream); | |
2018 | /* Destroy tracer buffers of the stream. */ | |
2019 | consumer_stream_destroy_buffers(stream); | |
fb3a43a9 DG |
2020 | |
2021 | /* Atomically decrement channel refcount since other threads can use it. */ | |
f2ad556d | 2022 | if (!uatomic_sub_return(&stream->chan->refcount, 1) |
ffe60014 | 2023 | && !uatomic_read(&stream->chan->nb_init_stream_left)) { |
c30aaa51 | 2024 | /* Go for channel deletion! */ |
e316aad5 | 2025 | free_chan = stream->chan; |
fb3a43a9 DG |
2026 | } |
2027 | ||
73811ecc DG |
2028 | /* |
2029 | * Nullify the stream reference so it is not used after deletion. The | |
6d574024 DG |
2030 | * channel lock MUST be acquired before being able to check for a NULL |
2031 | * pointer value. | |
73811ecc DG |
2032 | */ |
2033 | stream->chan->metadata_stream = NULL; | |
2034 | ||
8994307f | 2035 | pthread_mutex_unlock(&stream->lock); |
a9838785 | 2036 | pthread_mutex_unlock(&stream->chan->lock); |
74251bb8 | 2037 | pthread_mutex_unlock(&consumer_data.lock); |
e316aad5 DG |
2038 | |
2039 | if (free_chan) { | |
2040 | consumer_del_channel(free_chan); | |
2041 | } | |
2042 | ||
6d574024 | 2043 | consumer_stream_free(stream); |
fb3a43a9 DG |
2044 | } |
2045 | ||
2046 | /* | |
2047 | * Action done with the metadata stream when adding it to the consumer internal | |
2048 | * data structures to handle it. | |
2049 | */ | |
5ab66908 | 2050 | int consumer_add_metadata_stream(struct lttng_consumer_stream *stream) |
fb3a43a9 | 2051 | { |
5ab66908 | 2052 | struct lttng_ht *ht = metadata_ht; |
e316aad5 | 2053 | int ret = 0; |
76082088 | 2054 | struct lttng_ht_iter iter; |
d88aee68 | 2055 | struct lttng_ht_node_u64 *node; |
fb3a43a9 | 2056 | |
e316aad5 DG |
2057 | assert(stream); |
2058 | assert(ht); | |
2059 | ||
d88aee68 | 2060 | DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key); |
e316aad5 DG |
2061 | |
2062 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 2063 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 2064 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 2065 | pthread_mutex_lock(&stream->lock); |
e316aad5 | 2066 | |
e316aad5 DG |
2067 | /* |
2068 | * From here, refcounts are updated so be _careful_ when returning an error | |
2069 | * after this point. | |
2070 | */ | |
2071 | ||
fb3a43a9 | 2072 | rcu_read_lock(); |
76082088 DG |
2073 | |
2074 | /* | |
2075 | * Lookup the stream just to make sure it does not exist in our internal | |
2076 | * state. This should NEVER happen. | |
2077 | */ | |
d88aee68 DG |
2078 | lttng_ht_lookup(ht, &stream->key, &iter); |
2079 | node = lttng_ht_iter_get_node_u64(&iter); | |
76082088 DG |
2080 | assert(!node); |
2081 | ||
e316aad5 | 2082 | /* |
ffe60014 DG |
2083 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
2084 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
2085 | * causes the count to become 0 also causes a stream to be added. The |
2086 | * channel deletion will thus be triggered by the following removal of this | |
2087 | * stream. | |
2088 | */ | |
ffe60014 | 2089 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
2090 | /* Increment refcount before decrementing nb_init_stream_left */ |
2091 | cmm_smp_wmb(); | |
ffe60014 | 2092 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
2093 | } |
2094 | ||
d88aee68 | 2095 | lttng_ht_add_unique_u64(ht, &stream->node); |
ca22feea | 2096 | |
d8ef542d MD |
2097 | lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht, |
2098 | &stream->node_channel_id); | |
2099 | ||
ca22feea DG |
2100 | /* |
2101 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
2102 | * the key since the HT does not use it and we allow to add redundant keys | |
2103 | * into this table. | |
2104 | */ | |
d88aee68 | 2105 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 2106 | |
fb3a43a9 | 2107 | rcu_read_unlock(); |
e316aad5 | 2108 | |
2e818a6a | 2109 | pthread_mutex_unlock(&stream->lock); |
a9838785 | 2110 | pthread_mutex_unlock(&stream->chan->lock); |
ec6ea7d0 | 2111 | pthread_mutex_unlock(&stream->chan->timer_lock); |
e316aad5 DG |
2112 | pthread_mutex_unlock(&consumer_data.lock); |
2113 | return ret; | |
fb3a43a9 DG |
2114 | } |
2115 | ||
8994307f DG |
2116 | /* |
2117 | * Delete data stream that are flagged for deletion (endpoint_status). | |
2118 | */ | |
2119 | static void validate_endpoint_status_data_stream(void) | |
2120 | { | |
2121 | struct lttng_ht_iter iter; | |
2122 | struct lttng_consumer_stream *stream; | |
2123 | ||
2124 | DBG("Consumer delete flagged data stream"); | |
2125 | ||
2126 | rcu_read_lock(); | |
2127 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
2128 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2129 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2130 | continue; |
2131 | } | |
2132 | /* Delete it right now */ | |
2133 | consumer_del_stream(stream, data_ht); | |
2134 | } | |
2135 | rcu_read_unlock(); | |
2136 | } | |
2137 | ||
2138 | /* | |
2139 | * Delete metadata stream that are flagged for deletion (endpoint_status). | |
2140 | */ | |
2141 | static void validate_endpoint_status_metadata_stream( | |
2142 | struct lttng_poll_event *pollset) | |
2143 | { | |
2144 | struct lttng_ht_iter iter; | |
2145 | struct lttng_consumer_stream *stream; | |
2146 | ||
2147 | DBG("Consumer delete flagged metadata stream"); | |
2148 | ||
2149 | assert(pollset); | |
2150 | ||
2151 | rcu_read_lock(); | |
2152 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
2153 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2154 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2155 | continue; |
2156 | } | |
2157 | /* | |
2158 | * Remove from pollset so the metadata thread can continue without | |
2159 | * blocking on a deleted stream. | |
2160 | */ | |
2161 | lttng_poll_del(pollset, stream->wait_fd); | |
2162 | ||
2163 | /* Delete it right now */ | |
2164 | consumer_del_metadata_stream(stream, metadata_ht); | |
2165 | } | |
2166 | rcu_read_unlock(); | |
2167 | } | |
2168 | ||
fb3a43a9 DG |
2169 | /* |
2170 | * Thread polls on metadata file descriptor and write them on disk or on the | |
2171 | * network. | |
2172 | */ | |
7d980def | 2173 | void *consumer_thread_metadata_poll(void *data) |
fb3a43a9 | 2174 | { |
1fc79fb4 | 2175 | int ret, i, pollfd, err = -1; |
fb3a43a9 | 2176 | uint32_t revents, nb_fd; |
e316aad5 | 2177 | struct lttng_consumer_stream *stream = NULL; |
fb3a43a9 | 2178 | struct lttng_ht_iter iter; |
d88aee68 | 2179 | struct lttng_ht_node_u64 *node; |
fb3a43a9 DG |
2180 | struct lttng_poll_event events; |
2181 | struct lttng_consumer_local_data *ctx = data; | |
2182 | ssize_t len; | |
2183 | ||
2184 | rcu_register_thread(); | |
2185 | ||
1fc79fb4 MD |
2186 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA); |
2187 | ||
2d57de81 MD |
2188 | if (testpoint(consumerd_thread_metadata)) { |
2189 | goto error_testpoint; | |
2190 | } | |
2191 | ||
9ce5646a MD |
2192 | health_code_update(); |
2193 | ||
fb3a43a9 DG |
2194 | DBG("Thread metadata poll started"); |
2195 | ||
fb3a43a9 DG |
2196 | /* Size is set to 1 for the consumer_metadata pipe */ |
2197 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2198 | if (ret < 0) { | |
2199 | ERR("Poll set creation failed"); | |
d8ef542d | 2200 | goto end_poll; |
fb3a43a9 DG |
2201 | } |
2202 | ||
13886d2d DG |
2203 | ret = lttng_poll_add(&events, |
2204 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN); | |
fb3a43a9 DG |
2205 | if (ret < 0) { |
2206 | goto end; | |
2207 | } | |
2208 | ||
2209 | /* Main loop */ | |
2210 | DBG("Metadata main loop started"); | |
2211 | ||
2212 | while (1) { | |
fb3a43a9 | 2213 | restart: |
65da3cc7 | 2214 | health_code_update(); |
9ce5646a | 2215 | health_poll_entry(); |
65da3cc7 | 2216 | DBG("Metadata poll wait"); |
fb3a43a9 | 2217 | ret = lttng_poll_wait(&events, -1); |
65da3cc7 MD |
2218 | DBG("Metadata poll return from wait with %d fd(s)", |
2219 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2220 | health_poll_exit(); |
bfa78590 | 2221 | DBG("Metadata event caught in thread"); |
fb3a43a9 DG |
2222 | if (ret < 0) { |
2223 | if (errno == EINTR) { | |
bfa78590 | 2224 | ERR("Poll EINTR caught"); |
