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> | |
5feafd41 | 50 | #include <common/consumer/consumer-metadata-cache.h> |
3bd1e081 MD |
51 | |
52 | struct lttng_consumer_global_data consumer_data = { | |
3bd1e081 MD |
53 | .stream_count = 0, |
54 | .need_update = 1, | |
55 | .type = LTTNG_CONSUMER_UNKNOWN, | |
56 | }; | |
57 | ||
d8ef542d MD |
58 | enum consumer_channel_action { |
59 | CONSUMER_CHANNEL_ADD, | |
a0cbdd2e | 60 | CONSUMER_CHANNEL_DEL, |
d8ef542d MD |
61 | CONSUMER_CHANNEL_QUIT, |
62 | }; | |
63 | ||
64 | struct consumer_channel_msg { | |
65 | enum consumer_channel_action action; | |
a0cbdd2e MD |
66 | struct lttng_consumer_channel *chan; /* add */ |
67 | uint64_t key; /* del */ | |
d8ef542d MD |
68 | }; |
69 | ||
80957876 | 70 | /* Flag used to temporarily pause data consumption from testpoints. */ |
cf0bcb51 JG |
71 | int data_consumption_paused; |
72 | ||
3bd1e081 MD |
73 | /* |
74 | * Flag to inform the polling thread to quit when all fd hung up. Updated by | |
75 | * the consumer_thread_receive_fds when it notices that all fds has hung up. | |
76 | * Also updated by the signal handler (consumer_should_exit()). Read by the | |
77 | * polling threads. | |
78 | */ | |
10211f5c | 79 | int consumer_quit; |
3bd1e081 | 80 | |
43c34bc3 | 81 | /* |
43c34bc3 DG |
82 | * Global hash table containing respectively metadata and data streams. The |
83 | * stream element in this ht should only be updated by the metadata poll thread | |
84 | * for the metadata and the data poll thread for the data. | |
85 | */ | |
40dc48e0 DG |
86 | static struct lttng_ht *metadata_ht; |
87 | static struct lttng_ht *data_ht; | |
43c34bc3 | 88 | |
acdb9057 DG |
89 | /* |
90 | * Notify a thread lttng pipe to poll back again. This usually means that some | |
91 | * global state has changed so we just send back the thread in a poll wait | |
92 | * call. | |
93 | */ | |
94 | static void notify_thread_lttng_pipe(struct lttng_pipe *pipe) | |
95 | { | |
96 | struct lttng_consumer_stream *null_stream = NULL; | |
97 | ||
98 | assert(pipe); | |
99 | ||
100 | (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream)); | |
101 | } | |
102 | ||
5c635c72 MD |
103 | static void notify_health_quit_pipe(int *pipe) |
104 | { | |
6cd525e8 | 105 | ssize_t ret; |
5c635c72 | 106 | |
6cd525e8 MD |
107 | ret = lttng_write(pipe[1], "4", 1); |
108 | if (ret < 1) { | |
5c635c72 MD |
109 | PERROR("write consumer health quit"); |
110 | } | |
111 | } | |
112 | ||
d8ef542d MD |
113 | static void notify_channel_pipe(struct lttng_consumer_local_data *ctx, |
114 | struct lttng_consumer_channel *chan, | |
a0cbdd2e | 115 | uint64_t key, |
d8ef542d MD |
116 | enum consumer_channel_action action) |
117 | { | |
118 | struct consumer_channel_msg msg; | |
6cd525e8 | 119 | ssize_t ret; |
d8ef542d | 120 | |
e56251fc DG |
121 | memset(&msg, 0, sizeof(msg)); |
122 | ||
d8ef542d MD |
123 | msg.action = action; |
124 | msg.chan = chan; | |
f21dae48 | 125 | msg.key = key; |
6cd525e8 MD |
126 | ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg)); |
127 | if (ret < sizeof(msg)) { | |
128 | PERROR("notify_channel_pipe write error"); | |
129 | } | |
d8ef542d MD |
130 | } |
131 | ||
a0cbdd2e MD |
132 | void notify_thread_del_channel(struct lttng_consumer_local_data *ctx, |
133 | uint64_t key) | |
134 | { | |
135 | notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL); | |
136 | } | |
137 | ||
d8ef542d MD |
138 | static int read_channel_pipe(struct lttng_consumer_local_data *ctx, |
139 | struct lttng_consumer_channel **chan, | |
a0cbdd2e | 140 | uint64_t *key, |
d8ef542d MD |
141 | enum consumer_channel_action *action) |
142 | { | |
143 | struct consumer_channel_msg msg; | |
6cd525e8 | 144 | ssize_t ret; |
d8ef542d | 145 | |
6cd525e8 MD |
146 | ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg)); |
147 | if (ret < sizeof(msg)) { | |
148 | ret = -1; | |
149 | goto error; | |
d8ef542d | 150 | } |
6cd525e8 MD |
151 | *action = msg.action; |
152 | *chan = msg.chan; | |
153 | *key = msg.key; | |
154 | error: | |
155 | return (int) ret; | |
d8ef542d MD |
156 | } |
157 | ||
212d67a2 DG |
158 | /* |
159 | * Cleanup the stream list of a channel. Those streams are not yet globally | |
160 | * visible | |
161 | */ | |
162 | static void clean_channel_stream_list(struct lttng_consumer_channel *channel) | |
163 | { | |
164 | struct lttng_consumer_stream *stream, *stmp; | |
165 | ||
166 | assert(channel); | |
167 | ||
168 | /* Delete streams that might have been left in the stream list. */ | |
169 | cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head, | |
170 | send_node) { | |
171 | cds_list_del(&stream->send_node); | |
172 | /* | |
173 | * Once a stream is added to this list, the buffers were created so we | |
174 | * have a guarantee that this call will succeed. Setting the monitor | |
175 | * mode to 0 so we don't lock nor try to delete the stream from the | |
176 | * global hash table. | |
177 | */ | |
178 | stream->monitor = 0; | |
179 | consumer_stream_destroy(stream, NULL); | |
180 | } | |
181 | } | |
182 | ||
3bd1e081 MD |
183 | /* |
184 | * Find a stream. The consumer_data.lock must be locked during this | |
185 | * call. | |
186 | */ | |
d88aee68 | 187 | static struct lttng_consumer_stream *find_stream(uint64_t key, |
8389e4f8 | 188 | struct lttng_ht *ht) |
3bd1e081 | 189 | { |
e4421fec | 190 | struct lttng_ht_iter iter; |
d88aee68 | 191 | struct lttng_ht_node_u64 *node; |
e4421fec | 192 | struct lttng_consumer_stream *stream = NULL; |
3bd1e081 | 193 | |
8389e4f8 DG |
194 | assert(ht); |
195 | ||
d88aee68 DG |
196 | /* -1ULL keys are lookup failures */ |
197 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 198 | return NULL; |
7a57cf92 | 199 | } |
e4421fec | 200 | |
6065ceec DG |
201 | rcu_read_lock(); |
202 | ||
d88aee68 DG |
203 | lttng_ht_lookup(ht, &key, &iter); |
204 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
205 | if (node != NULL) { |
206 | stream = caa_container_of(node, struct lttng_consumer_stream, node); | |
3bd1e081 | 207 | } |
e4421fec | 208 | |
6065ceec DG |
209 | rcu_read_unlock(); |
210 | ||
e4421fec | 211 | return stream; |
3bd1e081 MD |
212 | } |
213 | ||
da009f2c | 214 | static void steal_stream_key(uint64_t key, struct lttng_ht *ht) |
7ad0a0cb MD |
215 | { |
216 | struct lttng_consumer_stream *stream; | |
217 | ||
04253271 | 218 | rcu_read_lock(); |
ffe60014 | 219 | stream = find_stream(key, ht); |
04253271 | 220 | if (stream) { |
da009f2c | 221 | stream->key = (uint64_t) -1ULL; |
04253271 MD |
222 | /* |
223 | * We don't want the lookup to match, but we still need | |
224 | * to iterate on this stream when iterating over the hash table. Just | |
225 | * change the node key. | |
226 | */ | |
da009f2c | 227 | stream->node.key = (uint64_t) -1ULL; |
04253271 MD |
228 | } |
229 | rcu_read_unlock(); | |
7ad0a0cb MD |
230 | } |
231 | ||
d56db448 DG |
232 | /* |
233 | * Return a channel object for the given key. | |
234 | * | |
235 | * RCU read side lock MUST be acquired before calling this function and | |
236 | * protects the channel ptr. | |
237 | */ | |
d88aee68 | 238 | struct lttng_consumer_channel *consumer_find_channel(uint64_t key) |
3bd1e081 | 239 | { |
e4421fec | 240 | struct lttng_ht_iter iter; |
d88aee68 | 241 | struct lttng_ht_node_u64 *node; |
e4421fec | 242 | struct lttng_consumer_channel *channel = NULL; |
3bd1e081 | 243 | |
d88aee68 DG |
244 | /* -1ULL keys are lookup failures */ |
245 | if (key == (uint64_t) -1ULL) { | |
7ad0a0cb | 246 | return NULL; |
7a57cf92 | 247 | } |
e4421fec | 248 | |
d88aee68 DG |
249 | lttng_ht_lookup(consumer_data.channel_ht, &key, &iter); |
250 | node = lttng_ht_iter_get_node_u64(&iter); | |
e4421fec DG |
251 | if (node != NULL) { |
252 | channel = caa_container_of(node, struct lttng_consumer_channel, node); | |
3bd1e081 | 253 | } |
e4421fec DG |
254 | |
255 | return channel; | |
3bd1e081 MD |
256 | } |
257 | ||
b5a6470f DG |
258 | /* |
259 | * There is a possibility that the consumer does not have enough time between | |
260 | * the close of the channel on the session daemon and the cleanup in here thus | |
261 | * once we have a channel add with an existing key, we know for sure that this | |
262 | * channel will eventually get cleaned up by all streams being closed. | |
263 | * | |
264 | * This function just nullifies the already existing channel key. | |
265 | */ | |
266 | static void steal_channel_key(uint64_t key) | |
267 | { | |
268 | struct lttng_consumer_channel *channel; | |
269 | ||
270 | rcu_read_lock(); | |
271 | channel = consumer_find_channel(key); | |
272 | if (channel) { | |
273 | channel->key = (uint64_t) -1ULL; | |
274 | /* | |
275 | * We don't want the lookup to match, but we still need to iterate on | |
276 | * this channel when iterating over the hash table. Just change the | |
277 | * node key. | |
278 | */ | |
279 | channel->node.key = (uint64_t) -1ULL; | |
280 | } | |
281 | rcu_read_unlock(); | |
282 | } | |
283 | ||
ffe60014 | 284 | static void free_channel_rcu(struct rcu_head *head) |
702b1ea4 | 285 | { |
d88aee68 DG |
286 | struct lttng_ht_node_u64 *node = |
287 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
ffe60014 DG |
288 | struct lttng_consumer_channel *channel = |
289 | caa_container_of(node, struct lttng_consumer_channel, node); | |
702b1ea4 | 290 | |
b83e03c4 MD |
291 | switch (consumer_data.type) { |
292 | case LTTNG_CONSUMER_KERNEL: | |
293 | break; | |
294 | case LTTNG_CONSUMER32_UST: | |
295 | case LTTNG_CONSUMER64_UST: | |
296 | lttng_ustconsumer_free_channel(channel); | |
297 | break; | |
298 | default: | |
299 | ERR("Unknown consumer_data type"); | |
300 | abort(); | |
301 | } | |
ffe60014 | 302 | free(channel); |
702b1ea4 MD |
303 | } |
304 | ||
00e2e675 DG |
305 | /* |
306 | * RCU protected relayd socket pair free. | |
307 | */ | |
ffe60014 | 308 | static void free_relayd_rcu(struct rcu_head *head) |
00e2e675 | 309 | { |
d88aee68 DG |
310 | struct lttng_ht_node_u64 *node = |
311 | caa_container_of(head, struct lttng_ht_node_u64, head); | |
00e2e675 DG |
312 | struct consumer_relayd_sock_pair *relayd = |
313 | caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
314 | ||
8994307f DG |
315 | /* |
316 | * Close all sockets. This is done in the call RCU since we don't want the | |
317 | * socket fds to be reassigned thus potentially creating bad state of the | |
318 | * relayd object. | |
319 | * | |
320 | * We do not have to lock the control socket mutex here since at this stage | |
321 | * there is no one referencing to this relayd object. | |
322 | */ | |
323 | (void) relayd_close(&relayd->control_sock); | |
324 | (void) relayd_close(&relayd->data_sock); | |
325 | ||
70053766 | 326 | pthread_mutex_destroy(&relayd->ctrl_sock_mutex); |
00e2e675 DG |
327 | free(relayd); |
328 | } | |
329 | ||
330 | /* | |
331 | * Destroy and free relayd socket pair object. | |
00e2e675 | 332 | */ |
51230d70 | 333 | void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
334 | { |
335 | int ret; | |
336 | struct lttng_ht_iter iter; | |
337 | ||
173af62f DG |
338 | if (relayd == NULL) { |
339 | return; | |
340 | } | |
341 | ||
00e2e675 DG |
342 | DBG("Consumer destroy and close relayd socket pair"); |
343 | ||
344 | iter.iter.node = &relayd->node.node; | |
345 | ret = lttng_ht_del(consumer_data.relayd_ht, &iter); | |
173af62f | 346 | if (ret != 0) { |
8994307f | 347 | /* We assume the relayd is being or is destroyed */ |
173af62f DG |
348 | return; |
349 | } | |
00e2e675 | 350 | |
00e2e675 | 351 | /* RCU free() call */ |
ffe60014 DG |
352 | call_rcu(&relayd->node.head, free_relayd_rcu); |
353 | } | |
354 | ||
355 | /* | |
356 | * Remove a channel from the global list protected by a mutex. This function is | |
357 | * also responsible for freeing its data structures. | |
358 | */ | |
359 | void consumer_del_channel(struct lttng_consumer_channel *channel) | |
360 | { | |
361 | int ret; | |
362 | struct lttng_ht_iter iter; | |
363 | ||
d88aee68 | 364 | DBG("Consumer delete channel key %" PRIu64, channel->key); |
ffe60014 DG |
365 | |
366 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 367 | pthread_mutex_lock(&channel->lock); |
ffe60014 | 368 | |
212d67a2 DG |
369 | /* Destroy streams that might have been left in the stream list. */ |
370 | clean_channel_stream_list(channel); | |
51e762e5 | 371 | |
d3e2ba59 JD |
372 | if (channel->live_timer_enabled == 1) { |
373 | consumer_timer_live_stop(channel); | |
374 | } | |
e9404c27 JG |
375 | if (channel->monitor_timer_enabled == 1) { |
376 | consumer_timer_monitor_stop(channel); | |
377 | } | |
d3e2ba59 | 378 | |
ffe60014 DG |
379 | switch (consumer_data.type) { |
380 | case LTTNG_CONSUMER_KERNEL: | |
381 | break; | |
382 | case LTTNG_CONSUMER32_UST: | |
383 | case LTTNG_CONSUMER64_UST: | |
384 | lttng_ustconsumer_del_channel(channel); | |
385 | break; | |
386 | default: | |
387 | ERR("Unknown consumer_data type"); | |
388 | assert(0); | |
389 | goto end; | |
390 | } | |
391 | ||
392 | rcu_read_lock(); | |
393 | iter.iter.node = &channel->node.node; | |
394 | ret = lttng_ht_del(consumer_data.channel_ht, &iter); | |
395 | assert(!ret); | |
0fc67d9b JG |
396 | |
397 | iter.iter.node = &channel->channels_by_session_id_ht_node.node; | |
398 | ret = lttng_ht_del(consumer_data.channels_by_session_id_ht, &iter); | |
399 | assert(!ret); | |
ffe60014 DG |
400 | rcu_read_unlock(); |
401 | ||
402 | call_rcu(&channel->node.head, free_channel_rcu); | |
403 | end: | |
a9838785 | 404 | pthread_mutex_unlock(&channel->lock); |
ffe60014 | 405 | pthread_mutex_unlock(&consumer_data.lock); |
00e2e675 DG |
406 | } |
407 | ||
228b5bf7 DG |
408 | /* |
409 | * Iterate over the relayd hash table and destroy each element. Finally, | |
410 | * destroy the whole hash table. | |
411 | */ | |
412 | static void cleanup_relayd_ht(void) | |
413 | { | |
414 | struct lttng_ht_iter iter; | |
415 | struct consumer_relayd_sock_pair *relayd; | |
416 | ||
417 | rcu_read_lock(); | |
418 | ||
419 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
420 | node.node) { | |
51230d70 | 421 | consumer_destroy_relayd(relayd); |
228b5bf7 DG |
422 | } |
423 | ||
228b5bf7 | 424 | rcu_read_unlock(); |
36b588ed MD |
425 | |
426 | lttng_ht_destroy(consumer_data.relayd_ht); | |
228b5bf7 DG |
427 | } |
428 | ||
8994307f DG |
429 | /* |
430 | * Update the end point status of all streams having the given network sequence | |
431 | * index (relayd index). | |
432 | * | |
433 | * It's atomically set without having the stream mutex locked which is fine | |
434 | * because we handle the write/read race with a pipe wakeup for each thread. | |
435 | */ | |
da009f2c | 436 | static void update_endpoint_status_by_netidx(uint64_t net_seq_idx, |
8994307f DG |
437 | enum consumer_endpoint_status status) |
438 | { | |
439 | struct lttng_ht_iter iter; | |
440 | struct lttng_consumer_stream *stream; | |
441 | ||
da009f2c | 442 | DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx); |
8994307f DG |
443 | |
444 | rcu_read_lock(); | |
445 | ||
446 | /* Let's begin with metadata */ | |
447 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
448 | if (stream->net_seq_idx == net_seq_idx) { | |
449 | uatomic_set(&stream->endpoint_status, status); | |
450 | DBG("Delete flag set to metadata stream %d", stream->wait_fd); | |
451 | } | |
452 | } | |
453 | ||
454 | /* Follow up by the data streams */ | |
455 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
456 | if (stream->net_seq_idx == net_seq_idx) { | |
457 | uatomic_set(&stream->endpoint_status, status); | |
458 | DBG("Delete flag set to data stream %d", stream->wait_fd); | |
459 | } | |
460 | } | |
461 | rcu_read_unlock(); | |
462 | } | |
463 | ||
464 | /* | |
465 | * Cleanup a relayd object by flagging every associated streams for deletion, | |
466 | * destroying the object meaning removing it from the relayd hash table, | |
467 | * closing the sockets and freeing the memory in a RCU call. | |
468 | * | |
469 | * If a local data context is available, notify the threads that the streams' | |
470 | * state have changed. | |
471 | */ | |
11413bb9 | 472 | void lttng_consumer_cleanup_relayd(struct consumer_relayd_sock_pair *relayd) |
8994307f | 473 | { |
da009f2c | 474 | uint64_t netidx; |
8994307f DG |
475 | |
476 | assert(relayd); | |
477 | ||
11413bb9 | 478 | DBG("Cleaning up relayd object ID %"PRIu64, relayd->net_seq_idx); |
9617607b | 479 | |
8994307f DG |
480 | /* Save the net sequence index before destroying the object */ |
481 | netidx = relayd->net_seq_idx; | |
482 | ||
483 | /* | |
484 | * Delete the relayd from the relayd hash table, close the sockets and free | |
485 | * the object in a RCU call. | |
486 | */ | |
51230d70 | 487 | consumer_destroy_relayd(relayd); |
8994307f DG |
488 | |
489 | /* Set inactive endpoint to all streams */ | |
490 | update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE); | |
491 | ||
492 | /* | |
493 | * With a local data context, notify the threads that the streams' state | |
494 | * have changed. The write() action on the pipe acts as an "implicit" | |
495 | * memory barrier ordering the updates of the end point status from the | |
496 | * read of this status which happens AFTER receiving this notify. | |
497 | */ | |
11413bb9 JR |
498 | notify_thread_lttng_pipe(relayd->ctx->consumer_data_pipe); |
499 | notify_thread_lttng_pipe(relayd->ctx->consumer_metadata_pipe); | |
8994307f DG |
500 | } |
501 | ||
a6ba4fe1 DG |
502 | /* |
503 | * Flag a relayd socket pair for destruction. Destroy it if the refcount | |
504 | * reaches zero. | |
505 | * | |
506 | * RCU read side lock MUST be aquired before calling this function. | |
507 | */ | |
508 | void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) | |
509 | { | |
510 | assert(relayd); | |
511 | ||
512 | /* Set destroy flag for this object */ | |
513 | uatomic_set(&relayd->destroy_flag, 1); | |
514 | ||
515 | /* Destroy the relayd if refcount is 0 */ | |
516 | if (uatomic_read(&relayd->refcount) == 0) { | |
51230d70 | 517 | consumer_destroy_relayd(relayd); |
a6ba4fe1 DG |
518 | } |
519 | } | |
520 | ||
3bd1e081 | 521 | /* |
1d1a276c DG |
522 | * Completly destroy stream from every visiable data structure and the given |
523 | * hash table if one. | |
524 | * | |
525 | * One this call returns, the stream object is not longer usable nor visible. | |
3bd1e081 | 526 | */ |
e316aad5 DG |
527 | void consumer_del_stream(struct lttng_consumer_stream *stream, |
528 | struct lttng_ht *ht) | |
3bd1e081 | 529 | { |
1d1a276c | 530 | consumer_stream_destroy(stream, ht); |
3bd1e081 MD |
531 | } |
532 | ||
5ab66908 MD |
533 | /* |
534 | * XXX naming of del vs destroy is all mixed up. | |
535 | */ | |
536 | void consumer_del_stream_for_data(struct lttng_consumer_stream *stream) | |
537 | { | |
538 | consumer_stream_destroy(stream, data_ht); | |
539 | } | |
540 | ||
541 | void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream) | |
542 | { | |
543 | consumer_stream_destroy(stream, metadata_ht); | |
544 | } | |
545 | ||
d9a2e16e JD |
546 | void consumer_stream_update_channel_attributes( |
547 | struct lttng_consumer_stream *stream, | |
548 | struct lttng_consumer_channel *channel) | |
549 | { | |
550 | stream->channel_read_only_attributes.tracefile_size = | |
551 | channel->tracefile_size; | |
552 | memcpy(stream->channel_read_only_attributes.path, channel->pathname, | |
553 | sizeof(stream->channel_read_only_attributes.path)); | |
554 | } | |
555 | ||
d88aee68 DG |
556 | struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, |
557 | uint64_t stream_key, | |
3bd1e081 | 558 | enum lttng_consumer_stream_state state, |
ffe60014 | 559 | const char *channel_name, |
6df2e2c9 | 560 | uid_t uid, |
00e2e675 | 561 | gid_t gid, |
57a269f2 | 562 | uint64_t relayd_id, |
53632229 | 563 | uint64_t session_id, |
ffe60014 DG |
564 | int cpu, |
565 | int *alloc_ret, | |
4891ece8 | 566 | enum consumer_channel_type type, |
e098433c JG |
567 | unsigned int monitor, |
568 | uint64_t trace_archive_id) | |
3bd1e081 | 569 | { |
ffe60014 | 570 | int ret; |
3bd1e081 | 571 | struct lttng_consumer_stream *stream; |
3bd1e081 | 572 | |
effcf122 | 573 | stream = zmalloc(sizeof(*stream)); |
3bd1e081 | 574 | if (stream == NULL) { |
7a57cf92 | 575 | PERROR("malloc struct lttng_consumer_stream"); |
ffe60014 | 576 | ret = -ENOMEM; |
7a57cf92 | 577 | goto end; |
3bd1e081 | 578 | } |
7a57cf92 | 579 | |
d56db448 DG |
580 | rcu_read_lock(); |
581 | ||
3bd1e081 | 582 | stream->key = stream_key; |
