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 | ||
6c1c0768 | 20 | #define _LGPL_SOURCE |
3bd1e081 | 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> |
c8fea79c JR |
45 | #include <common/consumer/consumer-timer.h> |
46 | #include <common/consumer/consumer.h> | |
47 | #include <common/consumer/consumer-stream.h> | |
48 | #include <common/consumer/consumer-testpoint.h> | |
49 | #include <common/align.h> | |
3bd1e081 MD |
50 | |
51 | struct lttng_consumer_global_data consumer_data = { | |
3bd1e081 MD |
52 | .stream_count = 0, |
53 | .need_update = 1, | |
54 | .type = LTTNG_CONSUMER_UNKNOWN, | |
55 | }; | |
56 | ||
d8ef542d MD |
57 | enum consumer_channel_action { |
58 | CONSUMER_CHANNEL_ADD, | |
a0cbdd2e | 59 | CONSUMER_CHANNEL_DEL, |
d8ef542d MD |
60 | CONSUMER_CHANNEL_QUIT, |
61 | }; | |
62 | ||
63 | struct consumer_channel_msg { | |
64 | enum consumer_channel_action action; | |
a0cbdd2e MD |
65 | struct lttng_consumer_channel *chan; /* add */ |
66 | uint64_t key; /* del */ | |
d8ef542d MD |
67 | }; |
68 | ||
3bd1e081 MD |
69 | /* |
70 | * Flag to inform the polling thread to quit when all fd hung up. Updated by | |
71 | * the consumer_thread_receive_fds when it notices that all fds has hung up. | |
72 | * Also updated by the signal handler (consumer_should_exit()). Read by the | |
73 | * polling threads. | |
74 | */ | |
a98dae5f | 75 | volatile int consumer_quit; |
3bd1e081 | 76 | |
43c34bc3 | 77 | /* |
43c34bc3 DG |
78 | * Global hash table containing respectively metadata and data streams. The |
79 | * stream element in this ht should only be updated by the metadata poll thread | |
80 | * for the metadata and the data poll thread for the data. | |
81 | */ | |
40dc48e0 DG |
82 | static struct lttng_ht *metadata_ht; |
83 | static struct lttng_ht *data_ht; | |
43c34bc3 | 84 | |
acdb9057 DG |
85 | /* |
86 | * Notify a thread lttng pipe to poll back again. This usually means that some | |
87 | * global state has changed so we just send back the thread in a poll wait | |
88 | * call. | |
89 | */ | |
90 | static void notify_thread_lttng_pipe(struct lttng_pipe *pipe) | |
91 | { | |
92 | struct lttng_consumer_stream *null_stream = NULL; | |
93 | ||
94 | assert(pipe); | |
95 | ||
96 | (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream)); | |
97 | } | |
98 | ||
5c635c72 MD |
99 | static void notify_health_quit_pipe(int *pipe) |
100 | { | |
6cd525e8 | 101 | ssize_t ret; |
5c635c72 | 102 | |
6cd525e8 MD |
103 | ret = lttng_write(pipe[1], "4", 1); |
104 | if (ret < 1) { | |
5c635c72 MD |
105 | PERROR("write consumer health quit"); |
106 | } | |
107 | } | |
108 | ||
d8ef542d MD |
109 | static void notify_channel_pipe(struct lttng_consumer_local_data *ctx, |
110 | struct lttng_consumer_channel *chan, | |
a0cbdd2e | 111 | uint64_t key, |
d8ef542d MD |
112 | enum consumer_channel_action action) |
113 | { | |
114 | struct consumer_channel_msg msg; | |
6cd525e8 | 115 | ssize_t ret; |
d8ef542d | 116 | |
e56251fc DG |
117 | memset(&msg, 0, sizeof(msg)); |
118 | ||
d8ef542d MD |
119 | msg.action = action; |
120 | msg.chan = chan; | |
f21dae48 | 121 | msg.key = key; |
6cd525e8 MD |
122 | ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg)); |
123 | if (ret < sizeof(msg)) { | |
124 | PERROR("notify_channel_pipe write error"); | |
125 | } | |
d8ef542d MD |
126 | } |
127 | ||
a0cbdd2e MD |
128 | void notify_thread_del_channel(struct lttng_consumer_local_data *ctx, |
129 | uint64_t key) | |
130 | { | |
131 | notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL); | |
132 | } | |
133 | ||
d8ef542d MD |
134 | static int read_channel_pipe(struct lttng_consumer_local_data *ctx, |
135 | struct lttng_consumer_channel **chan, | |
a0cbdd2e | 136 | uint64_t *key, |
d8ef542d MD |
137 | enum consumer_channel_action *action) |
138 | { | |
139 | struct consumer_channel_msg msg; | |
6cd525e8 | 140 | ssize_t ret; |
d8ef542d | 141 | |
6cd525e8 MD |
142 | ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg)); |
143 | if (ret < sizeof(msg)) { | |
144 | ret = -1; | |
145 | goto error; | |
d8ef542d | 146 | } |
6cd525e8 MD |
147 | *action = msg.action; |
148 | *chan = msg.chan; | |
149 | *key = msg.key; | |
150 | error: | |
151 | return (int) ret; | |
d8ef542d MD |
152 | } |
153 | ||
212d67a2 DG |
154 | /* |
155 | * Cleanup the stream list of a channel. Those streams are not yet globally | |
156 | * visible | |
157 | */ | |
158 | static void clean_channel_stream_list(struct lttng_consumer_channel *channel) | |
159 | { | |
160 | struct lttng_consumer_stream *stream, *stmp; | |
161 | ||
162 | assert(channel); | |
163 | ||
164 | /* Delete streams that might have been left in the stream list. */ | |
165 | cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head, | |
166 | send_node) { | |
167 | cds_list_del(&stream->send_node); | |
168 | /* | |
169 | * Once a stream is added to this list, the buffers were created so we | |
170 | * have a guarantee that this call will succeed. Setting the monitor | |
171 | * mode to 0 so we don't lock nor try to delete the stream from the | |
172 | * global hash table. | |
173 | */ | |
174 | stream->monitor = 0; | |
175 | consumer_stream_destroy(stream, NULL); | |
176 | } | |
177 | } | |
178 | ||
3bd1e081 MD |
179 | /* |
180 | * Find a stream. The consumer_data.lock must be locked during this | |
181 | * call. | |
182 | */ | |
d88aee68 | 183 | static struct lttng_consumer_stream *find_stream(uint64_t key, |
8389e4f8 | 184 | struct lttng_ht *ht) |
3bd1e081 | 185 | { |
e4421fec | 186 | struct lttng_ht_iter iter; |
d88aee68 | 187 | struct lttng_ht_node_u64 *node; |
e4421fec | 188 | struct lttng_consumer_stream *stream = NULL; |
3bd1e081 | 189 | |
8389e4f8 DG |
190 | assert(ht); |
191 | ||
d88aee68 DG |
192 | /* -1ULL keys are lookup failures */ |
193 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 194 | return NULL; |
7a57cf92 | 195 | } |
e4421fec | 196 | |
6065ceec DG |
197 | rcu_read_lock(); |
198 | ||
d88aee68 DG |
199 | lttng_ht_lookup(ht, &key, &iter); |
200 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
201 | if (node != NULL) { |
202 | stream = caa_container_of(node, struct lttng_consumer_stream, node); | |
3bd1e081 | 203 | } |
e4421fec | 204 | |
6065ceec DG |
205 | rcu_read_unlock(); |
206 | ||
e4421fec | 207 | return stream; |
3bd1e081 MD |
208 | } |
209 | ||
da009f2c | 210 | static void steal_stream_key(uint64_t key, struct lttng_ht *ht) |
7ad0a0cb MD |
211 | { |
212 | struct lttng_consumer_stream *stream; | |
213 | ||
04253271 | 214 | rcu_read_lock(); |
ffe60014 | 215 | stream = find_stream(key, ht); |
04253271 | 216 | if (stream) { |
da009f2c | 217 | stream->key = (uint64_t) -1ULL; |
04253271 MD |
218 | /* |
219 | * We don't want the lookup to match, but we still need | |
220 | * to iterate on this stream when iterating over the hash table. Just | |
221 | * change the node key. | |
222 | */ | |
da009f2c | 223 | stream->node.key = (uint64_t) -1ULL; |
04253271 MD |
224 | } |
225 | rcu_read_unlock(); | |
7ad0a0cb MD |
226 | } |
227 | ||
d56db448 DG |
228 | /* |
229 | * Return a channel object for the given key. | |
230 | * | |
231 | * RCU read side lock MUST be acquired before calling this function and | |
232 | * protects the channel ptr. | |
233 | */ | |
d88aee68 | 234 | struct lttng_consumer_channel *consumer_find_channel(uint64_t key) |
3bd1e081 | 235 | { |
e4421fec | 236 | struct lttng_ht_iter iter; |
d88aee68 | 237 | struct lttng_ht_node_u64 *node; |
e4421fec | 238 | struct lttng_consumer_channel *channel = NULL; |
3bd1e081 | 239 | |
d88aee68 DG |
240 | /* -1ULL keys are lookup failures */ |
241 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 242 | return NULL; |
7a57cf92 | 243 | } |
e4421fec | 244 | |
d88aee68 DG |
245 | lttng_ht_lookup(consumer_data.channel_ht, &key, &iter); |
246 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
247 | if (node != NULL) { |
248 | channel = caa_container_of(node, struct lttng_consumer_channel, node); | |
3bd1e081 | 249 | } |
e4421fec DG |
250 | |
251 | return channel; | |
3bd1e081 MD |
252 | } |
253 | ||
b5a6470f DG |
254 | /* |
255 | * There is a possibility that the consumer does not have enough time between | |
256 | * the close of the channel on the session daemon and the cleanup in here thus | |
257 | * once we have a channel add with an existing key, we know for sure that this | |
258 | * channel will eventually get cleaned up by all streams being closed. | |
259 | * | |
260 | * This function just nullifies the already existing channel key. | |
261 | */ | |
262 | static void steal_channel_key(uint64_t key) | |
263 | { | |
264 | struct lttng_consumer_channel *channel; | |
265 | ||
266 | rcu_read_lock(); | |
267 | channel = consumer_find_channel(key); | |
268 | if (channel) { | |
269 | channel->key = (uint64_t) -1ULL; | |
270 | /* | |
271 | * We don't want the lookup to match, but we still need to iterate on | |
272 | * this channel when iterating over the hash table. Just change the | |
273 | * node key. | |
274 | */ | |
275 | channel->node.key = (uint64_t) -1ULL; | |
276 | } | |
277 | rcu_read_unlock(); | |
278 | } | |
279 | ||
ffe60014 | 280 | static void free_channel_rcu(struct rcu_head *head) |
702b1ea4 | 281 | { |
d88aee68 DG |
282 | struct lttng_ht_node_u64 *node = |
283 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
ffe60014 DG |
284 | struct lttng_consumer_channel *channel = |
285 | caa_container_of(node, struct lttng_consumer_channel, node); | |
702b1ea4 | 286 | |
b83e03c4 MD |
287 | switch (consumer_data.type) { |
288 | case LTTNG_CONSUMER_KERNEL: | |
289 | break; | |
290 | case LTTNG_CONSUMER32_UST: | |
291 | case LTTNG_CONSUMER64_UST: | |
292 | lttng_ustconsumer_free_channel(channel); | |
293 | break; | |
294 | default: | |
295 | ERR("Unknown consumer_data type"); | |
296 | abort(); | |
297 | } | |
ffe60014 | 298 | free(channel); |
702b1ea4 MD |
299 | } |
300 | ||
00e2e675 DG |
301 | /* |
302 | * RCU protected relayd socket pair free. | |
303 | */ | |
ffe60014 | 304 | static void free_relayd_rcu(struct rcu_head *head) |
00e2e675 | 305 | { |
d88aee68 DG |
306 | struct lttng_ht_node_u64 *node = |
307 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
00e2e675 DG |
308 | struct consumer_relayd_sock_pair *relayd = |
309 | caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
310 | ||
8994307f DG |
311 | /* |
312 | * Close all sockets. This is done in the call RCU since we don't want the | |
313 | * socket fds to be reassigned thus potentially creating bad state of the | |
314 | * relayd object. | |
315 | * | |
316 | * We do not have to lock the control socket mutex here since at this stage | |
317 | * there is no one referencing to this relayd object. | |
318 | */ | |
319 | (void) relayd_close(&relayd->control_sock); | |
320 | (void) relayd_close(&relayd->data_sock); | |
321 | ||
00e2e675 DG |
322 | free(relayd); |
323 | } | |
324 | ||
325 | /* | |
326 | * Destroy and free relayd socket pair object. | |
00e2e675 | 327 | */ |
51230d70 | 328 | void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
329 | { |
330 | int ret; | |
331 | struct lttng_ht_iter iter; | |
332 | ||
173af62f DG |
333 | if (relayd == NULL) { |
334 | return; | |
335 | } | |
336 | ||
00e2e675 DG |
337 | DBG("Consumer destroy and close relayd socket pair"); |
338 | ||
339 | iter.iter.node = &relayd->node.node; | |
340 | ret = lttng_ht_del(consumer_data.relayd_ht, &iter); | |
173af62f | 341 | if (ret != 0) { |
8994307f | 342 | /* We assume the relayd is being or is destroyed */ |
173af62f DG |
343 | return; |
344 | } | |
00e2e675 | 345 | |
00e2e675 | 346 | /* RCU free() call */ |
ffe60014 DG |
347 | call_rcu(&relayd->node.head, free_relayd_rcu); |
348 | } | |
349 | ||
350 | /* | |
351 | * Remove a channel from the global list protected by a mutex. This function is | |
352 | * also responsible for freeing its data structures. | |
353 | */ | |
354 | void consumer_del_channel(struct lttng_consumer_channel *channel) | |
355 | { | |
356 | int ret; | |
357 | struct lttng_ht_iter iter; | |
358 | ||
d88aee68 | 359 | DBG("Consumer delete channel key %" PRIu64, channel->key); |
ffe60014 DG |
360 | |
361 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 362 | pthread_mutex_lock(&channel->lock); |
ffe60014 | 363 | |
212d67a2 DG |
364 | /* Destroy streams that might have been left in the stream list. */ |
365 | clean_channel_stream_list(channel); | |
51e762e5 | 366 | |
d3e2ba59 JD |
367 | if (channel->live_timer_enabled == 1) { |
368 | consumer_timer_live_stop(channel); | |
369 | } | |
370 | ||
ffe60014 DG |
371 | switch (consumer_data.type) { |
372 | case LTTNG_CONSUMER_KERNEL: | |
373 | break; | |
374 | case LTTNG_CONSUMER32_UST: | |
375 | case LTTNG_CONSUMER64_UST: | |
376 | lttng_ustconsumer_del_channel(channel); | |
377 | break; | |
378 | default: | |
379 | ERR("Unknown consumer_data type"); | |
380 | assert(0); | |
381 | goto end; | |
382 | } | |
383 | ||
384 | rcu_read_lock(); | |
385 | iter.iter.node = &channel->node.node; | |
386 | ret = lttng_ht_del(consumer_data.channel_ht, &iter); | |
387 | assert(!ret); | |
388 | rcu_read_unlock(); | |
389 | ||
390 | call_rcu(&channel->node.head, free_channel_rcu); | |
391 | end: | |
a9838785 | 392 | pthread_mutex_unlock(&channel->lock); |
ffe60014 | 393 | pthread_mutex_unlock(&consumer_data.lock); |
00e2e675 DG |
394 | } |
395 | ||
228b5bf7 DG |
396 | /* |
397 | * Iterate over the relayd hash table and destroy each element. Finally, | |
398 | * destroy the whole hash table. | |
399 | */ | |
400 | static void cleanup_relayd_ht(void) | |
401 | { | |
402 | struct lttng_ht_iter iter; | |
403 | struct consumer_relayd_sock_pair *relayd; | |
404 | ||
405 | rcu_read_lock(); | |
406 | ||
407 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
408 | node.node) { | |
51230d70 | 409 | consumer_destroy_relayd(relayd); |
228b5bf7 DG |
410 | } |
411 | ||
228b5bf7 | 412 | rcu_read_unlock(); |
36b588ed MD |
413 | |
414 | lttng_ht_destroy(consumer_data.relayd_ht); | |
228b5bf7 DG |
415 | } |
416 | ||
8994307f DG |
417 | /* |
418 | * Update the end point status of all streams having the given network sequence | |
419 | * index (relayd index). | |
420 | * | |
421 | * It's atomically set without having the stream mutex locked which is fine | |
422 | * because we handle the write/read race with a pipe wakeup for each thread. | |
423 | */ | |
da009f2c | 424 | static void update_endpoint_status_by_netidx(uint64_t net_seq_idx, |
8994307f DG |
425 | enum consumer_endpoint_status status) |
426 | { | |
427 | struct lttng_ht_iter iter; | |
428 | struct lttng_consumer_stream *stream; | |
429 | ||
da009f2c | 430 | DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx); |
8994307f DG |
431 | |
432 | rcu_read_lock(); | |
433 | ||
434 | /* Let's begin with metadata */ | |
435 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
436 | if (stream->net_seq_idx == net_seq_idx) { | |
437 | uatomic_set(&stream->endpoint_status, status); | |
438 | DBG("Delete flag set to metadata stream %d", stream->wait_fd); | |
439 | } | |
440 | } | |
441 | ||
442 | /* Follow up by the data streams */ | |
443 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
444 | if (stream->net_seq_idx == net_seq_idx) { | |
445 | uatomic_set(&stream->endpoint_status, status); | |
446 | DBG("Delete flag set to data stream %d", stream->wait_fd); | |
447 | } | |
448 | } | |
449 | rcu_read_unlock(); | |
450 | } | |
451 | ||
452 | /* | |
453 | * Cleanup a relayd object by flagging every associated streams for deletion, | |
454 | * destroying the object meaning removing it from the relayd hash table, | |
455 | * closing the sockets and freeing the memory in a RCU call. | |
456 | * | |
457 | * If a local data context is available, notify the threads that the streams' | |
458 | * state have changed. | |
459 | */ | |
460 | static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd, | |
461 | struct lttng_consumer_local_data *ctx) | |
462 | { | |
da009f2c | 463 | uint64_t netidx; |
8994307f DG |
464 | |
465 | assert(relayd); | |
466 | ||
9617607b DG |
467 | DBG("Cleaning up relayd sockets"); |
468 | ||
8994307f DG |
469 | /* Save the net sequence index before destroying the object */ |
470 | netidx = relayd->net_seq_idx; | |