fb3a43a9 DG |
2225 | goto restart; |
2226 | } | |
0a7555a7 MD |
2227 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2228 | err = 0; /* All is OK */ | |
2229 | } | |
2230 | goto end; | |
fb3a43a9 DG |
2231 | } |
2232 | ||
0d9c5d77 DG |
2233 | nb_fd = ret; |
2234 | ||
e316aad5 | 2235 | /* From here, the event is a metadata wait fd */ |
fb3a43a9 | 2236 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2237 | health_code_update(); |
2238 | ||
fb3a43a9 DG |
2239 | revents = LTTNG_POLL_GETEV(&events, i); |
2240 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2241 | ||
e0077699 MD |
2242 | if (!revents) { |
2243 | /* No activity for this FD (poll implementation). */ | |
2244 | continue; | |
2245 | } | |
2246 | ||
13886d2d | 2247 | if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) { |
584fc280 | 2248 | if (revents & LPOLLIN) { |
13886d2d DG |
2249 | ssize_t pipe_len; |
2250 | ||
2251 | pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe, | |
2252 | &stream, sizeof(stream)); | |
6cd525e8 | 2253 | if (pipe_len < sizeof(stream)) { |
584fc280 MD |
2254 | if (pipe_len < 0) { |
2255 | PERROR("read metadata stream"); | |
2256 | } | |
fb3a43a9 | 2257 | /* |
584fc280 MD |
2258 | * Remove the pipe from the poll set and continue the loop |
2259 | * since their might be data to consume. | |
fb3a43a9 | 2260 | */ |
584fc280 MD |
2261 | lttng_poll_del(&events, |
2262 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2263 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
fb3a43a9 DG |
2264 | continue; |
2265 | } | |
2266 | ||
8994307f DG |
2267 | /* A NULL stream means that the state has changed. */ |
2268 | if (stream == NULL) { | |
2269 | /* Check for deleted streams. */ | |
2270 | validate_endpoint_status_metadata_stream(&events); | |
3714380f | 2271 | goto restart; |
8994307f DG |
2272 | } |
2273 | ||
fb3a43a9 DG |
2274 | DBG("Adding metadata stream %d to poll set", |
2275 | stream->wait_fd); | |
2276 | ||
fb3a43a9 DG |
2277 | /* Add metadata stream to the global poll events list */ |
2278 | lttng_poll_add(&events, stream->wait_fd, | |
6d574024 | 2279 | LPOLLIN | LPOLLPRI | LPOLLHUP); |
584fc280 MD |
2280 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
2281 | DBG("Metadata thread pipe hung up"); | |
2282 | /* | |
2283 | * Remove the pipe from the poll set and continue the loop | |
2284 | * since their might be data to consume. | |
2285 | */ | |
2286 | lttng_poll_del(&events, | |
2287 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2288 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
2289 | continue; | |
2290 | } else { | |
2291 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
2292 | goto end; | |
fb3a43a9 DG |
2293 | } |
2294 | ||
e316aad5 | 2295 | /* Handle other stream */ |
fb3a43a9 DG |
2296 | continue; |
2297 | } | |
2298 | ||
d09e1200 | 2299 | rcu_read_lock(); |
d88aee68 DG |
2300 | { |
2301 | uint64_t tmp_id = (uint64_t) pollfd; | |
2302 | ||
2303 | lttng_ht_lookup(metadata_ht, &tmp_id, &iter); | |
2304 | } | |
2305 | node = lttng_ht_iter_get_node_u64(&iter); | |
e316aad5 | 2306 | assert(node); |
fb3a43a9 DG |
2307 | |
2308 | stream = caa_container_of(node, struct lttng_consumer_stream, | |
58b1f425 | 2309 | node); |
fb3a43a9 | 2310 | |
584fc280 MD |
2311 | if (revents & (LPOLLIN | LPOLLPRI)) { |
2312 | /* Get the data out of the metadata file descriptor */ | |
2313 | DBG("Metadata available on fd %d", pollfd); | |
2314 | assert(stream->wait_fd == pollfd); | |
2315 | ||
2316 | do { | |
2317 | health_code_update(); | |
2318 | ||
2319 | len = ctx->on_buffer_ready(stream, ctx); | |
2320 | /* | |
2321 | * We don't check the return value here since if we get | |
2322 | * a negative len, it means an error occured thus we | |
2323 | * simply remove it from the poll set and free the | |
2324 | * stream. | |
2325 | */ | |
2326 | } while (len > 0); | |
2327 | ||
2328 | /* It's ok to have an unavailable sub-buffer */ | |
2329 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { | |
2330 | /* Clean up stream from consumer and free it. */ | |
2331 | lttng_poll_del(&events, stream->wait_fd); | |
2332 | consumer_del_metadata_stream(stream, metadata_ht); | |
2333 | } | |
2334 | } else if (revents & (LPOLLERR | LPOLLHUP)) { | |
e316aad5 | 2335 | DBG("Metadata fd %d is hup|err.", pollfd); |
fb3a43a9 DG |
2336 | if (!stream->hangup_flush_done |
2337 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2338 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2339 | DBG("Attempting to flush and consume the UST buffers"); | |
2340 | lttng_ustconsumer_on_stream_hangup(stream); | |
2341 | ||
2342 | /* We just flushed the stream now read it. */ | |
4bb94b75 | 2343 | do { |
9ce5646a MD |
2344 | health_code_update(); |
2345 | ||
4bb94b75 DG |
2346 | len = ctx->on_buffer_ready(stream, ctx); |
2347 | /* | |
2348 | * We don't check the return value here since if we get | |
2349 | * a negative len, it means an error occured thus we | |
2350 | * simply remove it from the poll set and free the | |
2351 | * stream. | |
2352 | */ | |
2353 | } while (len > 0); | |
fb3a43a9 DG |
2354 | } |
2355 | ||
fb3a43a9 | 2356 | lttng_poll_del(&events, stream->wait_fd); |
e316aad5 DG |
2357 | /* |
2358 | * This call update the channel states, closes file descriptors | |
2359 | * and securely free the stream. | |
2360 | */ | |
2361 | consumer_del_metadata_stream(stream, metadata_ht); | |
584fc280 MD |
2362 | } else { |
2363 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
88d9d15a | 2364 | rcu_read_unlock(); |
584fc280 | 2365 | goto end; |
fb3a43a9 | 2366 | } |
e316aad5 | 2367 | /* Release RCU lock for the stream looked up */ |
d09e1200 | 2368 | rcu_read_unlock(); |
fb3a43a9 DG |
2369 | } |
2370 | } | |
2371 | ||
1fc79fb4 MD |
2372 | /* All is OK */ |
2373 | err = 0; | |
fb3a43a9 DG |
2374 | end: |
2375 | DBG("Metadata poll thread exiting"); | |
fb3a43a9 | 2376 | |
d8ef542d MD |
2377 | lttng_poll_clean(&events); |
2378 | end_poll: | |
2d57de81 | 2379 | error_testpoint: |
1fc79fb4 MD |
2380 | if (err) { |
2381 | health_error(); | |
2382 | ERR("Health error occurred in %s", __func__); | |
2383 | } | |
2384 | health_unregister(health_consumerd); | |
fb3a43a9 DG |
2385 | rcu_unregister_thread(); |
2386 | return NULL; | |
2387 | } | |
2388 | ||
3bd1e081 | 2389 | /* |
e4421fec | 2390 | * This thread polls the fds in the set to consume the data and write |
3bd1e081 MD |
2391 | * it to tracefile if necessary. |
2392 | */ | |
7d980def | 2393 | void *consumer_thread_data_poll(void *data) |
3bd1e081 | 2394 | { |
1fc79fb4 | 2395 | int num_rdy, num_hup, high_prio, ret, i, err = -1; |
3bd1e081 MD |
2396 | struct pollfd *pollfd = NULL; |
2397 | /* local view of the streams */ | |
c869f647 | 2398 | struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; |
3bd1e081 MD |
2399 | /* local view of consumer_data.fds_count */ |
2400 | int nb_fd = 0; | |
3bd1e081 | 2401 | struct lttng_consumer_local_data *ctx = data; |
00e2e675 | 2402 | ssize_t len; |
3bd1e081 | 2403 | |
e7b994a3 DG |
2404 | rcu_register_thread(); |
2405 | ||
1fc79fb4 MD |
2406 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA); |
2407 | ||
2d57de81 MD |
2408 | if (testpoint(consumerd_thread_data)) { |
2409 | goto error_testpoint; | |
2410 | } | |
2411 | ||
9ce5646a MD |
2412 | health_code_update(); |
2413 | ||
4df6c8cb MD |
2414 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream *)); |
2415 | if (local_stream == NULL) { | |
2416 | PERROR("local_stream malloc"); | |
2417 | goto end; | |
2418 | } | |
3bd1e081 MD |
2419 | |
2420 | while (1) { | |
9ce5646a MD |
2421 | health_code_update(); |
2422 | ||
3bd1e081 MD |
2423 | high_prio = 0; |
2424 | num_hup = 0; | |
2425 | ||
2426 | /* | |
e4421fec | 2427 | * the fds set has been updated, we need to update our |
3bd1e081 MD |
2428 | * local array as well |
2429 | */ | |
2430 | pthread_mutex_lock(&consumer_data.lock); | |
2431 | if (consumer_data.need_update) { | |
0e428499 DG |
2432 | free(pollfd); |
2433 | pollfd = NULL; | |
2434 | ||
2435 | free(local_stream); | |
2436 | local_stream = NULL; | |
3bd1e081 | 2437 | |