3bd1e081 MD |
583 | stream->out_fd = -1; |
584 | stream->out_fd_offset = 0; | |
e5d1a9b3 | 585 | stream->output_written = 0; |
3bd1e081 | 586 | stream->state = state; |
6df2e2c9 MD |
587 | stream->uid = uid; |
588 | stream->gid = gid; | |
ffe60014 | 589 | stream->net_seq_idx = relayd_id; |
53632229 | 590 | stream->session_id = session_id; |
4891ece8 | 591 | stream->monitor = monitor; |
774d490c | 592 | stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE; |
f8f3885c | 593 | stream->index_file = NULL; |
fb83fe64 | 594 | stream->last_sequence_number = -1ULL; |
e098433c | 595 | stream->trace_archive_id = trace_archive_id; |
53632229 | 596 | pthread_mutex_init(&stream->lock, NULL); |
c585821b | 597 | pthread_mutex_init(&stream->metadata_timer_lock, NULL); |
58b1f425 | 598 | |
ffe60014 DG |
599 | /* If channel is the metadata, flag this stream as metadata. */ |
600 | if (type == CONSUMER_CHANNEL_TYPE_METADATA) { | |
601 | stream->metadata_flag = 1; | |
602 | /* Metadata is flat out. */ | |
603 | strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name)); | |
94d49140 JD |
604 | /* Live rendez-vous point. */ |
605 | pthread_cond_init(&stream->metadata_rdv, NULL); | |
606 | pthread_mutex_init(&stream->metadata_rdv_lock, NULL); | |
58b1f425 | 607 | } else { |
ffe60014 DG |
608 | /* Format stream name to <channel_name>_<cpu_number> */ |
609 | ret = snprintf(stream->name, sizeof(stream->name), "%s_%d", | |
610 | channel_name, cpu); | |
611 | if (ret < 0) { | |
612 | PERROR("snprintf stream name"); | |
613 | goto error; | |
614 | } | |
58b1f425 | 615 | } |
c30aaa51 | 616 | |
ffe60014 | 617 | /* Key is always the wait_fd for streams. */ |
d88aee68 | 618 | lttng_ht_node_init_u64(&stream->node, stream->key); |
ffe60014 | 619 | |
d8ef542d MD |
620 | /* Init node per channel id key */ |
621 | lttng_ht_node_init_u64(&stream->node_channel_id, channel_key); | |
622 | ||
53632229 | 623 | /* Init session id node with the stream session id */ |
d88aee68 | 624 | lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id); |
53632229 | 625 | |
07b86b52 JD |
626 | DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64 |
627 | " relayd_id %" PRIu64 ", session_id %" PRIu64, | |
628 | stream->name, stream->key, channel_key, | |
629 | stream->net_seq_idx, stream->session_id); | |
d56db448 DG |
630 | |
631 | rcu_read_unlock(); | |
3bd1e081 | 632 | return stream; |
c80048c6 MD |
633 | |
634 | error: | |
d56db448 | 635 | rcu_read_unlock(); |
c80048c6 | 636 | free(stream); |
7a57cf92 | 637 | end: |
ffe60014 DG |
638 | if (alloc_ret) { |
639 | *alloc_ret = ret; | |
640 | } | |
c80048c6 | 641 | return NULL; |
3bd1e081 MD |
642 | } |
643 | ||
644 | /* | |
645 | * Add a stream to the global list protected by a mutex. | |
646 | */ | |
66d583dc | 647 | void consumer_add_data_stream(struct lttng_consumer_stream *stream) |
3bd1e081 | 648 | { |
5ab66908 | 649 | struct lttng_ht *ht = data_ht; |
3bd1e081 | 650 | |
e316aad5 | 651 | assert(stream); |
43c34bc3 | 652 | assert(ht); |
c77fc10a | 653 | |
d88aee68 | 654 | DBG3("Adding consumer stream %" PRIu64, stream->key); |
e316aad5 DG |
655 | |
656 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 657 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 658 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 659 | pthread_mutex_lock(&stream->lock); |
b0b335c8 | 660 | rcu_read_lock(); |
e316aad5 | 661 | |
43c34bc3 | 662 | /* Steal stream identifier to avoid having streams with the same key */ |
ffe60014 | 663 | steal_stream_key(stream->key, ht); |
43c34bc3 | 664 | |
d88aee68 | 665 | lttng_ht_add_unique_u64(ht, &stream->node); |
00e2e675 | 666 | |
d8ef542d MD |
667 | lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, |
668 | &stream->node_channel_id); | |
669 | ||
ca22feea DG |
670 | /* |
671 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
672 | * the key since the HT does not use it and we allow to add redundant keys | |
673 | * into this table. | |
674 | */ | |
d88aee68 | 675 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 676 | |
e316aad5 | 677 | /* |
ffe60014 DG |
678 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
679 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
680 | * causes the count to become 0 also causes a stream to be added. The |
681 | * channel deletion will thus be triggered by the following removal of this | |
682 | * stream. | |
683 | */ | |
ffe60014 | 684 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
685 | /* Increment refcount before decrementing nb_init_stream_left */ |
686 | cmm_smp_wmb(); | |
ffe60014 | 687 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
688 | } |
689 | ||
690 | /* Update consumer data once the node is inserted. */ | |
3bd1e081 MD |
691 | consumer_data.stream_count++; |
692 | consumer_data.need_update = 1; | |
693 | ||
e316aad5 | 694 | rcu_read_unlock(); |
2e818a6a | 695 | pthread_mutex_unlock(&stream->lock); |
ec6ea7d0 | 696 | pthread_mutex_unlock(&stream->chan->timer_lock); |
a9838785 | 697 | pthread_mutex_unlock(&stream->chan->lock); |
3bd1e081 | 698 | pthread_mutex_unlock(&consumer_data.lock); |
3bd1e081 MD |
699 | } |
700 | ||
5ab66908 MD |
701 | void consumer_del_data_stream(struct lttng_consumer_stream *stream) |
702 | { | |
703 | consumer_del_stream(stream, data_ht); | |
704 | } | |
705 | ||
00e2e675 | 706 | /* |
3f8e211f DG |
707 | * Add relayd socket to global consumer data hashtable. RCU read side lock MUST |
708 | * be acquired before calling this. | |
00e2e675 | 709 | */ |
d09e1200 | 710 | static int add_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
711 | { |
712 | int ret = 0; | |
d88aee68 | 713 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
714 | struct lttng_ht_iter iter; |
715 | ||
ffe60014 | 716 | assert(relayd); |
00e2e675 | 717 | |
00e2e675 | 718 | lttng_ht_lookup(consumer_data.relayd_ht, |
d88aee68 DG |
719 | &relayd->net_seq_idx, &iter); |
720 | node = lttng_ht_iter_get_node_u64(&iter); | |
00e2e675 | 721 | if (node != NULL) { |
00e2e675 DG |
722 | goto end; |
723 | } | |
d88aee68 | 724 | lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node); |
00e2e675 | 725 | |
00e2e675 DG |
726 | end: |
727 | return ret; | |
728 | } | |
729 | ||
730 | /* | |
731 | * Allocate and return a consumer relayd socket. | |
732 | */ | |
027a694f | 733 | static struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( |
da009f2c | 734 | uint64_t net_seq_idx) |
00e2e675 DG |
735 | { |
736 | struct consumer_relayd_sock_pair *obj = NULL; | |
737 | ||
da009f2c MD |
738 | /* net sequence index of -1 is a failure */ |
739 | if (net_seq_idx == (uint64_t) -1ULL) { | |
00e2e675 DG |
740 | goto error; |
741 | } | |
742 | ||
743 | obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); | |
744 | if (obj == NULL) { | |
745 | PERROR("zmalloc relayd sock"); | |
746 | goto error; | |
747 | } | |
748 | ||
749 | obj->net_seq_idx = net_seq_idx; | |
750 | obj->refcount = 0; | |
173af62f | 751 | obj->destroy_flag = 0; |
f96e4545 MD |
752 | obj->control_sock.sock.fd = -1; |
753 | obj->data_sock.sock.fd = -1; | |
d88aee68 | 754 | lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx); |
00e2e675 DG |
755 | pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); |
756 | ||
757 | error: | |
758 | return obj; | |
759 | } | |
760 | ||
761 | /* | |
762 | * Find a relayd socket pair in the global consumer data. | |
763 | * | |
764 | * Return the object if found else NULL. | |
b0b335c8 MD |
765 | * RCU read-side lock must be held across this call and while using the |
766 | * returned object. | |
00e2e675 | 767 | */ |
d88aee68 | 768 | struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key) |
00e2e675 DG |
769 | { |
770 | struct lttng_ht_iter iter; | |
d88aee68 | 771 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
772 | struct consumer_relayd_sock_pair *relayd = NULL; |
773 | ||
774 | /* Negative keys are lookup failures */ | |
d88aee68 | 775 | if (key == (uint64_t) -1ULL) { |
00e2e675 DG |
776 | goto error; |
777 | } | |
778 | ||
d88aee68 | 779 | lttng_ht_lookup(consumer_data.relayd_ht, &key, |
00e2e675 | 780 | &iter); |
d88aee68 | 781 | node = lttng_ht_iter_get_node_u64(&iter); |
00e2e675 DG |
782 | if (node != NULL) { |
783 | relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
784 | } | |
785 | ||
00e2e675 DG |
786 | error: |
787 | return relayd; | |
788 | } | |
789 | ||
10a50311 JD |
790 | /* |
791 | * Find a relayd and send the stream | |
792 | * | |
793 | * Returns 0 on success, < 0 on error | |
794 | */ | |
795 | int consumer_send_relayd_stream(struct lttng_consumer_stream *stream, | |
796 | char *path) | |
797 | { | |
798 | int ret = 0; | |
799 | struct consumer_relayd_sock_pair *relayd; | |
800 | ||
801 | assert(stream); | |
802 | assert(stream->net_seq_idx != -1ULL); | |
803 | assert(path); | |
804 | ||
805 | /* The stream is not metadata. Get relayd reference if exists. */ | |
806 | rcu_read_lock(); | |
807 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
808 | if (relayd != NULL) { | |
809 | /* Add stream on the relayd */ | |
810 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
811 | ret = relayd_add_stream(&relayd->control_sock, stream->name, | |
812 | path, &stream->relayd_stream_id, | |
0b50e4b3 JG |
813 | stream->chan->tracefile_size, stream->chan->tracefile_count, |
814 | stream->trace_archive_id); | |
10a50311 JD |
815 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
816 | if (ret < 0) { | |
11413bb9 JR |
817 | ERR("Relayd add stream failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
818 | lttng_consumer_cleanup_relayd(relayd); | |
10a50311 JD |
819 | goto end; |
820 | } | |
1c20f0e2 | 821 | |
10a50311 | 822 | uatomic_inc(&relayd->refcount); |
d01178b6 | 823 | stream->sent_to_relayd = 1; |
10a50311 JD |
824 | } else { |
825 | ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.", | |
826 | stream->key, stream->net_seq_idx); | |
827 | ret = -1; | |
828 | goto end; | |
829 | } | |
830 | ||
831 | DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64, | |
832 | stream->name, stream->key, stream->net_seq_idx); | |
833 | ||
834 | end: | |
835 | rcu_read_unlock(); | |
836 | return ret; | |
837 | } | |
838 | ||
a4baae1b JD |
839 | /* |
840 | * Find a relayd and send the streams sent message | |
841 | * | |
842 | * Returns 0 on success, < 0 on error | |
843 | */ | |
844 | int consumer_send_relayd_streams_sent(uint64_t net_seq_idx) | |
845 | { | |
846 | int ret = 0; | |
847 | struct consumer_relayd_sock_pair *relayd; | |
848 | ||
849 | assert(net_seq_idx != -1ULL); | |
850 | ||
851 | /* The stream is not metadata. Get relayd reference if exists. */ | |
852 | rcu_read_lock(); | |
853 | relayd = consumer_find_relayd(net_seq_idx); | |
854 | if (relayd != NULL) { | |
855 | /* Add stream on the relayd */ | |
856 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
857 | ret = relayd_streams_sent(&relayd->control_sock); | |
858 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
859 | if (ret < 0) { | |
11413bb9 JR |
860 | ERR("Relayd streams sent failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
861 | lttng_consumer_cleanup_relayd(relayd); | |
a4baae1b JD |
862 | goto end; |
863 | } | |
864 | } else { | |
865 | ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.", | |
866 | net_seq_idx); | |
867 | ret = -1; | |
868 | goto end; | |
869 | } | |
870 | ||
871 | ret = 0; | |
872 | DBG("All streams sent relayd id %" PRIu64, net_seq_idx); | |
873 | ||
874 | end: | |
875 | rcu_read_unlock(); | |
876 | return ret; | |
877 | } | |
878 | ||
10a50311 JD |
879 | /* |
880 | * Find a relayd and close the stream | |
881 | */ | |
882 | void close_relayd_stream(struct lttng_consumer_stream *stream) | |
883 | { | |
884 | struct consumer_relayd_sock_pair *relayd; | |
885 | ||
886 | /* The stream is not metadata. Get relayd reference if exists. */ | |
887 | rcu_read_lock(); | |
888 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
889 | if (relayd) { | |
890 | consumer_stream_relayd_close(stream, relayd); | |
891 | } | |
892 | rcu_read_unlock(); | |
893 | } | |
894 | ||
00e2e675 DG |
895 | /* |
896 | * Handle stream for relayd transmission if the stream applies for network | |
897 | * streaming where the net sequence index is set. | |
898 | * | |
899 | * Return destination file descriptor or negative value on error. | |
900 | */ | |
6197aea7 | 901 | static int write_relayd_stream_header(struct lttng_consumer_stream *stream, |
1d4dfdef DG |
902 | size_t data_size, unsigned long padding, |
903 | struct consumer_relayd_sock_pair *relayd) | |
00e2e675 DG |
904 | { |
905 | int outfd = -1, ret; | |
00e2e675 DG |
906 | struct lttcomm_relayd_data_hdr data_hdr; |
907 | ||
908 | /* Safety net */ | |
909 | assert(stream); | |
6197aea7 | 910 | assert(relayd); |
00e2e675 DG |
911 | |
912 | /* Reset data header */ | |
913 | memset(&data_hdr, 0, sizeof(data_hdr)); | |
914 | ||
00e2e675 DG |
915 | if (stream->metadata_flag) { |
916 | /* Caller MUST acquire the relayd control socket lock */ | |
917 | ret = relayd_send_metadata(&relayd->control_sock, data_size); | |
918 | if (ret < 0) { | |
919 | goto error; | |
920 | } | |
921 | ||
922 | /* Metadata are always sent on the control socket. */ | |
6151a90f | 923 | outfd = relayd->control_sock.sock.fd; |
00e2e675 DG |
924 | } else { |
925 | /* Set header with stream information */ | |
926 | data_hdr.stream_id = htobe64(stream->relayd_stream_id); | |
927 | data_hdr.data_size = htobe32(data_size); | |
1d4dfdef | 928 | data_hdr.padding_size = htobe32(padding); |
39df6d9f DG |
929 | /* |
930 | * Note that net_seq_num below is assigned with the *current* value of | |
931 | * next_net_seq_num and only after that the next_net_seq_num will be | |
932 | * increment. This is why when issuing a command on the relayd using | |
933 | * this next value, 1 should always be substracted in order to compare | |
934 | * the last seen sequence number on the relayd side to the last sent. | |
935 | */ | |
3604f373 | 936 | data_hdr.net_seq_num = htobe64(stream->next_net_seq_num); |
00e2e675 DG |
937 | /* Other fields are zeroed previously */ |
938 | ||
939 | ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, | |
940 | sizeof(data_hdr)); | |
941 | if (ret < 0) { | |
942 | goto error; | |
943 | } | |
944 | ||
3604f373 DG |
945 | ++stream->next_net_seq_num; |
946 | ||
00e2e675 | 947 | /* Set to go on data socket */ |
6151a90f | 948 | outfd = relayd->data_sock.sock.fd; |
00e2e675 DG |
949 | } |
950 | ||
951 | error: | |
952 | return outfd; | |
953 | } | |
954 | ||
3bd1e081 | 955 | /* |
ffe60014 DG |
956 | * Allocate and return a new lttng_consumer_channel object using the given key |
957 | * to initialize the hash table node. | |
958 | * | |
959 | * On error, return NULL. | |
3bd1e081 | 960 | */ |
886224ff | 961 | struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, |
ffe60014 DG |
962 | uint64_t session_id, |
963 | const char *pathname, | |
964 | const char *name, | |
965 | uid_t uid, | |
966 | gid_t gid, | |
57a269f2 | 967 | uint64_t relayd_id, |
1624d5b7 JD |
968 | enum lttng_event_output output, |
969 | uint64_t tracefile_size, | |
2bba9e53 | 970 | uint64_t tracefile_count, |
1950109e | 971 | uint64_t session_id_per_pid, |
ecc48a90 | 972 | unsigned int monitor, |
d7ba1388 | 973 | unsigned int live_timer_interval, |
3d071855 | 974 | const char *root_shm_path, |
d7ba1388 | 975 | const char *shm_path) |
3bd1e081 MD |
976 | { |
977 | struct lttng_consumer_channel *channel; | |
3bd1e081 | 978 | |
276b26d1 | 979 | channel = zmalloc(sizeof(*channel)); |
3bd1e081 | 980 | if (channel == NULL) { |
7a57cf92 | 981 | PERROR("malloc struct lttng_consumer_channel"); |
3bd1e081 MD |
982 | goto end; |
983 | } | |
ffe60014 DG |
984 | |
985 | channel->key = key; | |
3bd1e081 | 986 | channel->refcount = 0; |
ffe60014 | 987 | channel->session_id = session_id; |
1950109e | 988 | channel->session_id_per_pid = session_id_per_pid; |
ffe60014 DG |
989 | channel->uid = uid; |
990 | channel->gid = gid; | |
991 | channel->relayd_id = relayd_id; | |
1624d5b7 JD |
992 | channel->tracefile_size = tracefile_size; |
993 | channel->tracefile_count = tracefile_count; | |
2bba9e53 | 994 | channel->monitor = monitor; |
ecc48a90 | 995 | channel->live_timer_interval = live_timer_interval; |
a9838785 | 996 | pthread_mutex_init(&channel->lock, NULL); |
ec6ea7d0 | 997 | pthread_mutex_init(&channel->timer_lock, NULL); |
ffe60014 | 998 | |
0c759fc9 DG |
999 | switch (output) { |
1000 | case LTTNG_EVENT_SPLICE: | |
1001 | channel->output = CONSUMER_CHANNEL_SPLICE; | |
1002 | break; | |
1003 | case LTTNG_EVENT_MMAP: | |
1004 | channel->output = CONSUMER_CHANNEL_MMAP; | |
1005 | break; | |
1006 | default: | |
1007 | assert(0); | |
1008 | free(channel); | |
1009 | channel = NULL; | |
1010 | goto end; | |
1011 | } | |
1012 | ||
07b86b52 JD |
1013 | /* |
1014 | * In monitor mode, the streams associated with the channel will be put in | |
1015 | * a special list ONLY owned by this channel. So, the refcount is set to 1 | |
1016 | * here meaning that the channel itself has streams that are referenced. | |
1017 | * | |
1018 | * On a channel deletion, once the channel is no longer visible, the | |
1019 | * refcount is decremented and checked for a zero value to delete it. With | |
1020 | * streams in no monitor mode, it will now be safe to destroy the channel. | |
1021 | */ | |
1022 | if (!channel->monitor) { | |
1023 | channel->refcount = 1; | |
1024 | } | |
1025 | ||
ffe60014 DG |
1026 | strncpy(channel->pathname, pathname, sizeof(channel->pathname)); |
1027 | channel->pathname[sizeof(channel->pathname) - 1] = '\0'; | |
1028 | ||
1029 | strncpy(channel->name, name, sizeof(channel->name)); | |
1030 | channel->name[sizeof(channel->name) - 1] = '\0'; | |
1031 | ||
3d071855 MD |
1032 | if (root_shm_path) { |
1033 | strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path)); | |
1034 | channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0'; | |
1035 | } | |
d7ba1388 MD |
1036 | if (shm_path) { |
1037 | strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path)); | |
1038 | channel->shm_path[sizeof(channel->shm_path) - 1] = '\0'; | |
1039 | } | |
1040 | ||
d88aee68 | 1041 | lttng_ht_node_init_u64(&channel->node, channel->key); |
0fc67d9b JG |
1042 | lttng_ht_node_init_u64(&channel->channels_by_session_id_ht_node, |
1043 | channel->session_id); | |
d8ef542d MD |
1044 | |
1045 | channel->wait_fd = -1; | |
1046 | ||
ffe60014 DG |
1047 | CDS_INIT_LIST_HEAD(&channel->streams.head); |
1048 | ||
62a7b8ed | 1049 | DBG("Allocated channel (key %" PRIu64 ")", channel->key); |
3bd1e081 | 1050 | |
3bd1e081 MD |
1051 | end: |
1052 | return channel; | |
1053 | } | |
1054 | ||
1055 | /* | |
1056 | * Add a channel to the global list protected by a mutex. | |
821fffb2 | 1057 | * |
b5a6470f | 1058 | * Always return 0 indicating success. |
3bd1e081 | 1059 | */ |
d8ef542d MD |
1060 | int consumer_add_channel(struct lttng_consumer_channel *channel, |
1061 | struct lttng_consumer_local_data *ctx) | |
3bd1e081 | 1062 | { |
3bd1e081 | 1063 | pthread_mutex_lock(&consumer_data.lock); |
a9838785 | 1064 | pthread_mutex_lock(&channel->lock); |
ec6ea7d0 | 1065 | pthread_mutex_lock(&channel->timer_lock); |
c77fc10a | 1066 | |
b5a6470f DG |
1067 | /* |
1068 | * This gives us a guarantee that the channel we are about to add to the | |
1069 | * channel hash table will be unique. See this function comment on the why | |
1070 | * we need to steel the channel key at this stage. | |
1071 | */ | |
1072 | steal_channel_key(channel->key); | |
c77fc10a | 1073 | |
b5a6470f | 1074 | rcu_read_lock(); |
d88aee68 | 1075 | lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node); |
0fc67d9b JG |
1076 | lttng_ht_add_u64(consumer_data.channels_by_session_id_ht, |
1077 | &channel->channels_by_session_id_ht_node); | |
6065ceec | 1078 | rcu_read_unlock(); |
b5a6470f | 1079 | |
ec6ea7d0 | 1080 | pthread_mutex_unlock(&channel->timer_lock); |
a9838785 | 1081 | pthread_mutex_unlock(&channel->lock); |
3bd1e081 | 1082 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 1083 | |
b5a6470f | 1084 | if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) { |
a0cbdd2e | 1085 | notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD); |
d8ef542d | 1086 | } |
b5a6470f DG |
1087 | |
1088 | return 0; | |
3bd1e081 MD |
1089 | } |
1090 | ||
1091 | /* | |
1092 | * Allocate the pollfd structure and the local view of the out fds to avoid | |