471 | ||
472 | /* | |
473 | * Delete the relayd from the relayd hash table, close the sockets and free | |
474 | * the object in a RCU call. | |
475 | */ | |
51230d70 | 476 | consumer_destroy_relayd(relayd); |
8994307f DG |
477 | |
478 | /* Set inactive endpoint to all streams */ | |
479 | update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE); | |
480 | ||
481 | /* | |
482 | * With a local data context, notify the threads that the streams' state | |
483 | * have changed. The write() action on the pipe acts as an "implicit" | |
484 | * memory barrier ordering the updates of the end point status from the | |
485 | * read of this status which happens AFTER receiving this notify. | |
486 | */ | |
487 | if (ctx) { | |
acdb9057 | 488 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
13886d2d | 489 | notify_thread_lttng_pipe(ctx->consumer_metadata_pipe); |
8994307f DG |
490 | } |
491 | } | |
492 | ||
a6ba4fe1 DG |
493 | /* |
494 | * Flag a relayd socket pair for destruction. Destroy it if the refcount | |
495 | * reaches zero. | |
496 | * | |
497 | * RCU read side lock MUST be aquired before calling this function. | |
498 | */ | |
499 | void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) | |
500 | { | |
501 | assert(relayd); | |
502 | ||
503 | /* Set destroy flag for this object */ | |
504 | uatomic_set(&relayd->destroy_flag, 1); | |
505 | ||
506 | /* Destroy the relayd if refcount is 0 */ | |
507 | if (uatomic_read(&relayd->refcount) == 0) { | |
51230d70 | 508 | consumer_destroy_relayd(relayd); |
a6ba4fe1 DG |
509 | } |
510 | } | |
511 | ||
3bd1e081 | 512 | /* |
1d1a276c DG |
513 | * Completly destroy stream from every visiable data structure and the given |
514 | * hash table if one. | |
515 | * | |
516 | * One this call returns, the stream object is not longer usable nor visible. | |
3bd1e081 | 517 | */ |
e316aad5 DG |
518 | void consumer_del_stream(struct lttng_consumer_stream *stream, |
519 | struct lttng_ht *ht) | |
3bd1e081 | 520 | { |
1d1a276c | 521 | consumer_stream_destroy(stream, ht); |
3bd1e081 MD |
522 | } |
523 | ||
5ab66908 MD |
524 | /* |
525 | * XXX naming of del vs destroy is all mixed up. | |
526 | */ | |
527 | void consumer_del_stream_for_data(struct lttng_consumer_stream *stream) | |
528 | { | |
529 | consumer_stream_destroy(stream, data_ht); | |
530 | } | |
531 | ||
532 | void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream) | |
533 | { | |
534 | consumer_stream_destroy(stream, metadata_ht); | |
535 | } | |
536 | ||
d88aee68 DG |
537 | struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, |
538 | uint64_t stream_key, | |
3bd1e081 | 539 | enum lttng_consumer_stream_state state, |
ffe60014 | 540 | const char *channel_name, |
6df2e2c9 | 541 | uid_t uid, |
00e2e675 | 542 | gid_t gid, |
57a269f2 | 543 | uint64_t relayd_id, |
53632229 | 544 | uint64_t session_id, |
ffe60014 DG |
545 | int cpu, |
546 | int *alloc_ret, | |
4891ece8 DG |
547 | enum consumer_channel_type type, |
548 | unsigned int monitor) | |
3bd1e081 | 549 | { |
ffe60014 | 550 | int ret; |
3bd1e081 | 551 | struct lttng_consumer_stream *stream; |
3bd1e081 | 552 | |
effcf122 | 553 | stream = zmalloc(sizeof(*stream)); |
3bd1e081 | 554 | if (stream == NULL) { |
7a57cf92 | 555 | PERROR("malloc struct lttng_consumer_stream"); |
ffe60014 | 556 | ret = -ENOMEM; |
7a57cf92 | 557 | goto end; |
3bd1e081 | 558 | } |
7a57cf92 | 559 | |
d56db448 DG |
560 | rcu_read_lock(); |
561 | ||
3bd1e081 | 562 | stream->key = stream_key; |
3bd1e081 MD |
563 | stream->out_fd = -1; |
564 | stream->out_fd_offset = 0; | |
e5d1a9b3 | 565 | stream->output_written = 0; |
3bd1e081 | 566 | stream->state = state; |
6df2e2c9 MD |
567 | stream->uid = uid; |
568 | stream->gid = gid; | |
ffe60014 | 569 | stream->net_seq_idx = relayd_id; |
53632229 | 570 | stream->session_id = session_id; |
4891ece8 | 571 | stream->monitor = monitor; |
774d490c | 572 | stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE; |
8dbd7d83 | 573 | stream->index_file = NULL; |
fb83fe64 | 574 | stream->last_sequence_number = -1ULL; |
53632229 | 575 | pthread_mutex_init(&stream->lock, NULL); |
c585821b | 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 | */ | |
027a694f | 715 | static 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 | 949 | unsigned int monitor, |
d7ba1388 | 950 | unsigned int live_timer_interval, |
3d071855 | 951 | const char *root_shm_path, |
d7ba1388 | 952 | const char *shm_path) |
3bd1e081 MD |
953 | { |
954 | struct lttng_consumer_channel *channel; | |
3bd1e081 | 955 | |
276b26d1 | 956 | channel = zmalloc(sizeof(*channel)); |
3bd1e081 | 957 | if (channel == NULL) { |
7a57cf92 | 958 | PERROR("malloc struct lttng_consumer_channel"); |
3bd1e081 MD |
959 | goto end; |
960 | } | |
ffe60014 DG |
961 | |
962 | channel->key = key; | |
3bd1e081 | 963 | channel->refcount = 0; |
ffe60014 | 964 | channel->session_id = session_id; |
1950109e | 965 | channel->session_id_per_pid = session_id_per_pid; |
ffe60014 DG |
966 | channel->uid = uid; |
967 | channel->gid = gid; | |
968 | channel->relayd_id = relayd_id; | |
1624d5b7 JD |
969 | channel->tracefile_size = tracefile_size; |
970 | channel->tracefile_count = tracefile_count; | |
2bba9e53 | 971 | channel->monitor = monitor; |
ecc48a90 | 972 | channel->live_timer_interval = live_timer_interval; |
a9838785 | 973 | pthread_mutex_init(&channel->lock, NULL); |
ec6ea7d0 | 974 | pthread_mutex_init(&channel->timer_lock, NULL); |
ffe60014 | 975 | |
0c759fc9 DG |
976 | switch (output) { |
977 | case LTTNG_EVENT_SPLICE: | |
978 | channel->output = CONSUMER_CHANNEL_SPLICE; | |
979 | break; | |
980 | case LTTNG_EVENT_MMAP: | |
981 | channel->output = CONSUMER_CHANNEL_MMAP; | |
982 | break; | |
983 | default: | |
984 | assert(0); | |
985 | free(channel); | |
986 | channel = NULL; | |
987 | goto end; | |
988 | } | |
989 | ||
07b86b52 JD |
990 | /* |
991 | * In monitor mode, the streams associated with the channel will be put in | |
992 | * a special list ONLY owned by this channel. So, the refcount is set to 1 | |
993 | * here meaning that the channel itself has streams that are referenced. | |
994 | * | |
995 | * On a channel deletion, once the channel is no longer visible, the | |
996 | * refcount is decremented and checked for a zero value to delete it. With | |
997 | * streams in no monitor mode, it will now be safe to destroy the channel. | |
998 | */ | |
999 | if (!channel->monitor) { | |
1000 | channel->refcount = 1; | |
1001 | } | |
1002 | ||
ffe60014 DG |
1003 | strncpy(channel->pathname, pathname, sizeof(channel->pathname)); |
1004 | channel->pathname[sizeof(channel->pathname) - 1] = '\0'; | |
1005 | ||
1006 | strncpy(channel->name, name, sizeof(channel->name)); | |
1007 | channel->name[sizeof(channel->name) - 1] = '\0'; | |
1008 | ||
3d071855 MD |
1009 | if (root_shm_path) { |
1010 | strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path)); | |
1011 | channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0'; | |
1012 | } | |
d7ba1388 MD |
1013 | if (shm_path) { |
1014 | strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path)); | |
1015 | channel->shm_path[sizeof(channel->shm_path) - 1] = '\0'; | |
1016 | } | |
1017 | ||
d88aee68 | 1018 | lttng_ht_node_init_u64(&channel->node, channel->key); |
d8ef542d MD |
1019 | |
1020 | channel->wait_fd = -1; | |
1021 | ||
ffe60014 DG |
1022 | CDS_INIT_LIST_HEAD(&channel->streams.head); |
1023 | ||
cb1ce0aa | 1024 | DBG("Allocated channel (key %" PRIu64 ")", channel->key); |
3bd1e081 | 1025 | |
3bd1e081 MD |
1026 | end: |
1027 | return channel; | |
1028 | } | |
1029 | ||
1030 | /* | |
1031 | * Add a channel to the global list protected by a mutex. | |
821fffb2 | 1032 | * |
b5a6470f | 1033 | * Always return 0 indicating success. |
3bd1e081 | 1034 | */ |
d8ef542d MD |
1035 | int consumer_add_channel(struct lttng_consumer_channel *channel, |
1036 | struct lttng_consumer_local_data *ctx) | |
3bd1e081 | 1037 | { |
3bd1e081 | 1038 | pthread_mutex_lock(&consumer_data.lock); |
a9838785 | 1039 | pthread_mutex_lock(&channel->lock); |
ec6ea7d0 | 1040 | pthread_mutex_lock(&channel->timer_lock); |
c77fc10a | 1041 | |
b5a6470f DG |
1042 | /* |
1043 | * This gives us a guarantee that the channel we are about to add to the | |
1044 | * channel hash table will be unique. See this function comment on the why | |
1045 | * we need to steel the channel key at this stage. | |
1046 | */ | |
1047 | steal_channel_key(channel->key); | |
c77fc10a | 1048 | |
b5a6470f | 1049 | rcu_read_lock(); |
d88aee68 | 1050 | lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node); |
6065ceec | 1051 | rcu_read_unlock(); |
b5a6470f | 1052 | |
ec6ea7d0 | 1053 | pthread_mutex_unlock(&channel->timer_lock); |
a9838785 | 1054 | pthread_mutex_unlock(&channel->lock); |
3bd1e081 | 1055 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 1056 | |
b5a6470f | 1057 | if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) { |
a0cbdd2e | 1058 | notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD); |
d8ef542d | 1059 | } |
b5a6470f DG |
1060 | |
1061 | return 0; | |
3bd1e081 MD |
1062 | } |
1063 | ||
1064 | /* | |
1065 | * Allocate the pollfd structure and the local view of the out fds to avoid | |
1066 | * doing a lookup in the linked list and concurrency issues when writing is | |
1067 | * needed. Called with consumer_data.lock held. | |
1068 | * | |
1069 | * Returns the number of fds in the structures. | |
1070 | */ | |
ffe60014 DG |
1071 | static int update_poll_array(struct lttng_consumer_local_data *ctx, |
1072 | struct pollfd **pollfd, struct lttng_consumer_stream **local_stream, | |
1073 | struct lttng_ht *ht) | |
3bd1e081 | 1074 | { |
3bd1e081 | 1075 | int i = 0; |
e4421fec DG |
1076 | struct lttng_ht_iter iter; |
1077 | struct lttng_consumer_stream *stream; | |
3bd1e081 | 1078 | |
ffe60014 DG |
1079 | assert(ctx); |
1080 | assert(ht); | |
1081 | assert(pollfd); | |
1082 | assert(local_stream); | |
1083 | ||
3bd1e081 | 1084 | DBG("Updating poll fd array"); |
481d6c57 | 1085 | rcu_read_lock(); |
43c34bc3 | 1086 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
8994307f DG |
1087 | /* |
1088 | * Only active streams with an active end point can be added to the | |
1089 | * poll set and local stream storage of the thread. | |
1090 | * | |
1091 | * There is a potential race here for endpoint_status to be updated | |
1092 | * just after the check. However, this is OK since the stream(s) will | |
1093 | * be deleted once the thread is notified that the end point state has | |
1094 | * changed where this function will be called back again. | |
1095 | */ | |
1096 | if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM || | |
79d4ffb7 | 1097 | stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { |
3bd1e081 MD |
1098 | continue; |
1099 | } | |
7972aab2 DG |
1100 | /* |
1101 | * This clobbers way too much the debug output. Uncomment that if you | |
1102 | * need it for debugging purposes. | |
1103 | * | |
1104 | * DBG("Active FD %d", stream->wait_fd); | |
1105 | */ | |
e4421fec | 1106 | (*pollfd)[i].fd = stream->wait_fd; |
3bd1e081 | 1107 | (*pollfd)[i].events = POLLIN | POLLPRI; |
e4421fec | 1108 | local_stream[i] = stream; |
3bd1e081 MD |
1109 | i++; |
1110 | } | |
481d6c57 | 1111 | rcu_read_unlock(); |
3bd1e081 MD |
1112 | |
1113 | /* | |
50f8ae69 | 1114 | * Insert the consumer_data_pipe at the end of the array and don't |
3bd1e081 MD |
1115 | * increment i so nb_fd is the number of real FD. |
1116 | */ | |
acdb9057 | 1117 | (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe); |
509bb1cf | 1118 | (*pollfd)[i].events = POLLIN | POLLPRI; |
02b3d176 DG |
1119 | |
1120 | (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe); | |
1121 | (*pollfd)[i + 1].events = POLLIN | POLLPRI; | |
3bd1e081 MD |
1122 | return i; |
1123 | } | |
1124 | ||
1125 | /* | |
84382d49 MD |
1126 | * Poll on the should_quit pipe and the command socket return -1 on |
1127 | * error, 1 if should exit, 0 if data is available on the command socket | |
3bd1e081 MD |
1128 | */ |
1129 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) | |
1130 | { | |
1131 | int num_rdy; | |
1132 | ||
88f2b785 | 1133 | restart: |
3bd1e081 MD |
1134 | num_rdy = poll(consumer_sockpoll, 2, -1); |
1135 | if (num_rdy == -1) { | |
88f2b785 MD |
1136 | /* |
1137 | * Restart interrupted system call. | |
1138 | */ | |
1139 | if (errno == EINTR) { | |
1140 | goto restart; | |
1141 | } | |
7a57cf92 | 1142 | PERROR("Poll error"); |
84382d49 | 1143 | return -1; |
3bd1e081 | 1144 | } |
509bb1cf | 1145 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
3bd1e081 | 1146 | DBG("consumer_should_quit wake up"); |
84382d49 | 1147 | return 1; |
3bd1e081 MD |
1148 | } |
1149 | return 0; | |
3bd1e081 MD |
1150 | } |
1151 | ||
1152 | /* | |
1153 | * Set the error socket. | |
1154 | */ | |
ffe60014 DG |
1155 | void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx, |
1156 | int sock) | |
3bd1e081 MD |
1157 | { |
1158 | ctx->consumer_error_socket = sock; | |
1159 | } | |
1160 | ||
1161 | /* | |
1162 | * Set the command socket path. | |
1163 | */ | |
3bd1e081 MD |
1164 | void lttng_consumer_set_command_sock_path( |
1165 | struct lttng_consumer_local_data *ctx, char *sock) | |
1166 | { | |
1167 | ctx->consumer_command_sock_path = sock; | |
1168 | } | |
1169 | ||
1170 | /* | |
1171 | * Send return code to the session daemon. | |
1172 | * If the socket is not defined, we return 0, it is not a fatal error | |
1173 | */ | |
ffe60014 | 1174 | int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd) |
3bd1e081 MD |
1175 | { |
1176 | if (ctx->consumer_error_socket > 0) { | |
1177 | return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd, | |
1178 | sizeof(enum lttcomm_sessiond_command)); | |
1179 | } | |
1180 | ||
1181 | return 0; | |
1182 | } | |
1183 | ||
1184 | /* | |
228b5bf7 DG |
1185 | * Close all the tracefiles and stream fds and MUST be called when all |
1186 | * instances are destroyed i.e. when all threads were joined and are ended. | |
3bd1e081 MD |
1187 | */ |
1188 | void lttng_consumer_cleanup(void) | |
1189 | { | |
e4421fec | 1190 | struct lttng_ht_iter iter; |
ffe60014 | 1191 | struct lttng_consumer_channel *channel; |
6065ceec DG |
1192 | |
1193 | rcu_read_lock(); | |
3bd1e081 | 1194 | |
ffe60014 DG |
1195 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel, |
1196 | node.node) { | |
702b1ea4 | 1197 | consumer_del_channel(channel); |
3bd1e081 | 1198 | } |
6065ceec DG |
1199 | |
1200 | rcu_read_unlock(); | |
d6ce1df2 | 1201 | |
d6ce1df2 | 1202 | lttng_ht_destroy(consumer_data.channel_ht); |
228b5bf7 DG |
1203 | |
1204 | cleanup_relayd_ht(); | |
1205 | ||
d8ef542d MD |
1206 | lttng_ht_destroy(consumer_data.stream_per_chan_id_ht); |
1207 | ||
228b5bf7 DG |
1208 | /* |
1209 | * This HT contains streams that are freed by either the metadata thread or | |
1210 | * the data thread so we do *nothing* on the hash table and simply destroy | |
1211 | * it. | |
1212 | */ | |
1213 | lttng_ht_destroy(consumer_data.stream_list_ht); | |
3bd1e081 MD |
1214 | } |
1215 | ||
1216 | /* | |
1217 | * Called from signal handler. | |
1218 | */ | |
1219 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) | |
1220 | { | |
6cd525e8 MD |
1221 | ssize_t ret; |
1222 | ||
3bd1e081 | 1223 | consumer_quit = 1; |
6cd525e8 MD |
1224 | ret = lttng_write(ctx->consumer_should_quit[1], "4", 1); |
1225 | if (ret < 1) { | |
7a57cf92 | 1226 | PERROR("write consumer quit"); |
3bd1e081 | 1227 | } |
ab1027f4 DG |
1228 | |
1229 | DBG("Consumer flag that it should quit"); | |
3bd1e081 MD |
1230 | } |
1231 | ||
631d9212 JG |
1232 | |
1233 | /* | |
1234 | * Flush pending writes to trace output disk file. | |
1235 | */ | |
1236 | static | |
00e2e675 DG |
1237 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
1238 | off_t orig_offset) | |
3bd1e081 | 1239 | { |
ca03b51d | 1240 | int ret; |
3bd1e081 MD |
1241 | int outfd = stream->out_fd; |
1242 | ||
1243 | /* | |
1244 | * This does a blocking write-and-wait on any page that belongs to the | |
1245 | * subbuffer prior to the one we just wrote. | |
1246 | * Don't care about error values, as these are just hints and ways to | |
1247 | * limit the amount of page cache used. | |