02b3d176 DG |
2438 | /* |
2439 | * Allocate for all fds +1 for the consumer_data_pipe and +1 for | |
2440 | * wake up pipe. | |
2441 | */ | |
2442 | pollfd = zmalloc((consumer_data.stream_count + 2) * sizeof(struct pollfd)); | |
3bd1e081 | 2443 | if (pollfd == NULL) { |
7a57cf92 | 2444 | PERROR("pollfd malloc"); |
3bd1e081 MD |
2445 | pthread_mutex_unlock(&consumer_data.lock); |
2446 | goto end; | |
2447 | } | |
2448 | ||
02b3d176 | 2449 | local_stream = zmalloc((consumer_data.stream_count + 2) * |
747f8642 | 2450 | sizeof(struct lttng_consumer_stream *)); |
3bd1e081 | 2451 | if (local_stream == NULL) { |
7a57cf92 | 2452 | PERROR("local_stream malloc"); |
3bd1e081 MD |
2453 | pthread_mutex_unlock(&consumer_data.lock); |
2454 | goto end; | |
2455 | } | |
ffe60014 | 2456 | ret = update_poll_array(ctx, &pollfd, local_stream, |
43c34bc3 | 2457 | data_ht); |
3bd1e081 MD |
2458 | if (ret < 0) { |
2459 | ERR("Error in allocating pollfd or local_outfds"); | |
f73fabfd | 2460 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2461 | pthread_mutex_unlock(&consumer_data.lock); |
2462 | goto end; | |
2463 | } | |
2464 | nb_fd = ret; | |
2465 | consumer_data.need_update = 0; | |
2466 | } | |
2467 | pthread_mutex_unlock(&consumer_data.lock); | |
2468 | ||
4078b776 MD |
2469 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
2470 | if (nb_fd == 0 && consumer_quit == 1) { | |
1fc79fb4 | 2471 | err = 0; /* All is OK */ |
4078b776 MD |
2472 | goto end; |
2473 | } | |
3bd1e081 | 2474 | /* poll on the array of fds */ |
88f2b785 | 2475 | restart: |
02b3d176 | 2476 | DBG("polling on %d fd", nb_fd + 2); |
9ce5646a | 2477 | health_poll_entry(); |
02b3d176 | 2478 | num_rdy = poll(pollfd, nb_fd + 2, -1); |
9ce5646a | 2479 | health_poll_exit(); |
3bd1e081 MD |
2480 | DBG("poll num_rdy : %d", num_rdy); |
2481 | if (num_rdy == -1) { | |
88f2b785 MD |
2482 | /* |
2483 | * Restart interrupted system call. | |
2484 | */ | |
2485 | if (errno == EINTR) { | |
2486 | goto restart; | |
2487 | } | |
7a57cf92 | 2488 | PERROR("Poll error"); |
f73fabfd | 2489 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2490 | goto end; |
2491 | } else if (num_rdy == 0) { | |
2492 | DBG("Polling thread timed out"); | |
2493 | goto end; | |
2494 | } | |
2495 | ||
3bd1e081 | 2496 | /* |
50f8ae69 | 2497 | * If the consumer_data_pipe triggered poll go directly to the |
00e2e675 DG |
2498 | * beginning of the loop to update the array. We want to prioritize |
2499 | * array update over low-priority reads. | |
3bd1e081 | 2500 | */ |
509bb1cf | 2501 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
ab30f567 | 2502 | ssize_t pipe_readlen; |
04fdd819 | 2503 | |
50f8ae69 | 2504 | DBG("consumer_data_pipe wake up"); |
acdb9057 DG |
2505 | pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe, |
2506 | &new_stream, sizeof(new_stream)); | |
6cd525e8 MD |
2507 | if (pipe_readlen < sizeof(new_stream)) { |
2508 | PERROR("Consumer data pipe"); | |
23f5f35d DG |
2509 | /* Continue so we can at least handle the current stream(s). */ |
2510 | continue; | |
2511 | } | |
c869f647 DG |
2512 | |
2513 | /* | |
2514 | * If the stream is NULL, just ignore it. It's also possible that | |
2515 | * the sessiond poll thread changed the consumer_quit state and is | |
2516 | * waking us up to test it. | |
2517 | */ | |
2518 | if (new_stream == NULL) { | |
8994307f | 2519 | validate_endpoint_status_data_stream(); |
c869f647 DG |
2520 | continue; |
2521 | } | |
2522 | ||
c869f647 | 2523 | /* Continue to update the local streams and handle prio ones */ |
3bd1e081 MD |
2524 | continue; |
2525 | } | |
2526 | ||
02b3d176 DG |
2527 | /* Handle wakeup pipe. */ |
2528 | if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) { | |
2529 | char dummy; | |
2530 | ssize_t pipe_readlen; | |
2531 | ||
2532 | pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy, | |
2533 | sizeof(dummy)); | |
2534 | if (pipe_readlen < 0) { | |
2535 | PERROR("Consumer data wakeup pipe"); | |
2536 | } | |
2537 | /* We've been awakened to handle stream(s). */ | |
2538 | ctx->has_wakeup = 0; | |
2539 | } | |
2540 | ||
3bd1e081 MD |
2541 | /* Take care of high priority channels first. */ |
2542 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2543 | health_code_update(); |
2544 | ||
9617607b DG |
2545 | if (local_stream[i] == NULL) { |
2546 | continue; | |
2547 | } | |
fb3a43a9 | 2548 | if (pollfd[i].revents & POLLPRI) { |
d41f73b7 MD |
2549 | DBG("Urgent read on fd %d", pollfd[i].fd); |
2550 | high_prio = 1; | |
4078b776 | 2551 | len = ctx->on_buffer_ready(local_stream[i], ctx); |
d41f73b7 | 2552 | /* it's ok to have an unavailable sub-buffer */ |
b64403e3 | 2553 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2554 | /* Clean the stream and free it. */ |
2555 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2556 | local_stream[i] = NULL; |
4078b776 MD |
2557 | } else if (len > 0) { |
2558 | local_stream[i]->data_read = 1; | |
d41f73b7 | 2559 | } |
3bd1e081 MD |
2560 | } |
2561 | } | |
2562 | ||
4078b776 MD |
2563 | /* |
2564 | * If we read high prio channel in this loop, try again | |
2565 | * for more high prio data. | |
2566 | */ | |
2567 | if (high_prio) { | |
3bd1e081 MD |
2568 | continue; |
2569 | } | |
2570 | ||
2571 | /* Take care of low priority channels. */ | |
4078b776 | 2572 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2573 | health_code_update(); |
2574 | ||
9617607b DG |
2575 | if (local_stream[i] == NULL) { |
2576 | continue; | |
2577 | } | |
4078b776 | 2578 | if ((pollfd[i].revents & POLLIN) || |
02b3d176 DG |
2579 | local_stream[i]->hangup_flush_done || |
2580 | local_stream[i]->has_data) { | |
4078b776 MD |
2581 | DBG("Normal read on fd %d", pollfd[i].fd); |
2582 | len = ctx->on_buffer_ready(local_stream[i], ctx); | |
2583 | /* it's ok to have an unavailable sub-buffer */ | |
b64403e3 | 2584 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2585 | /* Clean the stream and free it. */ |
2586 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2587 | local_stream[i] = NULL; |
4078b776 MD |
2588 | } else if (len > 0) { |
2589 | local_stream[i]->data_read = 1; | |
2590 | } | |
2591 | } | |
2592 | } | |
2593 | ||
2594 | /* Handle hangup and errors */ | |
2595 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2596 | health_code_update(); |
2597 | ||
9617607b DG |
2598 | if (local_stream[i] == NULL) { |
2599 | continue; | |
2600 | } | |
4078b776 MD |
2601 | if (!local_stream[i]->hangup_flush_done |
2602 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) | |
2603 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2604 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2605 | DBG("fd %d is hup|err|nval. Attempting flush and read.", | |
9617607b | 2606 | pollfd[i].fd); |
4078b776 MD |
2607 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); |
2608 | /* Attempt read again, for the data we just flushed. */ | |
2609 | local_stream[i]->data_read = 1; | |
2610 | } | |
2611 | /* | |
2612 | * If the poll flag is HUP/ERR/NVAL and we have | |
2613 | * read no data in this pass, we can remove the | |
2614 | * stream from its hash table. | |
2615 | */ | |
2616 | if ((pollfd[i].revents & POLLHUP)) { | |
2617 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); | |
2618 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2619 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2620 | local_stream[i] = NULL; |
4078b776 MD |
2621 | num_hup++; |
2622 | } | |
2623 | } else if (pollfd[i].revents & POLLERR) { | |
2624 | ERR("Error returned in polling fd %d.", pollfd[i].fd); | |
2625 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2626 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2627 | local_stream[i] = NULL; |
4078b776 MD |
2628 | num_hup++; |
2629 | } | |
2630 | } else if (pollfd[i].revents & POLLNVAL) { | |
2631 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); | |
2632 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2633 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2634 | local_stream[i] = NULL; |
4078b776 | 2635 | num_hup++; |
3bd1e081 MD |
2636 | } |
2637 | } | |