1093 | * doing a lookup in the linked list and concurrency issues when writing is | |
1094 | * needed. Called with consumer_data.lock held. | |
1095 | * | |
1096 | * Returns the number of fds in the structures. | |
1097 | */ | |
ffe60014 DG |
1098 | static int update_poll_array(struct lttng_consumer_local_data *ctx, |
1099 | struct pollfd **pollfd, struct lttng_consumer_stream **local_stream, | |
9a2fcf78 | 1100 | struct lttng_ht *ht, int *nb_inactive_fd) |
3bd1e081 | 1101 | { |
3bd1e081 | 1102 | int i = 0; |
e4421fec DG |
1103 | struct lttng_ht_iter iter; |
1104 | struct lttng_consumer_stream *stream; | |
3bd1e081 | 1105 | |
ffe60014 DG |
1106 | assert(ctx); |
1107 | assert(ht); | |
1108 | assert(pollfd); | |
1109 | assert(local_stream); | |
1110 | ||
3bd1e081 | 1111 | DBG("Updating poll fd array"); |
9a2fcf78 | 1112 | *nb_inactive_fd = 0; |
481d6c57 | 1113 | rcu_read_lock(); |
43c34bc3 | 1114 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
8994307f DG |
1115 | /* |
1116 | * Only active streams with an active end point can be added to the | |
1117 | * poll set and local stream storage of the thread. | |
1118 | * | |
1119 | * There is a potential race here for endpoint_status to be updated | |
1120 | * just after the check. However, this is OK since the stream(s) will | |
1121 | * be deleted once the thread is notified that the end point state has | |
1122 | * changed where this function will be called back again. | |
9a2fcf78 JD |
1123 | * |
1124 | * We track the number of inactive FDs because they still need to be | |
1125 | * closed by the polling thread after a wakeup on the data_pipe or | |
1126 | * metadata_pipe. | |
8994307f DG |
1127 | */ |
1128 | if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM || | |
79d4ffb7 | 1129 | stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { |
9a2fcf78 | 1130 | (*nb_inactive_fd)++; |
3bd1e081 MD |
1131 | continue; |
1132 | } | |
7972aab2 DG |
1133 | /* |
1134 | * This clobbers way too much the debug output. Uncomment that if you | |
1135 | * need it for debugging purposes. | |
1136 | * | |
1137 | * DBG("Active FD %d", stream->wait_fd); | |
1138 | */ | |
e4421fec | 1139 | (*pollfd)[i].fd = stream->wait_fd; |
3bd1e081 | 1140 | (*pollfd)[i].events = POLLIN | POLLPRI; |
e4421fec | 1141 | local_stream[i] = stream; |
3bd1e081 MD |
1142 | i++; |
1143 | } | |
481d6c57 | 1144 | rcu_read_unlock(); |
3bd1e081 MD |
1145 | |
1146 | /* | |
50f8ae69 | 1147 | * Insert the consumer_data_pipe at the end of the array and don't |
3bd1e081 MD |
1148 | * increment i so nb_fd is the number of real FD. |
1149 | */ | |
acdb9057 | 1150 | (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe); |
509bb1cf | 1151 | (*pollfd)[i].events = POLLIN | POLLPRI; |
02b3d176 DG |
1152 | |
1153 | (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe); | |
1154 | (*pollfd)[i + 1].events = POLLIN | POLLPRI; | |
3bd1e081 MD |
1155 | return i; |
1156 | } | |
1157 | ||
1158 | /* | |
84382d49 MD |
1159 | * Poll on the should_quit pipe and the command socket return -1 on |
1160 | * error, 1 if should exit, 0 if data is available on the command socket | |
3bd1e081 MD |
1161 | */ |
1162 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) | |
1163 | { | |
1164 | int num_rdy; | |
1165 | ||
88f2b785 | 1166 | restart: |
3bd1e081 MD |
1167 | num_rdy = poll(consumer_sockpoll, 2, -1); |
1168 | if (num_rdy == -1) { | |
88f2b785 MD |
1169 | /* |
1170 | * Restart interrupted system call. | |
1171 | */ | |
1172 | if (errno == EINTR) { | |
1173 | goto restart; | |
1174 | } | |
7a57cf92 | 1175 | PERROR("Poll error"); |
84382d49 | 1176 | return -1; |
3bd1e081 | 1177 | } |
509bb1cf | 1178 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
3bd1e081 | 1179 | DBG("consumer_should_quit wake up"); |
84382d49 | 1180 | return 1; |
3bd1e081 MD |
1181 | } |
1182 | return 0; | |
3bd1e081 MD |
1183 | } |
1184 | ||
1185 | /* | |
1186 | * Set the error socket. | |
1187 | */ | |
ffe60014 DG |
1188 | void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx, |
1189 | int sock) | |
3bd1e081 MD |
1190 | { |
1191 | ctx->consumer_error_socket = sock; | |
1192 | } | |
1193 | ||
1194 | /* | |
1195 | * Set the command socket path. | |
1196 | */ | |
3bd1e081 MD |
1197 | void lttng_consumer_set_command_sock_path( |
1198 | struct lttng_consumer_local_data *ctx, char *sock) | |
1199 | { | |
1200 | ctx->consumer_command_sock_path = sock; | |
1201 | } | |
1202 | ||
1203 | /* | |
1204 | * Send return code to the session daemon. | |
1205 | * If the socket is not defined, we return 0, it is not a fatal error | |
1206 | */ | |
ffe60014 | 1207 | int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd) |
3bd1e081 MD |
1208 | { |
1209 | if (ctx->consumer_error_socket > 0) { | |
1210 | return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd, | |
1211 | sizeof(enum lttcomm_sessiond_command)); | |
1212 | } | |
1213 | ||
1214 | return 0; | |
1215 | } | |
1216 | ||
1217 | /* | |
228b5bf7 DG |
1218 | * Close all the tracefiles and stream fds and MUST be called when all |
1219 | * instances are destroyed i.e. when all threads were joined and are ended. | |
3bd1e081 MD |
1220 | */ |
1221 | void lttng_consumer_cleanup(void) | |
1222 | { | |
e4421fec | 1223 | struct lttng_ht_iter iter; |
ffe60014 | 1224 | struct lttng_consumer_channel *channel; |
6065ceec DG |
1225 | |
1226 | rcu_read_lock(); | |
3bd1e081 | 1227 | |
ffe60014 DG |
1228 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel, |
1229 | node.node) { | |
702b1ea4 | 1230 | consumer_del_channel(channel); |
3bd1e081 | 1231 | } |
6065ceec DG |
1232 | |
1233 | rcu_read_unlock(); | |
d6ce1df2 | 1234 | |
d6ce1df2 | 1235 | lttng_ht_destroy(consumer_data.channel_ht); |
0fc67d9b | 1236 | lttng_ht_destroy(consumer_data.channels_by_session_id_ht); |
228b5bf7 DG |
1237 | |
1238 | cleanup_relayd_ht(); | |
1239 | ||
d8ef542d MD |
1240 | lttng_ht_destroy(consumer_data.stream_per_chan_id_ht); |
1241 | ||
228b5bf7 DG |
1242 | /* |
1243 | * This HT contains streams that are freed by either the metadata thread or | |
1244 | * the data thread so we do *nothing* on the hash table and simply destroy | |
1245 | * it. | |
1246 | */ | |
1247 | lttng_ht_destroy(consumer_data.stream_list_ht); | |
3bd1e081 MD |
1248 | } |
1249 | ||
1250 | /* | |
1251 | * Called from signal handler. | |
1252 | */ | |
1253 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) | |
1254 | { | |
6cd525e8 MD |
1255 | ssize_t ret; |
1256 | ||
10211f5c | 1257 | CMM_STORE_SHARED(consumer_quit, 1); |
6cd525e8 MD |
1258 | ret = lttng_write(ctx->consumer_should_quit[1], "4", 1); |
1259 | if (ret < 1) { | |
7a57cf92 | 1260 | PERROR("write consumer quit"); |
3bd1e081 | 1261 | } |
ab1027f4 DG |
1262 | |
1263 | DBG("Consumer flag that it should quit"); | |
3bd1e081 MD |
1264 | } |
1265 | ||
5199ffc4 JG |
1266 | |
1267 | /* | |
1268 | * Flush pending writes to trace output disk file. | |
1269 | */ | |
1270 | static | |
00e2e675 DG |
1271 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
1272 | off_t orig_offset) | |
3bd1e081 | 1273 | { |
c7a78aab | 1274 | int ret; |
3bd1e081 MD |
1275 | int outfd = stream->out_fd; |
1276 | ||
1277 | /* | |
1278 | * This does a blocking write-and-wait on any page that belongs to the | |
1279 | * subbuffer prior to the one we just wrote. | |
1280 | * Don't care about error values, as these are just hints and ways to | |
1281 | * limit the amount of page cache used. | |
1282 | */ | |
ffe60014 | 1283 | if (orig_offset < stream->max_sb_size) { |
3bd1e081 MD |
1284 | return; |
1285 | } | |
ffe60014 DG |
1286 | lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size, |
1287 | stream->max_sb_size, | |
3bd1e081 MD |
1288 | SYNC_FILE_RANGE_WAIT_BEFORE |
1289 | | SYNC_FILE_RANGE_WRITE | |
1290 | | SYNC_FILE_RANGE_WAIT_AFTER); | |
1291 | /* | |
1292 | * Give hints to the kernel about how we access the file: | |
1293 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after | |
1294 | * we write it. | |
1295 | * | |
1296 | * We need to call fadvise again after the file grows because the | |
1297 | * kernel does not seem to apply fadvise to non-existing parts of the | |
1298 | * file. | |
1299 | * | |
1300 | * Call fadvise _after_ having waited for the page writeback to | |
1301 | * complete because the dirty page writeback semantic is not well | |
1302 | * defined. So it can be expected to lead to lower throughput in | |
1303 | * streaming. | |
1304 | */ | |
c7a78aab | 1305 | ret = posix_fadvise(outfd, orig_offset - stream->max_sb_size, |
ffe60014 | 1306 | stream->max_sb_size, POSIX_FADV_DONTNEED); |
a0d0e127 | 1307 | if (ret && ret != -ENOSYS) { |
a74a5f4a JG |
1308 | errno = ret; |
1309 | PERROR("posix_fadvise on fd %i", outfd); | |
c7a78aab | 1310 | } |
3bd1e081 MD |
1311 | } |
1312 | ||
1313 | /* | |
1314 | * Initialise the necessary environnement : | |
1315 | * - create a new context | |
1316 | * - create the poll_pipe | |
1317 | * - create the should_quit pipe (for signal handler) | |
1318 | * - create the thread pipe (for splice) | |
1319 | * | |
1320 | * Takes a function pointer as argument, this function is called when data is | |
1321 | * available on a buffer. This function is responsible to do the | |
1322 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the | |
1323 | * buffer configuration and then kernctl_put_next_subbuf at the end. | |
1324 | * | |
1325 | * Returns a pointer to the new context or NULL on error. | |
1326 | */ | |
1327 | struct lttng_consumer_local_data *lttng_consumer_create( | |
1328 | enum lttng_consumer_type type, | |
4078b776 | 1329 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
d41f73b7 | 1330 | struct lttng_consumer_local_data *ctx), |
3bd1e081 MD |
1331 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
1332 | int (*recv_stream)(struct lttng_consumer_stream *stream), | |
30319bcb | 1333 | int (*update_stream)(uint64_t stream_key, uint32_t state)) |
3bd1e081 | 1334 | { |
d8ef542d | 1335 | int ret; |
3bd1e081 MD |
1336 | struct lttng_consumer_local_data *ctx; |
1337 | ||
1338 | assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || | |
1339 | consumer_data.type == type); | |
1340 | consumer_data.type = type; | |
1341 | ||
effcf122 | 1342 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
3bd1e081 | 1343 | if (ctx == NULL) { |
7a57cf92 | 1344 | PERROR("allocating context"); |
3bd1e081 MD |
1345 | goto error; |
1346 | } | |
1347 | ||
1348 | ctx->consumer_error_socket = -1; | |
331744e3 | 1349 | ctx->consumer_metadata_socket = -1; |
75d83e50 | 1350 | pthread_mutex_init(&ctx->metadata_socket_lock, NULL); |
3bd1e081 MD |
1351 | /* assign the callbacks */ |
1352 | ctx->on_buffer_ready = buffer_ready; | |
1353 | ctx->on_recv_channel = recv_channel; | |
1354 | ctx->on_recv_stream = recv_stream; | |
1355 | ctx->on_update_stream = update_stream; | |
1356 | ||
acdb9057 DG |
1357 | ctx->consumer_data_pipe = lttng_pipe_open(0); |
1358 | if (!ctx->consumer_data_pipe) { | |
3bd1e081 MD |
1359 | goto error_poll_pipe; |
1360 | } | |
1361 | ||
02b3d176 DG |
1362 | ctx->consumer_wakeup_pipe = lttng_pipe_open(0); |
1363 | if (!ctx->consumer_wakeup_pipe) { | |
1364 | goto error_wakeup_pipe; | |
1365 | } | |
1366 | ||
3bd1e081 MD |
1367 | ret = pipe(ctx->consumer_should_quit); |
1368 | if (ret < 0) { | |
7a57cf92 | 1369 | PERROR("Error creating recv pipe"); |
3bd1e081 MD |
1370 | goto error_quit_pipe; |
1371 | } | |
1372 | ||
d8ef542d MD |
1373 | ret = pipe(ctx->consumer_channel_pipe); |
1374 | if (ret < 0) { | |
1375 | PERROR("Error creating channel pipe"); | |
1376 | goto error_channel_pipe; | |
1377 | } | |
1378 | ||
13886d2d DG |
1379 | ctx->consumer_metadata_pipe = lttng_pipe_open(0); |
1380 | if (!ctx->consumer_metadata_pipe) { | |
fb3a43a9 DG |
1381 | goto error_metadata_pipe; |
1382 | } | |
3bd1e081 | 1383 | |
e9404c27 JG |
1384 | ctx->channel_monitor_pipe = -1; |
1385 | ||
fb3a43a9 | 1386 | return ctx; |
3bd1e081 | 1387 | |
fb3a43a9 | 1388 | error_metadata_pipe: |
d8ef542d MD |
1389 | utils_close_pipe(ctx->consumer_channel_pipe); |
1390 | error_channel_pipe: | |
d8ef542d | 1391 | utils_close_pipe(ctx->consumer_should_quit); |
3bd1e081 | 1392 | error_quit_pipe: |
02b3d176 DG |
1393 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
1394 | error_wakeup_pipe: | |
acdb9057 | 1395 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
3bd1e081 MD |
1396 | error_poll_pipe: |
1397 | free(ctx); | |
1398 | error: | |
1399 | return NULL; | |
1400 | } | |
1401 | ||
282dadbc MD |
1402 | /* |
1403 | * Iterate over all streams of the hashtable and free them properly. | |
1404 | */ | |
1405 | static void destroy_data_stream_ht(struct lttng_ht *ht) | |
1406 | { | |
1407 | struct lttng_ht_iter iter; | |
1408 | struct lttng_consumer_stream *stream; | |
1409 | ||
1410 | if (ht == NULL) { | |
1411 | return; | |
1412 | } | |
1413 | ||
1414 | rcu_read_lock(); | |
1415 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1416 | /* | |
1417 | * Ignore return value since we are currently cleaning up so any error | |
1418 | * can't be handled. | |
1419 | */ | |
1420 | (void) consumer_del_stream(stream, ht); | |
1421 | } | |
1422 | rcu_read_unlock(); | |
1423 | ||
1424 | lttng_ht_destroy(ht); | |
1425 | } | |
1426 | ||
1427 | /* | |
1428 | * Iterate over all streams of the metadata hashtable and free them | |
1429 | * properly. | |
1430 | */ | |
1431 | static void destroy_metadata_stream_ht(struct lttng_ht *ht) | |
1432 | { | |
1433 | struct lttng_ht_iter iter; | |
1434 | struct lttng_consumer_stream *stream; | |
1435 | ||
1436 | if (ht == NULL) { | |
1437 | return; | |
1438 | } | |
1439 | ||
1440 | rcu_read_lock(); | |
1441 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1442 | /* | |
1443 | * Ignore return value since we are currently cleaning up so any error | |
1444 | * can't be handled. | |
1445 | */ | |
1446 | (void) consumer_del_metadata_stream(stream, ht); | |
1447 | } | |
1448 | rcu_read_unlock(); | |
1449 | ||
1450 | lttng_ht_destroy(ht); | |
1451 | } | |
1452 | ||
3bd1e081 MD |
1453 | /* |
1454 | * Close all fds associated with the instance and free the context. | |
1455 | */ | |
1456 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) | |
1457 | { | |
4c462e79 MD |
1458 | int ret; |
1459 | ||
ab1027f4 DG |
1460 | DBG("Consumer destroying it. Closing everything."); |
1461 | ||
4f2e75b9 DG |
1462 | if (!ctx) { |
1463 | return; | |
1464 | } | |
1465 | ||
282dadbc MD |
1466 | destroy_data_stream_ht(data_ht); |
1467 | destroy_metadata_stream_ht(metadata_ht); | |
1468 | ||
4c462e79 MD |
1469 | ret = close(ctx->consumer_error_socket); |
1470 | if (ret) { | |
1471 | PERROR("close"); | |
1472 | } | |
331744e3 JD |
1473 | ret = close(ctx->consumer_metadata_socket); |
1474 | if (ret) { | |
1475 | PERROR("close"); | |
1476 | } | |
d8ef542d | 1477 | utils_close_pipe(ctx->consumer_channel_pipe); |
acdb9057 | 1478 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
13886d2d | 1479 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
02b3d176 | 1480 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
d8ef542d | 1481 | utils_close_pipe(ctx->consumer_should_quit); |
fb3a43a9 | 1482 | |
3bd1e081 MD |
1483 | unlink(ctx->consumer_command_sock_path); |
1484 | free(ctx); | |
1485 | } | |
1486 | ||
6197aea7 DG |
1487 | /* |
1488 | * Write the metadata stream id on the specified file descriptor. | |
1489 | */ | |
1490 | static int write_relayd_metadata_id(int fd, | |
1491 | struct lttng_consumer_stream *stream, | |
239f61af | 1492 | unsigned long padding) |
6197aea7 | 1493 | { |
6cd525e8 | 1494 | ssize_t ret; |
1d4dfdef | 1495 | struct lttcomm_relayd_metadata_payload hdr; |
6197aea7 | 1496 | |
1d4dfdef DG |
1497 | hdr.stream_id = htobe64(stream->relayd_stream_id); |
1498 | hdr.padding_size = htobe32(padding); | |
6cd525e8 MD |
1499 | ret = lttng_write(fd, (void *) &hdr, sizeof(hdr)); |
1500 | if (ret < sizeof(hdr)) { | |
d7b75ec8 | 1501 | /* |
6f04ed72 | 1502 | * This error means that the fd's end is closed so ignore the PERROR |
d7b75ec8 DG |
1503 | * not to clubber the error output since this can happen in a normal |
1504 | * code path. | |
1505 | */ | |
1506 | if (errno != EPIPE) { | |
1507 | PERROR("write metadata stream id"); | |
1508 | } | |
1509 | DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno); | |
534d2592 DG |
1510 | /* |
1511 | * Set ret to a negative value because if ret != sizeof(hdr), we don't | |
1512 | * handle writting the missing part so report that as an error and | |
1513 | * don't lie to the caller. | |
1514 | */ | |
1515 | ret = -1; | |
6197aea7 DG |
1516 | goto end; |
1517 | } | |
1d4dfdef DG |
1518 | DBG("Metadata stream id %" PRIu64 " with padding %lu written before data", |
1519 | stream->relayd_stream_id, padding); | |
6197aea7 DG |
1520 | |
1521 | end: | |
6cd525e8 | 1522 | return (int) ret; |
6197aea7 DG |
1523 | } |
1524 | ||
3bd1e081 | 1525 | /* |
09e26845 DG |
1526 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
1527 | * core function for writing trace buffers to either the local filesystem or | |
1528 | * the network. | |
1529 | * | |
79d4ffb7 DG |
1530 | * It must be called with the stream lock held. |
1531 | * | |
09e26845 | 1532 | * Careful review MUST be put if any changes occur! |
3bd1e081 MD |
1533 | * |
1534 | * Returns the number of bytes written | |
1535 | */ | |
4078b776 | 1536 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
3bd1e081 | 1537 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1538 | struct lttng_consumer_stream *stream, unsigned long len, |
309167d2 | 1539 | unsigned long padding, |
50adc264 | 1540 | struct ctf_packet_index *index) |
3bd1e081 | 1541 | { |
f02e1e8a | 1542 | unsigned long mmap_offset; |
ffe60014 | 1543 | void *mmap_base; |
994ab360 | 1544 | ssize_t ret = 0; |
f02e1e8a DG |
1545 | off_t orig_offset = stream->out_fd_offset; |
1546 | /* Default is on the disk */ | |
1547 | int outfd = stream->out_fd; | |
f02e1e8a | 1548 | struct consumer_relayd_sock_pair *relayd = NULL; |
8994307f | 1549 | unsigned int relayd_hang_up = 0; |
f02e1e8a DG |
1550 | |
1551 | /* RCU lock for the relayd pointer */ | |
1552 | rcu_read_lock(); | |
1553 | ||
1554 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1555 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1556 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1557 | if (relayd == NULL) { | |
56591bac | 1558 | ret = -EPIPE; |
f02e1e8a DG |
1559 | goto end; |
1560 | } | |
1561 | } | |
1562 | ||
1563 | /* get the offset inside the fd to mmap */ | |
3bd1e081 MD |
1564 | switch (consumer_data.type) { |
1565 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1566 | mmap_base = stream->mmap_base; |
f02e1e8a | 1567 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); |
994ab360 | 1568 | if (ret < 0) { |
56591bac | 1569 | PERROR("tracer ctl get_mmap_read_offset"); |
56591bac MD |
1570 | goto end; |
1571 | } | |
f02e1e8a | 1572 | break; |
7753dea8 MD |
1573 | case LTTNG_CONSUMER32_UST: |
1574 | case LTTNG_CONSUMER64_UST: | |
ffe60014 DG |
1575 | mmap_base = lttng_ustctl_get_mmap_base(stream); |
1576 | if (!mmap_base) { | |
1577 | ERR("read mmap get mmap base for stream %s", stream->name); | |
994ab360 | 1578 | ret = -EPERM; |
ffe60014 DG |
1579 | goto end; |
1580 | } | |
1581 | ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset); | |
56591bac MD |
1582 | if (ret != 0) { |
1583 | PERROR("tracer ctl get_mmap_read_offset"); | |
994ab360 | 1584 | ret = -EINVAL; |
56591bac MD |
1585 | goto end; |
1586 | } | |
f02e1e8a | 1587 | break; |
3bd1e081 MD |
1588 | default: |
1589 | ERR("Unknown consumer_data type"); | |
1590 | assert(0); | |
1591 | } | |
b9182dd9 | 1592 | |
f02e1e8a DG |
1593 | /* Handle stream on the relayd if the output is on the network */ |
1594 | if (relayd) { | |
1595 | unsigned long netlen = len; | |
1596 | ||
1597 | /* | |
1598 | * Lock the control socket for the complete duration of the function | |
1599 | * since from this point on we will use the socket. | |
1600 | */ | |
1601 | if (stream->metadata_flag) { | |
1602 | /* Metadata requires the control socket. */ | |
1603 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
93ec662e JD |
1604 | if (stream->reset_metadata_flag) { |
1605 | ret = relayd_reset_metadata(&relayd->control_sock, | |
1606 | stream->relayd_stream_id, | |
1607 | stream->metadata_version); | |
1608 | if (ret < 0) { | |
1609 | relayd_hang_up = 1; | |
1610 | goto write_error; | |
1611 | } | |
1612 | stream->reset_metadata_flag = 0; | |
1613 | } | |