1248 | */ | |
ffe60014 | 1249 | if (orig_offset < stream->max_sb_size) { |
3bd1e081 MD |
1250 | return; |
1251 | } | |
ffe60014 DG |
1252 | lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size, |
1253 | stream->max_sb_size, | |
3bd1e081 MD |
1254 | SYNC_FILE_RANGE_WAIT_BEFORE |
1255 | | SYNC_FILE_RANGE_WRITE | |
1256 | | SYNC_FILE_RANGE_WAIT_AFTER); | |
1257 | /* | |
1258 | * Give hints to the kernel about how we access the file: | |
1259 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after | |
1260 | * we write it. | |
1261 | * | |
1262 | * We need to call fadvise again after the file grows because the | |
1263 | * kernel does not seem to apply fadvise to non-existing parts of the | |
1264 | * file. | |
1265 | * | |
1266 | * Call fadvise _after_ having waited for the page writeback to | |
1267 | * complete because the dirty page writeback semantic is not well | |
1268 | * defined. So it can be expected to lead to lower throughput in | |
1269 | * streaming. | |
1270 | */ | |
ca03b51d | 1271 | ret = posix_fadvise(outfd, orig_offset - stream->max_sb_size, |
ffe60014 | 1272 | stream->max_sb_size, POSIX_FADV_DONTNEED); |
5bf26938 | 1273 | if (ret && ret != -ENOSYS) { |
144c7f8a JG |
1274 | errno = ret; |
1275 | PERROR("posix_fadvise on fd %i", outfd); | |
ca03b51d | 1276 | } |
3bd1e081 MD |
1277 | } |
1278 | ||
1279 | /* | |
1280 | * Initialise the necessary environnement : | |
1281 | * - create a new context | |
1282 | * - create the poll_pipe | |
1283 | * - create the should_quit pipe (for signal handler) | |
1284 | * - create the thread pipe (for splice) | |
1285 | * | |
1286 | * Takes a function pointer as argument, this function is called when data is | |
1287 | * available on a buffer. This function is responsible to do the | |
1288 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the | |
1289 | * buffer configuration and then kernctl_put_next_subbuf at the end. | |
1290 | * | |
1291 | * Returns a pointer to the new context or NULL on error. | |
1292 | */ | |
1293 | struct lttng_consumer_local_data *lttng_consumer_create( | |
1294 | enum lttng_consumer_type type, | |
4078b776 | 1295 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
d41f73b7 | 1296 | struct lttng_consumer_local_data *ctx), |
3bd1e081 MD |
1297 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
1298 | int (*recv_stream)(struct lttng_consumer_stream *stream), | |
30319bcb | 1299 | int (*update_stream)(uint64_t stream_key, uint32_t state)) |
3bd1e081 | 1300 | { |
d8ef542d | 1301 | int ret; |
3bd1e081 MD |
1302 | struct lttng_consumer_local_data *ctx; |
1303 | ||
1304 | assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || | |
1305 | consumer_data.type == type); | |
1306 | consumer_data.type = type; | |
1307 | ||
effcf122 | 1308 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
3bd1e081 | 1309 | if (ctx == NULL) { |
7a57cf92 | 1310 | PERROR("allocating context"); |
3bd1e081 MD |
1311 | goto error; |
1312 | } | |
1313 | ||
1314 | ctx->consumer_error_socket = -1; | |
331744e3 | 1315 | ctx->consumer_metadata_socket = -1; |
75d83e50 | 1316 | pthread_mutex_init(&ctx->metadata_socket_lock, NULL); |
3bd1e081 MD |
1317 | /* assign the callbacks */ |
1318 | ctx->on_buffer_ready = buffer_ready; | |
1319 | ctx->on_recv_channel = recv_channel; | |
1320 | ctx->on_recv_stream = recv_stream; | |
1321 | ctx->on_update_stream = update_stream; | |
1322 | ||
acdb9057 DG |
1323 | ctx->consumer_data_pipe = lttng_pipe_open(0); |
1324 | if (!ctx->consumer_data_pipe) { | |
3bd1e081 MD |
1325 | goto error_poll_pipe; |
1326 | } | |
1327 | ||
02b3d176 DG |
1328 | ctx->consumer_wakeup_pipe = lttng_pipe_open(0); |
1329 | if (!ctx->consumer_wakeup_pipe) { | |
1330 | goto error_wakeup_pipe; | |
1331 | } | |
1332 | ||
3bd1e081 MD |
1333 | ret = pipe(ctx->consumer_should_quit); |
1334 | if (ret < 0) { | |
7a57cf92 | 1335 | PERROR("Error creating recv pipe"); |
3bd1e081 MD |
1336 | goto error_quit_pipe; |
1337 | } | |
1338 | ||
d8ef542d MD |
1339 | ret = pipe(ctx->consumer_channel_pipe); |
1340 | if (ret < 0) { | |
1341 | PERROR("Error creating channel pipe"); | |
1342 | goto error_channel_pipe; | |
1343 | } | |
1344 | ||
13886d2d DG |
1345 | ctx->consumer_metadata_pipe = lttng_pipe_open(0); |
1346 | if (!ctx->consumer_metadata_pipe) { | |
fb3a43a9 DG |
1347 | goto error_metadata_pipe; |
1348 | } | |
3bd1e081 | 1349 | |
fb3a43a9 | 1350 | return ctx; |
3bd1e081 | 1351 | |
fb3a43a9 | 1352 | error_metadata_pipe: |
d8ef542d MD |
1353 | utils_close_pipe(ctx->consumer_channel_pipe); |
1354 | error_channel_pipe: | |
d8ef542d | 1355 | utils_close_pipe(ctx->consumer_should_quit); |
3bd1e081 | 1356 | error_quit_pipe: |
02b3d176 DG |
1357 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
1358 | error_wakeup_pipe: | |
acdb9057 | 1359 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
3bd1e081 MD |
1360 | error_poll_pipe: |
1361 | free(ctx); | |
1362 | error: | |
1363 | return NULL; | |
1364 | } | |
1365 | ||
282dadbc MD |
1366 | /* |
1367 | * Iterate over all streams of the hashtable and free them properly. | |
1368 | */ | |
1369 | static void destroy_data_stream_ht(struct lttng_ht *ht) | |
1370 | { | |
1371 | struct lttng_ht_iter iter; | |
1372 | struct lttng_consumer_stream *stream; | |
1373 | ||
1374 | if (ht == NULL) { | |
1375 | return; | |
1376 | } | |
1377 | ||
1378 | rcu_read_lock(); | |
1379 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1380 | /* | |
1381 | * Ignore return value since we are currently cleaning up so any error | |
1382 | * can't be handled. | |
1383 | */ | |
1384 | (void) consumer_del_stream(stream, ht); | |
1385 | } | |
1386 | rcu_read_unlock(); | |
1387 | ||
1388 | lttng_ht_destroy(ht); | |
1389 | } | |
1390 | ||
1391 | /* | |
1392 | * Iterate over all streams of the metadata hashtable and free them | |
1393 | * properly. | |
1394 | */ | |
1395 | static void destroy_metadata_stream_ht(struct lttng_ht *ht) | |
1396 | { | |
1397 | struct lttng_ht_iter iter; | |
1398 | struct lttng_consumer_stream *stream; | |
1399 | ||
1400 | if (ht == NULL) { | |
1401 | return; | |
1402 | } | |
1403 | ||
1404 | rcu_read_lock(); | |
1405 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1406 | /* | |
1407 | * Ignore return value since we are currently cleaning up so any error | |
1408 | * can't be handled. | |
1409 | */ | |
1410 | (void) consumer_del_metadata_stream(stream, ht); | |
1411 | } | |
1412 | rcu_read_unlock(); | |
1413 | ||
1414 | lttng_ht_destroy(ht); | |
1415 | } | |
1416 | ||
3bd1e081 MD |
1417 | /* |
1418 | * Close all fds associated with the instance and free the context. | |
1419 | */ | |
1420 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) | |
1421 | { | |
4c462e79 MD |
1422 | int ret; |
1423 | ||
ab1027f4 DG |
1424 | DBG("Consumer destroying it. Closing everything."); |
1425 | ||
4f2e75b9 DG |
1426 | if (!ctx) { |
1427 | return; | |
1428 | } | |
1429 | ||
282dadbc MD |
1430 | destroy_data_stream_ht(data_ht); |
1431 | destroy_metadata_stream_ht(metadata_ht); | |
1432 | ||
4c462e79 MD |
1433 | ret = close(ctx->consumer_error_socket); |
1434 | if (ret) { | |
1435 | PERROR("close"); | |
1436 | } | |
331744e3 JD |
1437 | ret = close(ctx->consumer_metadata_socket); |
1438 | if (ret) { | |
1439 | PERROR("close"); | |
1440 | } | |
d8ef542d | 1441 | utils_close_pipe(ctx->consumer_channel_pipe); |
acdb9057 | 1442 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
13886d2d | 1443 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
02b3d176 | 1444 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
d8ef542d | 1445 | utils_close_pipe(ctx->consumer_should_quit); |
fb3a43a9 | 1446 | |
3bd1e081 MD |
1447 | unlink(ctx->consumer_command_sock_path); |
1448 | free(ctx); | |
1449 | } | |
1450 | ||
6197aea7 DG |
1451 | /* |
1452 | * Write the metadata stream id on the specified file descriptor. | |
1453 | */ | |
1454 | static int write_relayd_metadata_id(int fd, | |
1455 | struct lttng_consumer_stream *stream, | |
ffe60014 | 1456 | struct consumer_relayd_sock_pair *relayd, unsigned long padding) |
6197aea7 | 1457 | { |
6cd525e8 | 1458 | ssize_t ret; |
1d4dfdef | 1459 | struct lttcomm_relayd_metadata_payload hdr; |
6197aea7 | 1460 | |
1d4dfdef DG |
1461 | hdr.stream_id = htobe64(stream->relayd_stream_id); |
1462 | hdr.padding_size = htobe32(padding); | |
6cd525e8 MD |
1463 | ret = lttng_write(fd, (void *) &hdr, sizeof(hdr)); |
1464 | if (ret < sizeof(hdr)) { | |
d7b75ec8 | 1465 | /* |
6f04ed72 | 1466 | * This error means that the fd's end is closed so ignore the PERROR |
d7b75ec8 DG |
1467 | * not to clubber the error output since this can happen in a normal |
1468 | * code path. | |
1469 | */ | |
1470 | if (errno != EPIPE) { | |
1471 | PERROR("write metadata stream id"); | |
1472 | } | |
1473 | DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno); | |
534d2592 DG |
1474 | /* |
1475 | * Set ret to a negative value because if ret != sizeof(hdr), we don't | |
1476 | * handle writting the missing part so report that as an error and | |
1477 | * don't lie to the caller. | |
1478 | */ | |
1479 | ret = -1; | |
6197aea7 DG |
1480 | goto end; |
1481 | } | |
1d4dfdef DG |
1482 | DBG("Metadata stream id %" PRIu64 " with padding %lu written before data", |
1483 | stream->relayd_stream_id, padding); | |
6197aea7 DG |
1484 | |
1485 | end: | |
6cd525e8 | 1486 | return (int) ret; |
6197aea7 DG |
1487 | } |
1488 | ||
3bd1e081 | 1489 | /* |
09e26845 DG |
1490 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
1491 | * core function for writing trace buffers to either the local filesystem or | |
1492 | * the network. | |
1493 | * | |
79d4ffb7 DG |
1494 | * It must be called with the stream lock held. |
1495 | * | |
09e26845 | 1496 | * Careful review MUST be put if any changes occur! |
3bd1e081 MD |
1497 | * |
1498 | * Returns the number of bytes written | |
1499 | */ | |
4078b776 | 1500 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
3bd1e081 | 1501 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1502 | struct lttng_consumer_stream *stream, unsigned long len, |
309167d2 | 1503 | unsigned long padding, |
50adc264 | 1504 | struct ctf_packet_index *index) |
3bd1e081 | 1505 | { |
f02e1e8a | 1506 | unsigned long mmap_offset; |
ffe60014 | 1507 | void *mmap_base; |
994ab360 | 1508 | ssize_t ret = 0; |
f02e1e8a DG |
1509 | off_t orig_offset = stream->out_fd_offset; |
1510 | /* Default is on the disk */ | |
1511 | int outfd = stream->out_fd; | |
f02e1e8a | 1512 | struct consumer_relayd_sock_pair *relayd = NULL; |
8994307f | 1513 | unsigned int relayd_hang_up = 0; |
f02e1e8a DG |
1514 | |
1515 | /* RCU lock for the relayd pointer */ | |
1516 | rcu_read_lock(); | |
1517 | ||
1518 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1519 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1520 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1521 | if (relayd == NULL) { | |
56591bac | 1522 | ret = -EPIPE; |
f02e1e8a DG |
1523 | goto end; |
1524 | } | |
1525 | } | |
1526 | ||
1527 | /* get the offset inside the fd to mmap */ | |
3bd1e081 MD |
1528 | switch (consumer_data.type) { |
1529 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1530 | mmap_base = stream->mmap_base; |
f02e1e8a | 1531 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); |
994ab360 DG |
1532 | if (ret < 0) { |
1533 | ret = -errno; | |
56591bac | 1534 | PERROR("tracer ctl get_mmap_read_offset"); |
56591bac MD |
1535 | goto end; |
1536 | } | |
f02e1e8a | 1537 | break; |
7753dea8 MD |
1538 | case LTTNG_CONSUMER32_UST: |
1539 | case LTTNG_CONSUMER64_UST: | |
ffe60014 DG |
1540 | mmap_base = lttng_ustctl_get_mmap_base(stream); |
1541 | if (!mmap_base) { | |
1542 | ERR("read mmap get mmap base for stream %s", stream->name); | |
994ab360 | 1543 | ret = -EPERM; |
ffe60014 DG |
1544 | goto end; |
1545 | } | |
1546 | ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset); | |
56591bac MD |
1547 | if (ret != 0) { |
1548 | PERROR("tracer ctl get_mmap_read_offset"); | |
994ab360 | 1549 | ret = -EINVAL; |
56591bac MD |
1550 | goto end; |
1551 | } | |
f02e1e8a | 1552 | break; |
3bd1e081 MD |
1553 | default: |
1554 | ERR("Unknown consumer_data type"); | |
1555 | assert(0); | |
1556 | } | |
b9182dd9 | 1557 | |
f02e1e8a DG |
1558 | /* Handle stream on the relayd if the output is on the network */ |
1559 | if (relayd) { | |
1560 | unsigned long netlen = len; | |
1561 | ||
1562 | /* | |
1563 | * Lock the control socket for the complete duration of the function | |
1564 | * since from this point on we will use the socket. | |
1565 | */ | |
1566 | if (stream->metadata_flag) { | |
1567 | /* Metadata requires the control socket. */ | |
1568 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
93ec662e JD |
1569 | if (stream->reset_metadata_flag) { |
1570 | ret = relayd_reset_metadata(&relayd->control_sock, | |
1571 | stream->relayd_stream_id, | |
1572 | stream->metadata_version); | |
1573 | if (ret < 0) { | |
1574 | relayd_hang_up = 1; | |
1575 | goto write_error; | |
1576 | } | |
1577 | stream->reset_metadata_flag = 0; | |
1578 | } | |
1d4dfdef | 1579 | netlen += sizeof(struct lttcomm_relayd_metadata_payload); |
f02e1e8a DG |
1580 | } |
1581 | ||
1d4dfdef | 1582 | ret = write_relayd_stream_header(stream, netlen, padding, relayd); |
994ab360 DG |
1583 | if (ret < 0) { |
1584 | relayd_hang_up = 1; | |
1585 | goto write_error; | |
1586 | } | |
1587 | /* Use the returned socket. */ | |
1588 | outfd = ret; | |
f02e1e8a | 1589 | |
994ab360 DG |
1590 | /* Write metadata stream id before payload */ |
1591 | if (stream->metadata_flag) { | |
1592 | ret = write_relayd_metadata_id(outfd, stream, relayd, padding); | |
1593 | if (ret < 0) { | |
8994307f DG |
1594 | relayd_hang_up = 1; |
1595 | goto write_error; | |
1596 | } | |
f02e1e8a | 1597 | } |
1d4dfdef DG |
1598 | } else { |
1599 | /* No streaming, we have to set the len with the full padding */ | |
1600 | len += padding; | |
1624d5b7 | 1601 | |
93ec662e JD |
1602 | if (stream->metadata_flag && stream->reset_metadata_flag) { |
1603 | ret = utils_truncate_stream_file(stream->out_fd, 0); | |
1604 | if (ret < 0) { | |
1605 | ERR("Reset metadata file"); | |
1606 | goto end; | |
1607 | } | |
1608 | stream->reset_metadata_flag = 0; | |
1609 | } | |
1610 | ||
1624d5b7 JD |
1611 | /* |
1612 | * Check if we need to change the tracefile before writing the packet. | |
1613 | */ | |
1614 | if (stream->chan->tracefile_size > 0 && | |
1615 | (stream->tracefile_size_current + len) > | |
1616 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1617 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1618 | stream->name, stream->chan->tracefile_size, | |
1619 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
309167d2 JD |
1620 | stream->out_fd, &(stream->tracefile_count_current), |
1621 | &stream->out_fd); | |
1624d5b7 JD |
1622 | if (ret < 0) { |
1623 | ERR("Rotating output file"); | |
1624 | goto end; | |
1625 | } | |
309167d2 JD |
1626 | outfd = stream->out_fd; |
1627 | ||
8dbd7d83 MD |
1628 | if (stream->index_file) { |
1629 | lttng_index_file_put(stream->index_file); | |
1630 | stream->index_file = lttng_index_file_create(stream->chan->pathname, | |
309167d2 JD |
1631 | stream->name, stream->uid, stream->gid, |
1632 | stream->chan->tracefile_size, | |
8dbd7d83 MD |
1633 | stream->tracefile_count_current, |
1634 | CTF_INDEX_MAJOR, CTF_INDEX_MINOR); | |
1635 | if (!stream->index_file) { | |
309167d2 JD |
1636 | goto end; |
1637 | } | |
309167d2 JD |
1638 | } |
1639 | ||
a6976990 DG |
1640 | /* Reset current size because we just perform a rotation. */ |
1641 | stream->tracefile_size_current = 0; | |
a1ae300f JD |
1642 | stream->out_fd_offset = 0; |
1643 | orig_offset = 0; | |
1624d5b7 JD |
1644 | } |
1645 | stream->tracefile_size_current += len; | |
309167d2 JD |
1646 | if (index) { |
1647 | index->offset = htobe64(stream->out_fd_offset); | |
1648 | } | |
f02e1e8a DG |
1649 | } |
1650 | ||
d02b8372 DG |
1651 | /* |
1652 | * This call guarantee that len or less is returned. It's impossible to | |
1653 | * receive a ret value that is bigger than len. | |
1654 | */ | |
1655 | ret = lttng_write(outfd, mmap_base + mmap_offset, len); | |
1656 | DBG("Consumer mmap write() ret %zd (len %lu)", ret, len); | |