9617607b DG |
2638 | if (local_stream[i] != NULL) { |
2639 | local_stream[i]->data_read = 0; | |
2640 | } | |
3bd1e081 MD |
2641 | } |
2642 | } | |
1fc79fb4 MD |
2643 | /* All is OK */ |
2644 | err = 0; | |
3bd1e081 MD |
2645 | end: |
2646 | DBG("polling thread exiting"); | |
0e428499 DG |
2647 | free(pollfd); |
2648 | free(local_stream); | |
fb3a43a9 DG |
2649 | |
2650 | /* | |
2651 | * Close the write side of the pipe so epoll_wait() in | |
7d980def DG |
2652 | * consumer_thread_metadata_poll can catch it. The thread is monitoring the |
2653 | * read side of the pipe. If we close them both, epoll_wait strangely does | |
2654 | * not return and could create a endless wait period if the pipe is the | |
2655 | * only tracked fd in the poll set. The thread will take care of closing | |
2656 | * the read side. | |
fb3a43a9 | 2657 | */ |
13886d2d | 2658 | (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe); |
fb3a43a9 | 2659 | |
2d57de81 | 2660 | error_testpoint: |
1fc79fb4 MD |
2661 | if (err) { |
2662 | health_error(); | |
2663 | ERR("Health error occurred in %s", __func__); | |
2664 | } | |
2665 | health_unregister(health_consumerd); | |
2666 | ||
e7b994a3 | 2667 | rcu_unregister_thread(); |
3bd1e081 MD |
2668 | return NULL; |
2669 | } | |
2670 | ||
d8ef542d MD |
2671 | /* |
2672 | * Close wake-up end of each stream belonging to the channel. This will | |
2673 | * allow the poll() on the stream read-side to detect when the | |
2674 | * write-side (application) finally closes them. | |
2675 | */ | |
2676 | static | |
2677 | void consumer_close_channel_streams(struct lttng_consumer_channel *channel) | |
2678 | { | |
2679 | struct lttng_ht *ht; | |
2680 | struct lttng_consumer_stream *stream; | |
2681 | struct lttng_ht_iter iter; | |
2682 | ||
2683 | ht = consumer_data.stream_per_chan_id_ht; | |
2684 | ||
2685 | rcu_read_lock(); | |
2686 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
2687 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
2688 | ht->match_fct, &channel->key, | |
2689 | &iter.iter, stream, node_channel_id.node) { | |
f2ad556d MD |
2690 | /* |
2691 | * Protect against teardown with mutex. | |
2692 | */ | |
2693 | pthread_mutex_lock(&stream->lock); | |
2694 | if (cds_lfht_is_node_deleted(&stream->node.node)) { | |
2695 | goto next; | |
2696 | } | |
d8ef542d MD |
2697 | switch (consumer_data.type) { |
2698 | case LTTNG_CONSUMER_KERNEL: | |
2699 | break; | |
2700 | case LTTNG_CONSUMER32_UST: | |
2701 | case LTTNG_CONSUMER64_UST: | |
b4a650f3 DG |
2702 | if (stream->metadata_flag) { |
2703 | /* Safe and protected by the stream lock. */ | |
2704 | lttng_ustconsumer_close_metadata(stream->chan); | |
2705 | } else { | |
2706 | /* | |
2707 | * Note: a mutex is taken internally within | |
2708 | * liblttng-ust-ctl to protect timer wakeup_fd | |
2709 | * use from concurrent close. | |
2710 | */ | |
2711 | lttng_ustconsumer_close_stream_wakeup(stream); | |
2712 | } | |
d8ef542d MD |
2713 | break; |
2714 | default: | |
2715 | ERR("Unknown consumer_data type"); | |
2716 | assert(0); | |
2717 | } | |
f2ad556d MD |
2718 | next: |
2719 | pthread_mutex_unlock(&stream->lock); | |
d8ef542d MD |
2720 | } |
2721 | rcu_read_unlock(); | |
2722 | } | |
2723 | ||
2724 | static void destroy_channel_ht(struct lttng_ht *ht) | |
2725 | { | |
2726 | struct lttng_ht_iter iter; | |
2727 | struct lttng_consumer_channel *channel; | |
2728 | int ret; | |
2729 | ||
2730 | if (ht == NULL) { | |
2731 | return; | |
2732 | } | |
2733 | ||
2734 | rcu_read_lock(); | |
2735 | cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) { | |
2736 | ret = lttng_ht_del(ht, &iter); | |
2737 | assert(ret != 0); | |
2738 | } | |
2739 | rcu_read_unlock(); | |
2740 | ||
2741 | lttng_ht_destroy(ht); | |
2742 | } | |
2743 | ||
2744 | /* | |
2745 | * This thread polls the channel fds to detect when they are being | |
2746 | * closed. It closes all related streams if the channel is detected as | |
2747 | * closed. It is currently only used as a shim layer for UST because the | |
2748 | * consumerd needs to keep the per-stream wakeup end of pipes open for | |
2749 | * periodical flush. | |
2750 | */ | |
2751 | void *consumer_thread_channel_poll(void *data) | |
2752 | { | |
1fc79fb4 | 2753 | int ret, i, pollfd, err = -1; |
d8ef542d MD |
2754 | uint32_t revents, nb_fd; |
2755 | struct lttng_consumer_channel *chan = NULL; | |
2756 | struct lttng_ht_iter iter; | |
2757 | struct lttng_ht_node_u64 *node; | |
2758 | struct lttng_poll_event events; | |
2759 | struct lttng_consumer_local_data *ctx = data; | |
2760 | struct lttng_ht *channel_ht; | |
2761 | ||
2762 | rcu_register_thread(); | |
2763 | ||
1fc79fb4 MD |
2764 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL); |
2765 | ||
2d57de81 MD |
2766 | if (testpoint(consumerd_thread_channel)) { |
2767 | goto error_testpoint; | |
2768 | } | |
2769 | ||
9ce5646a MD |
2770 | health_code_update(); |
2771 | ||
d8ef542d MD |
2772 | channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
2773 | if (!channel_ht) { | |
2774 | /* ENOMEM at this point. Better to bail out. */ | |
2775 | goto end_ht; | |
2776 | } | |
2777 | ||
2778 | DBG("Thread channel poll started"); | |
2779 | ||
2780 | /* Size is set to 1 for the consumer_channel pipe */ | |
2781 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2782 | if (ret < 0) { | |
2783 | ERR("Poll set creation failed"); | |
2784 | goto end_poll; | |
2785 | } | |
2786 | ||
2787 | ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN); | |
2788 | if (ret < 0) { | |
2789 | goto end; | |
2790 | } | |
2791 | ||
2792 | /* Main loop */ | |
2793 | DBG("Channel main loop started"); | |
2794 | ||
2795 | while (1) { | |
d8ef542d | 2796 | restart: |
65da3cc7 MD |
2797 | health_code_update(); |
2798 | DBG("Channel poll wait"); | |
9ce5646a | 2799 | health_poll_entry(); |
d8ef542d | 2800 | ret = lttng_poll_wait(&events, -1); |
65da3cc7 MD |
2801 | DBG("Channel poll return from wait with %d fd(s)", |
2802 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2803 | health_poll_exit(); |
bfa78590 | 2804 | DBG("Channel event caught in thread"); |
d8ef542d MD |
2805 | if (ret < 0) { |
2806 | if (errno == EINTR) { | |
bfa78590 | 2807 | ERR("Poll EINTR caught"); |
d8ef542d MD |
2808 | goto restart; |
2809 | } | |
0a7555a7 MD |
2810 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2811 | err = 0; /* All is OK */ | |
2812 | } | |
d8ef542d MD |
2813 | goto end; |
2814 | } | |
2815 | ||
2816 | nb_fd = ret; | |
2817 | ||
2818 | /* From here, the event is a channel wait fd */ | |
2819 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2820 | health_code_update(); |
2821 | ||
d8ef542d MD |
2822 | revents = LTTNG_POLL_GETEV(&events, i); |
2823 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2824 | ||
d8ef542d | 2825 | if (!revents) { |
e0077699 | 2826 | /* No activity for this FD (poll implementation). */ |
d8ef542d MD |
2827 | continue; |
2828 | } | |
e0077699 | 2829 | |
d8ef542d | 2830 | if (pollfd == ctx->consumer_channel_pipe[0]) { |
584fc280 | 2831 | if (revents & LPOLLIN) { |
d8ef542d | 2832 | enum consumer_channel_action action; |
a0cbdd2e | 2833 | uint64_t key; |
d8ef542d | 2834 | |
a0cbdd2e | 2835 | ret = read_channel_pipe(ctx, &chan, &key, &action); |
d8ef542d | 2836 | if (ret <= 0) { |
584fc280 MD |
2837 | if (ret < 0) { |
2838 | ERR("Error reading channel pipe"); | |
2839 | } | |
2840 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
d8ef542d MD |
2841 | continue; |
2842 | } | |
2843 | ||
2844 | switch (action) { | |
2845 | case CONSUMER_CHANNEL_ADD: | |
2846 | DBG("Adding channel %d to poll set", | |
2847 | chan->wait_fd); | |
2848 | ||
2849 | lttng_ht_node_init_u64(&chan->wait_fd_node, | |
2850 | chan->wait_fd); | |
c7260a81 | 2851 | rcu_read_lock(); |
d8ef542d MD |
2852 | lttng_ht_add_unique_u64(channel_ht, |
2853 | &chan->wait_fd_node); | |
c7260a81 | 2854 | rcu_read_unlock(); |
d8ef542d MD |
2855 | /* Add channel to the global poll events list */ |
2856 | lttng_poll_add(&events, chan->wait_fd, | |
584fc280 | 2857 | LPOLLERR | LPOLLHUP); |
d8ef542d | 2858 | break; |
a0cbdd2e MD |