1d4dfdef | 1614 | netlen += sizeof(struct lttcomm_relayd_metadata_payload); |
f02e1e8a DG |
1615 | } |
1616 | ||
1d4dfdef | 1617 | ret = write_relayd_stream_header(stream, netlen, padding, relayd); |
994ab360 DG |
1618 | if (ret < 0) { |
1619 | relayd_hang_up = 1; | |
1620 | goto write_error; | |
1621 | } | |
1622 | /* Use the returned socket. */ | |
1623 | outfd = ret; | |
f02e1e8a | 1624 | |
994ab360 DG |
1625 | /* Write metadata stream id before payload */ |
1626 | if (stream->metadata_flag) { | |
239f61af | 1627 | ret = write_relayd_metadata_id(outfd, stream, padding); |
994ab360 | 1628 | if (ret < 0) { |
8994307f DG |
1629 | relayd_hang_up = 1; |
1630 | goto write_error; | |
1631 | } | |
f02e1e8a | 1632 | } |
1d4dfdef DG |
1633 | } else { |
1634 | /* No streaming, we have to set the len with the full padding */ | |
1635 | len += padding; | |
1624d5b7 | 1636 | |
93ec662e JD |
1637 | if (stream->metadata_flag && stream->reset_metadata_flag) { |
1638 | ret = utils_truncate_stream_file(stream->out_fd, 0); | |
1639 | if (ret < 0) { | |
1640 | ERR("Reset metadata file"); | |
1641 | goto end; | |
1642 | } | |
1643 | stream->reset_metadata_flag = 0; | |
1644 | } | |
1645 | ||
1624d5b7 JD |
1646 | /* |
1647 | * Check if we need to change the tracefile before writing the packet. | |
1648 | */ | |
1649 | if (stream->chan->tracefile_size > 0 && | |
1650 | (stream->tracefile_size_current + len) > | |
1651 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1652 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1653 | stream->name, stream->chan->tracefile_size, | |
1654 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
309167d2 JD |
1655 | stream->out_fd, &(stream->tracefile_count_current), |
1656 | &stream->out_fd); | |
1624d5b7 JD |
1657 | if (ret < 0) { |
1658 | ERR("Rotating output file"); | |
1659 | goto end; | |
1660 | } | |
309167d2 JD |
1661 | outfd = stream->out_fd; |
1662 | ||
f8f3885c MD |
1663 | if (stream->index_file) { |
1664 | lttng_index_file_put(stream->index_file); | |
1665 | stream->index_file = lttng_index_file_create(stream->chan->pathname, | |
309167d2 JD |
1666 | stream->name, stream->uid, stream->gid, |
1667 | stream->chan->tracefile_size, | |
f8f3885c MD |
1668 | stream->tracefile_count_current, |
1669 | CTF_INDEX_MAJOR, CTF_INDEX_MINOR); | |
1670 | if (!stream->index_file) { | |
309167d2 JD |
1671 | goto end; |
1672 | } | |
309167d2 JD |
1673 | } |
1674 | ||
a6976990 DG |
1675 | /* Reset current size because we just perform a rotation. */ |
1676 | stream->tracefile_size_current = 0; | |
a1ae300f JD |
1677 | stream->out_fd_offset = 0; |
1678 | orig_offset = 0; | |
1624d5b7 JD |
1679 | } |
1680 | stream->tracefile_size_current += len; | |
309167d2 JD |
1681 | if (index) { |
1682 | index->offset = htobe64(stream->out_fd_offset); | |
1683 | } | |
f02e1e8a DG |
1684 | } |
1685 | ||
d02b8372 DG |
1686 | /* |
1687 | * This call guarantee that len or less is returned. It's impossible to | |
1688 | * receive a ret value that is bigger than len. | |
1689 | */ | |
1690 | ret = lttng_write(outfd, mmap_base + mmap_offset, len); | |
1691 | DBG("Consumer mmap write() ret %zd (len %lu)", ret, len); | |
1692 | if (ret < 0 || ((size_t) ret != len)) { | |
1693 | /* | |
1694 | * Report error to caller if nothing was written else at least send the | |
1695 | * amount written. | |
1696 | */ | |
1697 | if (ret < 0) { | |
994ab360 | 1698 | ret = -errno; |
f02e1e8a | 1699 | } |
994ab360 | 1700 | relayd_hang_up = 1; |
f02e1e8a | 1701 | |
d02b8372 | 1702 | /* Socket operation failed. We consider the relayd dead */ |
994ab360 | 1703 | if (errno == EPIPE || errno == EINVAL || errno == EBADF) { |
d02b8372 DG |
1704 | /* |
1705 | * This is possible if the fd is closed on the other side | |
1706 | * (outfd) or any write problem. It can be verbose a bit for a | |
1707 | * normal execution if for instance the relayd is stopped | |
1708 | * abruptly. This can happen so set this to a DBG statement. | |
1709 | */ | |
1710 | DBG("Consumer mmap write detected relayd hang up"); | |
994ab360 DG |
1711 | } else { |
1712 | /* Unhandled error, print it and stop function right now. */ | |
1713 | PERROR("Error in write mmap (ret %zd != len %lu)", ret, len); | |
f02e1e8a | 1714 | } |
994ab360 | 1715 | goto write_error; |
d02b8372 DG |
1716 | } |
1717 | stream->output_written += ret; | |
d02b8372 DG |
1718 | |
1719 | /* This call is useless on a socket so better save a syscall. */ | |
1720 | if (!relayd) { | |
1721 | /* This won't block, but will start writeout asynchronously */ | |
1722 | lttng_sync_file_range(outfd, stream->out_fd_offset, len, | |
1723 | SYNC_FILE_RANGE_WRITE); | |
1724 | stream->out_fd_offset += len; | |
f5dbe415 | 1725 | lttng_consumer_sync_trace_file(stream, orig_offset); |
f02e1e8a | 1726 | } |
f02e1e8a | 1727 | |
8994307f DG |
1728 | write_error: |
1729 | /* | |
1730 | * This is a special case that the relayd has closed its socket. Let's | |
1731 | * cleanup the relayd object and all associated streams. | |
1732 | */ | |
1733 | if (relayd && relayd_hang_up) { | |
11413bb9 JR |
1734 | ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
1735 | lttng_consumer_cleanup_relayd(relayd); | |
8994307f DG |
1736 | } |
1737 | ||
f02e1e8a DG |
1738 | end: |
1739 | /* Unlock only if ctrl socket used */ | |
1740 | if (relayd && stream->metadata_flag) { | |
1741 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1742 | } | |
1743 | ||
1744 | rcu_read_unlock(); | |
994ab360 | 1745 | return ret; |
3bd1e081 MD |
1746 | } |
1747 | ||
1748 | /* | |
1749 | * Splice the data from the ring buffer to the tracefile. | |
1750 | * | |
79d4ffb7 DG |
1751 | * It must be called with the stream lock held. |
1752 | * | |
3bd1e081 MD |
1753 | * Returns the number of bytes spliced. |
1754 | */ | |
4078b776 | 1755 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
3bd1e081 | 1756 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1757 | struct lttng_consumer_stream *stream, unsigned long len, |
309167d2 | 1758 | unsigned long padding, |
50adc264 | 1759 | struct ctf_packet_index *index) |
3bd1e081 | 1760 | { |
f02e1e8a DG |
1761 | ssize_t ret = 0, written = 0, ret_splice = 0; |
1762 | loff_t offset = 0; | |
1763 | off_t orig_offset = stream->out_fd_offset; | |
1764 | int fd = stream->wait_fd; | |
1765 | /* Default is on the disk */ | |
1766 | int outfd = stream->out_fd; | |
f02e1e8a | 1767 | struct consumer_relayd_sock_pair *relayd = NULL; |
fb3a43a9 | 1768 | int *splice_pipe; |
8994307f | 1769 | unsigned int relayd_hang_up = 0; |
f02e1e8a | 1770 | |
3bd1e081 MD |
1771 | switch (consumer_data.type) { |
1772 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a | 1773 | break; |
7753dea8 MD |
1774 | case LTTNG_CONSUMER32_UST: |
1775 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a | 1776 | /* Not supported for user space tracing */ |
3bd1e081 MD |
1777 | return -ENOSYS; |
1778 | default: | |
1779 | ERR("Unknown consumer_data type"); | |
1780 | assert(0); | |
3bd1e081 MD |
1781 | } |
1782 | ||
f02e1e8a DG |
1783 | /* RCU lock for the relayd pointer */ |
1784 | rcu_read_lock(); | |
1785 | ||
1786 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1787 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1788 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1789 | if (relayd == NULL) { | |
ad0b0d23 | 1790 | written = -ret; |
f02e1e8a DG |
1791 | goto end; |
1792 | } | |
1793 | } | |
a2361a61 | 1794 | splice_pipe = stream->splice_pipe; |
fb3a43a9 | 1795 | |
f02e1e8a | 1796 | /* Write metadata stream id before payload */ |
1d4dfdef | 1797 | if (relayd) { |
ad0b0d23 | 1798 | unsigned long total_len = len; |
f02e1e8a | 1799 | |
1d4dfdef DG |
1800 | if (stream->metadata_flag) { |
1801 | /* | |
1802 | * Lock the control socket for the complete duration of the function | |
1803 | * since from this point on we will use the socket. | |
1804 | */ | |
1805 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1806 | ||
93ec662e JD |
1807 | if (stream->reset_metadata_flag) { |
1808 | ret = relayd_reset_metadata(&relayd->control_sock, | |
1809 | stream->relayd_stream_id, | |
1810 | stream->metadata_version); | |
1811 | if (ret < 0) { | |
1812 | relayd_hang_up = 1; | |
1813 | goto write_error; | |
1814 | } | |
1815 | stream->reset_metadata_flag = 0; | |
1816 | } | |
239f61af | 1817 | ret = write_relayd_metadata_id(splice_pipe[1], stream, |
1d4dfdef DG |
1818 | padding); |
1819 | if (ret < 0) { | |
1820 | written = ret; | |
ad0b0d23 DG |
1821 | relayd_hang_up = 1; |
1822 | goto write_error; | |
1d4dfdef DG |
1823 | } |
1824 | ||
1825 | total_len += sizeof(struct lttcomm_relayd_metadata_payload); | |
1826 | } | |
1827 | ||
1828 | ret = write_relayd_stream_header(stream, total_len, padding, relayd); | |
ad0b0d23 DG |
1829 | if (ret < 0) { |
1830 | written = ret; | |
1831 | relayd_hang_up = 1; | |
1832 | goto write_error; | |
f02e1e8a | 1833 | } |
ad0b0d23 DG |
1834 | /* Use the returned socket. */ |
1835 | outfd = ret; | |
1d4dfdef DG |
1836 | } else { |
1837 | /* No streaming, we have to set the len with the full padding */ | |
1838 | len += padding; | |
1624d5b7 | 1839 | |
93ec662e JD |
1840 | if (stream->metadata_flag && stream->reset_metadata_flag) { |
1841 | ret = utils_truncate_stream_file(stream->out_fd, 0); | |
1842 | if (ret < 0) { | |
1843 | ERR("Reset metadata file"); | |
1844 | goto end; | |
1845 | } | |
1846 | stream->reset_metadata_flag = 0; | |
1847 | } | |
1624d5b7 JD |
1848 | /* |
1849 | * Check if we need to change the tracefile before writing the packet. | |
1850 | */ | |
1851 | if (stream->chan->tracefile_size > 0 && | |
1852 | (stream->tracefile_size_current + len) > | |
1853 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1854 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1855 | stream->name, stream->chan->tracefile_size, | |
1856 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
309167d2 JD |
1857 | stream->out_fd, &(stream->tracefile_count_current), |
1858 | &stream->out_fd); | |
1624d5b7 | 1859 | if (ret < 0) { |
ad0b0d23 | 1860 | written = ret; |
1624d5b7 JD |
1861 | ERR("Rotating output file"); |
1862 | goto end; | |
1863 | } | |
309167d2 JD |
1864 | outfd = stream->out_fd; |
1865 | ||
f8f3885c MD |
1866 | if (stream->index_file) { |
1867 | lttng_index_file_put(stream->index_file); | |
1868 | stream->index_file = lttng_index_file_create(stream->chan->pathname, | |
309167d2 JD |
1869 | stream->name, stream->uid, stream->gid, |
1870 | stream->chan->tracefile_size, | |
f8f3885c MD |
1871 | stream->tracefile_count_current, |
1872 | CTF_INDEX_MAJOR, CTF_INDEX_MINOR); | |
1873 | if (!stream->index_file) { | |
309167d2 JD |
1874 | goto end; |
1875 | } | |
309167d2 JD |
1876 | } |
1877 | ||
a6976990 DG |
1878 | /* Reset current size because we just perform a rotation. */ |
1879 | stream->tracefile_size_current = 0; | |
a1ae300f JD |
1880 | stream->out_fd_offset = 0; |
1881 | orig_offset = 0; | |
1624d5b7 JD |
1882 | } |
1883 | stream->tracefile_size_current += len; | |
309167d2 | 1884 | index->offset = htobe64(stream->out_fd_offset); |
f02e1e8a DG |
1885 | } |
1886 | ||
1887 | while (len > 0) { | |
1d4dfdef DG |
1888 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)", |
1889 | (unsigned long)offset, len, fd, splice_pipe[1]); | |
fb3a43a9 | 1890 | ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len, |
f02e1e8a DG |
1891 | SPLICE_F_MOVE | SPLICE_F_MORE); |
1892 | DBG("splice chan to pipe, ret %zd", ret_splice); | |
1893 | if (ret_splice < 0) { | |
d02b8372 | 1894 | ret = errno; |
ad0b0d23 | 1895 | written = -ret; |
d02b8372 | 1896 | PERROR("Error in relay splice"); |
f02e1e8a DG |
1897 | goto splice_error; |
1898 | } | |
1899 | ||
1900 | /* Handle stream on the relayd if the output is on the network */ | |
ad0b0d23 DG |
1901 | if (relayd && stream->metadata_flag) { |
1902 | size_t metadata_payload_size = | |
1903 | sizeof(struct lttcomm_relayd_metadata_payload); | |
1904 | ||
1905 | /* Update counter to fit the spliced data */ | |
1906 | ret_splice += metadata_payload_size; | |
1907 | len += metadata_payload_size; | |
1908 | /* | |
1909 | * We do this so the return value can match the len passed as | |
1910 | * argument to this function. | |
1911 | */ | |
1912 | written -= metadata_payload_size; | |
f02e1e8a DG |
1913 | } |
1914 | ||
1915 | /* Splice data out */ | |
fb3a43a9 | 1916 | ret_splice = splice(splice_pipe[0], NULL, outfd, NULL, |
f02e1e8a | 1917 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); |
a2361a61 JD |
1918 | DBG("Consumer splice pipe to file (out_fd: %d), ret %zd", |
1919 | outfd, ret_splice); | |
f02e1e8a | 1920 | if (ret_splice < 0) { |
d02b8372 | 1921 | ret = errno; |
ad0b0d23 DG |
1922 | written = -ret; |
1923 | relayd_hang_up = 1; | |
1924 | goto write_error; | |
f02e1e8a | 1925 | } else if (ret_splice > len) { |
d02b8372 DG |
1926 | /* |
1927 | * We don't expect this code path to be executed but you never know | |
1928 | * so this is an extra protection agains a buggy splice(). | |
1929 | */ | |
f02e1e8a | 1930 | ret = errno; |
ad0b0d23 | 1931 | written += ret_splice; |
d02b8372 DG |
1932 | PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice, |
1933 | len); | |
f02e1e8a | 1934 | goto splice_error; |
d02b8372 DG |
1935 | } else { |
1936 | /* All good, update current len and continue. */ | |
1937 | len -= ret_splice; | |
f02e1e8a | 1938 | } |
f02e1e8a DG |
1939 | |
1940 | /* This call is useless on a socket so better save a syscall. */ | |
1941 | if (!relayd) { | |
1942 | /* This won't block, but will start writeout asynchronously */ | |
1943 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, | |
1944 | SYNC_FILE_RANGE_WRITE); | |
1945 | stream->out_fd_offset += ret_splice; | |
1946 | } | |
e5d1a9b3 | 1947 | stream->output_written += ret_splice; |
f02e1e8a DG |
1948 | written += ret_splice; |
1949 | } | |
f5dbe415 JG |
1950 | if (!relayd) { |
1951 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1952 | } | |
f02e1e8a DG |
1953 | goto end; |
1954 | ||
8994307f DG |
1955 | write_error: |
1956 | /* | |
1957 | * This is a special case that the relayd has closed its socket. Let's | |
1958 | * cleanup the relayd object and all associated streams. | |
1959 | */ | |
1960 | if (relayd && relayd_hang_up) { | |
11413bb9 JR |
1961 | ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
1962 | lttng_consumer_cleanup_relayd(relayd); | |
8994307f DG |
1963 | /* Skip splice error so the consumer does not fail */ |
1964 | goto end; | |
1965 | } | |
1966 | ||
f02e1e8a DG |
1967 | splice_error: |
1968 | /* send the appropriate error description to sessiond */ | |
1969 | switch (ret) { | |
f02e1e8a | 1970 | case EINVAL: |
f73fabfd | 1971 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); |
f02e1e8a DG |
1972 | break; |
1973 | case ENOMEM: | |
f73fabfd | 1974 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); |
f02e1e8a DG |
1975 | break; |
1976 | case ESPIPE: | |
f73fabfd | 1977 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); |
f02e1e8a DG |
1978 | break; |
1979 | } | |
1980 | ||
1981 | end: | |
1982 | if (relayd && stream->metadata_flag) { | |
1983 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1984 | } | |
1985 | ||
1986 | rcu_read_unlock(); | |
1987 | return written; | |
3bd1e081 MD |
1988 | } |
1989 | ||
15055ce5 JD |
1990 | /* |
1991 | * Sample the snapshot positions for a specific fd | |
1992 | * | |
1993 | * Returns 0 on success, < 0 on error | |
1994 | */ | |
1995 | int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream *stream) | |
1996 | { | |
1997 | switch (consumer_data.type) { | |
1998 | case LTTNG_CONSUMER_KERNEL: | |
1999 | return lttng_kconsumer_sample_snapshot_positions(stream); | |
2000 | case LTTNG_CONSUMER32_UST: | |
2001 | case LTTNG_CONSUMER64_UST: | |
2002 | return lttng_ustconsumer_sample_snapshot_positions(stream); | |
2003 | default: | |
2004 | ERR("Unknown consumer_data type"); | |
2005 | assert(0); | |
2006 | return -ENOSYS; | |
2007 | } | |
2008 | } | |
3bd1e081 MD |
2009 | /* |
2010 | * Take a snapshot for a specific fd | |
2011 | * | |
2012 | * Returns 0 on success, < 0 on error | |
2013 | */ | |
ffe60014 | 2014 | int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) |
3bd1e081 MD |
2015 | { |
2016 | switch (consumer_data.type) { | |
2017 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 2018 | return lttng_kconsumer_take_snapshot(stream); |
7753dea8 MD |
2019 | case LTTNG_CONSUMER32_UST: |
2020 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 2021 | return lttng_ustconsumer_take_snapshot(stream); |
3bd1e081 MD |
2022 | default: |
2023 | ERR("Unknown consumer_data type"); | |
2024 | assert(0); | |
2025 | return -ENOSYS; | |
2026 | } | |
3bd1e081 MD |
2027 | } |
2028 | ||
2029 | /* | |
2030 | * Get the produced position | |
2031 | * | |
2032 | * Returns 0 on success, < 0 on error | |
2033 | */ | |
ffe60014 | 2034 | int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, |
3bd1e081 MD |
2035 | unsigned long *pos) |
2036 | { | |
2037 | switch (consumer_data.type) { | |
2038 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 2039 | return lttng_kconsumer_get_produced_snapshot(stream, pos); |
7753dea8 MD |
2040 | case LTTNG_CONSUMER32_UST: |
2041 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 2042 | return lttng_ustconsumer_get_produced_snapshot(stream, pos); |
3bd1e081 MD |
2043 | default: |
2044 | ERR("Unknown consumer_data type"); | |
2045 | assert(0); | |
2046 | return -ENOSYS; | |
2047 | } | |
2048 | } | |
2049 | ||
15055ce5 JD |
2050 | /* |
2051 | * Get the consumed position (free-running counter position in bytes). | |
2052 | * | |
2053 | * Returns 0 on success, < 0 on error | |
2054 | */ | |
2055 | int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream *stream, | |
2056 | unsigned long *pos) | |
2057 | { | |
2058 | switch (consumer_data.type) { | |
2059 | case LTTNG_CONSUMER_KERNEL: | |
2060 | return lttng_kconsumer_get_consumed_snapshot(stream, pos); | |
2061 | case LTTNG_CONSUMER32_UST: | |
2062 | case LTTNG_CONSUMER64_UST: | |
2063 | return lttng_ustconsumer_get_consumed_snapshot(stream, pos); | |
2064 | default: | |
2065 | ERR("Unknown consumer_data type"); | |
2066 | assert(0); | |
2067 | return -ENOSYS; | |
2068 | } | |
2069 | } | |
2070 | ||
3bd1e081 MD |
2071 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, |
2072 | int sock, struct pollfd *consumer_sockpoll) | |
2073 | { | |
2074 | switch (consumer_data.type) { | |
2075 | case LTTNG_CONSUMER_KERNEL: | |
2076 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
7753dea8 MD |
2077 | case LTTNG_CONSUMER32_UST: |
2078 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
2079 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
2080 | default: | |
2081 | ERR("Unknown consumer_data type"); | |
2082 | assert(0); | |
2083 | return -ENOSYS; | |
2084 | } | |
2085 | } | |
2086 | ||
6d574024 | 2087 | void lttng_consumer_close_all_metadata(void) |
d88aee68 DG |
2088 | { |
2089 | switch (consumer_data.type) { | |
2090 | case LTTNG_CONSUMER_KERNEL: | |
2091 | /* | |
2092 | * The Kernel consumer has a different metadata scheme so we don't | |
2093 | * close anything because the stream will be closed by the session | |
2094 | * daemon. | |
2095 | */ | |
2096 | break; | |
2097 | case LTTNG_CONSUMER32_UST: | |
2098 | case LTTNG_CONSUMER64_UST: | |
2099 | /* | |
2100 | * Close all metadata streams. The metadata hash table is passed and | |
2101 | * this call iterates over it by closing all wakeup fd. This is safe | |
2102 | * because at this point we are sure that the metadata producer is | |
2103 | * either dead or blocked. | |
2104 | */ | |
6d574024 | 2105 | lttng_ustconsumer_close_all_metadata(metadata_ht); |
d88aee68 DG |
2106 | break; |
2107 | default: | |
2108 | ERR("Unknown consumer_data type"); | |
2109 | assert(0); | |
2110 | } | |
2111 | } | |
2112 | ||
fb3a43a9 DG |
2113 | /* |
2114 | * Clean up a metadata stream and free its memory. | |
2115 | */ | |
e316aad5 DG |
2116 | void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, |
2117 | struct lttng_ht *ht) | |
fb3a43a9 | 2118 | { |
e316aad5 | 2119 | struct lttng_consumer_channel *free_chan = NULL; |
fb3a43a9 DG |
2120 | |
2121 | assert(stream); | |
2122 | /* | |
2123 | * This call should NEVER receive regular stream. It must always be | |
2124 | * metadata stream and this is crucial for data structure synchronization. | |
2125 | */ | |
2126 | assert(stream->metadata_flag); | |
2127 | ||