1657 | if (ret < 0 || ((size_t) ret != len)) { | |
1658 | /* | |
1659 | * Report error to caller if nothing was written else at least send the | |
1660 | * amount written. | |
1661 | */ | |
1662 | if (ret < 0) { | |
994ab360 | 1663 | ret = -errno; |
f02e1e8a | 1664 | } |
994ab360 | 1665 | relayd_hang_up = 1; |
f02e1e8a | 1666 | |
d02b8372 | 1667 | /* Socket operation failed. We consider the relayd dead */ |
994ab360 | 1668 | if (errno == EPIPE || errno == EINVAL || errno == EBADF) { |
d02b8372 DG |
1669 | /* |
1670 | * This is possible if the fd is closed on the other side | |
1671 | * (outfd) or any write problem. It can be verbose a bit for a | |
1672 | * normal execution if for instance the relayd is stopped | |
1673 | * abruptly. This can happen so set this to a DBG statement. | |
1674 | */ | |
1675 | DBG("Consumer mmap write detected relayd hang up"); | |
994ab360 DG |
1676 | } else { |
1677 | /* Unhandled error, print it and stop function right now. */ | |
1678 | PERROR("Error in write mmap (ret %zd != len %lu)", ret, len); | |
f02e1e8a | 1679 | } |
994ab360 | 1680 | goto write_error; |
d02b8372 DG |
1681 | } |
1682 | stream->output_written += ret; | |
d02b8372 DG |
1683 | |
1684 | /* This call is useless on a socket so better save a syscall. */ | |
1685 | if (!relayd) { | |
1686 | /* This won't block, but will start writeout asynchronously */ | |
1687 | lttng_sync_file_range(outfd, stream->out_fd_offset, len, | |
1688 | SYNC_FILE_RANGE_WRITE); | |
1689 | stream->out_fd_offset += len; | |
141a9783 | 1690 | lttng_consumer_sync_trace_file(stream, orig_offset); |
f02e1e8a | 1691 | } |
f02e1e8a | 1692 | |
8994307f DG |
1693 | write_error: |
1694 | /* | |
1695 | * This is a special case that the relayd has closed its socket. Let's | |
1696 | * cleanup the relayd object and all associated streams. | |
1697 | */ | |
1698 | if (relayd && relayd_hang_up) { | |
1699 | cleanup_relayd(relayd, ctx); | |
1700 | } | |
1701 | ||
f02e1e8a DG |
1702 | end: |
1703 | /* Unlock only if ctrl socket used */ | |
1704 | if (relayd && stream->metadata_flag) { | |
1705 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1706 | } | |
1707 | ||
1708 | rcu_read_unlock(); | |
994ab360 | 1709 | return ret; |
3bd1e081 MD |
1710 | } |
1711 | ||
1712 | /* | |
1713 | * Splice the data from the ring buffer to the tracefile. | |
1714 | * | |
79d4ffb7 DG |
1715 | * It must be called with the stream lock held. |
1716 | * | |
3bd1e081 MD |
1717 | * Returns the number of bytes spliced. |
1718 | */ | |
4078b776 | 1719 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
3bd1e081 | 1720 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1721 | struct lttng_consumer_stream *stream, unsigned long len, |
309167d2 | 1722 | unsigned long padding, |
50adc264 | 1723 | struct ctf_packet_index *index) |
3bd1e081 | 1724 | { |
f02e1e8a DG |
1725 | ssize_t ret = 0, written = 0, ret_splice = 0; |
1726 | loff_t offset = 0; | |
1727 | off_t orig_offset = stream->out_fd_offset; | |
1728 | int fd = stream->wait_fd; | |
1729 | /* Default is on the disk */ | |
1730 | int outfd = stream->out_fd; | |
f02e1e8a | 1731 | struct consumer_relayd_sock_pair *relayd = NULL; |
fb3a43a9 | 1732 | int *splice_pipe; |
8994307f | 1733 | unsigned int relayd_hang_up = 0; |
f02e1e8a | 1734 | |
3bd1e081 MD |
1735 | switch (consumer_data.type) { |
1736 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a | 1737 | break; |
7753dea8 MD |
1738 | case LTTNG_CONSUMER32_UST: |
1739 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a | 1740 | /* Not supported for user space tracing */ |
3bd1e081 MD |
1741 | return -ENOSYS; |
1742 | default: | |
1743 | ERR("Unknown consumer_data type"); | |
1744 | assert(0); | |
3bd1e081 MD |
1745 | } |
1746 | ||
f02e1e8a DG |
1747 | /* RCU lock for the relayd pointer */ |
1748 | rcu_read_lock(); | |
1749 | ||
1750 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1751 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1752 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1753 | if (relayd == NULL) { | |
ad0b0d23 | 1754 | written = -ret; |
f02e1e8a DG |
1755 | goto end; |
1756 | } | |
1757 | } | |
a2361a61 | 1758 | splice_pipe = stream->splice_pipe; |
fb3a43a9 | 1759 | |
f02e1e8a | 1760 | /* Write metadata stream id before payload */ |
1d4dfdef | 1761 | if (relayd) { |
ad0b0d23 | 1762 | unsigned long total_len = len; |
f02e1e8a | 1763 | |
1d4dfdef DG |
1764 | if (stream->metadata_flag) { |
1765 | /* | |
1766 | * Lock the control socket for the complete duration of the function | |
1767 | * since from this point on we will use the socket. | |
1768 | */ | |
1769 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1770 | ||
93ec662e JD |
1771 | if (stream->reset_metadata_flag) { |
1772 | ret = relayd_reset_metadata(&relayd->control_sock, | |
1773 | stream->relayd_stream_id, | |
1774 | stream->metadata_version); | |
1775 | if (ret < 0) { | |
1776 | relayd_hang_up = 1; | |
1777 | goto write_error; | |
1778 | } | |
1779 | stream->reset_metadata_flag = 0; | |
1780 | } | |
1d4dfdef DG |
1781 | ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd, |
1782 | padding); | |
1783 | if (ret < 0) { | |
1784 | written = ret; | |
ad0b0d23 DG |
1785 | relayd_hang_up = 1; |
1786 | goto write_error; | |
1d4dfdef DG |
1787 | } |
1788 | ||
1789 | total_len += sizeof(struct lttcomm_relayd_metadata_payload); | |
1790 | } | |
1791 | ||
1792 | ret = write_relayd_stream_header(stream, total_len, padding, relayd); | |
ad0b0d23 DG |
1793 | if (ret < 0) { |
1794 | written = ret; | |
1795 | relayd_hang_up = 1; | |
1796 | goto write_error; | |
f02e1e8a | 1797 | } |
ad0b0d23 DG |
1798 | /* Use the returned socket. */ |
1799 | outfd = ret; | |
1d4dfdef DG |
1800 | } else { |
1801 | /* No streaming, we have to set the len with the full padding */ | |
1802 | len += padding; | |
1624d5b7 | 1803 | |
93ec662e JD |
1804 | if (stream->metadata_flag && stream->reset_metadata_flag) { |
1805 | ret = utils_truncate_stream_file(stream->out_fd, 0); | |
1806 | if (ret < 0) { | |
1807 | ERR("Reset metadata file"); | |
1808 | goto end; | |
1809 | } | |
1810 | stream->reset_metadata_flag = 0; | |
1811 | } | |
1624d5b7 JD |
1812 | /* |
1813 | * Check if we need to change the tracefile before writing the packet. | |
1814 | */ | |
1815 | if (stream->chan->tracefile_size > 0 && | |
1816 | (stream->tracefile_size_current + len) > | |
1817 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1818 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1819 | stream->name, stream->chan->tracefile_size, | |
1820 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
309167d2 JD |
1821 | stream->out_fd, &(stream->tracefile_count_current), |
1822 | &stream->out_fd); | |
1624d5b7 | 1823 | if (ret < 0) { |
ad0b0d23 | 1824 | written = ret; |
1624d5b7 JD |
1825 | ERR("Rotating output file"); |
1826 | goto end; | |
1827 | } | |
309167d2 JD |
1828 | outfd = stream->out_fd; |
1829 | ||
8dbd7d83 MD |
1830 | if (stream->index_file) { |
1831 | lttng_index_file_put(stream->index_file); | |
1832 | stream->index_file = lttng_index_file_create(stream->chan->pathname, | |
309167d2 JD |
1833 | stream->name, stream->uid, stream->gid, |
1834 | stream->chan->tracefile_size, | |
8dbd7d83 MD |
1835 | stream->tracefile_count_current, |
1836 | CTF_INDEX_MAJOR, CTF_INDEX_MINOR); | |
1837 | if (!stream->index_file) { | |
309167d2 JD |
1838 | goto end; |
1839 | } | |
309167d2 JD |
1840 | } |
1841 | ||
a6976990 DG |
1842 | /* Reset current size because we just perform a rotation. */ |
1843 | stream->tracefile_size_current = 0; | |
a1ae300f JD |
1844 | stream->out_fd_offset = 0; |
1845 | orig_offset = 0; | |
1624d5b7 JD |
1846 | } |
1847 | stream->tracefile_size_current += len; | |
309167d2 | 1848 | index->offset = htobe64(stream->out_fd_offset); |
f02e1e8a DG |
1849 | } |
1850 | ||
1851 | while (len > 0) { | |
1d4dfdef DG |
1852 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)", |
1853 | (unsigned long)offset, len, fd, splice_pipe[1]); | |
fb3a43a9 | 1854 | ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len, |
f02e1e8a DG |
1855 | SPLICE_F_MOVE | SPLICE_F_MORE); |
1856 | DBG("splice chan to pipe, ret %zd", ret_splice); | |
1857 | if (ret_splice < 0) { | |
d02b8372 | 1858 | ret = errno; |
ad0b0d23 | 1859 | written = -ret; |
d02b8372 | 1860 | PERROR("Error in relay splice"); |
f02e1e8a DG |
1861 | goto splice_error; |
1862 | } | |
1863 | ||
1864 | /* Handle stream on the relayd if the output is on the network */ | |
ad0b0d23 DG |
1865 | if (relayd && stream->metadata_flag) { |
1866 | size_t metadata_payload_size = | |
1867 | sizeof(struct lttcomm_relayd_metadata_payload); | |
1868 | ||
1869 | /* Update counter to fit the spliced data */ | |
1870 | ret_splice += metadata_payload_size; | |
1871 | len += metadata_payload_size; | |
1872 | /* | |
1873 | * We do this so the return value can match the len passed as | |
1874 | * argument to this function. | |
1875 | */ | |
1876 | written -= metadata_payload_size; | |
f02e1e8a DG |
1877 | } |
1878 | ||
1879 | /* Splice data out */ | |
fb3a43a9 | 1880 | ret_splice = splice(splice_pipe[0], NULL, outfd, NULL, |
f02e1e8a | 1881 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); |
a2361a61 JD |
1882 | DBG("Consumer splice pipe to file (out_fd: %d), ret %zd", |
1883 | outfd, ret_splice); | |
f02e1e8a | 1884 | if (ret_splice < 0) { |
d02b8372 | 1885 | ret = errno; |
ad0b0d23 DG |
1886 | written = -ret; |
1887 | relayd_hang_up = 1; | |
1888 | goto write_error; | |
f02e1e8a | 1889 | } else if (ret_splice > len) { |
d02b8372 DG |
1890 | /* |
1891 | * We don't expect this code path to be executed but you never know | |
1892 | * so this is an extra protection agains a buggy splice(). | |
1893 | */ | |
f02e1e8a | 1894 | ret = errno; |
ad0b0d23 | 1895 | written += ret_splice; |
d02b8372 DG |
1896 | PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice, |
1897 | len); | |
f02e1e8a | 1898 | goto splice_error; |
d02b8372 DG |
1899 | } else { |
1900 | /* All good, update current len and continue. */ | |
1901 | len -= ret_splice; | |
f02e1e8a | 1902 | } |
f02e1e8a DG |
1903 | |
1904 | /* This call is useless on a socket so better save a syscall. */ | |
1905 | if (!relayd) { | |
1906 | /* This won't block, but will start writeout asynchronously */ | |
1907 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, | |
1908 | SYNC_FILE_RANGE_WRITE); | |
1909 | stream->out_fd_offset += ret_splice; | |
1910 | } | |
e5d1a9b3 | 1911 | stream->output_written += ret_splice; |
f02e1e8a DG |
1912 | written += ret_splice; |
1913 | } | |
141a9783 JG |
1914 | if (!relayd) { |
1915 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1916 | } | |
f02e1e8a DG |
1917 | goto end; |
1918 | ||
8994307f DG |
1919 | write_error: |
1920 | /* | |
1921 | * This is a special case that the relayd has closed its socket. Let's | |
1922 | * cleanup the relayd object and all associated streams. | |
1923 | */ | |
1924 | if (relayd && relayd_hang_up) { | |
1925 | cleanup_relayd(relayd, ctx); | |
1926 | /* Skip splice error so the consumer does not fail */ | |
1927 | goto end; | |
1928 | } | |
1929 | ||
f02e1e8a DG |
1930 | splice_error: |
1931 | /* send the appropriate error description to sessiond */ | |
1932 | switch (ret) { | |
f02e1e8a | 1933 | case EINVAL: |
f73fabfd | 1934 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); |
f02e1e8a DG |
1935 | break; |
1936 | case ENOMEM: | |
f73fabfd | 1937 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); |
f02e1e8a DG |
1938 | break; |
1939 | case ESPIPE: | |
f73fabfd | 1940 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); |
f02e1e8a DG |
1941 | break; |
1942 | } | |
1943 | ||
1944 | end: | |
1945 | if (relayd && stream->metadata_flag) { | |
1946 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1947 | } | |
1948 | ||
1949 | rcu_read_unlock(); | |
1950 | return written; | |
3bd1e081 MD |
1951 | } |
1952 | ||
1953 | /* | |
1954 | * Take a snapshot for a specific fd | |
1955 | * | |
1956 | * Returns 0 on success, < 0 on error | |
1957 | */ | |
ffe60014 | 1958 | int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) |
3bd1e081 MD |
1959 | { |
1960 | switch (consumer_data.type) { | |
1961 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1962 | return lttng_kconsumer_take_snapshot(stream); |
7753dea8 MD |
1963 | case LTTNG_CONSUMER32_UST: |
1964 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 1965 | return lttng_ustconsumer_take_snapshot(stream); |
3bd1e081 MD |
1966 | default: |
1967 | ERR("Unknown consumer_data type"); | |
1968 | assert(0); | |
1969 | return -ENOSYS; | |
1970 | } | |
3bd1e081 MD |
1971 | } |
1972 | ||
1973 | /* | |
1974 | * Get the produced position | |
1975 | * | |
1976 | * Returns 0 on success, < 0 on error | |
1977 | */ | |
ffe60014 | 1978 | int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, |
3bd1e081 MD |
1979 | unsigned long *pos) |
1980 | { | |
1981 | switch (consumer_data.type) { | |
1982 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1983 | return lttng_kconsumer_get_produced_snapshot(stream, pos); |
7753dea8 MD |
1984 | case LTTNG_CONSUMER32_UST: |
1985 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 1986 | return lttng_ustconsumer_get_produced_snapshot(stream, pos); |
3bd1e081 MD |
1987 | default: |
1988 | ERR("Unknown consumer_data type"); | |
1989 | assert(0); | |
1990 | return -ENOSYS; | |
1991 | } | |
1992 | } | |
1993 | ||
1994 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, | |
1995 | int sock, struct pollfd *consumer_sockpoll) | |
1996 | { | |
1997 | switch (consumer_data.type) { | |
1998 | case LTTNG_CONSUMER_KERNEL: | |
1999 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
7753dea8 MD |
2000 | case LTTNG_CONSUMER32_UST: |
2001 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
2002 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
2003 | default: | |
2004 | ERR("Unknown consumer_data type"); | |
2005 | assert(0); | |
2006 | return -ENOSYS; | |
2007 | } | |
2008 | } | |
2009 | ||
6d574024 | 2010 | void lttng_consumer_close_all_metadata(void) |
d88aee68 DG |
2011 | { |
2012 | switch (consumer_data.type) { | |
2013 | case LTTNG_CONSUMER_KERNEL: | |
2014 | /* | |
2015 | * The Kernel consumer has a different metadata scheme so we don't | |
2016 | * close anything because the stream will be closed by the session | |
2017 | * daemon. | |
2018 | */ | |
2019 | break; | |
2020 | case LTTNG_CONSUMER32_UST: | |
2021 | case LTTNG_CONSUMER64_UST: | |
2022 | /* | |
2023 | * Close all metadata streams. The metadata hash table is passed and | |
2024 | * this call iterates over it by closing all wakeup fd. This is safe | |
2025 | * because at this point we are sure that the metadata producer is | |
2026 | * either dead or blocked. | |
2027 | */ | |
6d574024 | 2028 | lttng_ustconsumer_close_all_metadata(metadata_ht); |
d88aee68 DG |
2029 | break; |
2030 | default: | |
2031 | ERR("Unknown consumer_data type"); | |
2032 | assert(0); | |
2033 | } | |
2034 | } | |
2035 | ||
fb3a43a9 DG |
2036 | /* |
2037 | * Clean up a metadata stream and free its memory. | |
2038 | */ | |
e316aad5 DG |
2039 | void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, |
2040 | struct lttng_ht *ht) | |
fb3a43a9 | 2041 | { |
e316aad5 | 2042 | struct lttng_consumer_channel *free_chan = NULL; |
fb3a43a9 DG |
2043 | |
2044 | assert(stream); | |
2045 | /* | |
2046 | * This call should NEVER receive regular stream. It must always be | |
2047 | * metadata stream and this is crucial for data structure synchronization. | |
2048 | */ | |
2049 | assert(stream->metadata_flag); | |
2050 | ||
e316aad5 DG |
2051 | DBG3("Consumer delete metadata stream %d", stream->wait_fd); |
2052 | ||
74251bb8 | 2053 | pthread_mutex_lock(&consumer_data.lock); |
a9838785 | 2054 | pthread_mutex_lock(&stream->chan->lock); |
8994307f DG |
2055 | pthread_mutex_lock(&stream->lock); |
2056 | ||
6d574024 DG |
2057 | /* Remove any reference to that stream. */ |
2058 | consumer_stream_delete(stream, ht); | |
ca22feea | 2059 | |
6d574024 DG |
2060 | /* Close down everything including the relayd if one. */ |
2061 | consumer_stream_close(stream); | |
2062 | /* Destroy tracer buffers of the stream. */ | |