2859 | case CONSUMER_CHANNEL_DEL: |
2860 | { | |
b4a650f3 DG |
2861 | /* |
2862 | * This command should never be called if the channel | |
2863 | * has streams monitored by either the data or metadata | |
2864 | * thread. The consumer only notify this thread with a | |
2865 | * channel del. command if it receives a destroy | |
2866 | * channel command from the session daemon that send it | |
2867 | * if a command prior to the GET_CHANNEL failed. | |
2868 | */ | |
2869 | ||
c7260a81 | 2870 | rcu_read_lock(); |
a0cbdd2e MD |
2871 | chan = consumer_find_channel(key); |
2872 | if (!chan) { | |
c7260a81 | 2873 | rcu_read_unlock(); |
a0cbdd2e MD |
2874 | ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key); |
2875 | break; | |
2876 | } | |
2877 | lttng_poll_del(&events, chan->wait_fd); | |
f623cc0b | 2878 | iter.iter.node = &chan->wait_fd_node.node; |
a0cbdd2e MD |
2879 | ret = lttng_ht_del(channel_ht, &iter); |
2880 | assert(ret == 0); | |
a0cbdd2e | 2881 | |
f2a444f1 DG |
2882 | switch (consumer_data.type) { |
2883 | case LTTNG_CONSUMER_KERNEL: | |
2884 | break; | |
2885 | case LTTNG_CONSUMER32_UST: | |
2886 | case LTTNG_CONSUMER64_UST: | |
212d67a2 DG |
2887 | health_code_update(); |
2888 | /* Destroy streams that might have been left in the stream list. */ | |
2889 | clean_channel_stream_list(chan); | |
f2a444f1 DG |
2890 | break; |
2891 | default: | |
2892 | ERR("Unknown consumer_data type"); | |
2893 | assert(0); | |
2894 | } | |
2895 | ||
a0cbdd2e MD |
2896 | /* |
2897 | * Release our own refcount. Force channel deletion even if | |
2898 | * streams were not initialized. | |
2899 | */ | |
2900 | if (!uatomic_sub_return(&chan->refcount, 1)) { | |
2901 | consumer_del_channel(chan); | |
2902 | } | |
c7260a81 | 2903 | rcu_read_unlock(); |
a0cbdd2e MD |
2904 | goto restart; |
2905 | } | |
d8ef542d MD |
2906 | case CONSUMER_CHANNEL_QUIT: |
2907 | /* | |
2908 | * Remove the pipe from the poll set and continue the loop | |
2909 | * since their might be data to consume. | |
2910 | */ | |
2911 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
2912 | continue; | |
2913 | default: | |
2914 | ERR("Unknown action"); | |
2915 | break; | |
2916 | } | |
584fc280 MD |
2917 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
2918 | DBG("Channel thread pipe hung up"); | |
2919 | /* | |
2920 | * Remove the pipe from the poll set and continue the loop | |
2921 | * since their might be data to consume. | |
2922 | */ | |
2923 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
2924 | continue; | |
2925 | } else { | |
2926 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
2927 | goto end; | |
d8ef542d MD |
2928 | } |
2929 | ||
2930 | /* Handle other stream */ | |
2931 | continue; | |
2932 | } | |
2933 | ||
2934 | rcu_read_lock(); | |
2935 | { | |
2936 | uint64_t tmp_id = (uint64_t) pollfd; | |
2937 | ||
2938 | lttng_ht_lookup(channel_ht, &tmp_id, &iter); | |
2939 | } | |
2940 | node = lttng_ht_iter_get_node_u64(&iter); | |
2941 | assert(node); | |
2942 | ||
2943 | chan = caa_container_of(node, struct lttng_consumer_channel, | |
2944 | wait_fd_node); | |
2945 | ||
2946 | /* Check for error event */ | |
2947 | if (revents & (LPOLLERR | LPOLLHUP)) { | |
2948 | DBG("Channel fd %d is hup|err.", pollfd); | |
2949 | ||
2950 | lttng_poll_del(&events, chan->wait_fd); | |
2951 | ret = lttng_ht_del(channel_ht, &iter); | |
2952 | assert(ret == 0); | |
b4a650f3 DG |
2953 | |
2954 | /* | |
2955 | * This will close the wait fd for each stream associated to | |
2956 | * this channel AND monitored by the data/metadata thread thus | |
2957 | * will be clean by the right thread. | |
2958 | */ | |
d8ef542d | 2959 | consumer_close_channel_streams(chan); |
f2ad556d MD |
2960 | |
2961 | /* Release our own refcount */ | |
2962 | if (!uatomic_sub_return(&chan->refcount, 1) | |
2963 | && !uatomic_read(&chan->nb_init_stream_left)) { | |
2964 | consumer_del_channel(chan); | |
2965 | } | |
584fc280 MD |
2966 | } else { |
2967 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
2968 | rcu_read_unlock(); | |
2969 | goto end; | |
d8ef542d MD |
2970 | } |
2971 | ||
2972 | /* Release RCU lock for the channel looked up */ | |
2973 | rcu_read_unlock(); | |
2974 | } | |
2975 | } | |
2976 | ||
1fc79fb4 MD |
2977 | /* All is OK */ |
2978 | err = 0; | |
d8ef542d MD |
2979 | end: |
2980 | lttng_poll_clean(&events); | |
2981 | end_poll: | |
2982 | destroy_channel_ht(channel_ht); | |
2983 | end_ht: | |
2d57de81 | 2984 | error_testpoint: |
d8ef542d | 2985 | DBG("Channel poll thread exiting"); |
1fc79fb4 MD |
2986 | if (err) { |
2987 | health_error(); | |
2988 | ERR("Health error occurred in %s", __func__); | |
2989 | } | |
2990 | health_unregister(health_consumerd); | |
d8ef542d MD |
2991 | rcu_unregister_thread(); |
2992 | return NULL; | |
2993 | } | |
2994 | ||
331744e3 JD |
2995 | static int set_metadata_socket(struct lttng_consumer_local_data *ctx, |
2996 | struct pollfd *sockpoll, int client_socket) | |
2997 | { | |
2998 | int ret; | |
2999 | ||
3000 | assert(ctx); | |
3001 | assert(sockpoll); | |
3002 | ||
84382d49 MD |
3003 | ret = lttng_consumer_poll_socket(sockpoll); |
3004 | if (ret) { | |
331744e3 JD |
3005 | goto error; |
3006 | } | |
3007 | DBG("Metadata connection on client_socket"); | |
3008 | ||
3009 | /* Blocking call, waiting for transmission */ | |
3010 | ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket); | |
3011 | if (ctx->consumer_metadata_socket < 0) { | |
3012 | WARN("On accept metadata"); | |
3013 | ret = -1; | |
3014 | goto error; | |
3015 | } | |
3016 | ret = 0; | |
3017 | ||
3018 | error: | |
3019 | return ret; | |
3020 | } | |
3021 | ||
3bd1e081 MD |
3022 | /* |
3023 | * This thread listens on the consumerd socket and receives the file | |
3024 | * descriptors from the session daemon. | |
3025 | */ | |
7d980def | 3026 | void *consumer_thread_sessiond_poll(void *data) |
3bd1e081 | 3027 | { |
1fc79fb4 | 3028 | int sock = -1, client_socket, ret, err = -1; |
3bd1e081 MD |
3029 | /* |
3030 | * structure to poll for incoming data on communication socket avoids | |
3031 | * making blocking sockets. | |
3032 | */ | |
3033 | struct pollfd consumer_sockpoll[2]; | |
3034 | struct lttng_consumer_local_data *ctx = data; | |
3035 | ||
e7b994a3 DG |
3036 | rcu_register_thread(); |
3037 | ||
1fc79fb4 MD |
3038 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND); |
3039 | ||
2d57de81 MD |
3040 | if (testpoint(consumerd_thread_sessiond)) { |
3041 | goto error_testpoint; | |
3042 | } | |
3043 | ||
9ce5646a MD |
3044 | health_code_update(); |
3045 | ||
3bd1e081 MD |
3046 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
3047 | unlink(ctx->consumer_command_sock_path); | |
3048 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); | |
3049 | if (client_socket < 0) { | |
3050 | ERR("Cannot create command socket"); | |
3051 | goto end; | |
3052 | } | |
3053 | ||
3054 | ret = lttcomm_listen_unix_sock(client_socket); | |
3055 | if (ret < 0) { | |
3056 | goto end; | |
3057 | } | |
3058 | ||
32258573 | 3059 | DBG("Sending ready command to lttng-sessiond"); |
f73fabfd | 3060 | ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); |
3bd1e081 MD |
3061 | /* return < 0 on error, but == 0 is not fatal */ |
3062 | if (ret < 0) { | |
32258573 | 3063 | ERR("Error sending ready command to lttng-sessiond"); |
3bd1e081 MD |
3064 | goto end; |
3065 | } | |
3066 | ||
3bd1e081 MD |
3067 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ |
3068 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; | |
3069 | consumer_sockpoll[0].events = POLLIN | POLLPRI; | |
3070 | consumer_sockpoll[1].fd = client_socket; | |
3071 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3072 | ||
84382d49 MD |
3073 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3074 | if (ret) { | |
3075 | if (ret > 0) { | |
3076 | /* should exit */ | |
3077 | err = 0; | |
3078 | } | |
3bd1e081 MD |
3079 | goto end; |
3080 | } | |
3081 | DBG("Connection on client_socket"); | |
3082 | ||
3083 | /* Blocking call, waiting for transmission */ | |
3084 | sock = lttcomm_accept_unix_sock(client_socket); | |