e316aad5 DG |
2128 | DBG3("Consumer delete metadata stream %d", stream->wait_fd); |
2129 | ||
74251bb8 | 2130 | pthread_mutex_lock(&consumer_data.lock); |
fb549e7a | 2131 | pthread_mutex_lock(&stream->chan->lock); |
3dad2c0f | 2132 | pthread_mutex_lock(&stream->lock); |
081424af JG |
2133 | if (stream->chan->metadata_cache) { |
2134 | /* Only applicable to userspace consumers. */ | |
2135 | pthread_mutex_lock(&stream->chan->metadata_cache->lock); | |
2136 | } | |
8994307f | 2137 | |
6d574024 DG |
2138 | /* Remove any reference to that stream. */ |
2139 | consumer_stream_delete(stream, ht); | |
ca22feea | 2140 | |
6d574024 DG |
2141 | /* Close down everything including the relayd if one. */ |
2142 | consumer_stream_close(stream); | |
2143 | /* Destroy tracer buffers of the stream. */ | |
2144 | consumer_stream_destroy_buffers(stream); | |
fb3a43a9 DG |
2145 | |
2146 | /* Atomically decrement channel refcount since other threads can use it. */ | |
f2ad556d | 2147 | if (!uatomic_sub_return(&stream->chan->refcount, 1) |
ffe60014 | 2148 | && !uatomic_read(&stream->chan->nb_init_stream_left)) { |
c30aaa51 | 2149 | /* Go for channel deletion! */ |
e316aad5 | 2150 | free_chan = stream->chan; |
fb3a43a9 DG |
2151 | } |
2152 | ||
73811ecc DG |
2153 | /* |
2154 | * Nullify the stream reference so it is not used after deletion. The | |
6d574024 DG |
2155 | * channel lock MUST be acquired before being able to check for a NULL |
2156 | * pointer value. | |
73811ecc DG |
2157 | */ |
2158 | stream->chan->metadata_stream = NULL; | |
2159 | ||
081424af JG |
2160 | if (stream->chan->metadata_cache) { |
2161 | pthread_mutex_unlock(&stream->chan->metadata_cache->lock); | |
2162 | } | |
3dad2c0f | 2163 | pthread_mutex_unlock(&stream->lock); |
fb549e7a | 2164 | pthread_mutex_unlock(&stream->chan->lock); |
74251bb8 | 2165 | pthread_mutex_unlock(&consumer_data.lock); |
e316aad5 DG |
2166 | |
2167 | if (free_chan) { | |
2168 | consumer_del_channel(free_chan); | |
2169 | } | |
2170 | ||
6d574024 | 2171 | consumer_stream_free(stream); |
fb3a43a9 DG |
2172 | } |
2173 | ||
2174 | /* | |
2175 | * Action done with the metadata stream when adding it to the consumer internal | |
2176 | * data structures to handle it. | |
2177 | */ | |
66d583dc | 2178 | void consumer_add_metadata_stream(struct lttng_consumer_stream *stream) |
fb3a43a9 | 2179 | { |
5ab66908 | 2180 | struct lttng_ht *ht = metadata_ht; |
76082088 | 2181 | struct lttng_ht_iter iter; |
d88aee68 | 2182 | struct lttng_ht_node_u64 *node; |
fb3a43a9 | 2183 | |
e316aad5 DG |
2184 | assert(stream); |
2185 | assert(ht); | |
2186 | ||
d88aee68 | 2187 | DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key); |
e316aad5 DG |
2188 | |
2189 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 2190 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 2191 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 2192 | pthread_mutex_lock(&stream->lock); |
e316aad5 | 2193 | |
e316aad5 DG |
2194 | /* |
2195 | * From here, refcounts are updated so be _careful_ when returning an error | |
2196 | * after this point. | |
2197 | */ | |
2198 | ||
fb3a43a9 | 2199 | rcu_read_lock(); |
76082088 DG |
2200 | |
2201 | /* | |
2202 | * Lookup the stream just to make sure it does not exist in our internal | |
2203 | * state. This should NEVER happen. | |
2204 | */ | |
d88aee68 DG |
2205 | lttng_ht_lookup(ht, &stream->key, &iter); |
2206 | node = lttng_ht_iter_get_node_u64(&iter); | |
76082088 DG |
2207 | assert(!node); |
2208 | ||
e316aad5 | 2209 | /* |
ffe60014 DG |
2210 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
2211 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
2212 | * causes the count to become 0 also causes a stream to be added. The |
2213 | * channel deletion will thus be triggered by the following removal of this | |
2214 | * stream. | |
2215 | */ | |
ffe60014 | 2216 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
2217 | /* Increment refcount before decrementing nb_init_stream_left */ |
2218 | cmm_smp_wmb(); | |
ffe60014 | 2219 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
2220 | } |
2221 | ||
d88aee68 | 2222 | lttng_ht_add_unique_u64(ht, &stream->node); |
ca22feea | 2223 | |
446156b4 | 2224 | lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, |
d8ef542d MD |
2225 | &stream->node_channel_id); |
2226 | ||
ca22feea DG |
2227 | /* |
2228 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
2229 | * the key since the HT does not use it and we allow to add redundant keys | |
2230 | * into this table. | |
2231 | */ | |
d88aee68 | 2232 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 2233 | |
fb3a43a9 | 2234 | rcu_read_unlock(); |
e316aad5 | 2235 | |
2e818a6a | 2236 | pthread_mutex_unlock(&stream->lock); |
a9838785 | 2237 | pthread_mutex_unlock(&stream->chan->lock); |
ec6ea7d0 | 2238 | pthread_mutex_unlock(&stream->chan->timer_lock); |
e316aad5 | 2239 | pthread_mutex_unlock(&consumer_data.lock); |
fb3a43a9 DG |
2240 | } |
2241 | ||
8994307f DG |
2242 | /* |
2243 | * Delete data stream that are flagged for deletion (endpoint_status). | |
2244 | */ | |
2245 | static void validate_endpoint_status_data_stream(void) | |
2246 | { | |
2247 | struct lttng_ht_iter iter; | |
2248 | struct lttng_consumer_stream *stream; | |
2249 | ||
2250 | DBG("Consumer delete flagged data stream"); | |
2251 | ||
2252 | rcu_read_lock(); | |
2253 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
2254 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2255 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2256 | continue; |
2257 | } | |
2258 | /* Delete it right now */ | |
2259 | consumer_del_stream(stream, data_ht); | |
2260 | } | |
2261 | rcu_read_unlock(); | |
2262 | } | |
2263 | ||
2264 | /* | |
2265 | * Delete metadata stream that are flagged for deletion (endpoint_status). | |
2266 | */ | |
2267 | static void validate_endpoint_status_metadata_stream( | |
2268 | struct lttng_poll_event *pollset) | |
2269 | { | |
2270 | struct lttng_ht_iter iter; | |
2271 | struct lttng_consumer_stream *stream; | |
2272 | ||
2273 | DBG("Consumer delete flagged metadata stream"); | |
2274 | ||
2275 | assert(pollset); | |
2276 | ||
2277 | rcu_read_lock(); | |
2278 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
2279 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2280 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2281 | continue; |
2282 | } | |
2283 | /* | |
2284 | * Remove from pollset so the metadata thread can continue without | |
2285 | * blocking on a deleted stream. | |
2286 | */ | |
2287 | lttng_poll_del(pollset, stream->wait_fd); | |
2288 | ||
2289 | /* Delete it right now */ | |
2290 | consumer_del_metadata_stream(stream, metadata_ht); | |
2291 | } | |
2292 | rcu_read_unlock(); | |
2293 | } | |
2294 | ||
d73bf3d7 JD |
2295 | /* |
2296 | * Perform operations that need to be done after a stream has | |
2297 | * rotated and released the stream lock. | |
2298 | * | |
2299 | * Multiple rotations cannot occur simultaneously, so we know the state of the | |
2300 | * "rotated" stream flag cannot change. | |
2301 | * | |
2302 | * This MUST be called WITHOUT the stream lock held. | |
2303 | */ | |
2304 | static | |
2305 | int consumer_post_rotation(struct lttng_consumer_stream *stream, | |
2306 | struct lttng_consumer_local_data *ctx) | |
2307 | { | |
2308 | int ret = 0; | |
2309 | ||
2310 | pthread_mutex_lock(&stream->chan->lock); | |
2311 | ||
2312 | switch (consumer_data.type) { | |
2313 | case LTTNG_CONSUMER_KERNEL: | |
2314 | break; | |
2315 | case LTTNG_CONSUMER32_UST: | |
2316 | case LTTNG_CONSUMER64_UST: | |
2317 | /* | |
2318 | * The ust_metadata_pushed counter has been reset to 0, so now | |
2319 | * we can wakeup the metadata thread so it dumps the metadata | |
2320 | * cache to the new file. | |
2321 | */ | |
2322 | if (stream->metadata_flag) { | |
2323 | consumer_metadata_wakeup_pipe(stream->chan); | |
2324 | } | |
2325 | break; | |
2326 | default: | |
2327 | ERR("Unknown consumer_data type"); | |
2328 | abort(); | |
2329 | } | |
2330 | ||
d73bf3d7 | 2331 | pthread_mutex_unlock(&stream->chan->lock); |
d73bf3d7 JD |
2332 | return ret; |
2333 | } | |
2334 | ||
fb3a43a9 DG |
2335 | /* |
2336 | * Thread polls on metadata file descriptor and write them on disk or on the | |
2337 | * network. | |
2338 | */ | |
7d980def | 2339 | void *consumer_thread_metadata_poll(void *data) |
fb3a43a9 | 2340 | { |
1fc79fb4 | 2341 | int ret, i, pollfd, err = -1; |
fb3a43a9 | 2342 | uint32_t revents, nb_fd; |
e316aad5 | 2343 | struct lttng_consumer_stream *stream = NULL; |
fb3a43a9 | 2344 | struct lttng_ht_iter iter; |
d88aee68 | 2345 | struct lttng_ht_node_u64 *node; |
fb3a43a9 DG |
2346 | struct lttng_poll_event events; |
2347 | struct lttng_consumer_local_data *ctx = data; | |
2348 | ssize_t len; | |
2349 | ||
2350 | rcu_register_thread(); | |
2351 | ||
1fc79fb4 MD |
2352 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA); |
2353 | ||
2d57de81 MD |
2354 | if (testpoint(consumerd_thread_metadata)) { |
2355 | goto error_testpoint; | |
2356 | } | |
2357 | ||
9ce5646a MD |
2358 | health_code_update(); |
2359 | ||
fb3a43a9 DG |
2360 | DBG("Thread metadata poll started"); |
2361 | ||
fb3a43a9 DG |
2362 | /* Size is set to 1 for the consumer_metadata pipe */ |
2363 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2364 | if (ret < 0) { | |
2365 | ERR("Poll set creation failed"); | |
d8ef542d | 2366 | goto end_poll; |
fb3a43a9 DG |
2367 | } |
2368 | ||
13886d2d DG |
2369 | ret = lttng_poll_add(&events, |
2370 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN); | |
fb3a43a9 DG |
2371 | if (ret < 0) { |
2372 | goto end; | |
2373 | } | |
2374 | ||
2375 | /* Main loop */ | |
2376 | DBG("Metadata main loop started"); | |
2377 | ||
2378 | while (1) { | |
fb3a43a9 | 2379 | restart: |
7fa2082e | 2380 | health_code_update(); |
9ce5646a | 2381 | health_poll_entry(); |
7fa2082e | 2382 | DBG("Metadata poll wait"); |
fb3a43a9 | 2383 | ret = lttng_poll_wait(&events, -1); |
7fa2082e MD |
2384 | DBG("Metadata poll return from wait with %d fd(s)", |
2385 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2386 | health_poll_exit(); |
40063ead | 2387 | DBG("Metadata event caught in thread"); |
fb3a43a9 DG |
2388 | if (ret < 0) { |
2389 | if (errno == EINTR) { | |
40063ead | 2390 | ERR("Poll EINTR caught"); |
fb3a43a9 DG |
2391 | goto restart; |
2392 | } | |
d9607cd7 MD |
2393 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2394 | err = 0; /* All is OK */ | |
2395 | } | |
2396 | goto end; | |
fb3a43a9 DG |
2397 | } |
2398 | ||
0d9c5d77 DG |
2399 | nb_fd = ret; |
2400 | ||
e316aad5 | 2401 | /* From here, the event is a metadata wait fd */ |
fb3a43a9 | 2402 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2403 | health_code_update(); |
2404 | ||
fb3a43a9 DG |
2405 | revents = LTTNG_POLL_GETEV(&events, i); |
2406 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2407 | ||
13886d2d | 2408 | if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) { |
03e43155 | 2409 | if (revents & LPOLLIN) { |
13886d2d DG |
2410 | ssize_t pipe_len; |
2411 | ||
2412 | pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe, | |
2413 | &stream, sizeof(stream)); | |
6cd525e8 | 2414 | if (pipe_len < sizeof(stream)) { |
03e43155 MD |
2415 | if (pipe_len < 0) { |
2416 | PERROR("read metadata stream"); | |
2417 | } | |
fb3a43a9 | 2418 | /* |
03e43155 MD |
2419 | * Remove the pipe from the poll set and continue the loop |
2420 | * since their might be data to consume. | |
fb3a43a9 | 2421 | */ |
03e43155 MD |
2422 | lttng_poll_del(&events, |
2423 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2424 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
fb3a43a9 DG |
2425 | continue; |
2426 | } | |
2427 | ||
8994307f DG |
2428 | /* A NULL stream means that the state has changed. */ |
2429 | if (stream == NULL) { | |
2430 | /* Check for deleted streams. */ | |
2431 | validate_endpoint_status_metadata_stream(&events); | |
3714380f | 2432 | goto restart; |
8994307f DG |
2433 | } |
2434 | ||
fb3a43a9 DG |
2435 | DBG("Adding metadata stream %d to poll set", |
2436 | stream->wait_fd); | |
2437 | ||
fb3a43a9 DG |
2438 | /* Add metadata stream to the global poll events list */ |
2439 | lttng_poll_add(&events, stream->wait_fd, | |
6d574024 | 2440 | LPOLLIN | LPOLLPRI | LPOLLHUP); |
03e43155 MD |
2441 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
2442 | DBG("Metadata thread pipe hung up"); | |
2443 | /* | |
2444 | * Remove the pipe from the poll set and continue the loop | |
2445 | * since their might be data to consume. | |
2446 | */ | |
2447 | lttng_poll_del(&events, | |
2448 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2449 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
2450 | continue; | |
2451 | } else { | |
2452 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
2453 | goto end; | |
fb3a43a9 DG |
2454 | } |
2455 | ||
e316aad5 | 2456 | /* Handle other stream */ |
fb3a43a9 DG |
2457 | continue; |
2458 | } | |
2459 | ||
d09e1200 | 2460 | rcu_read_lock(); |
d88aee68 DG |
2461 | { |
2462 | uint64_t tmp_id = (uint64_t) pollfd; | |
2463 | ||
2464 | lttng_ht_lookup(metadata_ht, &tmp_id, &iter); | |
2465 | } | |
2466 | node = lttng_ht_iter_get_node_u64(&iter); | |
e316aad5 | 2467 | assert(node); |
fb3a43a9 DG |
2468 | |
2469 | stream = caa_container_of(node, struct lttng_consumer_stream, | |
58b1f425 | 2470 | node); |
fb3a43a9 | 2471 | |
03e43155 MD |
2472 | if (revents & (LPOLLIN | LPOLLPRI)) { |
2473 | /* Get the data out of the metadata file descriptor */ | |
2474 | DBG("Metadata available on fd %d", pollfd); | |
2475 | assert(stream->wait_fd == pollfd); | |
2476 | ||
2477 | do { | |
2478 | health_code_update(); | |
2479 | ||
2480 | len = ctx->on_buffer_ready(stream, ctx); | |
2481 | /* | |
2482 | * We don't check the return value here since if we get | |
83f4233d | 2483 | * a negative len, it means an error occurred thus we |
03e43155 MD |
2484 | * simply remove it from the poll set and free the |
2485 | * stream. | |
2486 | */ | |
2487 | } while (len > 0); | |
2488 | ||
2489 | /* It's ok to have an unavailable sub-buffer */ | |
2490 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { | |
2491 | /* Clean up stream from consumer and free it. */ | |
2492 | lttng_poll_del(&events, stream->wait_fd); | |
2493 | consumer_del_metadata_stream(stream, metadata_ht); | |
2494 | } | |
2495 | } else if (revents & (LPOLLERR | LPOLLHUP)) { | |
e316aad5 | 2496 | DBG("Metadata fd %d is hup|err.", pollfd); |
fb3a43a9 DG |
2497 | if (!stream->hangup_flush_done |
2498 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2499 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2500 | DBG("Attempting to flush and consume the UST buffers"); | |
2501 | lttng_ustconsumer_on_stream_hangup(stream); | |
2502 | ||
2503 | /* We just flushed the stream now read it. */ | |
4bb94b75 | 2504 | do { |
9ce5646a MD |
2505 | health_code_update(); |
2506 | ||
4bb94b75 DG |
2507 | len = ctx->on_buffer_ready(stream, ctx); |
2508 | /* | |
2509 | * We don't check the return value here since if we get | |
83f4233d | 2510 | * a negative len, it means an error occurred thus we |
4bb94b75 DG |
2511 | * simply remove it from the poll set and free the |
2512 | * stream. | |
2513 | */ | |
2514 | } while (len > 0); | |
fb3a43a9 DG |
2515 | } |
2516 | ||
fb3a43a9 | 2517 | lttng_poll_del(&events, stream->wait_fd); |
e316aad5 DG |
2518 | /* |
2519 | * This call update the channel states, closes file descriptors | |
2520 | * and securely free the stream. | |
2521 | */ | |
2522 | consumer_del_metadata_stream(stream, metadata_ht); | |
03e43155 MD |
2523 | } else { |
2524 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
6f2f1a70 | 2525 | rcu_read_unlock(); |
03e43155 | 2526 | goto end; |
fb3a43a9 | 2527 | } |
e316aad5 | 2528 | /* Release RCU lock for the stream looked up */ |
d09e1200 | 2529 | rcu_read_unlock(); |
fb3a43a9 DG |
2530 | } |
2531 | } | |
2532 | ||
1fc79fb4 MD |
2533 | /* All is OK */ |
2534 | err = 0; | |
fb3a43a9 DG |
2535 | end: |
2536 | DBG("Metadata poll thread exiting"); | |
fb3a43a9 | 2537 | |
d8ef542d MD |
2538 | lttng_poll_clean(&events); |
2539 | end_poll: | |
2d57de81 | 2540 | error_testpoint: |
1fc79fb4 MD |
2541 | if (err) { |
2542 | health_error(); | |
2543 | ERR("Health error occurred in %s", __func__); | |
2544 | } | |
2545 | health_unregister(health_consumerd); | |
fb3a43a9 DG |
2546 | rcu_unregister_thread(); |
2547 | return NULL; | |
2548 | } | |
2549 | ||
3bd1e081 | 2550 | /* |
e4421fec | 2551 | * This thread polls the fds in the set to consume the data and write |
3bd1e081 MD |
2552 | * it to tracefile if necessary. |
2553 | */ | |
7d980def | 2554 | void *consumer_thread_data_poll(void *data) |
3bd1e081 | 2555 | { |
1fc79fb4 | 2556 | int num_rdy, num_hup, high_prio, ret, i, err = -1; |
3bd1e081 MD |
2557 | struct pollfd *pollfd = NULL; |
2558 | /* local view of the streams */ | |
c869f647 | 2559 | struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; |
3bd1e081 | 2560 | /* local view of consumer_data.fds_count */ |
8bdcc002 JG |
2561 | int nb_fd = 0; |
2562 | /* 2 for the consumer_data_pipe and wake up pipe */ | |
2563 | const int nb_pipes_fd = 2; | |
9a2fcf78 JD |
2564 | /* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */ |
2565 | int nb_inactive_fd = 0; | |
3bd1e081 | 2566 | struct lttng_consumer_local_data *ctx = data; |
00e2e675 | 2567 | ssize_t len; |
3bd1e081 | 2568 | |
e7b994a3 DG |
2569 | rcu_register_thread(); |
2570 | ||
1fc79fb4 MD |
2571 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA); |
2572 | ||
2d57de81 MD |
2573 | if (testpoint(consumerd_thread_data)) { |
2574 | goto error_testpoint; | |
2575 | } | |
2576 | ||
9ce5646a MD |
2577 | health_code_update(); |
2578 | ||
4df6c8cb MD |
2579 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream *)); |
2580 | if (local_stream == NULL) { | |
2581 | PERROR("local_stream malloc"); | |
2582 | goto end; | |
2583 | } | |
3bd1e081 MD |
2584 | |
2585 | while (1) { | |
9ce5646a MD |
2586 | health_code_update(); |
2587 | ||
3bd1e081 MD |
2588 | high_prio = 0; |
2589 | num_hup = 0; | |
2590 | ||
2591 | /* | |
e4421fec | 2592 | * the fds set has been updated, we need to update our |
3bd1e081 MD |
2593 | * local array as well |
2594 | */ | |
2595 | pthread_mutex_lock(&consumer_data.lock); | |
2596 | if (consumer_data.need_update) { | |
0e428499 DG |
2597 | free(pollfd); |
2598 | pollfd = NULL; | |
2599 | ||
2600 | free(local_stream); | |
2601 | local_stream = NULL; | |
3bd1e081 | 2602 | |
8bdcc002 | 2603 | /* Allocate for all fds */ |
261de637 | 2604 | pollfd = zmalloc((consumer_data.stream_count + nb_pipes_fd) * sizeof(struct pollfd)); |
3bd1e081 | 2605 | if (pollfd == NULL) { |
7a57cf92 | 2606 | PERROR("pollfd malloc"); |
3bd1e081 MD |
2607 | pthread_mutex_unlock(&consumer_data.lock); |
2608 | goto end; | |
2609 | } | |
2610 | ||
261de637 | 2611 | local_stream = zmalloc((consumer_data.stream_count + nb_pipes_fd) * |
747f8642 | 2612 | sizeof(struct lttng_consumer_stream *)); |
3bd1e081 | 2613 | if (local_stream == NULL) { |
7a57cf92 | 2614 | PERROR("local_stream malloc"); |
3bd1e081 MD |
2615 | pthread_mutex_unlock(&consumer_data.lock); |
2616 | goto end; | |
2617 | } | |
ffe60014 | 2618 | ret = update_poll_array(ctx, &pollfd, local_stream, |
9a2fcf78 | 2619 | data_ht, &nb_inactive_fd); |
3bd1e081 MD |
2620 | if (ret < 0) { |
2621 | ERR("Error in allocating pollfd or local_outfds"); | |
f73fabfd | 2622 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2623 | pthread_mutex_unlock(&consumer_data.lock); |
2624 | goto end; | |
2625 | } | |
2626 | nb_fd = ret; | |
2627 | consumer_data.need_update = 0; | |
2628 | } | |
2629 | pthread_mutex_unlock(&consumer_data.lock); | |
2630 | ||
4078b776 | 2631 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
9a2fcf78 JD |
2632 | if (nb_fd == 0 && nb_inactive_fd == 0 && |
2633 | CMM_LOAD_SHARED(consumer_quit) == 1) { | |
1fc79fb4 | 2634 | err = 0; /* All is OK */ |
4078b776 MD |
2635 | goto end; |
2636 | } | |
3bd1e081 | 2637 | /* poll on the array of fds */ |
88f2b785 | 2638 | restart: |
261de637 | 2639 | DBG("polling on %d fd", nb_fd + nb_pipes_fd); |
cf0bcb51 JG |
2640 | if (testpoint(consumerd_thread_data_poll)) { |
2641 | goto end; | |
2642 | } | |
9ce5646a | 2643 | health_poll_entry(); |
261de637 | 2644 | num_rdy = poll(pollfd, nb_fd + nb_pipes_fd, -1); |
9ce5646a | 2645 | health_poll_exit(); |
3bd1e081 MD |