2063 | consumer_stream_destroy_buffers(stream); | |
fb3a43a9 DG |
2064 | |
2065 | /* Atomically decrement channel refcount since other threads can use it. */ | |
f2ad556d | 2066 | if (!uatomic_sub_return(&stream->chan->refcount, 1) |
ffe60014 | 2067 | && !uatomic_read(&stream->chan->nb_init_stream_left)) { |
c30aaa51 | 2068 | /* Go for channel deletion! */ |
e316aad5 | 2069 | free_chan = stream->chan; |
fb3a43a9 DG |
2070 | } |
2071 | ||
73811ecc DG |
2072 | /* |
2073 | * Nullify the stream reference so it is not used after deletion. The | |
6d574024 DG |
2074 | * channel lock MUST be acquired before being able to check for a NULL |
2075 | * pointer value. | |
73811ecc DG |
2076 | */ |
2077 | stream->chan->metadata_stream = NULL; | |
2078 | ||
8994307f | 2079 | pthread_mutex_unlock(&stream->lock); |
a9838785 | 2080 | pthread_mutex_unlock(&stream->chan->lock); |
74251bb8 | 2081 | pthread_mutex_unlock(&consumer_data.lock); |
e316aad5 DG |
2082 | |
2083 | if (free_chan) { | |
2084 | consumer_del_channel(free_chan); | |
2085 | } | |
2086 | ||
6d574024 | 2087 | consumer_stream_free(stream); |
fb3a43a9 DG |
2088 | } |
2089 | ||
2090 | /* | |
2091 | * Action done with the metadata stream when adding it to the consumer internal | |
2092 | * data structures to handle it. | |
2093 | */ | |
5ab66908 | 2094 | int consumer_add_metadata_stream(struct lttng_consumer_stream *stream) |
fb3a43a9 | 2095 | { |
5ab66908 | 2096 | struct lttng_ht *ht = metadata_ht; |
e316aad5 | 2097 | int ret = 0; |
76082088 | 2098 | struct lttng_ht_iter iter; |
d88aee68 | 2099 | struct lttng_ht_node_u64 *node; |
fb3a43a9 | 2100 | |
e316aad5 DG |
2101 | assert(stream); |
2102 | assert(ht); | |
2103 | ||
d88aee68 | 2104 | DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key); |
e316aad5 DG |
2105 | |
2106 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 2107 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 2108 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 2109 | pthread_mutex_lock(&stream->lock); |
e316aad5 | 2110 | |
e316aad5 DG |
2111 | /* |
2112 | * From here, refcounts are updated so be _careful_ when returning an error | |
2113 | * after this point. | |
2114 | */ | |
2115 | ||
fb3a43a9 | 2116 | rcu_read_lock(); |
76082088 DG |
2117 | |
2118 | /* | |
2119 | * Lookup the stream just to make sure it does not exist in our internal | |
2120 | * state. This should NEVER happen. | |
2121 | */ | |
d88aee68 DG |
2122 | lttng_ht_lookup(ht, &stream->key, &iter); |
2123 | node = lttng_ht_iter_get_node_u64(&iter); | |
76082088 DG |
2124 | assert(!node); |
2125 | ||
e316aad5 | 2126 | /* |
ffe60014 DG |
2127 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
2128 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
2129 | * causes the count to become 0 also causes a stream to be added. The |
2130 | * channel deletion will thus be triggered by the following removal of this | |
2131 | * stream. | |
2132 | */ | |
ffe60014 | 2133 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
2134 | /* Increment refcount before decrementing nb_init_stream_left */ |
2135 | cmm_smp_wmb(); | |
ffe60014 | 2136 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
2137 | } |
2138 | ||
d88aee68 | 2139 | lttng_ht_add_unique_u64(ht, &stream->node); |
ca22feea | 2140 | |
d8ef542d MD |
2141 | lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht, |
2142 | &stream->node_channel_id); | |
2143 | ||
ca22feea DG |
2144 | /* |
2145 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
2146 | * the key since the HT does not use it and we allow to add redundant keys | |
2147 | * into this table. | |
2148 | */ | |
d88aee68 | 2149 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 2150 | |
fb3a43a9 | 2151 | rcu_read_unlock(); |
e316aad5 | 2152 | |
2e818a6a | 2153 | pthread_mutex_unlock(&stream->lock); |
a9838785 | 2154 | pthread_mutex_unlock(&stream->chan->lock); |
ec6ea7d0 | 2155 | pthread_mutex_unlock(&stream->chan->timer_lock); |
e316aad5 DG |
2156 | pthread_mutex_unlock(&consumer_data.lock); |
2157 | return ret; | |
fb3a43a9 DG |
2158 | } |
2159 | ||
8994307f DG |
2160 | /* |
2161 | * Delete data stream that are flagged for deletion (endpoint_status). | |
2162 | */ | |
2163 | static void validate_endpoint_status_data_stream(void) | |
2164 | { | |
2165 | struct lttng_ht_iter iter; | |
2166 | struct lttng_consumer_stream *stream; | |
2167 | ||
2168 | DBG("Consumer delete flagged data stream"); | |
2169 | ||
2170 | rcu_read_lock(); | |
2171 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
2172 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2173 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2174 | continue; |
2175 | } | |
2176 | /* Delete it right now */ | |
2177 | consumer_del_stream(stream, data_ht); | |
2178 | } | |
2179 | rcu_read_unlock(); | |
2180 | } | |
2181 | ||
2182 | /* | |
2183 | * Delete metadata stream that are flagged for deletion (endpoint_status). | |
2184 | */ | |
2185 | static void validate_endpoint_status_metadata_stream( | |
2186 | struct lttng_poll_event *pollset) | |
2187 | { | |
2188 | struct lttng_ht_iter iter; | |
2189 | struct lttng_consumer_stream *stream; | |
2190 | ||
2191 | DBG("Consumer delete flagged metadata stream"); | |
2192 | ||
2193 | assert(pollset); | |
2194 | ||
2195 | rcu_read_lock(); | |
2196 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
2197 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2198 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2199 | continue; |
2200 | } | |
2201 | /* | |
2202 | * Remove from pollset so the metadata thread can continue without | |
2203 | * blocking on a deleted stream. | |
2204 | */ | |
2205 | lttng_poll_del(pollset, stream->wait_fd); | |
2206 | ||
2207 | /* Delete it right now */ | |
2208 | consumer_del_metadata_stream(stream, metadata_ht); | |
2209 | } | |
2210 | rcu_read_unlock(); | |
2211 | } | |
2212 | ||
fb3a43a9 DG |
2213 | /* |
2214 | * Thread polls on metadata file descriptor and write them on disk or on the | |
2215 | * network. | |
2216 | */ | |
7d980def | 2217 | void *consumer_thread_metadata_poll(void *data) |
fb3a43a9 | 2218 | { |
1fc79fb4 | 2219 | int ret, i, pollfd, err = -1; |
fb3a43a9 | 2220 | uint32_t revents, nb_fd; |
e316aad5 | 2221 | struct lttng_consumer_stream *stream = NULL; |
fb3a43a9 | 2222 | struct lttng_ht_iter iter; |
d88aee68 | 2223 | struct lttng_ht_node_u64 *node; |
fb3a43a9 DG |
2224 | struct lttng_poll_event events; |
2225 | struct lttng_consumer_local_data *ctx = data; | |
2226 | ssize_t len; | |
2227 | ||
2228 | rcu_register_thread(); | |
2229 | ||
1fc79fb4 MD |
2230 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA); |
2231 | ||
2d57de81 MD |
2232 | if (testpoint(consumerd_thread_metadata)) { |
2233 | goto error_testpoint; | |
2234 | } | |
2235 | ||
9ce5646a MD |
2236 | health_code_update(); |
2237 | ||
fb3a43a9 DG |
2238 | DBG("Thread metadata poll started"); |
2239 | ||
fb3a43a9 DG |
2240 | /* Size is set to 1 for the consumer_metadata pipe */ |
2241 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2242 | if (ret < 0) { | |
2243 | ERR("Poll set creation failed"); | |
d8ef542d | 2244 | goto end_poll; |
fb3a43a9 DG |
2245 | } |
2246 | ||
13886d2d DG |
2247 | ret = lttng_poll_add(&events, |
2248 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN); | |
fb3a43a9 DG |
2249 | if (ret < 0) { |
2250 | goto end; | |
2251 | } | |
2252 | ||
2253 | /* Main loop */ | |
2254 | DBG("Metadata main loop started"); | |
2255 | ||
2256 | while (1) { | |
fb3a43a9 | 2257 | restart: |
7fa2082e | 2258 | health_code_update(); |
9ce5646a | 2259 | health_poll_entry(); |
7fa2082e | 2260 | DBG("Metadata poll wait"); |
fb3a43a9 | 2261 | ret = lttng_poll_wait(&events, -1); |
7fa2082e MD |
2262 | DBG("Metadata poll return from wait with %d fd(s)", |
2263 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2264 | health_poll_exit(); |
7e4ebd00 | 2265 | DBG("Metadata event caught in thread"); |
fb3a43a9 DG |
2266 | if (ret < 0) { |
2267 | if (errno == EINTR) { | |
7e4ebd00 | 2268 | ERR("Poll EINTR caught"); |
fb3a43a9 DG |
2269 | goto restart; |
2270 | } | |
d9607cd7 MD |
2271 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2272 | err = 0; /* All is OK */ | |
2273 | } | |
2274 | goto end; | |
fb3a43a9 DG |
2275 | } |
2276 | ||
0d9c5d77 DG |
2277 | nb_fd = ret; |
2278 | ||
e316aad5 | 2279 | /* From here, the event is a metadata wait fd */ |
fb3a43a9 | 2280 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2281 | health_code_update(); |
2282 | ||
fb3a43a9 DG |
2283 | revents = LTTNG_POLL_GETEV(&events, i); |
2284 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2285 | ||
fd20dac9 MD |
2286 | if (!revents) { |
2287 | /* No activity for this FD (poll implementation). */ | |
2288 | continue; | |
2289 | } | |
2290 | ||
13886d2d | 2291 | if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) { |
03e43155 | 2292 | if (revents & LPOLLIN) { |
13886d2d DG |
2293 | ssize_t pipe_len; |
2294 | ||
2295 | pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe, | |
2296 | &stream, sizeof(stream)); | |
6cd525e8 | 2297 | if (pipe_len < sizeof(stream)) { |
03e43155 MD |
2298 | if (pipe_len < 0) { |
2299 | PERROR("read metadata stream"); | |
2300 | } | |
fb3a43a9 | 2301 | /* |
03e43155 MD |
2302 | * Remove the pipe from the poll set and continue the loop |
2303 | * since their might be data to consume. | |
fb3a43a9 | 2304 | */ |
03e43155 MD |
2305 | lttng_poll_del(&events, |
2306 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2307 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
fb3a43a9 DG |
2308 | continue; |
2309 | } | |
2310 | ||
8994307f DG |
2311 | /* A NULL stream means that the state has changed. */ |
2312 | if (stream == NULL) { | |
2313 | /* Check for deleted streams. */ | |
2314 | validate_endpoint_status_metadata_stream(&events); | |
3714380f | 2315 | goto restart; |
8994307f DG |
2316 | } |
2317 | ||
fb3a43a9 DG |
2318 | DBG("Adding metadata stream %d to poll set", |
2319 | stream->wait_fd); | |
2320 | ||
fb3a43a9 DG |
2321 | /* Add metadata stream to the global poll events list */ |
2322 | lttng_poll_add(&events, stream->wait_fd, | |
6d574024 | 2323 | LPOLLIN | LPOLLPRI | LPOLLHUP); |
03e43155 MD |
2324 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
2325 | DBG("Metadata thread pipe hung up"); | |
2326 | /* | |
2327 | * Remove the pipe from the poll set and continue the loop | |
2328 | * since their might be data to consume. | |
2329 | */ | |
2330 | lttng_poll_del(&events, | |
2331 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2332 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
2333 | continue; | |
2334 | } else { | |
2335 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
2336 | goto end; | |
fb3a43a9 DG |
2337 | } |
2338 | ||
e316aad5 | 2339 | /* Handle other stream */ |
fb3a43a9 DG |
2340 | continue; |
2341 | } | |
2342 | ||
d09e1200 | 2343 | rcu_read_lock(); |
d88aee68 DG |
2344 | { |
2345 | uint64_t tmp_id = (uint64_t) pollfd; | |
2346 | ||
2347 | lttng_ht_lookup(metadata_ht, &tmp_id, &iter); | |
2348 | } | |
2349 | node = lttng_ht_iter_get_node_u64(&iter); | |
e316aad5 | 2350 | assert(node); |
fb3a43a9 DG |
2351 | |
2352 | stream = caa_container_of(node, struct lttng_consumer_stream, | |
58b1f425 | 2353 | node); |
fb3a43a9 | 2354 | |
03e43155 MD |
2355 | if (revents & (LPOLLIN | LPOLLPRI)) { |
2356 | /* Get the data out of the metadata file descriptor */ | |
2357 | DBG("Metadata available on fd %d", pollfd); | |
2358 | assert(stream->wait_fd == pollfd); | |
2359 | ||
2360 | do { | |
2361 | health_code_update(); | |
2362 | ||
2363 | len = ctx->on_buffer_ready(stream, ctx); | |
2364 | /* | |
2365 | * We don't check the return value here since if we get | |
fae1e3cc | 2366 | * a negative len, it means an error occurred thus we |
03e43155 MD |
2367 | * simply remove it from the poll set and free the |
2368 | * stream. | |
2369 | */ | |
2370 | } while (len > 0); | |
2371 | ||
2372 | /* It's ok to have an unavailable sub-buffer */ | |
2373 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { | |
2374 | /* Clean up stream from consumer and free it. */ | |
2375 | lttng_poll_del(&events, stream->wait_fd); | |
2376 | consumer_del_metadata_stream(stream, metadata_ht); | |
2377 | } | |
2378 | } else if (revents & (LPOLLERR | LPOLLHUP)) { | |
e316aad5 | 2379 | DBG("Metadata fd %d is hup|err.", pollfd); |
fb3a43a9 DG |
2380 | if (!stream->hangup_flush_done |
2381 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2382 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2383 | DBG("Attempting to flush and consume the UST buffers"); | |
2384 | lttng_ustconsumer_on_stream_hangup(stream); | |
2385 | ||
2386 | /* We just flushed the stream now read it. */ | |
4bb94b75 | 2387 | do { |
9ce5646a MD |
2388 | health_code_update(); |
2389 | ||
4bb94b75 DG |
2390 | len = ctx->on_buffer_ready(stream, ctx); |
2391 | /* | |
2392 | * We don't check the return value here since if we get | |
fae1e3cc | 2393 | * a negative len, it means an error occurred thus we |
4bb94b75 DG |
2394 | * simply remove it from the poll set and free the |
2395 | * stream. | |
2396 | */ | |
2397 | } while (len > 0); | |
fb3a43a9 DG |
2398 | } |
2399 | ||
fb3a43a9 | 2400 | lttng_poll_del(&events, stream->wait_fd); |
e316aad5 DG |
2401 | /* |
2402 | * This call update the channel states, closes file descriptors | |
2403 | * and securely free the stream. | |
2404 | */ | |
2405 | consumer_del_metadata_stream(stream, metadata_ht); | |
03e43155 MD |
2406 | } else { |
2407 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
6f2f1a70 | 2408 | rcu_read_unlock(); |
03e43155 | 2409 | goto end; |
fb3a43a9 | 2410 | } |
e316aad5 | 2411 | /* Release RCU lock for the stream looked up */ |
d09e1200 | 2412 | rcu_read_unlock(); |
fb3a43a9 DG |
2413 | } |
2414 | } | |
2415 | ||
1fc79fb4 MD |
2416 | /* All is OK */ |
2417 | err = 0; | |
fb3a43a9 DG |
2418 | end: |
2419 | DBG("Metadata poll thread exiting"); | |
fb3a43a9 | 2420 | |
d8ef542d MD |
2421 | lttng_poll_clean(&events); |
2422 | end_poll: | |
2d57de81 | 2423 | error_testpoint: |
1fc79fb4 MD |
2424 | if (err) { |
2425 | health_error(); | |
2426 | ERR("Health error occurred in %s", __func__); | |
2427 | } | |
2428 | health_unregister(health_consumerd); | |
fb3a43a9 DG |
2429 | rcu_unregister_thread(); |
2430 | return NULL; | |
2431 | } | |
2432 | ||
3bd1e081 | 2433 | /* |
e4421fec | 2434 | * This thread polls the fds in the set to consume the data and write |
3bd1e081 MD |
2435 | * it to tracefile if necessary. |
2436 | */ | |
7d980def | 2437 | void *consumer_thread_data_poll(void *data) |
3bd1e081 | 2438 | { |
1fc79fb4 | 2439 | int num_rdy, num_hup, high_prio, ret, i, err = -1; |
3bd1e081 MD |
2440 | struct pollfd *pollfd = NULL; |
2441 | /* local view of the streams */ | |
c869f647 | 2442 | struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; |
3bd1e081 MD |
2443 | /* local view of consumer_data.fds_count */ |
2444 | int nb_fd = 0; | |
3bd1e081 | 2445 | struct lttng_consumer_local_data *ctx = data; |
00e2e675 | 2446 | ssize_t len; |
3bd1e081 | 2447 | |
e7b994a3 DG |
2448 | rcu_register_thread(); |
2449 | ||
1fc79fb4 MD |
2450 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA); |
2451 | ||
2d57de81 MD |
2452 | if (testpoint(consumerd_thread_data)) { |
2453 | goto error_testpoint; | |
2454 | } | |
2455 | ||
9ce5646a MD |
2456 | health_code_update(); |
2457 | ||
4df6c8cb MD |
2458 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream *)); |
2459 | if (local_stream == NULL) { | |
2460 | PERROR("local_stream malloc"); | |
2461 | goto end; | |
2462 | } | |
3bd1e081 MD |
2463 | |
2464 | while (1) { | |
9ce5646a MD |
2465 | health_code_update(); |
2466 | ||
3bd1e081 MD |
2467 | high_prio = 0; |
2468 | num_hup = 0; | |
2469 | ||
2470 | /* | |
e4421fec | 2471 | * the fds set has been updated, we need to update our |
3bd1e081 MD |
2472 | * local array as well |
2473 | */ | |
2474 | pthread_mutex_lock(&consumer_data.lock); | |
2475 | if (consumer_data.need_update) { | |
0e428499 DG |
2476 | free(pollfd); |
2477 | pollfd = NULL; | |
2478 | ||
2479 | free(local_stream); | |
2480 | local_stream = NULL; | |
3bd1e081 | 2481 | |
02b3d176 DG |