534d2592 | 3085 | if (sock < 0) { |
3bd1e081 MD |
3086 | WARN("On accept"); |
3087 | goto end; | |
3088 | } | |
3bd1e081 | 3089 | |
331744e3 JD |
3090 | /* |
3091 | * Setup metadata socket which is the second socket connection on the | |
3092 | * command unix socket. | |
3093 | */ | |
3094 | ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket); | |
84382d49 MD |
3095 | if (ret) { |
3096 | if (ret > 0) { | |
3097 | /* should exit */ | |
3098 | err = 0; | |
3099 | } | |
331744e3 JD |
3100 | goto end; |
3101 | } | |
3102 | ||
d96f09c6 DG |
3103 | /* This socket is not useful anymore. */ |
3104 | ret = close(client_socket); | |
3105 | if (ret < 0) { | |
3106 | PERROR("close client_socket"); | |
3107 | } | |
3108 | client_socket = -1; | |
3109 | ||
3bd1e081 MD |
3110 | /* update the polling structure to poll on the established socket */ |
3111 | consumer_sockpoll[1].fd = sock; | |
3112 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3113 | ||
3114 | while (1) { | |
9ce5646a MD |
3115 | health_code_update(); |
3116 | ||
3117 | health_poll_entry(); | |
3118 | ret = lttng_consumer_poll_socket(consumer_sockpoll); | |
3119 | health_poll_exit(); | |
84382d49 MD |
3120 | if (ret) { |
3121 | if (ret > 0) { | |
3122 | /* should exit */ | |
3123 | err = 0; | |
3124 | } | |
3bd1e081 MD |
3125 | goto end; |
3126 | } | |
3127 | DBG("Incoming command on sock"); | |
3128 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
4cbc1a04 DG |
3129 | if (ret <= 0) { |
3130 | /* | |
3131 | * This could simply be a session daemon quitting. Don't output | |
3132 | * ERR() here. | |
3133 | */ | |
3134 | DBG("Communication interrupted on command socket"); | |
41ba6035 | 3135 | err = 0; |
3bd1e081 MD |
3136 | goto end; |
3137 | } | |
3138 | if (consumer_quit) { | |
3139 | DBG("consumer_thread_receive_fds received quit from signal"); | |
1fc79fb4 | 3140 | err = 0; /* All is OK */ |
3bd1e081 MD |
3141 | goto end; |
3142 | } | |
ffe60014 | 3143 | DBG("received command on sock"); |
3bd1e081 | 3144 | } |
1fc79fb4 MD |
3145 | /* All is OK */ |
3146 | err = 0; | |
3147 | ||
3bd1e081 | 3148 | end: |
ffe60014 | 3149 | DBG("Consumer thread sessiond poll exiting"); |
3bd1e081 | 3150 | |
d88aee68 DG |
3151 | /* |
3152 | * Close metadata streams since the producer is the session daemon which | |
3153 | * just died. | |
3154 | * | |
3155 | * NOTE: for now, this only applies to the UST tracer. | |
3156 | */ | |
6d574024 | 3157 | lttng_consumer_close_all_metadata(); |
d88aee68 | 3158 | |
3bd1e081 MD |
3159 | /* |
3160 | * when all fds have hung up, the polling thread | |
3161 | * can exit cleanly | |
3162 | */ | |
3163 | consumer_quit = 1; | |
3164 | ||
04fdd819 | 3165 | /* |
c869f647 | 3166 | * Notify the data poll thread to poll back again and test the |
8994307f | 3167 | * consumer_quit state that we just set so to quit gracefully. |
04fdd819 | 3168 | */ |
acdb9057 | 3169 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
c869f647 | 3170 | |
a0cbdd2e | 3171 | notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT); |
d8ef542d | 3172 | |
5c635c72 MD |
3173 | notify_health_quit_pipe(health_quit_pipe); |
3174 | ||
d96f09c6 DG |
3175 | /* Cleaning up possibly open sockets. */ |
3176 | if (sock >= 0) { | |
3177 | ret = close(sock); | |
3178 | if (ret < 0) { | |
3179 | PERROR("close sock sessiond poll"); | |
3180 | } | |
3181 | } | |
3182 | if (client_socket >= 0) { | |
38476d24 | 3183 | ret = close(client_socket); |
d96f09c6 DG |
3184 | if (ret < 0) { |
3185 | PERROR("close client_socket sessiond poll"); | |
3186 | } | |
3187 | } | |
3188 | ||
2d57de81 | 3189 | error_testpoint: |
1fc79fb4 MD |
3190 | if (err) { |
3191 | health_error(); | |
3192 | ERR("Health error occurred in %s", __func__); | |
3193 | } | |
3194 | health_unregister(health_consumerd); | |
3195 | ||
e7b994a3 | 3196 | rcu_unregister_thread(); |
3bd1e081 MD |
3197 | return NULL; |
3198 | } | |
d41f73b7 | 3199 | |
4078b776 | 3200 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
d41f73b7 MD |
3201 | struct lttng_consumer_local_data *ctx) |
3202 | { | |
74251bb8 DG |
3203 | ssize_t ret; |
3204 | ||
3205 | pthread_mutex_lock(&stream->lock); | |
94d49140 JD |
3206 | if (stream->metadata_flag) { |
3207 | pthread_mutex_lock(&stream->metadata_rdv_lock); | |
3208 | } | |
74251bb8 | 3209 | |
d41f73b7 MD |
3210 | switch (consumer_data.type) { |
3211 | case LTTNG_CONSUMER_KERNEL: | |
74251bb8 DG |
3212 | ret = lttng_kconsumer_read_subbuffer(stream, ctx); |
3213 | break; | |
7753dea8 MD |
3214 | case LTTNG_CONSUMER32_UST: |
3215 | case LTTNG_CONSUMER64_UST: | |
74251bb8 DG |
3216 | ret = lttng_ustconsumer_read_subbuffer(stream, ctx); |
3217 | break; | |
d41f73b7 MD |
3218 | default: |
3219 | ERR("Unknown consumer_data type"); | |
3220 | assert(0); | |
74251bb8 DG |
3221 | ret = -ENOSYS; |
3222 | break; | |
d41f73b7 | 3223 | } |
74251bb8 | 3224 | |
94d49140 JD |
3225 | if (stream->metadata_flag) { |
3226 | pthread_cond_broadcast(&stream->metadata_rdv); | |
3227 | pthread_mutex_unlock(&stream->metadata_rdv_lock); | |
3228 | } | |
74251bb8 DG |
3229 | pthread_mutex_unlock(&stream->lock); |
3230 | return ret; | |
d41f73b7 MD |
3231 | } |
3232 | ||
3233 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
3234 | { | |
3235 | switch (consumer_data.type) { | |
3236 | case LTTNG_CONSUMER_KERNEL: | |
3237 | return lttng_kconsumer_on_recv_stream(stream); | |
7753dea8 MD |
3238 | case LTTNG_CONSUMER32_UST: |
3239 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
3240 | return lttng_ustconsumer_on_recv_stream(stream); |
3241 | default: | |
3242 | ERR("Unknown consumer_data type"); | |
3243 | assert(0); | |
3244 | return -ENOSYS; | |
3245 | } | |
3246 | } | |
e4421fec DG |
3247 | |
3248 | /* | |
3249 | * Allocate and set consumer data hash tables. | |
3250 | */ | |
282dadbc | 3251 | int lttng_consumer_init(void) |
e4421fec | 3252 | { |
d88aee68 | 3253 | consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3254 | if (!consumer_data.channel_ht) { |
3255 | goto error; | |
3256 | } | |
3257 | ||
d88aee68 | 3258 | consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3259 | if (!consumer_data.relayd_ht) { |
3260 | goto error; | |
3261 | } | |
3262 | ||
d88aee68 | 3263 | consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3264 | if (!consumer_data.stream_list_ht) { |
3265 | goto error; | |
3266 | } | |
3267 | ||
d8ef542d | 3268 | consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3269 | if (!consumer_data.stream_per_chan_id_ht) { |
3270 | goto error; | |
3271 | } | |
3272 | ||
3273 | data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3274 | if (!data_ht) { | |
3275 | goto error; | |
3276 | } | |
3277 | ||
3278 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3279 | if (!metadata_ht) { | |
3280 | goto error; | |
3281 | } | |
3282 | ||
3283 | return 0; | |
3284 | ||
3285 | error: | |
3286 | return -1; | |
e4421fec | 3287 | } |
7735ef9e DG |
3288 | |
3289 | /* | |
3290 | * Process the ADD_RELAYD command receive by a consumer. | |
3291 | * | |
3292 | * This will create a relayd socket pair and add it to the relayd hash table. | |
3293 | * The caller MUST acquire a RCU read side lock before calling it. | |
3294 | */ | |
da009f2c | 3295 | int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type, |
7735ef9e | 3296 | struct lttng_consumer_local_data *ctx, int sock, |
6151a90f | 3297 | struct pollfd *consumer_sockpoll, |
d3e2ba59 JD |
3298 | struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id, |
3299 | uint64_t relayd_session_id) | |
7735ef9e | 3300 | { |
cd2b09ed | 3301 | int fd = -1, ret = -1, relayd_created = 0; |
0c759fc9 | 3302 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
d4298c99 | 3303 | struct consumer_relayd_sock_pair *relayd = NULL; |
7735ef9e | 3304 | |
6151a90f JD |
3305 | assert(ctx); |
3306 | assert(relayd_sock); | |
3307 | ||
da009f2c | 3308 | DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx); |
7735ef9e DG |
3309 | |
3310 | /* Get relayd reference if exists. */ | |
3311 | relayd = consumer_find_relayd(net_seq_idx); | |
3312 | if (relayd == NULL) { | |