2646 | DBG("poll num_rdy : %d", num_rdy); |
2647 | if (num_rdy == -1) { | |
88f2b785 MD |
2648 | /* |
2649 | * Restart interrupted system call. | |
2650 | */ | |
2651 | if (errno == EINTR) { | |
2652 | goto restart; | |
2653 | } | |
7a57cf92 | 2654 | PERROR("Poll error"); |
f73fabfd | 2655 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2656 | goto end; |
2657 | } else if (num_rdy == 0) { | |
2658 | DBG("Polling thread timed out"); | |
2659 | goto end; | |
2660 | } | |
2661 | ||
80957876 JG |
2662 | if (caa_unlikely(data_consumption_paused)) { |
2663 | DBG("Data consumption paused, sleeping..."); | |
2664 | sleep(1); | |
2665 | goto restart; | |
2666 | } | |
2667 | ||
3bd1e081 | 2668 | /* |
50f8ae69 | 2669 | * If the consumer_data_pipe triggered poll go directly to the |
00e2e675 DG |
2670 | * beginning of the loop to update the array. We want to prioritize |
2671 | * array update over low-priority reads. | |
3bd1e081 | 2672 | */ |
509bb1cf | 2673 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
ab30f567 | 2674 | ssize_t pipe_readlen; |
04fdd819 | 2675 | |
50f8ae69 | 2676 | DBG("consumer_data_pipe wake up"); |
acdb9057 DG |
2677 | pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe, |
2678 | &new_stream, sizeof(new_stream)); | |
6cd525e8 MD |
2679 | if (pipe_readlen < sizeof(new_stream)) { |
2680 | PERROR("Consumer data pipe"); | |
23f5f35d DG |
2681 | /* Continue so we can at least handle the current stream(s). */ |
2682 | continue; | |
2683 | } | |
c869f647 DG |
2684 | |
2685 | /* | |
2686 | * If the stream is NULL, just ignore it. It's also possible that | |
2687 | * the sessiond poll thread changed the consumer_quit state and is | |
2688 | * waking us up to test it. | |
2689 | */ | |
2690 | if (new_stream == NULL) { | |
8994307f | 2691 | validate_endpoint_status_data_stream(); |
c869f647 DG |
2692 | continue; |
2693 | } | |
2694 | ||
c869f647 | 2695 | /* Continue to update the local streams and handle prio ones */ |
3bd1e081 MD |
2696 | continue; |
2697 | } | |
2698 | ||
02b3d176 DG |
2699 | /* Handle wakeup pipe. */ |
2700 | if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) { | |
2701 | char dummy; | |
2702 | ssize_t pipe_readlen; | |
2703 | ||
2704 | pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy, | |
2705 | sizeof(dummy)); | |
2706 | if (pipe_readlen < 0) { | |
2707 | PERROR("Consumer data wakeup pipe"); | |
2708 | } | |
2709 | /* We've been awakened to handle stream(s). */ | |
2710 | ctx->has_wakeup = 0; | |
2711 | } | |
2712 | ||
3bd1e081 MD |
2713 | /* Take care of high priority channels first. */ |
2714 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2715 | health_code_update(); |
2716 | ||
9617607b DG |
2717 | if (local_stream[i] == NULL) { |
2718 | continue; | |
2719 | } | |
fb3a43a9 | 2720 | if (pollfd[i].revents & POLLPRI) { |
d41f73b7 MD |
2721 | DBG("Urgent read on fd %d", pollfd[i].fd); |
2722 | high_prio = 1; | |
4078b776 | 2723 | len = ctx->on_buffer_ready(local_stream[i], ctx); |
d41f73b7 | 2724 | /* it's ok to have an unavailable sub-buffer */ |
b64403e3 | 2725 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2726 | /* Clean the stream and free it. */ |
2727 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2728 | local_stream[i] = NULL; |
4078b776 MD |
2729 | } else if (len > 0) { |
2730 | local_stream[i]->data_read = 1; | |
d41f73b7 | 2731 | } |
3bd1e081 MD |
2732 | } |
2733 | } | |
2734 | ||
4078b776 MD |
2735 | /* |
2736 | * If we read high prio channel in this loop, try again | |
2737 | * for more high prio data. | |
2738 | */ | |
2739 | if (high_prio) { | |
3bd1e081 MD |
2740 | continue; |
2741 | } | |
2742 | ||
2743 | /* Take care of low priority channels. */ | |
4078b776 | 2744 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2745 | health_code_update(); |
2746 | ||
9617607b DG |
2747 | if (local_stream[i] == NULL) { |
2748 | continue; | |
2749 | } | |
4078b776 | 2750 | if ((pollfd[i].revents & POLLIN) || |
02b3d176 DG |
2751 | local_stream[i]->hangup_flush_done || |
2752 | local_stream[i]->has_data) { | |
4078b776 MD |
2753 | DBG("Normal read on fd %d", pollfd[i].fd); |
2754 | len = ctx->on_buffer_ready(local_stream[i], ctx); | |
2755 | /* it's ok to have an unavailable sub-buffer */ | |
b64403e3 | 2756 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2757 | /* Clean the stream and free it. */ |
2758 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2759 | local_stream[i] = NULL; |
4078b776 MD |
2760 | } else if (len > 0) { |
2761 | local_stream[i]->data_read = 1; | |
2762 | } | |
2763 | } | |
2764 | } | |
2765 | ||
2766 | /* Handle hangup and errors */ | |
2767 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2768 | health_code_update(); |
2769 | ||
9617607b DG |
2770 | if (local_stream[i] == NULL) { |
2771 | continue; | |
2772 | } | |
4078b776 MD |
2773 | if (!local_stream[i]->hangup_flush_done |
2774 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) | |
2775 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2776 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2777 | DBG("fd %d is hup|err|nval. Attempting flush and read.", | |
9617607b | 2778 | pollfd[i].fd); |
4078b776 MD |
2779 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); |
2780 | /* Attempt read again, for the data we just flushed. */ | |
2781 | local_stream[i]->data_read = 1; | |
2782 | } | |
2783 | /* | |
2784 | * If the poll flag is HUP/ERR/NVAL and we have | |
2785 | * read no data in this pass, we can remove the | |
2786 | * stream from its hash table. | |
2787 | */ | |
2788 | if ((pollfd[i].revents & POLLHUP)) { | |
2789 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); | |
2790 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2791 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2792 | local_stream[i] = NULL; |
4078b776 MD |
2793 | num_hup++; |
2794 | } | |
2795 | } else if (pollfd[i].revents & POLLERR) { | |
2796 | ERR("Error returned in polling fd %d.", pollfd[i].fd); | |
2797 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2798 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2799 | local_stream[i] = NULL; |
4078b776 MD |
2800 | num_hup++; |
2801 | } | |
2802 | } else if (pollfd[i].revents & POLLNVAL) { | |
2803 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); | |
2804 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2805 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2806 | local_stream[i] = NULL; |
4078b776 | 2807 | num_hup++; |
3bd1e081 MD |
2808 | } |
2809 | } | |
9617607b DG |
2810 | if (local_stream[i] != NULL) { |
2811 | local_stream[i]->data_read = 0; | |
2812 | } | |
3bd1e081 MD |
2813 | } |
2814 | } | |
1fc79fb4 MD |
2815 | /* All is OK */ |
2816 | err = 0; | |
3bd1e081 MD |
2817 | end: |
2818 | DBG("polling thread exiting"); | |
0e428499 DG |
2819 | free(pollfd); |
2820 | free(local_stream); | |
fb3a43a9 DG |
2821 | |
2822 | /* | |
2823 | * Close the write side of the pipe so epoll_wait() in | |
7d980def DG |
2824 | * consumer_thread_metadata_poll can catch it. The thread is monitoring the |
2825 | * read side of the pipe. If we close them both, epoll_wait strangely does | |
2826 | * not return and could create a endless wait period if the pipe is the | |
2827 | * only tracked fd in the poll set. The thread will take care of closing | |
2828 | * the read side. | |
fb3a43a9 | 2829 | */ |
13886d2d | 2830 | (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe); |
fb3a43a9 | 2831 | |
2d57de81 | 2832 | error_testpoint: |
1fc79fb4 MD |
2833 | if (err) { |
2834 | health_error(); | |
2835 | ERR("Health error occurred in %s", __func__); | |
2836 | } | |
2837 | health_unregister(health_consumerd); | |
2838 | ||
e7b994a3 | 2839 | rcu_unregister_thread(); |
3bd1e081 MD |
2840 | return NULL; |
2841 | } | |
2842 | ||
d8ef542d MD |
2843 | /* |
2844 | * Close wake-up end of each stream belonging to the channel. This will | |
2845 | * allow the poll() on the stream read-side to detect when the | |
2846 | * write-side (application) finally closes them. | |
2847 | */ | |
2848 | static | |
2849 | void consumer_close_channel_streams(struct lttng_consumer_channel *channel) | |
2850 | { | |
2851 | struct lttng_ht *ht; | |
2852 | struct lttng_consumer_stream *stream; | |
2853 | struct lttng_ht_iter iter; | |
2854 | ||
2855 | ht = consumer_data.stream_per_chan_id_ht; | |
2856 | ||
2857 | rcu_read_lock(); | |
2858 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
2859 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
2860 | ht->match_fct, &channel->key, | |
2861 | &iter.iter, stream, node_channel_id.node) { | |
f2ad556d MD |
2862 | /* |
2863 | * Protect against teardown with mutex. | |
2864 | */ | |
2865 | pthread_mutex_lock(&stream->lock); | |
2866 | if (cds_lfht_is_node_deleted(&stream->node.node)) { | |
2867 | goto next; | |
2868 | } | |
d8ef542d MD |
2869 | switch (consumer_data.type) { |
2870 | case LTTNG_CONSUMER_KERNEL: | |
2871 | break; | |
2872 | case LTTNG_CONSUMER32_UST: | |
2873 | case LTTNG_CONSUMER64_UST: | |
b4a650f3 DG |
2874 | if (stream->metadata_flag) { |
2875 | /* Safe and protected by the stream lock. */ | |
2876 | lttng_ustconsumer_close_metadata(stream->chan); | |
2877 | } else { | |
2878 | /* | |
2879 | * Note: a mutex is taken internally within | |
2880 | * liblttng-ust-ctl to protect timer wakeup_fd | |
2881 | * use from concurrent close. | |
2882 | */ | |
2883 | lttng_ustconsumer_close_stream_wakeup(stream); | |
2884 | } | |
d8ef542d MD |
2885 | break; |
2886 | default: | |
2887 | ERR("Unknown consumer_data type"); | |
2888 | assert(0); | |
2889 | } | |
f2ad556d MD |
2890 | next: |
2891 | pthread_mutex_unlock(&stream->lock); | |
d8ef542d MD |
2892 | } |
2893 | rcu_read_unlock(); | |
2894 | } | |
2895 | ||
2896 | static void destroy_channel_ht(struct lttng_ht *ht) | |
2897 | { | |
2898 | struct lttng_ht_iter iter; | |
2899 | struct lttng_consumer_channel *channel; | |
2900 | int ret; | |
2901 | ||
2902 | if (ht == NULL) { | |
2903 | return; | |
2904 | } | |
2905 | ||
2906 | rcu_read_lock(); | |
2907 | cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) { | |
2908 | ret = lttng_ht_del(ht, &iter); | |
2909 | assert(ret != 0); | |
2910 | } | |
2911 | rcu_read_unlock(); | |
2912 | ||
2913 | lttng_ht_destroy(ht); | |
2914 | } | |
2915 | ||
2916 | /* | |
2917 | * This thread polls the channel fds to detect when they are being | |
2918 | * closed. It closes all related streams if the channel is detected as | |
2919 | * closed. It is currently only used as a shim layer for UST because the | |
2920 | * consumerd needs to keep the per-stream wakeup end of pipes open for | |
2921 | * periodical flush. | |
2922 | */ | |
2923 | void *consumer_thread_channel_poll(void *data) | |
2924 | { | |
1fc79fb4 | 2925 | int ret, i, pollfd, err = -1; |
d8ef542d MD |
2926 | uint32_t revents, nb_fd; |
2927 | struct lttng_consumer_channel *chan = NULL; | |
2928 | struct lttng_ht_iter iter; | |
2929 | struct lttng_ht_node_u64 *node; | |
2930 | struct lttng_poll_event events; | |
2931 | struct lttng_consumer_local_data *ctx = data; | |
2932 | struct lttng_ht *channel_ht; | |
2933 | ||
2934 | rcu_register_thread(); | |
2935 | ||
1fc79fb4 MD |
2936 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL); |
2937 | ||
2d57de81 MD |
2938 | if (testpoint(consumerd_thread_channel)) { |
2939 | goto error_testpoint; | |
2940 | } | |
2941 | ||
9ce5646a MD |
2942 | health_code_update(); |
2943 | ||
d8ef542d MD |
2944 | channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
2945 | if (!channel_ht) { | |
2946 | /* ENOMEM at this point. Better to bail out. */ | |
2947 | goto end_ht; | |
2948 | } | |
2949 | ||
2950 | DBG("Thread channel poll started"); | |
2951 | ||
2952 | /* Size is set to 1 for the consumer_channel pipe */ | |
2953 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2954 | if (ret < 0) { | |
2955 | ERR("Poll set creation failed"); | |
2956 | goto end_poll; | |
2957 | } | |
2958 | ||
2959 | ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN); | |
2960 | if (ret < 0) { | |
2961 | goto end; | |
2962 | } | |
2963 | ||
2964 | /* Main loop */ | |
2965 | DBG("Channel main loop started"); | |
2966 | ||
2967 | while (1) { | |
d8ef542d | 2968 | restart: |
7fa2082e MD |
2969 | health_code_update(); |
2970 | DBG("Channel poll wait"); | |
9ce5646a | 2971 | health_poll_entry(); |
d8ef542d | 2972 | ret = lttng_poll_wait(&events, -1); |
7fa2082e MD |
2973 | DBG("Channel poll return from wait with %d fd(s)", |
2974 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2975 | health_poll_exit(); |
40063ead | 2976 | DBG("Channel event caught in thread"); |
d8ef542d MD |
2977 | if (ret < 0) { |
2978 | if (errno == EINTR) { | |
40063ead | 2979 | ERR("Poll EINTR caught"); |
d8ef542d MD |
2980 | goto restart; |
2981 | } | |
d9607cd7 MD |
2982 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2983 | err = 0; /* All is OK */ | |
2984 | } | |
d8ef542d MD |
2985 | goto end; |
2986 | } | |
2987 | ||
2988 | nb_fd = ret; | |
2989 | ||
2990 | /* From here, the event is a channel wait fd */ | |
2991 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2992 | health_code_update(); |
2993 | ||
d8ef542d MD |
2994 | revents = LTTNG_POLL_GETEV(&events, i); |
2995 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2996 | ||
d8ef542d | 2997 | if (pollfd == ctx->consumer_channel_pipe[0]) { |
03e43155 | 2998 | if (revents & LPOLLIN) { |
d8ef542d | 2999 | enum consumer_channel_action action; |
a0cbdd2e | 3000 | uint64_t key; |
d8ef542d | 3001 | |
a0cbdd2e | 3002 | ret = read_channel_pipe(ctx, &chan, &key, &action); |
d8ef542d | 3003 | if (ret <= 0) { |
03e43155 MD |
3004 | if (ret < 0) { |
3005 | ERR("Error reading channel pipe"); | |
3006 | } | |
3007 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
d8ef542d MD |
3008 | continue; |
3009 | } | |
3010 | ||
3011 | switch (action) { | |
3012 | case CONSUMER_CHANNEL_ADD: | |
3013 | DBG("Adding channel %d to poll set", | |
3014 | chan->wait_fd); | |
3015 | ||
3016 | lttng_ht_node_init_u64(&chan->wait_fd_node, | |
3017 | chan->wait_fd); | |
c7260a81 | 3018 | rcu_read_lock(); |
d8ef542d MD |
3019 | lttng_ht_add_unique_u64(channel_ht, |
3020 | &chan->wait_fd_node); | |
c7260a81 | 3021 | rcu_read_unlock(); |
d8ef542d MD |
3022 | /* Add channel to the global poll events list */ |
3023 | lttng_poll_add(&events, chan->wait_fd, | |
03e43155 | 3024 | LPOLLERR | LPOLLHUP); |
d8ef542d | 3025 | break; |
a0cbdd2e MD |
3026 | case CONSUMER_CHANNEL_DEL: |
3027 | { | |
b4a650f3 DG |
3028 | /* |
3029 | * This command should never be called if the channel | |
3030 | * has streams monitored by either the data or metadata | |
3031 | * thread. The consumer only notify this thread with a | |
3032 | * channel del. command if it receives a destroy | |
3033 | * channel command from the session daemon that send it | |
3034 | * if a command prior to the GET_CHANNEL failed. | |
3035 | */ | |
3036 | ||
c7260a81 | 3037 | rcu_read_lock(); |
a0cbdd2e MD |
3038 | chan = consumer_find_channel(key); |
3039 | if (!chan) { | |
c7260a81 | 3040 | rcu_read_unlock(); |
a0cbdd2e MD |
3041 | ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key); |
3042 | break; | |
3043 | } | |
3044 | lttng_poll_del(&events, chan->wait_fd); | |
f623cc0b | 3045 | iter.iter.node = &chan->wait_fd_node.node; |
a0cbdd2e MD |
3046 | ret = lttng_ht_del(channel_ht, &iter); |
3047 | assert(ret == 0); | |
a0cbdd2e | 3048 | |
f2a444f1 DG |
3049 | switch (consumer_data.type) { |
3050 | case LTTNG_CONSUMER_KERNEL: | |
3051 | break; | |
3052 | case LTTNG_CONSUMER32_UST: | |
3053 | case LTTNG_CONSUMER64_UST: | |
212d67a2 DG |
3054 | health_code_update(); |
3055 | /* Destroy streams that might have been left in the stream list. */ | |
3056 | clean_channel_stream_list(chan); | |
f2a444f1 DG |
3057 | break; |
3058 | default: | |
3059 | ERR("Unknown consumer_data type"); | |
3060 | assert(0); | |
3061 | } | |
3062 | ||
a0cbdd2e MD |
3063 | /* |
3064 | * Release our own refcount. Force channel deletion even if | |
3065 | * streams were not initialized. | |
3066 | */ | |
3067 | if (!uatomic_sub_return(&chan->refcount, 1)) { | |
3068 | consumer_del_channel(chan); | |
3069 | } | |
c7260a81 | 3070 | rcu_read_unlock(); |
a0cbdd2e MD |
3071 | goto restart; |
3072 | } | |
d8ef542d MD |
3073 | case CONSUMER_CHANNEL_QUIT: |
3074 | /* | |
3075 | * Remove the pipe from the poll set and continue the loop | |
3076 | * since their might be data to consume. | |
3077 | */ | |
3078 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
3079 | continue; | |
3080 | default: | |
3081 | ERR("Unknown action"); | |
3082 | break; | |
3083 | } | |
03e43155 MD |
3084 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
3085 | DBG("Channel thread pipe hung up"); | |
3086 | /* | |
3087 | * Remove the pipe from the poll set and continue the loop | |
3088 | * since their might be data to consume. | |
3089 | */ | |
3090 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
3091 | continue; | |
3092 | } else { | |
3093 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
3094 | goto end; | |
d8ef542d MD |
3095 | } |
3096 | ||
3097 | /* Handle other stream */ | |
3098 | continue; | |
3099 | } | |
3100 | ||
3101 | rcu_read_lock(); | |
3102 | { | |
3103 | uint64_t tmp_id = (uint64_t) pollfd; | |
3104 | ||
3105 | lttng_ht_lookup(channel_ht, &tmp_id, &iter); | |
3106 | } | |
3107 | node = lttng_ht_iter_get_node_u64(&iter); | |
3108 | assert(node); | |
3109 | ||
3110 | chan = caa_container_of(node, struct lttng_consumer_channel, | |
3111 | wait_fd_node); | |
3112 | ||
3113 | /* Check for error event */ | |
3114 | if (revents & (LPOLLERR | LPOLLHUP)) { | |
3115 | DBG("Channel fd %d is hup|err.", pollfd); | |
3116 | ||
3117 | lttng_poll_del(&events, chan->wait_fd); | |
3118 | ret = lttng_ht_del(channel_ht, &iter); | |
3119 | assert(ret == 0); | |
b4a650f3 DG |
3120 | |
3121 | /* | |
3122 | * This will close the wait fd for each stream associated to | |
3123 | * this channel AND monitored by the data/metadata thread thus | |
3124 | * will be clean by the right thread. | |
3125 | */ | |
d8ef542d | 3126 | consumer_close_channel_streams(chan); |
f2ad556d MD |
3127 | |
3128 | /* Release our own refcount */ | |
3129 | if (!uatomic_sub_return(&chan->refcount, 1) | |
3130 | && !uatomic_read(&chan->nb_init_stream_left)) { | |
3131 | consumer_del_channel(chan); | |
3132 | } | |
03e43155 MD |
3133 | } else { |
3134 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
3135 | rcu_read_unlock(); | |
3136 | goto end; | |
d8ef542d MD |
3137 | } |
3138 | ||
3139 | /* Release RCU lock for the channel looked up */ | |
3140 | rcu_read_unlock(); | |
3141 | } | |
3142 | } | |
3143 | ||
1fc79fb4 MD |
3144 | /* All is OK */ |
3145 | err = 0; | |
d8ef542d MD |
3146 | end: |
3147 | lttng_poll_clean(&events); | |
3148 | end_poll: | |
3149 | destroy_channel_ht(channel_ht); | |
3150 | end_ht: | |
2d57de81 | 3151 | error_testpoint: |
d8ef542d | 3152 | DBG("Channel poll thread exiting"); |
1fc79fb4 MD |
3153 | if (err) { |
3154 | health_error(); | |
3155 | ERR("Health error occurred in %s", __func__); | |
3156 | } | |
3157 | health_unregister(health_consumerd); | |
d8ef542d MD |
3158 | rcu_unregister_thread(); |
3159 | return NULL; | |
3160 | } | |
3161 | ||
331744e3 JD |
3162 | static int set_metadata_socket(struct lttng_consumer_local_data *ctx, |
3163 | struct pollfd *sockpoll, int client_socket) | |
3164 | { | |
3165 | int ret; | |
3166 | ||
3167 | assert(ctx); | |
3168 | assert(sockpoll); | |
3169 | ||
84382d49 MD |
3170 | ret = lttng_consumer_poll_socket(sockpoll); |
3171 | if (ret) { | |
331744e3 JD |
3172 | goto error; |
3173 | } | |
3174 | DBG("Metadata connection on client_socket"); | |
3175 | ||
3176 | /* Blocking call, waiting for transmission */ | |
3177 | ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket); | |
3178 | if (ctx->consumer_metadata_socket < 0) { | |
3179 | WARN("On accept metadata"); | |
3180 | ret = -1; | |
3181 | goto error; | |
3182 | } | |
3183 | ret = 0; | |
3184 | ||
3185 | error: | |
3186 | return ret; | |
3187 | } | |
3188 | ||
3bd1e081 MD |
3189 | /* |
3190 | * This thread listens on the consumerd socket and receives the file | |
3191 | * descriptors from the session daemon. | |
3192 | */ | |
7d980def | 3193 | void *consumer_thread_sessiond_poll(void *data) |
3bd1e081 | 3194 | { |
1fc79fb4 | 3195 | int sock = -1, client_socket, ret, err = -1; |
3bd1e081 MD |
3196 | /* |