2482 | /* |
2483 | * Allocate for all fds +1 for the consumer_data_pipe and +1 for | |
2484 | * wake up pipe. | |
2485 | */ | |
2486 | pollfd = zmalloc((consumer_data.stream_count + 2) * sizeof(struct pollfd)); | |
3bd1e081 | 2487 | if (pollfd == NULL) { |
7a57cf92 | 2488 | PERROR("pollfd malloc"); |
3bd1e081 MD |
2489 | pthread_mutex_unlock(&consumer_data.lock); |
2490 | goto end; | |
2491 | } | |
2492 | ||
02b3d176 | 2493 | local_stream = zmalloc((consumer_data.stream_count + 2) * |
747f8642 | 2494 | sizeof(struct lttng_consumer_stream *)); |
3bd1e081 | 2495 | if (local_stream == NULL) { |
7a57cf92 | 2496 | PERROR("local_stream malloc"); |
3bd1e081 MD |
2497 | pthread_mutex_unlock(&consumer_data.lock); |
2498 | goto end; | |
2499 | } | |
ffe60014 | 2500 | ret = update_poll_array(ctx, &pollfd, local_stream, |
43c34bc3 | 2501 | data_ht); |
3bd1e081 MD |
2502 | if (ret < 0) { |
2503 | ERR("Error in allocating pollfd or local_outfds"); | |
f73fabfd | 2504 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2505 | pthread_mutex_unlock(&consumer_data.lock); |
2506 | goto end; | |
2507 | } | |
2508 | nb_fd = ret; | |
2509 | consumer_data.need_update = 0; | |
2510 | } | |
2511 | pthread_mutex_unlock(&consumer_data.lock); | |
2512 | ||
4078b776 MD |
2513 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
2514 | if (nb_fd == 0 && consumer_quit == 1) { | |
1fc79fb4 | 2515 | err = 0; /* All is OK */ |
4078b776 MD |
2516 | goto end; |
2517 | } | |
3bd1e081 | 2518 | /* poll on the array of fds */ |
88f2b785 | 2519 | restart: |
02b3d176 | 2520 | DBG("polling on %d fd", nb_fd + 2); |
9ce5646a | 2521 | health_poll_entry(); |
02b3d176 | 2522 | num_rdy = poll(pollfd, nb_fd + 2, -1); |
9ce5646a | 2523 | health_poll_exit(); |
3bd1e081 MD |
2524 | DBG("poll num_rdy : %d", num_rdy); |
2525 | if (num_rdy == -1) { | |
88f2b785 MD |
2526 | /* |
2527 | * Restart interrupted system call. | |
2528 | */ | |
2529 | if (errno == EINTR) { | |
2530 | goto restart; | |
2531 | } | |
7a57cf92 | 2532 | PERROR("Poll error"); |
f73fabfd | 2533 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2534 | goto end; |
2535 | } else if (num_rdy == 0) { | |
2536 | DBG("Polling thread timed out"); | |
2537 | goto end; | |
2538 | } | |
2539 | ||
3bd1e081 | 2540 | /* |
50f8ae69 | 2541 | * If the consumer_data_pipe triggered poll go directly to the |
00e2e675 DG |
2542 | * beginning of the loop to update the array. We want to prioritize |
2543 | * array update over low-priority reads. | |
3bd1e081 | 2544 | */ |
509bb1cf | 2545 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
ab30f567 | 2546 | ssize_t pipe_readlen; |
04fdd819 | 2547 | |
50f8ae69 | 2548 | DBG("consumer_data_pipe wake up"); |
acdb9057 DG |
2549 | pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe, |
2550 | &new_stream, sizeof(new_stream)); | |
6cd525e8 MD |
2551 | if (pipe_readlen < sizeof(new_stream)) { |
2552 | PERROR("Consumer data pipe"); | |
23f5f35d DG |
2553 | /* Continue so we can at least handle the current stream(s). */ |
2554 | continue; | |
2555 | } | |
c869f647 DG |
2556 | |
2557 | /* | |
2558 | * If the stream is NULL, just ignore it. It's also possible that | |
2559 | * the sessiond poll thread changed the consumer_quit state and is | |
2560 | * waking us up to test it. | |
2561 | */ | |
2562 | if (new_stream == NULL) { | |
8994307f | 2563 | validate_endpoint_status_data_stream(); |
c869f647 DG |
2564 | continue; |
2565 | } | |
2566 | ||
c869f647 | 2567 | /* Continue to update the local streams and handle prio ones */ |
3bd1e081 MD |
2568 | continue; |
2569 | } | |
2570 | ||
02b3d176 DG |
2571 | /* Handle wakeup pipe. */ |
2572 | if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) { | |
2573 | char dummy; | |
2574 | ssize_t pipe_readlen; | |
2575 | ||
2576 | pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy, | |
2577 | sizeof(dummy)); | |
2578 | if (pipe_readlen < 0) { | |
2579 | PERROR("Consumer data wakeup pipe"); | |
2580 | } | |
2581 | /* We've been awakened to handle stream(s). */ | |
2582 | ctx->has_wakeup = 0; | |
2583 | } | |
2584 | ||
3bd1e081 MD |
2585 | /* Take care of high priority channels first. */ |
2586 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2587 | health_code_update(); |
2588 | ||
9617607b DG |
2589 | if (local_stream[i] == NULL) { |
2590 | continue; | |
2591 | } | |
fb3a43a9 | 2592 | if (pollfd[i].revents & POLLPRI) { |
d41f73b7 MD |
2593 | DBG("Urgent read on fd %d", pollfd[i].fd); |
2594 | high_prio = 1; | |
4078b776 | 2595 | len = ctx->on_buffer_ready(local_stream[i], ctx); |
d41f73b7 | 2596 | /* it's ok to have an unavailable sub-buffer */ |
b64403e3 | 2597 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2598 | /* Clean the stream and free it. */ |
2599 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2600 | local_stream[i] = NULL; |
4078b776 MD |
2601 | } else if (len > 0) { |
2602 | local_stream[i]->data_read = 1; | |
d41f73b7 | 2603 | } |
3bd1e081 MD |
2604 | } |
2605 | } | |
2606 | ||
4078b776 MD |
2607 | /* |
2608 | * If we read high prio channel in this loop, try again | |
2609 | * for more high prio data. | |
2610 | */ | |
2611 | if (high_prio) { | |
3bd1e081 MD |
2612 | continue; |
2613 | } | |
2614 | ||
2615 | /* Take care of low priority channels. */ | |
4078b776 | 2616 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2617 | health_code_update(); |
2618 | ||
9617607b DG |
2619 | if (local_stream[i] == NULL) { |
2620 | continue; | |
2621 | } | |
4078b776 | 2622 | if ((pollfd[i].revents & POLLIN) || |
02b3d176 DG |
2623 | local_stream[i]->hangup_flush_done || |
2624 | local_stream[i]->has_data) { | |
4078b776 MD |
2625 | DBG("Normal read on fd %d", pollfd[i].fd); |
2626 | len = ctx->on_buffer_ready(local_stream[i], ctx); | |
2627 | /* it's ok to have an unavailable sub-buffer */ | |
b64403e3 | 2628 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2629 | /* Clean the stream and free it. */ |
2630 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2631 | local_stream[i] = NULL; |
4078b776 MD |
2632 | } else if (len > 0) { |
2633 | local_stream[i]->data_read = 1; | |
2634 | } | |
2635 | } | |
2636 | } | |
2637 | ||
2638 | /* Handle hangup and errors */ | |
2639 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2640 | health_code_update(); |
2641 | ||
9617607b DG |
2642 | if (local_stream[i] == NULL) { |
2643 | continue; | |
2644 | } | |
4078b776 MD |
2645 | if (!local_stream[i]->hangup_flush_done |
2646 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) | |
2647 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2648 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2649 | DBG("fd %d is hup|err|nval. Attempting flush and read.", | |
9617607b | 2650 | pollfd[i].fd); |
4078b776 MD |
2651 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); |
2652 | /* Attempt read again, for the data we just flushed. */ | |
2653 | local_stream[i]->data_read = 1; | |
2654 | } | |
2655 | /* | |
2656 | * If the poll flag is HUP/ERR/NVAL and we have | |
2657 | * read no data in this pass, we can remove the | |
2658 | * stream from its hash table. | |
2659 | */ | |
2660 | if ((pollfd[i].revents & POLLHUP)) { | |
2661 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); | |
2662 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2663 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2664 | local_stream[i] = NULL; |
4078b776 MD |
2665 | num_hup++; |
2666 | } | |
2667 | } else if (pollfd[i].revents & POLLERR) { | |
2668 | ERR("Error returned in polling fd %d.", pollfd[i].fd); | |
2669 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2670 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2671 | local_stream[i] = NULL; |
4078b776 MD |
2672 | num_hup++; |
2673 | } | |
2674 | } else if (pollfd[i].revents & POLLNVAL) { | |
2675 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); | |
2676 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2677 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2678 | local_stream[i] = NULL; |
4078b776 | 2679 | num_hup++; |
3bd1e081 MD |
2680 | } |
2681 | } | |
9617607b DG |
2682 | if (local_stream[i] != NULL) { |
2683 | local_stream[i]->data_read = 0; | |
2684 | } | |
3bd1e081 MD |
2685 | } |
2686 | } | |
1fc79fb4 MD |
2687 | /* All is OK */ |
2688 | err = 0; | |
3bd1e081 MD |
2689 | end: |
2690 | DBG("polling thread exiting"); | |
0e428499 DG |
2691 | free(pollfd); |
2692 | free(local_stream); | |
fb3a43a9 DG |
2693 | |
2694 | /* | |
2695 | * Close the write side of the pipe so epoll_wait() in | |
7d980def DG |
2696 | * consumer_thread_metadata_poll can catch it. The thread is monitoring the |
2697 | * read side of the pipe. If we close them both, epoll_wait strangely does | |
2698 | * not return and could create a endless wait period if the pipe is the | |
2699 | * only tracked fd in the poll set. The thread will take care of closing | |
2700 | * the read side. | |
fb3a43a9 | 2701 | */ |
13886d2d | 2702 | (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe); |
fb3a43a9 | 2703 | |
2d57de81 | 2704 | error_testpoint: |
1fc79fb4 MD |
2705 | if (err) { |
2706 | health_error(); | |
2707 | ERR("Health error occurred in %s", __func__); | |
2708 | } | |
2709 | health_unregister(health_consumerd); | |
2710 | ||
e7b994a3 | 2711 | rcu_unregister_thread(); |
3bd1e081 MD |
2712 | return NULL; |
2713 | } | |
2714 | ||
d8ef542d MD |
2715 | /* |
2716 | * Close wake-up end of each stream belonging to the channel. This will | |
2717 | * allow the poll() on the stream read-side to detect when the | |
2718 | * write-side (application) finally closes them. | |
2719 | */ | |
2720 | static | |
2721 | void consumer_close_channel_streams(struct lttng_consumer_channel *channel) | |
2722 | { | |
2723 | struct lttng_ht *ht; | |
2724 | struct lttng_consumer_stream *stream; | |
2725 | struct lttng_ht_iter iter; | |
2726 | ||
2727 | ht = consumer_data.stream_per_chan_id_ht; | |
2728 | ||
2729 | rcu_read_lock(); | |
2730 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
2731 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
2732 | ht->match_fct, &channel->key, | |
2733 | &iter.iter, stream, node_channel_id.node) { | |
f2ad556d MD |
2734 | /* |
2735 | * Protect against teardown with mutex. | |
2736 | */ | |
2737 | pthread_mutex_lock(&stream->lock); | |
2738 | if (cds_lfht_is_node_deleted(&stream->node.node)) { | |
2739 | goto next; | |
2740 | } | |
d8ef542d MD |
2741 | switch (consumer_data.type) { |
2742 | case LTTNG_CONSUMER_KERNEL: | |
2743 | break; | |
2744 | case LTTNG_CONSUMER32_UST: | |
2745 | case LTTNG_CONSUMER64_UST: | |
b4a650f3 DG |
2746 | if (stream->metadata_flag) { |
2747 | /* Safe and protected by the stream lock. */ | |
2748 | lttng_ustconsumer_close_metadata(stream->chan); | |
2749 | } else { | |
2750 | /* | |
2751 | * Note: a mutex is taken internally within | |
2752 | * liblttng-ust-ctl to protect timer wakeup_fd | |
2753 | * use from concurrent close. | |
2754 | */ | |
2755 | lttng_ustconsumer_close_stream_wakeup(stream); | |
2756 | } | |
d8ef542d MD |
2757 | break; |
2758 | default: | |
2759 | ERR("Unknown consumer_data type"); | |
2760 | assert(0); | |
2761 | } | |
f2ad556d MD |
2762 | next: |
2763 | pthread_mutex_unlock(&stream->lock); | |
d8ef542d MD |
2764 | } |
2765 | rcu_read_unlock(); | |
2766 | } | |
2767 | ||
2768 | static void destroy_channel_ht(struct lttng_ht *ht) | |
2769 | { | |
2770 | struct lttng_ht_iter iter; | |
2771 | struct lttng_consumer_channel *channel; | |
2772 | int ret; | |
2773 | ||
2774 | if (ht == NULL) { | |
2775 | return; | |
2776 | } | |
2777 | ||
2778 | rcu_read_lock(); | |
2779 | cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) { | |
2780 | ret = lttng_ht_del(ht, &iter); | |
2781 | assert(ret != 0); | |
2782 | } | |
2783 | rcu_read_unlock(); | |
2784 | ||
2785 | lttng_ht_destroy(ht); | |
2786 | } | |
2787 | ||
2788 | /* | |
2789 | * This thread polls the channel fds to detect when they are being | |
2790 | * closed. It closes all related streams if the channel is detected as | |
2791 | * closed. It is currently only used as a shim layer for UST because the | |
2792 | * consumerd needs to keep the per-stream wakeup end of pipes open for | |
2793 | * periodical flush. | |
2794 | */ | |
2795 | void *consumer_thread_channel_poll(void *data) | |
2796 | { | |
1fc79fb4 | 2797 | int ret, i, pollfd, err = -1; |
d8ef542d MD |
2798 | uint32_t revents, nb_fd; |
2799 | struct lttng_consumer_channel *chan = NULL; | |
2800 | struct lttng_ht_iter iter; | |
2801 | struct lttng_ht_node_u64 *node; | |
2802 | struct lttng_poll_event events; | |
2803 | struct lttng_consumer_local_data *ctx = data; | |
2804 | struct lttng_ht *channel_ht; | |
2805 | ||
2806 | rcu_register_thread(); | |
2807 | ||
1fc79fb4 MD |
2808 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL); |
2809 | ||
2d57de81 MD |
2810 | if (testpoint(consumerd_thread_channel)) { |
2811 | goto error_testpoint; | |
2812 | } | |
2813 | ||
9ce5646a MD |
2814 | health_code_update(); |
2815 | ||
d8ef542d MD |
2816 | channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
2817 | if (!channel_ht) { | |
2818 | /* ENOMEM at this point. Better to bail out. */ | |
2819 | goto end_ht; | |
2820 | } | |
2821 | ||
2822 | DBG("Thread channel poll started"); | |
2823 | ||
2824 | /* Size is set to 1 for the consumer_channel pipe */ | |
2825 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2826 | if (ret < 0) { | |
2827 | ERR("Poll set creation failed"); | |
2828 | goto end_poll; | |
2829 | } | |
2830 | ||
2831 | ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN); | |
2832 | if (ret < 0) { | |
2833 | goto end; | |
2834 | } | |
2835 | ||
2836 | /* Main loop */ | |
2837 | DBG("Channel main loop started"); | |
2838 | ||
2839 | while (1) { | |
d8ef542d | 2840 | restart: |
7fa2082e MD |
2841 | health_code_update(); |
2842 | DBG("Channel poll wait"); | |
9ce5646a | 2843 | health_poll_entry(); |
d8ef542d | 2844 | ret = lttng_poll_wait(&events, -1); |
7fa2082e MD |
2845 | DBG("Channel poll return from wait with %d fd(s)", |
2846 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2847 | health_poll_exit(); |
7e4ebd00 | 2848 | DBG("Channel event caught in thread"); |
d8ef542d MD |
2849 | if (ret < 0) { |
2850 | if (errno == EINTR) { | |
7e4ebd00 | 2851 | ERR("Poll EINTR caught"); |
d8ef542d MD |
2852 | goto restart; |
2853 | } | |
d9607cd7 MD |
2854 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2855 | err = 0; /* All is OK */ | |
2856 | } | |
d8ef542d MD |
2857 | goto end; |
2858 | } | |
2859 | ||
2860 | nb_fd = ret; | |
2861 | ||
2862 | /* From here, the event is a channel wait fd */ | |
2863 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2864 | health_code_update(); |
2865 | ||
d8ef542d MD |
2866 | revents = LTTNG_POLL_GETEV(&events, i); |
2867 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2868 | ||
d8ef542d | 2869 | if (!revents) { |
fd20dac9 | 2870 | /* No activity for this FD (poll implementation). */ |
d8ef542d MD |
2871 | continue; |
2872 | } | |
fd20dac9 | 2873 | |
d8ef542d | 2874 | if (pollfd == ctx->consumer_channel_pipe[0]) { |
03e43155 | 2875 | if (revents & LPOLLIN) { |
d8ef542d | 2876 | enum consumer_channel_action action; |
a0cbdd2e | 2877 | uint64_t key; |
d8ef542d | 2878 | |
a0cbdd2e | 2879 | ret = read_channel_pipe(ctx, &chan, &key, &action); |
d8ef542d | 2880 | if (ret <= 0) { |
03e43155 MD |
2881 | if (ret < 0) { |
2882 | ERR("Error reading channel pipe"); | |
2883 | } | |
2884 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
d8ef542d MD |
2885 | continue; |
2886 | } | |
2887 | ||
2888 | switch (action) { | |
2889 | case CONSUMER_CHANNEL_ADD: | |
2890 | DBG("Adding channel %d to poll set", | |
2891 | chan->wait_fd); | |
2892 | ||
2893 | lttng_ht_node_init_u64(&chan->wait_fd_node, | |
2894 | chan->wait_fd); | |
c7260a81 | 2895 | rcu_read_lock(); |
d8ef542d MD |
2896 | lttng_ht_add_unique_u64(channel_ht, |
2897 | &chan->wait_fd_node); | |
c7260a81 | 2898 | rcu_read_unlock(); |
d8ef542d MD |
2899 | /* Add channel to the global poll events list */ |
2900 | lttng_poll_add(&events, chan->wait_fd, | |
03e43155 | 2901 | LPOLLERR | LPOLLHUP); |
d8ef542d | 2902 | break; |
a0cbdd2e MD |