da009f2c | 3313 | assert(sock_type == LTTNG_STREAM_CONTROL); |
7735ef9e DG |
3314 | /* Not found. Allocate one. */ |
3315 | relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); | |
3316 | if (relayd == NULL) { | |
0d08d75e | 3317 | ret = -ENOMEM; |
618a6a28 MD |
3318 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
3319 | goto error; | |
0d08d75e | 3320 | } else { |
30319bcb | 3321 | relayd->sessiond_session_id = sessiond_id; |
0d08d75e | 3322 | relayd_created = 1; |
7735ef9e | 3323 | } |
0d08d75e DG |
3324 | |
3325 | /* | |
3326 | * This code path MUST continue to the consumer send status message to | |
3327 | * we can notify the session daemon and continue our work without | |
3328 | * killing everything. | |
3329 | */ | |
da009f2c MD |
3330 | } else { |
3331 | /* | |
3332 | * relayd key should never be found for control socket. | |
3333 | */ | |
3334 | assert(sock_type != LTTNG_STREAM_CONTROL); | |
0d08d75e DG |
3335 | } |
3336 | ||
3337 | /* First send a status message before receiving the fds. */ | |
0c759fc9 | 3338 | ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS); |
618a6a28 | 3339 | if (ret < 0) { |
0d08d75e | 3340 | /* Somehow, the session daemon is not responding anymore. */ |
618a6a28 MD |
3341 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); |
3342 | goto error_nosignal; | |
7735ef9e DG |
3343 | } |
3344 | ||
3345 | /* Poll on consumer socket. */ | |
84382d49 MD |
3346 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3347 | if (ret) { | |
3348 | /* Needing to exit in the middle of a command: error. */ | |
0d08d75e | 3349 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
7735ef9e | 3350 | ret = -EINTR; |
618a6a28 | 3351 | goto error_nosignal; |
7735ef9e DG |
3352 | } |
3353 | ||
3354 | /* Get relayd socket from session daemon */ | |
3355 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); | |
3356 | if (ret != sizeof(fd)) { | |
7735ef9e | 3357 | ret = -1; |
4028eeb9 | 3358 | fd = -1; /* Just in case it gets set with an invalid value. */ |
0d08d75e DG |
3359 | |
3360 | /* | |
3361 | * Failing to receive FDs might indicate a major problem such as | |
3362 | * reaching a fd limit during the receive where the kernel returns a | |
3363 | * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we | |
3364 | * don't take any chances and stop everything. | |
3365 | * | |
3366 | * XXX: Feature request #558 will fix that and avoid this possible | |
3367 | * issue when reaching the fd limit. | |
3368 | */ | |
3369 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); | |
618a6a28 | 3370 | ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD; |
f50f23d9 DG |
3371 | goto error; |
3372 | } | |
3373 | ||
7735ef9e DG |
3374 | /* Copy socket information and received FD */ |
3375 | switch (sock_type) { | |
3376 | case LTTNG_STREAM_CONTROL: | |
3377 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3378 | lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock); |
3379 | ret = lttcomm_create_sock(&relayd->control_sock.sock); | |
4028eeb9 | 3380 | /* Handle create_sock error. */ |
f66c074c | 3381 | if (ret < 0) { |
618a6a28 | 3382 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3383 | goto error; |
f66c074c | 3384 | } |
da009f2c MD |
3385 | /* |
3386 | * Close the socket created internally by | |
3387 | * lttcomm_create_sock, so we can replace it by the one | |
3388 | * received from sessiond. | |
3389 | */ | |
3390 | if (close(relayd->control_sock.sock.fd)) { | |
3391 | PERROR("close"); | |
3392 | } | |
7735ef9e DG |
3393 | |
3394 | /* Assign new file descriptor */ | |
6151a90f | 3395 | relayd->control_sock.sock.fd = fd; |
4b29f1ce | 3396 | fd = -1; /* For error path */ |
6151a90f JD |
3397 | /* Assign version values. */ |
3398 | relayd->control_sock.major = relayd_sock->major; | |
3399 | relayd->control_sock.minor = relayd_sock->minor; | |
c5b6f4f0 | 3400 | |
d3e2ba59 | 3401 | relayd->relayd_session_id = relayd_session_id; |
c5b6f4f0 | 3402 | |
7735ef9e DG |
3403 | break; |
3404 | case LTTNG_STREAM_DATA: | |
3405 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3406 | lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock); |
3407 | ret = lttcomm_create_sock(&relayd->data_sock.sock); | |
4028eeb9 | 3408 | /* Handle create_sock error. */ |
f66c074c | 3409 | if (ret < 0) { |
618a6a28 | 3410 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3411 | goto error; |
f66c074c | 3412 | } |
da009f2c MD |
3413 | /* |
3414 | * Close the socket created internally by | |
3415 | * lttcomm_create_sock, so we can replace it by the one | |
3416 | * received from sessiond. | |
3417 | */ | |
3418 | if (close(relayd->data_sock.sock.fd)) { | |
3419 | PERROR("close"); | |
3420 | } | |
7735ef9e DG |
3421 | |
3422 | /* Assign new file descriptor */ | |
6151a90f | 3423 | relayd->data_sock.sock.fd = fd; |
4b29f1ce | 3424 | fd = -1; /* for eventual error paths */ |
6151a90f JD |
3425 | /* Assign version values. */ |
3426 | relayd->data_sock.major = relayd_sock->major; | |
3427 | relayd->data_sock.minor = relayd_sock->minor; | |
7735ef9e DG |
3428 | break; |
3429 | default: | |
3430 | ERR("Unknown relayd socket type (%d)", sock_type); | |
59e71485 | 3431 | ret = -1; |
618a6a28 | 3432 | ret_code = LTTCOMM_CONSUMERD_FATAL; |
7735ef9e DG |
3433 | goto error; |
3434 | } | |
3435 | ||
d88aee68 | 3436 | DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)", |
7735ef9e DG |
3437 | sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", |
3438 | relayd->net_seq_idx, fd); | |
3439 | ||
618a6a28 MD |
3440 | /* We successfully added the socket. Send status back. */ |
3441 | ret = consumer_send_status_msg(sock, ret_code); | |
3442 | if (ret < 0) { | |
3443 | /* Somehow, the session daemon is not responding anymore. */ | |
3444 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3445 | goto error_nosignal; | |
3446 | } | |
3447 | ||
7735ef9e DG |
3448 | /* |
3449 | * Add relayd socket pair to consumer data hashtable. If object already | |
3450 | * exists or on error, the function gracefully returns. | |
3451 | */ | |
d09e1200 | 3452 | add_relayd(relayd); |
7735ef9e DG |
3453 | |
3454 | /* All good! */ | |
4028eeb9 | 3455 | return 0; |
7735ef9e DG |
3456 | |
3457 | error: | |
618a6a28 MD |
3458 | if (consumer_send_status_msg(sock, ret_code) < 0) { |
3459 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3460 | } | |
3461 | ||
3462 | error_nosignal: | |
4028eeb9 DG |
3463 | /* Close received socket if valid. */ |
3464 | if (fd >= 0) { | |
3465 | if (close(fd)) { | |
3466 | PERROR("close received socket"); | |
3467 | } | |
3468 | } | |
cd2b09ed DG |
3469 | |
3470 | if (relayd_created) { | |
cd2b09ed DG |
3471 | free(relayd); |
3472 | } | |
3473 | ||
7735ef9e DG |
3474 | return ret; |
3475 | } | |
ca22feea | 3476 | |
4e9a4686 DG |
3477 | /* |
3478 | * Try to lock the stream mutex. | |
3479 | * | |
3480 | * On success, 1 is returned else 0 indicating that the mutex is NOT lock. | |
3481 | */ | |
3482 | static int stream_try_lock(struct lttng_consumer_stream *stream) | |
3483 | { | |
3484 | int ret; | |
3485 | ||
3486 | assert(stream); | |
3487 | ||
3488 | /* | |
3489 | * Try to lock the stream mutex. On failure, we know that the stream is | |
3490 | * being used else where hence there is data still being extracted. | |
3491 | */ | |
3492 | ret = pthread_mutex_trylock(&stream->lock); | |
3493 | if (ret) { | |
3494 | /* For both EBUSY and EINVAL error, the mutex is NOT locked. */ | |
3495 | ret = 0; | |
3496 | goto end; | |
3497 | } | |
3498 | ||
3499 | ret = 1; | |
3500 | ||
3501 | end: | |
3502 | return ret; | |
3503 | } | |
3504 | ||
f7079f67 DG |
3505 | /* |
3506 | * Search for a relayd associated to the session id and return the reference. | |
3507 | * | |
3508 | * A rcu read side lock MUST be acquire before calling this function and locked | |
3509 | * until the relayd object is no longer necessary. | |
3510 | */ | |
3511 | static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id) | |
3512 | { | |
3513 | struct lttng_ht_iter iter; | |
f7079f67 | 3514 | struct consumer_relayd_sock_pair *relayd = NULL; |
f7079f67 DG |
3515 | |