3197 | * structure to poll for incoming data on communication socket avoids | |
3198 | * making blocking sockets. | |
3199 | */ | |
3200 | struct pollfd consumer_sockpoll[2]; | |
3201 | struct lttng_consumer_local_data *ctx = data; | |
3202 | ||
e7b994a3 DG |
3203 | rcu_register_thread(); |
3204 | ||
1fc79fb4 MD |
3205 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND); |
3206 | ||
2d57de81 MD |
3207 | if (testpoint(consumerd_thread_sessiond)) { |
3208 | goto error_testpoint; | |
3209 | } | |
3210 | ||
9ce5646a MD |
3211 | health_code_update(); |
3212 | ||
3bd1e081 MD |
3213 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
3214 | unlink(ctx->consumer_command_sock_path); | |
3215 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); | |
3216 | if (client_socket < 0) { | |
3217 | ERR("Cannot create command socket"); | |
3218 | goto end; | |
3219 | } | |
3220 | ||
3221 | ret = lttcomm_listen_unix_sock(client_socket); | |
3222 | if (ret < 0) { | |
3223 | goto end; | |
3224 | } | |
3225 | ||
32258573 | 3226 | DBG("Sending ready command to lttng-sessiond"); |
f73fabfd | 3227 | ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); |
3bd1e081 MD |
3228 | /* return < 0 on error, but == 0 is not fatal */ |
3229 | if (ret < 0) { | |
32258573 | 3230 | ERR("Error sending ready command to lttng-sessiond"); |
3bd1e081 MD |
3231 | goto end; |
3232 | } | |
3233 | ||
3bd1e081 MD |
3234 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ |
3235 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; | |
3236 | consumer_sockpoll[0].events = POLLIN | POLLPRI; | |
3237 | consumer_sockpoll[1].fd = client_socket; | |
3238 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3239 | ||
84382d49 MD |
3240 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3241 | if (ret) { | |
3242 | if (ret > 0) { | |
3243 | /* should exit */ | |
3244 | err = 0; | |
3245 | } | |
3bd1e081 MD |
3246 | goto end; |
3247 | } | |
3248 | DBG("Connection on client_socket"); | |
3249 | ||
3250 | /* Blocking call, waiting for transmission */ | |
3251 | sock = lttcomm_accept_unix_sock(client_socket); | |
534d2592 | 3252 | if (sock < 0) { |
3bd1e081 MD |
3253 | WARN("On accept"); |
3254 | goto end; | |
3255 | } | |
3bd1e081 | 3256 | |
331744e3 JD |
3257 | /* |
3258 | * Setup metadata socket which is the second socket connection on the | |
3259 | * command unix socket. | |
3260 | */ | |
3261 | ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket); | |
84382d49 MD |
3262 | if (ret) { |
3263 | if (ret > 0) { | |
3264 | /* should exit */ | |
3265 | err = 0; | |
3266 | } | |
331744e3 JD |
3267 | goto end; |
3268 | } | |
3269 | ||
d96f09c6 DG |
3270 | /* This socket is not useful anymore. */ |
3271 | ret = close(client_socket); | |
3272 | if (ret < 0) { | |
3273 | PERROR("close client_socket"); | |
3274 | } | |
3275 | client_socket = -1; | |
3276 | ||
3bd1e081 MD |
3277 | /* update the polling structure to poll on the established socket */ |
3278 | consumer_sockpoll[1].fd = sock; | |
3279 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3280 | ||
3281 | while (1) { | |
9ce5646a MD |
3282 | health_code_update(); |
3283 | ||
3284 | health_poll_entry(); | |
3285 | ret = lttng_consumer_poll_socket(consumer_sockpoll); | |
3286 | health_poll_exit(); | |
84382d49 MD |
3287 | if (ret) { |
3288 | if (ret > 0) { | |
3289 | /* should exit */ | |
3290 | err = 0; | |
3291 | } | |
3bd1e081 MD |
3292 | goto end; |
3293 | } | |
3294 | DBG("Incoming command on sock"); | |
3295 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
4cbc1a04 DG |
3296 | if (ret <= 0) { |
3297 | /* | |
3298 | * This could simply be a session daemon quitting. Don't output | |
3299 | * ERR() here. | |
3300 | */ | |
3301 | DBG("Communication interrupted on command socket"); | |
41ba6035 | 3302 | err = 0; |
3bd1e081 MD |
3303 | goto end; |
3304 | } | |
10211f5c | 3305 | if (CMM_LOAD_SHARED(consumer_quit)) { |
3bd1e081 | 3306 | DBG("consumer_thread_receive_fds received quit from signal"); |
1fc79fb4 | 3307 | err = 0; /* All is OK */ |
3bd1e081 MD |
3308 | goto end; |
3309 | } | |
ffe60014 | 3310 | DBG("received command on sock"); |
3bd1e081 | 3311 | } |
1fc79fb4 MD |
3312 | /* All is OK */ |
3313 | err = 0; | |
3314 | ||
3bd1e081 | 3315 | end: |
ffe60014 | 3316 | DBG("Consumer thread sessiond poll exiting"); |
3bd1e081 | 3317 | |
d88aee68 DG |
3318 | /* |
3319 | * Close metadata streams since the producer is the session daemon which | |
3320 | * just died. | |
3321 | * | |
3322 | * NOTE: for now, this only applies to the UST tracer. | |
3323 | */ | |
6d574024 | 3324 | lttng_consumer_close_all_metadata(); |
d88aee68 | 3325 | |
3bd1e081 MD |
3326 | /* |
3327 | * when all fds have hung up, the polling thread | |
3328 | * can exit cleanly | |
3329 | */ | |
10211f5c | 3330 | CMM_STORE_SHARED(consumer_quit, 1); |
3bd1e081 | 3331 | |
04fdd819 | 3332 | /* |
c869f647 | 3333 | * Notify the data poll thread to poll back again and test the |
8994307f | 3334 | * consumer_quit state that we just set so to quit gracefully. |
04fdd819 | 3335 | */ |
acdb9057 | 3336 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
c869f647 | 3337 | |
a0cbdd2e | 3338 | notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT); |
d8ef542d | 3339 | |
5c635c72 MD |
3340 | notify_health_quit_pipe(health_quit_pipe); |
3341 | ||
d96f09c6 DG |
3342 | /* Cleaning up possibly open sockets. */ |
3343 | if (sock >= 0) { | |
3344 | ret = close(sock); | |
3345 | if (ret < 0) { | |
3346 | PERROR("close sock sessiond poll"); | |
3347 | } | |
3348 | } | |
3349 | if (client_socket >= 0) { | |
38476d24 | 3350 | ret = close(client_socket); |
d96f09c6 DG |
3351 | if (ret < 0) { |
3352 | PERROR("close client_socket sessiond poll"); | |
3353 | } | |
3354 | } | |
3355 | ||
2d57de81 | 3356 | error_testpoint: |
1fc79fb4 MD |
3357 | if (err) { |
3358 | health_error(); | |
3359 | ERR("Health error occurred in %s", __func__); | |
3360 | } | |
3361 | health_unregister(health_consumerd); | |
3362 | ||
e7b994a3 | 3363 | rcu_unregister_thread(); |
3bd1e081 MD |
3364 | return NULL; |
3365 | } | |
d41f73b7 | 3366 | |
4078b776 | 3367 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
d41f73b7 MD |
3368 | struct lttng_consumer_local_data *ctx) |
3369 | { | |
74251bb8 | 3370 | ssize_t ret; |
02d02e31 JD |
3371 | int rotate_ret; |
3372 | bool rotated = false; | |
74251bb8 DG |
3373 | |
3374 | pthread_mutex_lock(&stream->lock); | |
94d49140 JD |
3375 | if (stream->metadata_flag) { |
3376 | pthread_mutex_lock(&stream->metadata_rdv_lock); | |
3377 | } | |
74251bb8 | 3378 | |
d41f73b7 MD |
3379 | switch (consumer_data.type) { |
3380 | case LTTNG_CONSUMER_KERNEL: | |
02d02e31 | 3381 | ret = lttng_kconsumer_read_subbuffer(stream, ctx, &rotated); |
74251bb8 | 3382 | break; |
7753dea8 MD |
3383 | case LTTNG_CONSUMER32_UST: |
3384 | case LTTNG_CONSUMER64_UST: | |
02d02e31 | 3385 | ret = lttng_ustconsumer_read_subbuffer(stream, ctx, &rotated); |
74251bb8 | 3386 | break; |
d41f73b7 MD |
3387 | default: |
3388 | ERR("Unknown consumer_data type"); | |
3389 | assert(0); | |
74251bb8 DG |
3390 | ret = -ENOSYS; |
3391 | break; | |
d41f73b7 | 3392 | } |
74251bb8 | 3393 | |
94d49140 JD |
3394 | if (stream->metadata_flag) { |
3395 | pthread_cond_broadcast(&stream->metadata_rdv); | |
3396 | pthread_mutex_unlock(&stream->metadata_rdv_lock); | |
3397 | } | |
74251bb8 | 3398 | pthread_mutex_unlock(&stream->lock); |
02d02e31 JD |
3399 | if (rotated) { |
3400 | rotate_ret = consumer_post_rotation(stream, ctx); | |
3401 | if (rotate_ret < 0) { | |
3402 | ERR("Failed after a rotation"); | |
3403 | ret = -1; | |
3404 | } | |
3405 | } | |
3406 | ||
74251bb8 | 3407 | return ret; |
d41f73b7 MD |
3408 | } |
3409 | ||
3410 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
3411 | { | |
3412 | switch (consumer_data.type) { | |
3413 | case LTTNG_CONSUMER_KERNEL: | |
3414 | return lttng_kconsumer_on_recv_stream(stream); | |
7753dea8 MD |
3415 | case LTTNG_CONSUMER32_UST: |
3416 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
3417 | return lttng_ustconsumer_on_recv_stream(stream); |
3418 | default: | |
3419 | ERR("Unknown consumer_data type"); | |
3420 | assert(0); | |
3421 | return -ENOSYS; | |
3422 | } | |
3423 | } | |
e4421fec DG |
3424 | |
3425 | /* | |
3426 | * Allocate and set consumer data hash tables. | |
3427 | */ | |
282dadbc | 3428 | int lttng_consumer_init(void) |
e4421fec | 3429 | { |
d88aee68 | 3430 | consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3431 | if (!consumer_data.channel_ht) { |
3432 | goto error; | |
3433 | } | |
3434 | ||
0fc67d9b JG |
3435 | consumer_data.channels_by_session_id_ht = |
3436 | lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3437 | if (!consumer_data.channels_by_session_id_ht) { | |
3438 | goto error; | |
3439 | } | |
3440 | ||
d88aee68 | 3441 | consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3442 | if (!consumer_data.relayd_ht) { |
3443 | goto error; | |
3444 | } | |
3445 | ||
d88aee68 | 3446 | consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3447 | if (!consumer_data.stream_list_ht) { |
3448 | goto error; | |
3449 | } | |
3450 | ||
d8ef542d | 3451 | consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3452 | if (!consumer_data.stream_per_chan_id_ht) { |
3453 | goto error; | |
3454 | } | |
3455 | ||
3456 | data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3457 | if (!data_ht) { | |
3458 | goto error; | |
3459 | } | |
3460 | ||
3461 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3462 | if (!metadata_ht) { | |
3463 | goto error; | |
3464 | } | |
3465 | ||
3466 | return 0; | |
3467 | ||
3468 | error: | |
3469 | return -1; | |
e4421fec | 3470 | } |
7735ef9e DG |
3471 | |
3472 | /* | |
3473 | * Process the ADD_RELAYD command receive by a consumer. | |
3474 | * | |
3475 | * This will create a relayd socket pair and add it to the relayd hash table. | |
3476 | * The caller MUST acquire a RCU read side lock before calling it. | |
3477 | */ | |
2527bf85 | 3478 | void consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type, |
7735ef9e | 3479 | struct lttng_consumer_local_data *ctx, int sock, |
6151a90f | 3480 | struct pollfd *consumer_sockpoll, |
d3e2ba59 JD |
3481 | struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id, |
3482 | uint64_t relayd_session_id) | |
7735ef9e | 3483 | { |
cd2b09ed | 3484 | int fd = -1, ret = -1, relayd_created = 0; |
0c759fc9 | 3485 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
d4298c99 | 3486 | struct consumer_relayd_sock_pair *relayd = NULL; |
7735ef9e | 3487 | |
6151a90f JD |
3488 | assert(ctx); |
3489 | assert(relayd_sock); | |
3490 | ||
da009f2c | 3491 | DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx); |
7735ef9e DG |
3492 | |
3493 | /* Get relayd reference if exists. */ | |
3494 | relayd = consumer_find_relayd(net_seq_idx); | |
3495 | if (relayd == NULL) { | |
da009f2c | 3496 | assert(sock_type == LTTNG_STREAM_CONTROL); |
7735ef9e DG |
3497 | /* Not found. Allocate one. */ |
3498 | relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); | |
3499 | if (relayd == NULL) { | |
618a6a28 MD |
3500 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
3501 | goto error; | |
0d08d75e | 3502 | } else { |
30319bcb | 3503 | relayd->sessiond_session_id = sessiond_id; |
0d08d75e | 3504 | relayd_created = 1; |
7735ef9e | 3505 | } |
0d08d75e DG |
3506 | |
3507 | /* | |
3508 | * This code path MUST continue to the consumer send status message to | |
3509 | * we can notify the session daemon and continue our work without | |
3510 | * killing everything. | |
3511 | */ | |
da009f2c MD |
3512 | } else { |
3513 | /* | |
3514 | * relayd key should never be found for control socket. | |
3515 | */ | |
3516 | assert(sock_type != LTTNG_STREAM_CONTROL); | |
0d08d75e DG |
3517 | } |
3518 | ||
3519 | /* First send a status message before receiving the fds. */ | |
0c759fc9 | 3520 | ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS); |
618a6a28 | 3521 | if (ret < 0) { |
0d08d75e | 3522 | /* Somehow, the session daemon is not responding anymore. */ |
618a6a28 MD |
3523 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); |
3524 | goto error_nosignal; | |
7735ef9e DG |
3525 | } |
3526 | ||
3527 | /* Poll on consumer socket. */ | |
84382d49 MD |
3528 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3529 | if (ret) { | |
3530 | /* Needing to exit in the middle of a command: error. */ | |
0d08d75e | 3531 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
618a6a28 | 3532 | goto error_nosignal; |
7735ef9e DG |
3533 | } |
3534 | ||
3535 | /* Get relayd socket from session daemon */ | |
3536 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); | |
3537 | if (ret != sizeof(fd)) { | |
4028eeb9 | 3538 | fd = -1; /* Just in case it gets set with an invalid value. */ |
0d08d75e DG |
3539 | |
3540 | /* | |
3541 | * Failing to receive FDs might indicate a major problem such as | |
3542 | * reaching a fd limit during the receive where the kernel returns a | |
3543 | * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we | |
3544 | * don't take any chances and stop everything. | |
3545 | * | |
3546 | * XXX: Feature request #558 will fix that and avoid this possible | |
3547 | * issue when reaching the fd limit. | |
3548 | */ | |
3549 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); | |
618a6a28 | 3550 | ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD; |
f50f23d9 DG |
3551 | goto error; |
3552 | } | |
3553 | ||
7735ef9e DG |
3554 | /* Copy socket information and received FD */ |
3555 | switch (sock_type) { | |
3556 | case LTTNG_STREAM_CONTROL: | |
3557 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3558 | lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock); |
3559 | ret = lttcomm_create_sock(&relayd->control_sock.sock); | |
4028eeb9 | 3560 | /* Handle create_sock error. */ |
f66c074c | 3561 | if (ret < 0) { |
618a6a28 | 3562 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3563 | goto error; |
f66c074c | 3564 | } |
da009f2c MD |
3565 | /* |
3566 | * Close the socket created internally by | |
3567 | * lttcomm_create_sock, so we can replace it by the one | |
3568 | * received from sessiond. | |
3569 | */ | |
3570 | if (close(relayd->control_sock.sock.fd)) { | |
3571 | PERROR("close"); | |
3572 | } | |
7735ef9e DG |
3573 | |
3574 | /* Assign new file descriptor */ | |
6151a90f JD |
3575 | relayd->control_sock.sock.fd = fd; |
3576 | /* Assign version values. */ | |
3577 | relayd->control_sock.major = relayd_sock->major; | |
3578 | relayd->control_sock.minor = relayd_sock->minor; | |
c5b6f4f0 | 3579 | |
d3e2ba59 | 3580 | relayd->relayd_session_id = relayd_session_id; |
c5b6f4f0 | 3581 | |
7735ef9e DG |
3582 | break; |
3583 | case LTTNG_STREAM_DATA: | |
3584 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3585 | lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock); |
3586 | ret = lttcomm_create_sock(&relayd->data_sock.sock); | |
4028eeb9 | 3587 | /* Handle create_sock error. */ |
f66c074c | 3588 | if (ret < 0) { |
618a6a28 | 3589 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3590 | goto error; |
f66c074c | 3591 | } |
da009f2c MD |
3592 | /* |
3593 | * Close the socket created internally by | |
3594 | * lttcomm_create_sock, so we can replace it by the one | |
3595 | * received from sessiond. | |
3596 | */ | |
3597 | if (close(relayd->data_sock.sock.fd)) { | |
3598 | PERROR("close"); | |
3599 | } | |
7735ef9e DG |
3600 | |
3601 | /* Assign new file descriptor */ | |
6151a90f JD |
3602 | relayd->data_sock.sock.fd = fd; |
3603 | /* Assign version values. */ | |
3604 | relayd->data_sock.major = relayd_sock->major; | |
3605 | relayd->data_sock.minor = relayd_sock->minor; | |
7735ef9e DG |
3606 | break; |
3607 | default: | |
3608 | ERR("Unknown relayd socket type (%d)", sock_type); | |
618a6a28 | 3609 | ret_code = LTTCOMM_CONSUMERD_FATAL; |
7735ef9e DG |
3610 | goto error; |
3611 | } | |
3612 | ||
d88aee68 | 3613 | DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)", |
7735ef9e DG |
3614 | sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", |
3615 | relayd->net_seq_idx, fd); | |
e303e76b FD |
3616 | /* |
3617 | * We gave the ownership of the fd to the relayd structure. Set the | |
3618 | * fd to -1 so we don't call close() on it in the error path below. | |
3619 | */ | |
3620 | fd = -1; | |
7735ef9e | 3621 | |
618a6a28 MD |
3622 | /* We successfully added the socket. Send status back. */ |
3623 | ret = consumer_send_status_msg(sock, ret_code); | |
3624 | if (ret < 0) { | |
3625 | /* Somehow, the session daemon is not responding anymore. */ | |
3626 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3627 | goto error_nosignal; | |
3628 | } | |
3629 | ||
7735ef9e DG |
3630 | /* |
3631 | * Add relayd socket pair to consumer data hashtable. If object already | |
3632 | * exists or on error, the function gracefully returns. | |
3633 | */ | |
11413bb9 | 3634 | relayd->ctx = ctx; |
d09e1200 | 3635 | add_relayd(relayd); |
7735ef9e DG |
3636 | |
3637 | /* All good! */ | |
2527bf85 | 3638 | return; |
7735ef9e DG |
3639 | |
3640 | error: | |
618a6a28 MD |
3641 | if (consumer_send_status_msg(sock, ret_code) < 0) { |
3642 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3643 | } | |
3644 | ||
3645 | error_nosignal: | |
4028eeb9 DG |
3646 | /* Close received socket if valid. */ |
3647 | if (fd >= 0) { | |
3648 | if (close(fd)) { | |
3649 | PERROR("close received socket"); | |
3650 | } | |
3651 | } | |
cd2b09ed DG |
3652 | |
3653 | if (relayd_created) { | |
cd2b09ed DG |
3654 | free(relayd); |
3655 | } | |
7735ef9e | 3656 | } |
ca22feea | 3657 | |
f7079f67 DG |
3658 | /* |
3659 | * Search for a relayd associated to the session id and return the reference. | |
3660 | * | |
3661 | * A rcu read side lock MUST be acquire before calling this function and locked | |
3662 | * until the relayd object is no longer necessary. | |
3663 | */ | |
3664 | static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id) | |
3665 | { | |
3666 | struct lttng_ht_iter iter; | |
f7079f67 | 3667 | struct consumer_relayd_sock_pair *relayd = NULL; |
f7079f67 DG |
3668 | |
3669 | /* Iterate over all relayd since they are indexed by net_seq_idx. */ | |
3670 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
3671 | node.node) { | |
18261bd1 DG |
3672 | /* |
3673 | * Check by sessiond id which is unique here where the relayd session | |
3674 | * id might not be when having multiple relayd. | |
3675 | */ | |
3676 | if (relayd->sessiond_session_id == id) { | |
f7079f67 | 3677 | /* Found the relayd. There can be only one per id. */ |
18261bd1 | 3678 | goto found; |
f7079f67 DG |
3679 | } |
3680 | } | |
3681 | ||
18261bd1 DG |
3682 | return NULL; |
3683 | ||
3684 | found: | |
f7079f67 DG |
3685 | return relayd; |
3686 | } | |
3687 | ||
ca22feea DG |
3688 | /* |
3689 | * Check if for a given session id there is still data needed to be extract | |
3690 | * from the buffers. | |
3691 | * | |
6d805429 | 3692 | * Return 1 if data is pending or else 0 meaning ready to be read. |
ca22feea | 3693 | */ |
6d805429 | 3694 | int consumer_data_pending(uint64_t id) |
ca22feea DG |
3695 | { |
3696 | int ret; | |
3697 | struct lttng_ht_iter iter; | |
3698 | struct lttng_ht *ht; | |
3699 | struct lttng_consumer_stream *stream; | |
f7079f67 | 3700 | struct consumer_relayd_sock_pair *relayd = NULL; |
6d805429 | 3701 | int (*data_pending)(struct lttng_consumer_stream *); |
ca22feea | 3702 | |
6d805429 | 3703 | DBG("Consumer data pending command on session id %" PRIu64, id); |
ca22feea | 3704 | |
6f6eda74 | 3705 | rcu_read_lock(); |
ca22feea DG |
3706 | pthread_mutex_lock(&consumer_data.lock); |
3707 | ||
3708 | switch (consumer_data.type) { | |
3709 | case LTTNG_CONSUMER_KERNEL: | |
6d805429 | 3710 | data_pending = lttng_kconsumer_data_pending; |
ca22feea DG |
3711 | break; |
3712 | case LTTNG_CONSUMER32_UST: | |
3713 | case LTTNG_CONSUMER64_UST: | |
6d805429 | 3714 | data_pending = lttng_ustconsumer_data_pending; |
ca22feea DG |
3715 | break; |
3716 | default: | |
3717 | ERR("Unknown consumer data type"); | |
3718 | assert(0); | |
3719 | } | |
3720 | ||
3721 | /* Ease our life a bit */ | |
3722 | ht = consumer_data.stream_list_ht; | |
3723 | ||
c8f59ee5 | 3724 | cds_lfht_for_each_entry_duplicate(ht->ht, |
d88aee68 DG |
3725 | ht->hash_fct(&id, lttng_ht_seed), |
3726 | ht->match_fct, &id, | |
ca22feea | 3727 | &iter.iter, stream, node_session_id.node) { |
7c87fb16 | 3728 | pthread_mutex_lock(&stream->lock); |