2903 | case CONSUMER_CHANNEL_DEL: |
2904 | { | |
b4a650f3 DG |
2905 | /* |
2906 | * This command should never be called if the channel | |
2907 | * has streams monitored by either the data or metadata | |
2908 | * thread. The consumer only notify this thread with a | |
2909 | * channel del. command if it receives a destroy | |
2910 | * channel command from the session daemon that send it | |
2911 | * if a command prior to the GET_CHANNEL failed. | |
2912 | */ | |
2913 | ||
c7260a81 | 2914 | rcu_read_lock(); |
a0cbdd2e MD |
2915 | chan = consumer_find_channel(key); |
2916 | if (!chan) { | |
c7260a81 | 2917 | rcu_read_unlock(); |
a0cbdd2e MD |
2918 | ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key); |
2919 | break; | |
2920 | } | |
2921 | lttng_poll_del(&events, chan->wait_fd); | |
f623cc0b | 2922 | iter.iter.node = &chan->wait_fd_node.node; |
a0cbdd2e MD |
2923 | ret = lttng_ht_del(channel_ht, &iter); |
2924 | assert(ret == 0); | |
a0cbdd2e | 2925 | |
f2a444f1 DG |
2926 | switch (consumer_data.type) { |
2927 | case LTTNG_CONSUMER_KERNEL: | |
2928 | break; | |
2929 | case LTTNG_CONSUMER32_UST: | |
2930 | case LTTNG_CONSUMER64_UST: | |
212d67a2 DG |
2931 | health_code_update(); |
2932 | /* Destroy streams that might have been left in the stream list. */ | |
2933 | clean_channel_stream_list(chan); | |
f2a444f1 DG |
2934 | break; |
2935 | default: | |
2936 | ERR("Unknown consumer_data type"); | |
2937 | assert(0); | |
2938 | } | |
2939 | ||
a0cbdd2e MD |
2940 | /* |
2941 | * Release our own refcount. Force channel deletion even if | |
2942 | * streams were not initialized. | |
2943 | */ | |
2944 | if (!uatomic_sub_return(&chan->refcount, 1)) { | |
2945 | consumer_del_channel(chan); | |
2946 | } | |
c7260a81 | 2947 | rcu_read_unlock(); |
a0cbdd2e MD |
2948 | goto restart; |
2949 | } | |
d8ef542d MD |
2950 | case CONSUMER_CHANNEL_QUIT: |
2951 | /* | |
2952 | * Remove the pipe from the poll set and continue the loop | |
2953 | * since their might be data to consume. | |
2954 | */ | |
2955 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
2956 | continue; | |
2957 | default: | |
2958 | ERR("Unknown action"); | |
2959 | break; | |
2960 | } | |
03e43155 MD |
2961 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
2962 | DBG("Channel thread pipe hung up"); | |
2963 | /* | |
2964 | * Remove the pipe from the poll set and continue the loop | |
2965 | * since their might be data to consume. | |
2966 | */ | |
2967 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
2968 | continue; | |
2969 | } else { | |
2970 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
2971 | goto end; | |
d8ef542d MD |
2972 | } |
2973 | ||
2974 | /* Handle other stream */ | |
2975 | continue; | |
2976 | } | |
2977 | ||
2978 | rcu_read_lock(); | |
2979 | { | |
2980 | uint64_t tmp_id = (uint64_t) pollfd; | |
2981 | ||
2982 | lttng_ht_lookup(channel_ht, &tmp_id, &iter); | |
2983 | } | |
2984 | node = lttng_ht_iter_get_node_u64(&iter); | |
2985 | assert(node); | |
2986 | ||
2987 | chan = caa_container_of(node, struct lttng_consumer_channel, | |
2988 | wait_fd_node); | |
2989 | ||
2990 | /* Check for error event */ | |
2991 | if (revents & (LPOLLERR | LPOLLHUP)) { | |
2992 | DBG("Channel fd %d is hup|err.", pollfd); | |
2993 | ||
2994 | lttng_poll_del(&events, chan->wait_fd); | |
2995 | ret = lttng_ht_del(channel_ht, &iter); | |
2996 | assert(ret == 0); | |
b4a650f3 DG |
2997 | |
2998 | /* | |
2999 | * This will close the wait fd for each stream associated to | |
3000 | * this channel AND monitored by the data/metadata thread thus | |
3001 | * will be clean by the right thread. | |
3002 | */ | |
d8ef542d | 3003 | consumer_close_channel_streams(chan); |
f2ad556d MD |
3004 | |
3005 | /* Release our own refcount */ | |
3006 | if (!uatomic_sub_return(&chan->refcount, 1) | |
3007 | && !uatomic_read(&chan->nb_init_stream_left)) { | |
3008 | consumer_del_channel(chan); | |
3009 | } | |
03e43155 MD |
3010 | } else { |
3011 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
3012 | rcu_read_unlock(); | |
3013 | goto end; | |
d8ef542d MD |
3014 | } |
3015 | ||
3016 | /* Release RCU lock for the channel looked up */ | |
3017 | rcu_read_unlock(); | |
3018 | } | |
3019 | } | |
3020 | ||
1fc79fb4 MD |
3021 | /* All is OK */ |
3022 | err = 0; | |
d8ef542d MD |
3023 | end: |
3024 | lttng_poll_clean(&events); | |
3025 | end_poll: | |
3026 | destroy_channel_ht(channel_ht); | |
3027 | end_ht: | |
2d57de81 | 3028 | error_testpoint: |
d8ef542d | 3029 | DBG("Channel poll thread exiting"); |
1fc79fb4 MD |
3030 | if (err) { |
3031 | health_error(); | |
3032 | ERR("Health error occurred in %s", __func__); | |
3033 | } | |
3034 | health_unregister(health_consumerd); | |
d8ef542d MD |
3035 | rcu_unregister_thread(); |
3036 | return NULL; | |
3037 | } | |
3038 | ||
331744e3 JD |
3039 | static int set_metadata_socket(struct lttng_consumer_local_data *ctx, |
3040 | struct pollfd *sockpoll, int client_socket) | |
3041 | { | |
3042 | int ret; | |
3043 | ||
3044 | assert(ctx); | |
3045 | assert(sockpoll); | |
3046 | ||
84382d49 MD |
3047 | ret = lttng_consumer_poll_socket(sockpoll); |
3048 | if (ret) { | |
331744e3 JD |
3049 | goto error; |
3050 | } | |
3051 | DBG("Metadata connection on client_socket"); | |
3052 | ||
3053 | /* Blocking call, waiting for transmission */ | |
3054 | ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket); | |
3055 | if (ctx->consumer_metadata_socket < 0) { | |
3056 | WARN("On accept metadata"); | |
3057 | ret = -1; | |
3058 | goto error; | |
3059 | } | |
3060 | ret = 0; | |
3061 | ||
3062 | error: | |
3063 | return ret; | |
3064 | } | |
3065 | ||
3bd1e081 MD |
3066 | /* |
3067 | * This thread listens on the consumerd socket and receives the file | |
3068 | * descriptors from the session daemon. | |
3069 | */ | |
7d980def | 3070 | void *consumer_thread_sessiond_poll(void *data) |
3bd1e081 | 3071 | { |
1fc79fb4 | 3072 | int sock = -1, client_socket, ret, err = -1; |
3bd1e081 MD |
3073 | /* |
3074 | * structure to poll for incoming data on communication socket avoids | |
3075 | * making blocking sockets. | |
3076 | */ | |
3077 | struct pollfd consumer_sockpoll[2]; | |
3078 | struct lttng_consumer_local_data *ctx = data; | |
3079 | ||
e7b994a3 DG |
3080 | rcu_register_thread(); |
3081 | ||
1fc79fb4 MD |
3082 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND); |
3083 | ||
2d57de81 MD |
3084 | if (testpoint(consumerd_thread_sessiond)) { |
3085 | goto error_testpoint; | |
3086 | } | |
3087 | ||
9ce5646a MD |
3088 | health_code_update(); |
3089 | ||
3bd1e081 MD |
3090 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
3091 | unlink(ctx->consumer_command_sock_path); | |
3092 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); | |
3093 | if (client_socket < 0) { | |
3094 | ERR("Cannot create command socket"); | |
3095 | goto end; | |
3096 | } | |
3097 | ||
3098 | ret = lttcomm_listen_unix_sock(client_socket); | |
3099 | if (ret < 0) { | |
3100 | goto end; | |
3101 | } | |
3102 | ||
32258573 | 3103 | DBG("Sending ready command to lttng-sessiond"); |
f73fabfd | 3104 | ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); |
3bd1e081 MD |
3105 | /* return < 0 on error, but == 0 is not fatal */ |
3106 | if (ret < 0) { | |
32258573 | 3107 | ERR("Error sending ready command to lttng-sessiond"); |
3bd1e081 MD |
3108 | goto end; |
3109 | } | |
3110 | ||
3bd1e081 MD |
3111 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ |
3112 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; | |
3113 | consumer_sockpoll[0].events = POLLIN | POLLPRI; | |
3114 | consumer_sockpoll[1].fd = client_socket; | |
3115 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3116 | ||
84382d49 MD |
3117 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3118 | if (ret) { | |
3119 | if (ret > 0) { | |
3120 | /* should exit */ | |
3121 | err = 0; | |
3122 | } | |
3bd1e081 MD |
3123 | goto end; |
3124 | } | |
3125 | DBG("Connection on client_socket"); | |
3126 | ||
3127 | /* Blocking call, waiting for transmission */ | |
3128 | sock = lttcomm_accept_unix_sock(client_socket); | |
534d2592 | 3129 | if (sock < 0) { |
3bd1e081 MD |
3130 | WARN("On accept"); |
3131 | goto end; | |
3132 | } | |
3bd1e081 | 3133 | |
331744e3 JD |
3134 | /* |
3135 | * Setup metadata socket which is the second socket connection on the | |
3136 | * command unix socket. | |
3137 | */ | |
3138 | ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket); | |
84382d49 MD |
3139 | if (ret) { |
3140 | if (ret > 0) { | |
3141 | /* should exit */ | |
3142 | err = 0; | |
3143 | } | |
331744e3 JD |
3144 | goto end; |
3145 | } | |
3146 | ||
d96f09c6 DG |
3147 | /* This socket is not useful anymore. */ |
3148 | ret = close(client_socket); | |
3149 | if (ret < 0) { | |
3150 | PERROR("close client_socket"); | |
3151 | } | |
3152 | client_socket = -1; | |
3153 | ||
3bd1e081 MD |
3154 | /* update the polling structure to poll on the established socket */ |
3155 | consumer_sockpoll[1].fd = sock; | |
3156 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3157 | ||
3158 | while (1) { | |
9ce5646a MD |
3159 | health_code_update(); |
3160 | ||
3161 | health_poll_entry(); | |
3162 | ret = lttng_consumer_poll_socket(consumer_sockpoll); | |
3163 | health_poll_exit(); | |
84382d49 MD |
3164 | if (ret) { |
3165 | if (ret > 0) { | |
3166 | /* should exit */ | |
3167 | err = 0; | |
3168 | } | |
3bd1e081 MD |
3169 | goto end; |
3170 | } | |
3171 | DBG("Incoming command on sock"); | |
3172 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
4cbc1a04 DG |
3173 | if (ret <= 0) { |
3174 | /* | |
3175 | * This could simply be a session daemon quitting. Don't output | |
3176 | * ERR() here. | |
3177 | */ | |
3178 | DBG("Communication interrupted on command socket"); | |
41ba6035 | 3179 | err = 0; |
3bd1e081 MD |
3180 | goto end; |
3181 | } | |
3182 | if (consumer_quit) { | |
3183 | DBG("consumer_thread_receive_fds received quit from signal"); | |
1fc79fb4 | 3184 | err = 0; /* All is OK */ |
3bd1e081 MD |
3185 | goto end; |
3186 | } | |
ffe60014 | 3187 | DBG("received command on sock"); |
3bd1e081 | 3188 | } |
1fc79fb4 MD |
3189 | /* All is OK */ |
3190 | err = 0; | |
3191 | ||
3bd1e081 | 3192 | end: |
ffe60014 | 3193 | DBG("Consumer thread sessiond poll exiting"); |
3bd1e081 | 3194 | |
d88aee68 DG |
3195 | /* |
3196 | * Close metadata streams since the producer is the session daemon which | |
3197 | * just died. | |
3198 | * | |
3199 | * NOTE: for now, this only applies to the UST tracer. | |
3200 | */ | |
6d574024 | 3201 | lttng_consumer_close_all_metadata(); |
d88aee68 | 3202 | |
3bd1e081 MD |
3203 | /* |
3204 | * when all fds have hung up, the polling thread | |
3205 | * can exit cleanly | |
3206 | */ | |
3207 | consumer_quit = 1; | |
3208 | ||
04fdd819 | 3209 | /* |
c869f647 | 3210 | * Notify the data poll thread to poll back again and test the |
8994307f | 3211 | * consumer_quit state that we just set so to quit gracefully. |
04fdd819 | 3212 | */ |
acdb9057 | 3213 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
c869f647 | 3214 | |
a0cbdd2e | 3215 | notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT); |
d8ef542d | 3216 | |
5c635c72 MD |
3217 | notify_health_quit_pipe(health_quit_pipe); |
3218 | ||
d96f09c6 DG |
3219 | /* Cleaning up possibly open sockets. */ |
3220 | if (sock >= 0) { | |
3221 | ret = close(sock); | |
3222 | if (ret < 0) { | |
3223 | PERROR("close sock sessiond poll"); | |
3224 | } | |
3225 | } | |
3226 | if (client_socket >= 0) { | |
38476d24 | 3227 | ret = close(client_socket); |
d96f09c6 DG |
3228 | if (ret < 0) { |
3229 | PERROR("close client_socket sessiond poll"); | |
3230 | } | |
3231 | } | |
3232 | ||
2d57de81 | 3233 | error_testpoint: |
1fc79fb4 MD |
3234 | if (err) { |
3235 | health_error(); | |
3236 | ERR("Health error occurred in %s", __func__); | |
3237 | } | |
3238 | health_unregister(health_consumerd); | |
3239 | ||
e7b994a3 | 3240 | rcu_unregister_thread(); |
3bd1e081 MD |
3241 | return NULL; |
3242 | } | |
d41f73b7 | 3243 | |
4078b776 | 3244 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
d41f73b7 MD |
3245 | struct lttng_consumer_local_data *ctx) |
3246 | { | |
74251bb8 DG |
3247 | ssize_t ret; |
3248 | ||
3249 | pthread_mutex_lock(&stream->lock); | |
94d49140 JD |
3250 | if (stream->metadata_flag) { |
3251 | pthread_mutex_lock(&stream->metadata_rdv_lock); | |
3252 | } | |
74251bb8 | 3253 | |
d41f73b7 MD |
3254 | switch (consumer_data.type) { |
3255 | case LTTNG_CONSUMER_KERNEL: | |
74251bb8 DG |
3256 | ret = lttng_kconsumer_read_subbuffer(stream, ctx); |
3257 | break; | |
7753dea8 MD |
3258 | case LTTNG_CONSUMER32_UST: |
3259 | case LTTNG_CONSUMER64_UST: | |
74251bb8 DG |
3260 | ret = lttng_ustconsumer_read_subbuffer(stream, ctx); |
3261 | break; | |
d41f73b7 MD |
3262 | default: |
3263 | ERR("Unknown consumer_data type"); | |
3264 | assert(0); | |
74251bb8 DG |
3265 | ret = -ENOSYS; |
3266 | break; | |
d41f73b7 | 3267 | } |
74251bb8 | 3268 | |
94d49140 JD |
3269 | if (stream->metadata_flag) { |
3270 | pthread_cond_broadcast(&stream->metadata_rdv); | |
3271 | pthread_mutex_unlock(&stream->metadata_rdv_lock); | |
3272 | } | |
74251bb8 DG |
3273 | pthread_mutex_unlock(&stream->lock); |
3274 | return ret; | |
d41f73b7 MD |
3275 | } |
3276 | ||
3277 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
3278 | { | |
3279 | switch (consumer_data.type) { | |
3280 | case LTTNG_CONSUMER_KERNEL: | |
3281 | return lttng_kconsumer_on_recv_stream(stream); | |
7753dea8 MD |
3282 | case LTTNG_CONSUMER32_UST: |
3283 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
3284 | return lttng_ustconsumer_on_recv_stream(stream); |
3285 | default: | |
3286 | ERR("Unknown consumer_data type"); | |
3287 | assert(0); | |
3288 | return -ENOSYS; | |
3289 | } | |
3290 | } | |
e4421fec DG |
3291 | |
3292 | /* | |
3293 | * Allocate and set consumer data hash tables. | |
3294 | */ | |
282dadbc | 3295 | int lttng_consumer_init(void) |
e4421fec | 3296 | { |
d88aee68 | 3297 | consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3298 | if (!consumer_data.channel_ht) { |
3299 | goto error; | |
3300 | } | |
3301 | ||
d88aee68 | 3302 | consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3303 | if (!consumer_data.relayd_ht) { |
3304 | goto error; | |
3305 | } | |
3306 | ||
d88aee68 | 3307 | consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3308 | if (!consumer_data.stream_list_ht) { |
3309 | goto error; | |
3310 | } | |
3311 | ||
d8ef542d | 3312 | consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3313 | if (!consumer_data.stream_per_chan_id_ht) { |
3314 | goto error; | |
3315 | } | |
3316 | ||
3317 | data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3318 | if (!data_ht) { | |
3319 | goto error; | |
3320 | } | |
3321 | ||
3322 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3323 | if (!metadata_ht) { | |
3324 | goto error; | |
3325 | } | |
3326 | ||
3327 | return 0; | |
3328 | ||
3329 | error: | |
3330 | return -1; | |
e4421fec | 3331 | } |
7735ef9e DG |
3332 | |
3333 | /* | |
3334 | * Process the ADD_RELAYD command receive by a consumer. | |
3335 | * | |
3336 | * This will create a relayd socket pair and add it to the relayd hash table. | |
3337 | * The caller MUST acquire a RCU read side lock before calling it. | |
3338 | */ | |
da009f2c | 3339 | int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type, |
7735ef9e | 3340 | struct lttng_consumer_local_data *ctx, int sock, |
6151a90f | 3341 | struct pollfd *consumer_sockpoll, |
d3e2ba59 JD |
3342 | struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id, |
3343 | uint64_t relayd_session_id) | |
7735ef9e | 3344 | { |
cd2b09ed | 3345 | int fd = -1, ret = -1, relayd_created = 0; |
0c759fc9 | 3346 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
d4298c99 | 3347 | struct consumer_relayd_sock_pair *relayd = NULL; |
7735ef9e | 3348 | |
6151a90f JD |
3349 | assert(ctx); |
3350 | assert(relayd_sock); | |
3351 | ||
da009f2c | 3352 | DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx); |
7735ef9e DG |
3353 | |
3354 | /* Get relayd reference if exists. */ | |
3355 | relayd = consumer_find_relayd(net_seq_idx); | |
3356 | if (relayd == NULL) { | |