3516 | /* Iterate over all relayd since they are indexed by net_seq_idx. */ | |
3517 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
3518 | node.node) { | |
18261bd1 DG |
3519 | /* |
3520 | * Check by sessiond id which is unique here where the relayd session | |
3521 | * id might not be when having multiple relayd. | |
3522 | */ | |
3523 | if (relayd->sessiond_session_id == id) { | |
f7079f67 | 3524 | /* Found the relayd. There can be only one per id. */ |
18261bd1 | 3525 | goto found; |
f7079f67 DG |
3526 | } |
3527 | } | |
3528 | ||
18261bd1 DG |
3529 | return NULL; |
3530 | ||
3531 | found: | |
f7079f67 DG |
3532 | return relayd; |
3533 | } | |
3534 | ||
ca22feea DG |
3535 | /* |
3536 | * Check if for a given session id there is still data needed to be extract | |
3537 | * from the buffers. | |
3538 | * | |
6d805429 | 3539 | * Return 1 if data is pending or else 0 meaning ready to be read. |
ca22feea | 3540 | */ |
6d805429 | 3541 | int consumer_data_pending(uint64_t id) |
ca22feea DG |
3542 | { |
3543 | int ret; | |
3544 | struct lttng_ht_iter iter; | |
3545 | struct lttng_ht *ht; | |
3546 | struct lttng_consumer_stream *stream; | |
f7079f67 | 3547 | struct consumer_relayd_sock_pair *relayd = NULL; |
6d805429 | 3548 | int (*data_pending)(struct lttng_consumer_stream *); |
ca22feea | 3549 | |
6d805429 | 3550 | DBG("Consumer data pending command on session id %" PRIu64, id); |
ca22feea | 3551 | |
6f6eda74 | 3552 | rcu_read_lock(); |
ca22feea DG |
3553 | pthread_mutex_lock(&consumer_data.lock); |
3554 | ||
3555 | switch (consumer_data.type) { | |
3556 | case LTTNG_CONSUMER_KERNEL: | |
6d805429 | 3557 | data_pending = lttng_kconsumer_data_pending; |
ca22feea DG |
3558 | break; |
3559 | case LTTNG_CONSUMER32_UST: | |
3560 | case LTTNG_CONSUMER64_UST: | |
6d805429 | 3561 | data_pending = lttng_ustconsumer_data_pending; |
ca22feea DG |
3562 | break; |
3563 | default: | |
3564 | ERR("Unknown consumer data type"); | |
3565 | assert(0); | |
3566 | } | |
3567 | ||
3568 | /* Ease our life a bit */ | |
3569 | ht = consumer_data.stream_list_ht; | |
3570 | ||
f7079f67 DG |
3571 | relayd = find_relayd_by_session_id(id); |
3572 | if (relayd) { | |
3573 | /* Send init command for data pending. */ | |
3574 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3575 | ret = relayd_begin_data_pending(&relayd->control_sock, | |
3576 | relayd->relayd_session_id); | |
3577 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3578 | if (ret < 0) { | |
3579 | /* Communication error thus the relayd so no data pending. */ | |
3580 | goto data_not_pending; | |
3581 | } | |
3582 | } | |
3583 | ||
c8f59ee5 | 3584 | cds_lfht_for_each_entry_duplicate(ht->ht, |
d88aee68 DG |
3585 | ht->hash_fct(&id, lttng_ht_seed), |
3586 | ht->match_fct, &id, | |
ca22feea | 3587 | &iter.iter, stream, node_session_id.node) { |
4e9a4686 DG |
3588 | /* If this call fails, the stream is being used hence data pending. */ |
3589 | ret = stream_try_lock(stream); | |
3590 | if (!ret) { | |
f7079f67 | 3591 | goto data_pending; |
ca22feea | 3592 | } |
ca22feea | 3593 | |
4e9a4686 DG |
3594 | /* |
3595 | * A removed node from the hash table indicates that the stream has | |
3596 | * been deleted thus having a guarantee that the buffers are closed | |
3597 | * on the consumer side. However, data can still be transmitted | |
3598 | * over the network so don't skip the relayd check. | |
3599 | */ | |
3600 | ret = cds_lfht_is_node_deleted(&stream->node.node); | |
3601 | if (!ret) { | |
3602 | /* Check the stream if there is data in the buffers. */ | |
6d805429 DG |
3603 | ret = data_pending(stream); |
3604 | if (ret == 1) { | |
4e9a4686 | 3605 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3606 | goto data_pending; |
4e9a4686 DG |
3607 | } |
3608 | } | |
3609 | ||
3610 | /* Relayd check */ | |
f7079f67 | 3611 | if (relayd) { |
c8f59ee5 DG |
3612 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
3613 | if (stream->metadata_flag) { | |
ad7051c0 DG |
3614 | ret = relayd_quiescent_control(&relayd->control_sock, |
3615 | stream->relayd_stream_id); | |
c8f59ee5 | 3616 | } else { |
6d805429 | 3617 | ret = relayd_data_pending(&relayd->control_sock, |
39df6d9f DG |
3618 | stream->relayd_stream_id, |
3619 | stream->next_net_seq_num - 1); | |
c8f59ee5 DG |
3620 | } |
3621 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
6d805429 | 3622 | if (ret == 1) { |
4e9a4686 | 3623 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3624 | goto data_pending; |
c8f59ee5 DG |
3625 | } |
3626 | } | |
4e9a4686 | 3627 | pthread_mutex_unlock(&stream->lock); |
c8f59ee5 | 3628 | } |
ca22feea | 3629 | |
f7079f67 DG |
3630 | if (relayd) { |
3631 | unsigned int is_data_inflight = 0; | |
3632 | ||
3633 | /* Send init command for data pending. */ | |
3634 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3635 | ret = relayd_end_data_pending(&relayd->control_sock, | |
3636 | relayd->relayd_session_id, &is_data_inflight); | |
3637 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
bdd88757 | 3638 | if (ret < 0) { |
f7079f67 DG |
3639 | goto data_not_pending; |
3640 | } | |
bdd88757 DG |
3641 | if (is_data_inflight) { |
3642 | goto data_pending; | |
3643 | } | |
f7079f67 DG |
3644 | } |
3645 | ||
ca22feea | 3646 | /* |
f7079f67 DG |
3647 | * Finding _no_ node in the hash table and no inflight data means that the |
3648 | * stream(s) have been removed thus data is guaranteed to be available for | |
3649 | * analysis from the trace files. | |
ca22feea DG |
3650 | */ |
3651 | ||
f7079f67 | 3652 | data_not_pending: |
ca22feea DG |
3653 | /* Data is available to be read by a viewer. */ |
3654 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3655 | rcu_read_unlock(); |
6d805429 | 3656 | return 0; |
ca22feea | 3657 | |
f7079f67 | 3658 | data_pending: |
ca22feea DG |
3659 | /* Data is still being extracted from buffers. */ |
3660 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3661 | rcu_read_unlock(); |
6d805429 | 3662 | return 1; |
ca22feea | 3663 | } |
f50f23d9 DG |
3664 | |
3665 | /* | |
3666 | * Send a ret code status message to the sessiond daemon. | |
3667 | * | |
3668 | * Return the sendmsg() return value. | |
3669 | */ | |
3670 | int consumer_send_status_msg(int sock, int ret_code) | |
3671 | { | |
3672 | struct lttcomm_consumer_status_msg msg; | |
3673 | ||
53efb85a | 3674 | memset(&msg, 0, sizeof(msg)); |
f50f23d9 DG |
3675 | msg.ret_code = ret_code; |
3676 | ||
3677 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3678 | } | |
ffe60014 DG |
3679 | |
3680 | /* | |
3681 | * Send a channel status message to the sessiond daemon. | |
3682 | * | |
3683 | * Return the sendmsg() return value. | |
3684 | */ | |
3685 | int consumer_send_status_channel(int sock, | |
3686 | struct lttng_consumer_channel *channel) | |
3687 | { | |
3688 | struct lttcomm_consumer_status_channel msg; | |
3689 | ||
3690 | assert(sock >= 0); | |
3691 | ||
53efb85a | 3692 | memset(&msg, 0, sizeof(msg)); |
ffe60014 | 3693 | if (!channel) { |
0c759fc9 | 3694 | msg.ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL; |
ffe60014 | 3695 | } else { |
0c759fc9 | 3696 | msg.ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
ffe60014 DG |
3697 | msg.key = channel->key; |
3698 | msg.stream_count = channel->streams.count; | |
3699 | } | |
3700 | ||
3701 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3702 | } | |
5c786ded | 3703 | |
135347bd MD |
3704 | unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos, |
3705 | unsigned long produced_pos, uint64_t nb_packets_per_stream, | |
3706 | uint64_t max_sb_size) | |
5c786ded | 3707 | { |
135347bd | 3708 | unsigned long start_pos; |
5c786ded | 3709 | |
135347bd MD |
3710 | if (!nb_packets_per_stream) { |
3711 | return consumed_pos; /* Grab everything */ | |
3712 | } | |
3713 | start_pos = produced_pos - offset_align_floor(produced_pos, max_sb_size); | |
3714 | start_pos -= max_sb_size * nb_packets_per_stream; | |
3715 | if ((long) (start_pos - consumed_pos) < 0) { | |
3716 | return consumed_pos; /* Grab everything */ | |
3717 | } | |
3718 | return start_pos; | |
5c786ded | 3719 | } |