da009f2c | 3357 | assert(sock_type == LTTNG_STREAM_CONTROL); |
7735ef9e DG |
3358 | /* Not found. Allocate one. */ |
3359 | relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); | |
3360 | if (relayd == NULL) { | |
0d08d75e | 3361 | ret = -ENOMEM; |
618a6a28 MD |
3362 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
3363 | goto error; | |
0d08d75e | 3364 | } else { |
30319bcb | 3365 | relayd->sessiond_session_id = sessiond_id; |
0d08d75e | 3366 | relayd_created = 1; |
7735ef9e | 3367 | } |
0d08d75e DG |
3368 | |
3369 | /* | |
3370 | * This code path MUST continue to the consumer send status message to | |
3371 | * we can notify the session daemon and continue our work without | |
3372 | * killing everything. | |
3373 | */ | |
da009f2c MD |
3374 | } else { |
3375 | /* | |
3376 | * relayd key should never be found for control socket. | |
3377 | */ | |
3378 | assert(sock_type != LTTNG_STREAM_CONTROL); | |
0d08d75e DG |
3379 | } |
3380 | ||
3381 | /* First send a status message before receiving the fds. */ | |
0c759fc9 | 3382 | ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS); |
618a6a28 | 3383 | if (ret < 0) { |
0d08d75e | 3384 | /* Somehow, the session daemon is not responding anymore. */ |
618a6a28 MD |
3385 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); |
3386 | goto error_nosignal; | |
7735ef9e DG |
3387 | } |
3388 | ||
3389 | /* Poll on consumer socket. */ | |
84382d49 MD |
3390 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3391 | if (ret) { | |
3392 | /* Needing to exit in the middle of a command: error. */ | |
0d08d75e | 3393 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
7735ef9e | 3394 | ret = -EINTR; |
618a6a28 | 3395 | goto error_nosignal; |
7735ef9e DG |
3396 | } |
3397 | ||
3398 | /* Get relayd socket from session daemon */ | |
3399 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); | |
3400 | if (ret != sizeof(fd)) { | |
7735ef9e | 3401 | ret = -1; |
4028eeb9 | 3402 | fd = -1; /* Just in case it gets set with an invalid value. */ |
0d08d75e DG |
3403 | |
3404 | /* | |
3405 | * Failing to receive FDs might indicate a major problem such as | |
3406 | * reaching a fd limit during the receive where the kernel returns a | |
3407 | * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we | |
3408 | * don't take any chances and stop everything. | |
3409 | * | |
3410 | * XXX: Feature request #558 will fix that and avoid this possible | |
3411 | * issue when reaching the fd limit. | |
3412 | */ | |
3413 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); | |
618a6a28 | 3414 | ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD; |
f50f23d9 DG |
3415 | goto error; |
3416 | } | |
3417 | ||
7735ef9e DG |
3418 | /* Copy socket information and received FD */ |
3419 | switch (sock_type) { | |
3420 | case LTTNG_STREAM_CONTROL: | |
3421 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3422 | lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock); |
3423 | ret = lttcomm_create_sock(&relayd->control_sock.sock); | |
4028eeb9 | 3424 | /* Handle create_sock error. */ |
f66c074c | 3425 | if (ret < 0) { |
618a6a28 | 3426 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3427 | goto error; |
f66c074c | 3428 | } |
da009f2c MD |
3429 | /* |
3430 | * Close the socket created internally by | |
3431 | * lttcomm_create_sock, so we can replace it by the one | |
3432 | * received from sessiond. | |
3433 | */ | |
3434 | if (close(relayd->control_sock.sock.fd)) { | |
3435 | PERROR("close"); | |
3436 | } | |
7735ef9e DG |
3437 | |
3438 | /* Assign new file descriptor */ | |
6151a90f | 3439 | relayd->control_sock.sock.fd = fd; |
4b29f1ce | 3440 | fd = -1; /* For error path */ |
6151a90f JD |
3441 | /* Assign version values. */ |
3442 | relayd->control_sock.major = relayd_sock->major; | |
3443 | relayd->control_sock.minor = relayd_sock->minor; | |
c5b6f4f0 | 3444 | |
d3e2ba59 | 3445 | relayd->relayd_session_id = relayd_session_id; |
c5b6f4f0 | 3446 | |
7735ef9e DG |
3447 | break; |
3448 | case LTTNG_STREAM_DATA: | |
3449 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3450 | lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock); |
3451 | ret = lttcomm_create_sock(&relayd->data_sock.sock); | |
4028eeb9 | 3452 | /* Handle create_sock error. */ |
f66c074c | 3453 | if (ret < 0) { |
618a6a28 | 3454 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3455 | goto error; |
f66c074c | 3456 | } |
da009f2c MD |
3457 | /* |
3458 | * Close the socket created internally by | |
3459 | * lttcomm_create_sock, so we can replace it by the one | |
3460 | * received from sessiond. | |
3461 | */ | |
3462 | if (close(relayd->data_sock.sock.fd)) { | |
3463 | PERROR("close"); | |
3464 | } | |
7735ef9e DG |
3465 | |
3466 | /* Assign new file descriptor */ | |
6151a90f | 3467 | relayd->data_sock.sock.fd = fd; |
4b29f1ce | 3468 | fd = -1; /* for eventual error paths */ |
6151a90f JD |
3469 | /* Assign version values. */ |
3470 | relayd->data_sock.major = relayd_sock->major; | |
3471 | relayd->data_sock.minor = relayd_sock->minor; | |
7735ef9e DG |
3472 | break; |
3473 | default: | |
3474 | ERR("Unknown relayd socket type (%d)", sock_type); | |
59e71485 | 3475 | ret = -1; |
618a6a28 | 3476 | ret_code = LTTCOMM_CONSUMERD_FATAL; |
7735ef9e DG |
3477 | goto error; |
3478 | } | |
3479 | ||
d88aee68 | 3480 | DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)", |
7735ef9e DG |
3481 | sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", |
3482 | relayd->net_seq_idx, fd); | |
3483 | ||
618a6a28 MD |
3484 | /* We successfully added the socket. Send status back. */ |
3485 | ret = consumer_send_status_msg(sock, ret_code); | |
3486 | if (ret < 0) { | |
3487 | /* Somehow, the session daemon is not responding anymore. */ | |
3488 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3489 | goto error_nosignal; | |
3490 | } | |
3491 | ||
7735ef9e DG |
3492 | /* |
3493 | * Add relayd socket pair to consumer data hashtable. If object already | |
3494 | * exists or on error, the function gracefully returns. | |
3495 | */ | |
d09e1200 | 3496 | add_relayd(relayd); |
7735ef9e DG |
3497 | |
3498 | /* All good! */ | |
4028eeb9 | 3499 | return 0; |
7735ef9e DG |
3500 | |
3501 | error: | |
618a6a28 MD |
3502 | if (consumer_send_status_msg(sock, ret_code) < 0) { |
3503 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3504 | } | |
3505 | ||
3506 | error_nosignal: | |
4028eeb9 DG |
3507 | /* Close received socket if valid. */ |
3508 | if (fd >= 0) { | |
3509 | if (close(fd)) { | |
3510 | PERROR("close received socket"); | |
3511 | } | |
3512 | } | |
cd2b09ed DG |
3513 | |
3514 | if (relayd_created) { | |
cd2b09ed DG |
3515 | free(relayd); |
3516 | } | |
3517 | ||
7735ef9e DG |
3518 | return ret; |
3519 | } | |
ca22feea | 3520 | |
4e9a4686 DG |
3521 | /* |
3522 | * Try to lock the stream mutex. | |
3523 | * | |
3524 | * On success, 1 is returned else 0 indicating that the mutex is NOT lock. | |
3525 | */ | |
3526 | static int stream_try_lock(struct lttng_consumer_stream *stream) | |
3527 | { | |
3528 | int ret; | |
3529 | ||
3530 | assert(stream); | |
3531 | ||
3532 | /* | |
3533 | * Try to lock the stream mutex. On failure, we know that the stream is | |
3534 | * being used else where hence there is data still being extracted. | |
3535 | */ | |
3536 | ret = pthread_mutex_trylock(&stream->lock); | |
3537 | if (ret) { | |
3538 | /* For both EBUSY and EINVAL error, the mutex is NOT locked. */ | |
3539 | ret = 0; | |
3540 | goto end; | |
3541 | } | |
3542 | ||
3543 | ret = 1; | |
3544 | ||
3545 | end: | |
3546 | return ret; | |
3547 | } | |
3548 | ||
f7079f67 DG |
3549 | /* |
3550 | * Search for a relayd associated to the session id and return the reference. | |
3551 | * | |
3552 | * A rcu read side lock MUST be acquire before calling this function and locked | |
3553 | * until the relayd object is no longer necessary. | |
3554 | */ | |
3555 | static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id) | |
3556 | { | |
3557 | struct lttng_ht_iter iter; | |
f7079f67 | 3558 | struct consumer_relayd_sock_pair *relayd = NULL; |
f7079f67 DG |
3559 | |
3560 | /* Iterate over all relayd since they are indexed by net_seq_idx. */ | |
3561 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
3562 | node.node) { | |
18261bd1 DG |
3563 | /* |
3564 | * Check by sessiond id which is unique here where the relayd session | |
3565 | * id might not be when having multiple relayd. | |
3566 | */ | |
3567 | if (relayd->sessiond_session_id == id) { | |
f7079f67 | 3568 | /* Found the relayd. There can be only one per id. */ |
18261bd1 | 3569 | goto found; |
f7079f67 DG |
3570 | } |
3571 | } | |
3572 | ||
18261bd1 DG |
3573 | return NULL; |
3574 | ||
3575 | found: | |
f7079f67 DG |
3576 | return relayd; |
3577 | } | |
3578 | ||
ca22feea DG |
3579 | /* |
3580 | * Check if for a given session id there is still data needed to be extract | |
3581 | * from the buffers. | |
3582 | * | |
6d805429 | 3583 | * Return 1 if data is pending or else 0 meaning ready to be read. |
ca22feea | 3584 | */ |
6d805429 | 3585 | int consumer_data_pending(uint64_t id) |
ca22feea DG |
3586 | { |
3587 | int ret; | |
3588 | struct lttng_ht_iter iter; | |
3589 | struct lttng_ht *ht; | |
3590 | struct lttng_consumer_stream *stream; | |
f7079f67 | 3591 | struct consumer_relayd_sock_pair *relayd = NULL; |
6d805429 | 3592 | int (*data_pending)(struct lttng_consumer_stream *); |
ca22feea | 3593 | |
6d805429 | 3594 | DBG("Consumer data pending command on session id %" PRIu64, id); |
ca22feea | 3595 | |
6f6eda74 | 3596 | rcu_read_lock(); |
ca22feea DG |
3597 | pthread_mutex_lock(&consumer_data.lock); |
3598 | ||
3599 | switch (consumer_data.type) { | |
3600 | case LTTNG_CONSUMER_KERNEL: | |
6d805429 | 3601 | data_pending = lttng_kconsumer_data_pending; |
ca22feea DG |
3602 | break; |
3603 | case LTTNG_CONSUMER32_UST: | |
3604 | case LTTNG_CONSUMER64_UST: | |
6d805429 | 3605 | data_pending = lttng_ustconsumer_data_pending; |
ca22feea DG |
3606 | break; |
3607 | default: | |
3608 | ERR("Unknown consumer data type"); | |
3609 | assert(0); | |
3610 | } | |
3611 | ||
3612 | /* Ease our life a bit */ | |
3613 | ht = consumer_data.stream_list_ht; | |
3614 | ||
f7079f67 DG |
3615 | relayd = find_relayd_by_session_id(id); |
3616 | if (relayd) { | |
3617 | /* Send init command for data pending. */ | |
3618 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3619 | ret = relayd_begin_data_pending(&relayd->control_sock, | |
3620 | relayd->relayd_session_id); | |
3621 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3622 | if (ret < 0) { | |
3623 | /* Communication error thus the relayd so no data pending. */ | |
3624 | goto data_not_pending; | |
3625 | } | |
3626 | } | |
3627 | ||
c8f59ee5 | 3628 | cds_lfht_for_each_entry_duplicate(ht->ht, |
d88aee68 DG |
3629 | ht->hash_fct(&id, lttng_ht_seed), |
3630 | ht->match_fct, &id, | |
ca22feea | 3631 | &iter.iter, stream, node_session_id.node) { |
4e9a4686 DG |
3632 | /* If this call fails, the stream is being used hence data pending. */ |
3633 | ret = stream_try_lock(stream); | |
3634 | if (!ret) { | |
f7079f67 | 3635 | goto data_pending; |
ca22feea | 3636 | } |
ca22feea | 3637 | |
4e9a4686 DG |
3638 | /* |
3639 | * A removed node from the hash table indicates that the stream has | |
3640 | * been deleted thus having a guarantee that the buffers are closed | |
3641 | * on the consumer side. However, data can still be transmitted | |
3642 | * over the network so don't skip the relayd check. | |
3643 | */ | |
3644 | ret = cds_lfht_is_node_deleted(&stream->node.node); | |
3645 | if (!ret) { | |
3646 | /* Check the stream if there is data in the buffers. */ | |
6d805429 DG |
3647 | ret = data_pending(stream); |
3648 | if (ret == 1) { | |
4e9a4686 | 3649 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3650 | goto data_pending; |
4e9a4686 DG |
3651 | } |
3652 | } | |
3653 | ||
3654 | /* Relayd check */ | |
f7079f67 | 3655 | if (relayd) { |
c8f59ee5 DG |
3656 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
3657 | if (stream->metadata_flag) { | |
ad7051c0 DG |
3658 | ret = relayd_quiescent_control(&relayd->control_sock, |
3659 | stream->relayd_stream_id); | |
c8f59ee5 | 3660 | } else { |
6d805429 | 3661 | ret = relayd_data_pending(&relayd->control_sock, |
39df6d9f DG |
3662 | stream->relayd_stream_id, |
3663 | stream->next_net_seq_num - 1); | |
c8f59ee5 DG |
3664 | } |
3665 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
6d805429 | 3666 | if (ret == 1) { |
4e9a4686 | 3667 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3668 | goto data_pending; |
c8f59ee5 DG |
3669 | } |
3670 | } | |
4e9a4686 | 3671 | pthread_mutex_unlock(&stream->lock); |
c8f59ee5 | 3672 | } |
ca22feea | 3673 | |
f7079f67 DG |
3674 | if (relayd) { |
3675 | unsigned int is_data_inflight = 0; | |
3676 | ||
3677 | /* Send init command for data pending. */ | |
3678 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3679 | ret = relayd_end_data_pending(&relayd->control_sock, | |
3680 | relayd->relayd_session_id, &is_data_inflight); | |
3681 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
bdd88757 | 3682 | if (ret < 0) { |
f7079f67 DG |
3683 | goto data_not_pending; |
3684 | } | |
bdd88757 DG |
3685 | if (is_data_inflight) { |
3686 | goto data_pending; | |
3687 | } | |
f7079f67 DG |
3688 | } |
3689 | ||
ca22feea | 3690 | /* |
f7079f67 DG |
3691 | * Finding _no_ node in the hash table and no inflight data means that the |
3692 | * stream(s) have been removed thus data is guaranteed to be available for | |
3693 | * analysis from the trace files. | |
ca22feea DG |
3694 | */ |
3695 | ||
f7079f67 | 3696 | data_not_pending: |
ca22feea DG |
3697 | /* Data is available to be read by a viewer. */ |
3698 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3699 | rcu_read_unlock(); |
6d805429 | 3700 | return 0; |
ca22feea | 3701 | |
f7079f67 | 3702 | data_pending: |
ca22feea DG |
3703 | /* Data is still being extracted from buffers. */ |
3704 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3705 | rcu_read_unlock(); |
6d805429 | 3706 | return 1; |
ca22feea | 3707 | } |
f50f23d9 DG |
3708 | |
3709 | /* | |
3710 | * Send a ret code status message to the sessiond daemon. | |
3711 | * | |
3712 | * Return the sendmsg() return value. | |
3713 | */ | |
3714 | int consumer_send_status_msg(int sock, int ret_code) | |
3715 | { | |
3716 | struct lttcomm_consumer_status_msg msg; | |
3717 | ||
53efb85a | 3718 | memset(&msg, 0, sizeof(msg)); |
f50f23d9 DG |
3719 | msg.ret_code = ret_code; |
3720 | ||
3721 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3722 | } | |
ffe60014 DG |
3723 | |
3724 | /* | |
3725 | * Send a channel status message to the sessiond daemon. | |
3726 | * | |
3727 | * Return the sendmsg() return value. | |
3728 | */ | |
3729 | int consumer_send_status_channel(int sock, | |
3730 | struct lttng_consumer_channel *channel) | |
3731 | { | |
3732 | struct lttcomm_consumer_status_channel msg; | |
3733 | ||
3734 | assert(sock >= 0); | |
3735 | ||
53efb85a | 3736 | memset(&msg, 0, sizeof(msg)); |
ffe60014 | 3737 | if (!channel) { |
0c759fc9 | 3738 | msg.ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL; |
ffe60014 | 3739 | } else { |
0c759fc9 | 3740 | msg.ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
ffe60014 DG |
3741 | msg.key = channel->key; |
3742 | msg.stream_count = channel->streams.count; | |
3743 | } | |
3744 | ||
3745 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3746 | } | |
5c786ded | 3747 | |
d07ceecd MD |
3748 | unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos, |
3749 | unsigned long produced_pos, uint64_t nb_packets_per_stream, | |
3750 | uint64_t max_sb_size) | |
5c786ded | 3751 | { |
d07ceecd | 3752 | unsigned long start_pos; |
5c786ded | 3753 | |
d07ceecd MD |
3754 | if (!nb_packets_per_stream) { |
3755 | return consumed_pos; /* Grab everything */ | |
3756 | } | |
3757 | start_pos = produced_pos - offset_align_floor(produced_pos, max_sb_size); | |
3758 | start_pos -= max_sb_size * nb_packets_per_stream; | |
3759 | if ((long) (start_pos - consumed_pos) < 0) { | |
3760 | return consumed_pos; /* Grab everything */ | |
3761 | } | |
3762 | return start_pos; | |
5c786ded | 3763 | } |