ca22feea | 3729 | |
4e9a4686 DG |
3730 | /* |
3731 | * A removed node from the hash table indicates that the stream has | |
3732 | * been deleted thus having a guarantee that the buffers are closed | |
3733 | * on the consumer side. However, data can still be transmitted | |
3734 | * over the network so don't skip the relayd check. | |
3735 | */ | |
3736 | ret = cds_lfht_is_node_deleted(&stream->node.node); | |
3737 | if (!ret) { | |
3738 | /* Check the stream if there is data in the buffers. */ | |
6d805429 DG |
3739 | ret = data_pending(stream); |
3740 | if (ret == 1) { | |
4e9a4686 | 3741 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3742 | goto data_pending; |
4e9a4686 DG |
3743 | } |
3744 | } | |
3745 | ||
78f18a75 JR |
3746 | pthread_mutex_unlock(&stream->lock); |
3747 | } | |
3748 | ||
3749 | relayd = find_relayd_by_session_id(id); | |
3750 | if (relayd) { | |
3751 | unsigned int is_data_inflight = 0; | |
3752 | ||
3753 | /* Send init command for data pending. */ | |
3754 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3755 | ret = relayd_begin_data_pending(&relayd->control_sock, | |
3756 | relayd->relayd_session_id); | |
3757 | if (ret < 0) { | |
3758 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3759 | /* Communication error thus the relayd so no data pending. */ | |
3760 | goto data_not_pending; | |
3761 | } | |
3762 | ||
3763 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
3764 | ht->hash_fct(&id, lttng_ht_seed), | |
3765 | ht->match_fct, &id, | |
3766 | &iter.iter, stream, node_session_id.node) { | |
c8f59ee5 | 3767 | if (stream->metadata_flag) { |
ad7051c0 DG |
3768 | ret = relayd_quiescent_control(&relayd->control_sock, |
3769 | stream->relayd_stream_id); | |
c8f59ee5 | 3770 | } else { |
6d805429 | 3771 | ret = relayd_data_pending(&relayd->control_sock, |
39df6d9f DG |
3772 | stream->relayd_stream_id, |
3773 | stream->next_net_seq_num - 1); | |
c8f59ee5 | 3774 | } |
78f18a75 JR |
3775 | |
3776 | if (ret == 1) { | |
3777 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3778 | goto data_pending; | |
3779 | } else if (ret < 0) { | |
11413bb9 JR |
3780 | ERR("Relayd data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
3781 | lttng_consumer_cleanup_relayd(relayd); | |
3782 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
11413bb9 JR |
3783 | goto data_not_pending; |
3784 | } | |
c8f59ee5 | 3785 | } |
f7079f67 | 3786 | |
78f18a75 | 3787 | /* Send end command for data pending. */ |
f7079f67 DG |
3788 | ret = relayd_end_data_pending(&relayd->control_sock, |
3789 | relayd->relayd_session_id, &is_data_inflight); | |
3790 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
bdd88757 | 3791 | if (ret < 0) { |
11413bb9 JR |
3792 | ERR("Relayd end data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
3793 | lttng_consumer_cleanup_relayd(relayd); | |
f7079f67 DG |
3794 | goto data_not_pending; |
3795 | } | |
bdd88757 DG |
3796 | if (is_data_inflight) { |
3797 | goto data_pending; | |
3798 | } | |
f7079f67 DG |
3799 | } |
3800 | ||
ca22feea | 3801 | /* |
f7079f67 DG |
3802 | * Finding _no_ node in the hash table and no inflight data means that the |
3803 | * stream(s) have been removed thus data is guaranteed to be available for | |
3804 | * analysis from the trace files. | |
ca22feea DG |
3805 | */ |
3806 | ||
f7079f67 | 3807 | data_not_pending: |
ca22feea DG |
3808 | /* Data is available to be read by a viewer. */ |
3809 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3810 | rcu_read_unlock(); |
6d805429 | 3811 | return 0; |
ca22feea | 3812 | |
f7079f67 | 3813 | data_pending: |
ca22feea DG |
3814 | /* Data is still being extracted from buffers. */ |
3815 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3816 | rcu_read_unlock(); |
6d805429 | 3817 | return 1; |
ca22feea | 3818 | } |
f50f23d9 DG |
3819 | |
3820 | /* | |
3821 | * Send a ret code status message to the sessiond daemon. | |
3822 | * | |
3823 | * Return the sendmsg() return value. | |
3824 | */ | |
3825 | int consumer_send_status_msg(int sock, int ret_code) | |
3826 | { | |
3827 | struct lttcomm_consumer_status_msg msg; | |
3828 | ||
53efb85a | 3829 | memset(&msg, 0, sizeof(msg)); |
f50f23d9 DG |
3830 | msg.ret_code = ret_code; |
3831 | ||
3832 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3833 | } | |
ffe60014 DG |
3834 | |
3835 | /* | |
3836 | * Send a channel status message to the sessiond daemon. | |
3837 | * | |
3838 | * Return the sendmsg() return value. | |
3839 | */ | |
3840 | int consumer_send_status_channel(int sock, | |
3841 | struct lttng_consumer_channel *channel) | |
3842 | { | |
3843 | struct lttcomm_consumer_status_channel msg; | |
3844 | ||
3845 | assert(sock >= 0); | |
3846 | ||
53efb85a | 3847 | memset(&msg, 0, sizeof(msg)); |
ffe60014 | 3848 | if (!channel) { |
0c759fc9 | 3849 | msg.ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL; |
ffe60014 | 3850 | } else { |
0c759fc9 | 3851 | msg.ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
ffe60014 DG |
3852 | msg.key = channel->key; |
3853 | msg.stream_count = channel->streams.count; | |
3854 | } | |
3855 | ||
3856 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3857 | } | |
5c786ded | 3858 | |
d07ceecd MD |
3859 | unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos, |
3860 | unsigned long produced_pos, uint64_t nb_packets_per_stream, | |
3861 | uint64_t max_sb_size) | |
5c786ded | 3862 | { |
d07ceecd | 3863 | unsigned long start_pos; |
5c786ded | 3864 | |
d07ceecd MD |
3865 | if (!nb_packets_per_stream) { |
3866 | return consumed_pos; /* Grab everything */ | |
3867 | } | |
3868 | start_pos = produced_pos - offset_align_floor(produced_pos, max_sb_size); | |
3869 | start_pos -= max_sb_size * nb_packets_per_stream; | |
3870 | if ((long) (start_pos - consumed_pos) < 0) { | |
3871 | return consumed_pos; /* Grab everything */ | |
3872 | } | |
3873 | return start_pos; | |
5c786ded | 3874 | } |
a1ae2ea5 | 3875 | |
b99a8d42 JD |
3876 | static |
3877 | int consumer_flush_buffer(struct lttng_consumer_stream *stream, int producer_active) | |
3878 | { | |
3879 | int ret = 0; | |
3880 | ||
3881 | switch (consumer_data.type) { | |
3882 | case LTTNG_CONSUMER_KERNEL: | |
3883 | ret = kernctl_buffer_flush(stream->wait_fd); | |
3884 | if (ret < 0) { | |
3885 | ERR("Failed to flush kernel stream"); | |
3886 | goto end; | |
3887 | } | |
3888 | break; | |
3889 | case LTTNG_CONSUMER32_UST: | |
3890 | case LTTNG_CONSUMER64_UST: | |
3891 | lttng_ustctl_flush_buffer(stream, producer_active); | |
3892 | break; | |
3893 | default: | |
3894 | ERR("Unknown consumer_data type"); | |
3895 | abort(); | |
3896 | } | |
3897 | ||
3898 | end: | |
3899 | return ret; | |
3900 | } | |
3901 | ||
3902 | /* | |
3903 | * Sample the rotate position for all the streams of a channel. If a stream | |
3904 | * is already at the rotate position (produced == consumed), we flag it as | |
3905 | * ready for rotation. The rotation of ready streams occurs after we have | |
3906 | * replied to the session daemon that we have finished sampling the positions. | |
e96d66b4 | 3907 | * Must be called with RCU read-side lock held to ensure existence of channel. |
b99a8d42 JD |
3908 | * |
3909 | * Returns 0 on success, < 0 on error | |
3910 | */ | |
e96d66b4 MD |
3911 | int lttng_consumer_rotate_channel(struct lttng_consumer_channel *channel, |
3912 | uint64_t key, const char *path, uint64_t relayd_id, | |
3913 | uint32_t metadata, uint64_t new_chunk_id, | |
b99a8d42 JD |
3914 | struct lttng_consumer_local_data *ctx) |
3915 | { | |
3916 | int ret; | |
b99a8d42 JD |
3917 | struct lttng_consumer_stream *stream; |
3918 | struct lttng_ht_iter iter; | |
3919 | struct lttng_ht *ht = consumer_data.stream_per_chan_id_ht; | |
3920 | ||
3921 | DBG("Consumer sample rotate position for channel %" PRIu64, key); | |
3922 | ||
3923 | rcu_read_lock(); | |
3924 | ||
b99a8d42 JD |
3925 | pthread_mutex_lock(&channel->lock); |
3926 | channel->current_chunk_id = new_chunk_id; | |
3927 | ||
3928 | ret = lttng_strncpy(channel->pathname, path, sizeof(channel->pathname)); | |
3929 | if (ret) { | |
3930 | ERR("Failed to copy new path to channel during channel rotation"); | |
3931 | ret = -1; | |
3932 | goto end_unlock_channel; | |
3933 | } | |
3934 | ||
3935 | if (relayd_id == -1ULL) { | |
3936 | /* | |
3937 | * The domain path (/ust or /kernel) has been created before, we | |
3938 | * now need to create the last part of the path: the application/user | |
3939 | * specific section (uid/1000/64-bit). | |
3940 | */ | |
3941 | ret = utils_mkdir_recursive(channel->pathname, S_IRWXU | S_IRWXG, | |
3942 | channel->uid, channel->gid); | |
3943 | if (ret < 0) { | |
3944 | ERR("Failed to create trace directory at %s during rotation", | |
3945 | channel->pathname); | |
3946 | ret = -1; | |
3947 | goto end_unlock_channel; | |
3948 | } | |
3949 | } | |
3950 | ||
3951 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
3952 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
3953 | ht->match_fct, &channel->key, &iter.iter, | |
3954 | stream, node_channel_id.node) { | |
3955 | unsigned long consumed_pos; | |
3956 | ||
3957 | health_code_update(); | |
3958 | ||
3959 | /* | |
3960 | * Lock stream because we are about to change its state. | |
3961 | */ | |
3962 | pthread_mutex_lock(&stream->lock); | |
3963 | ||
3964 | ret = lttng_strncpy(stream->channel_read_only_attributes.path, | |
3965 | channel->pathname, | |
3966 | sizeof(stream->channel_read_only_attributes.path)); | |
3967 | if (ret) { | |
3968 | ERR("Failed to sample channel path name during channel rotation"); | |
3969 | goto end_unlock_stream; | |
3970 | } | |
3971 | ret = lttng_consumer_sample_snapshot_positions(stream); | |
3972 | if (ret < 0) { | |
3973 | ERR("Failed to sample snapshot position during channel rotation"); | |
3974 | goto end_unlock_stream; | |
3975 | } | |
3976 | ||
3977 | ret = lttng_consumer_get_produced_snapshot(stream, | |
3978 | &stream->rotate_position); | |
3979 | if (ret < 0) { | |
3980 | ERR("Failed to sample produced position during channel rotation"); | |
3981 | goto end_unlock_stream; | |
3982 | } | |
3983 | ||
3984 | lttng_consumer_get_consumed_snapshot(stream, | |
3985 | &consumed_pos); | |
3986 | if (consumed_pos == stream->rotate_position) { | |
3987 | stream->rotate_ready = true; | |
3988 | } | |
b99a8d42 JD |
3989 | |
3990 | ret = consumer_flush_buffer(stream, 1); | |
3991 | if (ret < 0) { | |
3992 | ERR("Failed to flush stream %" PRIu64 " during channel rotation", | |
3993 | stream->key); | |
3994 | goto end_unlock_stream; | |
3995 | } | |
3996 | ||
3997 | pthread_mutex_unlock(&stream->lock); | |
3998 | } | |
3999 | pthread_mutex_unlock(&channel->lock); | |
4000 | ||
4001 | ret = 0; | |
4002 | goto end; | |
4003 | ||
4004 | end_unlock_stream: | |
4005 | pthread_mutex_unlock(&stream->lock); | |
4006 | end_unlock_channel: | |
4007 | pthread_mutex_unlock(&channel->lock); | |
4008 | end: | |
4009 | rcu_read_unlock(); | |
4010 | return ret; | |
4011 | } | |
4012 | ||
02d02e31 JD |
4013 | /* |
4014 | * Check if a stream is ready to be rotated after extracting it. | |
4015 | * | |
4016 | * Return 1 if it is ready for rotation, 0 if it is not, a negative value on | |
4017 | * error. Stream lock must be held. | |
4018 | */ | |
4019 | int lttng_consumer_stream_is_rotate_ready(struct lttng_consumer_stream *stream) | |
4020 | { | |
4021 | int ret; | |
4022 | unsigned long consumed_pos; | |
4023 | ||
4024 | if (!stream->rotate_position && !stream->rotate_ready) { | |
4025 | ret = 0; | |
4026 | goto end; | |
4027 | } | |
4028 | ||
4029 | if (stream->rotate_ready) { | |
4030 | ret = 1; | |
4031 | goto end; | |
4032 | } | |
4033 | ||
4034 | /* | |
4035 | * If we don't have the rotate_ready flag, check the consumed position | |
4036 | * to determine if we need to rotate. | |
4037 | */ | |
4038 | ret = lttng_consumer_sample_snapshot_positions(stream); | |
4039 | if (ret < 0) { | |
4040 | ERR("Taking snapshot positions"); | |
4041 | goto end; | |
4042 | } | |
4043 | ||
4044 | ret = lttng_consumer_get_consumed_snapshot(stream, &consumed_pos); | |
4045 | if (ret < 0) { | |
4046 | ERR("Consumed snapshot position"); | |
4047 | goto end; | |
4048 | } | |
4049 | ||
4050 | /* Rotate position not reached yet (with check for overflow). */ | |
4051 | if ((long) (consumed_pos - stream->rotate_position) < 0) { | |
4052 | ret = 0; | |
4053 | goto end; | |
4054 | } | |
4055 | ret = 1; | |
4056 | ||
4057 | end: | |
4058 | return ret; | |
4059 | } | |
4060 | ||
d73bf3d7 JD |
4061 | /* |
4062 | * Reset the state for a stream after a rotation occurred. | |
4063 | */ | |
4064 | void lttng_consumer_reset_stream_rotate_state(struct lttng_consumer_stream *stream) | |
4065 | { | |
4066 | stream->rotate_position = 0; | |
4067 | stream->rotate_ready = false; | |
4068 | } | |
4069 | ||
4070 | /* | |
4071 | * Perform the rotation a local stream file. | |
4072 | */ | |
4073 | int rotate_local_stream(struct lttng_consumer_local_data *ctx, | |
4074 | struct lttng_consumer_stream *stream) | |
4075 | { | |
4076 | int ret; | |
4077 | ||
4078 | DBG("Rotate local stream: stream key %" PRIu64 ", channel key %" PRIu64 " at path %s", | |
4079 | stream->key, | |
4080 | stream->chan->key, | |
4081 | stream->channel_read_only_attributes.path); | |
4082 | ||
4083 | ret = close(stream->out_fd); | |
4084 | if (ret < 0) { | |
4085 | PERROR("Closing trace file (fd %d), stream %" PRIu64, | |
4086 | stream->out_fd, stream->key); | |
4087 | assert(0); | |
4088 | goto error; | |
4089 | } | |
4090 | ||
4091 | ret = utils_create_stream_file( | |
4092 | stream->channel_read_only_attributes.path, | |
4093 | stream->name, | |
4094 | stream->channel_read_only_attributes.tracefile_size, | |
4095 | stream->tracefile_count_current, | |
4096 | stream->uid, stream->gid, NULL); | |
4097 | if (ret < 0) { | |
4098 | ERR("Rotate create stream file"); | |
4099 | goto error; | |
4100 | } | |
4101 | stream->out_fd = ret; | |
4102 | stream->tracefile_size_current = 0; | |
4103 | ||
4104 | if (!stream->metadata_flag) { | |
4105 | struct lttng_index_file *index_file; | |
4106 | ||
4107 | lttng_index_file_put(stream->index_file); | |
4108 | ||
4109 | index_file = lttng_index_file_create( | |
4110 | stream->channel_read_only_attributes.path, | |
4111 | stream->name, stream->uid, stream->gid, | |
4112 | stream->channel_read_only_attributes.tracefile_size, | |
4113 | stream->tracefile_count_current, | |
4114 | CTF_INDEX_MAJOR, CTF_INDEX_MINOR); | |
4115 | if (!index_file) { | |
4116 | ERR("Create index file during rotation"); | |
4117 | goto error; | |
4118 | } | |
4119 | stream->index_file = index_file; | |
4120 | stream->out_fd_offset = 0; | |
4121 | } | |
4122 | ||
4123 | ret = 0; | |
4124 | goto end; | |
4125 | ||
4126 | error: | |
4127 | ret = -1; | |
4128 | end: | |
4129 | return ret; | |
4130 | ||
4131 | } | |
4132 | ||
4133 | /* | |
4134 | * Perform the rotation a stream file on the relay. | |
4135 | */ | |
4136 | int rotate_relay_stream(struct lttng_consumer_local_data *ctx, | |
4137 | struct lttng_consumer_stream *stream) | |
4138 | { | |
4139 | int ret; | |
4140 | struct consumer_relayd_sock_pair *relayd; | |
4141 | ||
4142 | DBG("Rotate relay stream"); | |
4143 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
4144 | if (!relayd) { | |
4145 | ERR("Failed to find relayd"); | |
4146 | ret = -1; | |
4147 | goto end; | |
4148 | } | |
4149 | ||
4150 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
4151 | ret = relayd_rotate_stream(&relayd->control_sock, | |
4152 | stream->relayd_stream_id, | |
4153 | stream->channel_read_only_attributes.path, | |
4154 | stream->chan->current_chunk_id, | |
4155 | stream->last_sequence_number); | |
4156 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
768f6a48 JR |
4157 | if (ret < 0) { |
4158 | ERR("Relayd rotate stream failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); | |
4159 | lttng_consumer_cleanup_relayd(relayd); | |
4160 | } | |
d73bf3d7 JD |
4161 | if (ret) { |
4162 | ERR("Rotate relay stream"); | |
4163 | } | |
4164 | ||
4165 | end: | |
4166 | return ret; | |
4167 | } | |
4168 | ||
4169 | /* | |
4170 | * Performs the stream rotation for the rotate session feature if needed. | |
4171 | * It must be called with the stream lock held. | |
4172 | * | |
4173 | * Return 0 on success, a negative number of error. | |
4174 | */ | |
4175 | int lttng_consumer_rotate_stream(struct lttng_consumer_local_data *ctx, | |
02d02e31 | 4176 | struct lttng_consumer_stream *stream, bool *rotated) |
d73bf3d7 JD |
4177 | { |
4178 | int ret; | |
4179 | ||
4180 | DBG("Consumer rotate stream %" PRIu64, stream->key); | |
4181 | ||
4182 | if (stream->net_seq_idx != (uint64_t) -1ULL) { | |
4183 | ret = rotate_relay_stream(ctx, stream); | |
4184 | } else { | |
4185 | ret = rotate_local_stream(ctx, stream); | |
4186 | } | |
82528808 | 4187 | stream->trace_archive_id++; |
d73bf3d7 | 4188 | if (ret < 0) { |
82528808 | 4189 | ERR("Failed to rotate stream, ret = %i", ret); |
d73bf3d7 JD |
4190 | goto error; |
4191 | } | |
4192 | ||
4193 | if (stream->metadata_flag) { | |
4194 | switch (consumer_data.type) { | |
4195 | case LTTNG_CONSUMER_KERNEL: | |
4196 | /* | |
4197 | * Reset the position of what has been read from the metadata | |
4198 | * cache to 0 so we can dump it again. | |
4199 | */ | |
4200 | ret = kernctl_metadata_cache_dump(stream->wait_fd); | |
4201 | if (ret < 0) { | |
4202 | ERR("Failed to dump the kernel metadata cache after rotation"); | |
4203 | goto error; | |
4204 | } | |
4205 | break; | |
4206 | case LTTNG_CONSUMER32_UST: | |
4207 | case LTTNG_CONSUMER64_UST: | |
4208 | /* | |
4209 | * Reset the position pushed from the metadata cache so it | |
4210 | * will write from the beginning on the next push. | |
4211 | */ | |
4212 | stream->ust_metadata_pushed = 0; | |
4213 | break; | |
4214 | default: | |
4215 | ERR("Unknown consumer_data type"); | |
4216 | abort(); | |
4217 | } | |
4218 | } | |
4219 | lttng_consumer_reset_stream_rotate_state(stream); | |
4220 | ||
02d02e31 JD |
4221 | if (rotated) { |
4222 | *rotated = true; | |
4223 | } | |
4224 | ||
d73bf3d7 JD |
4225 | ret = 0; |
4226 | ||
4227 | error: | |
4228 | return ret; | |
4229 | } | |
4230 | ||
b99a8d42 JD |
4231 | /* |
4232 | * Rotate all the ready streams now. | |
4233 | * | |
4234 | * This is especially important for low throughput streams that have already | |
4235 | * been consumed, we cannot wait for their next packet to perform the | |
4236 | * rotation. | |
e96d66b4 MD |
4237 | * Need to be called with RCU read-side lock held to ensure existence of |
4238 | * channel. | |
b99a8d42 JD |
4239 | * |
4240 | * Returns 0 on success, < 0 on error | |
4241 | */ | |
e96d66b4 MD |
4242 | int lttng_consumer_rotate_ready_streams(struct lttng_consumer_channel *channel, |
4243 | uint64_t key, struct lttng_consumer_local_data *ctx) | |
b99a8d42 JD |
4244 | { |
4245 | int ret; | |
b99a8d42 JD |
4246 | struct lttng_consumer_stream *stream; |
4247 | struct lttng_ht_iter iter; | |
4248 | struct lttng_ht *ht = consumer_data.stream_per_chan_id_ht; | |
4249 | ||
4250 | rcu_read_lock(); | |
4251 | ||
4252 | DBG("Consumer rotate ready streams in channel %" PRIu64, key); | |
4253 | ||
b99a8d42 JD |
4254 | cds_lfht_for_each_entry_duplicate(ht->ht, |
4255 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
4256 | ht->match_fct, &channel->key, &iter.iter, | |
4257 | stream, node_channel_id.node) { | |
4258 | health_code_update(); | |
4259 | ||
4260 | pthread_mutex_lock(&stream->lock); | |
4261 | ||
4262 | if (!stream->rotate_ready) { | |
4263 | pthread_mutex_unlock(&stream->lock); | |
4264 | continue; | |
4265 | } | |
4266 | DBG("Consumer rotate ready stream %" PRIu64, stream->key); | |
4267 | ||
4268 | ret = lttng_consumer_rotate_stream(ctx, stream, NULL); | |
4269 | pthread_mutex_unlock(&stream->lock); | |
4270 | if (ret) { | |
4271 | goto end; | |
4272 | } | |
4273 | ||
4274 | ret = consumer_post_rotation(stream, ctx); | |
4275 | if (ret) { | |
4276 | goto end; | |
4277 | } | |
4278 | } | |
4279 | ||
4280 | ret = 0; | |
4281 | ||
4282 | end: | |
4283 | rcu_read_unlock(); | |
4284 | return ret; | |
4285 | } | |
4286 | ||
00fb02ac JD |
4287 | static |
4288 | int rotate_rename_local(const char *old_path, const char *new_path, | |
4289 | uid_t uid, gid_t gid) | |
4290 | { | |
4291 | int ret; | |
4292 | ||
4293 | assert(old_path); | |
4294 | assert(new_path); | |
4295 | ||
4296 | ret = utils_mkdir_recursive(new_path, S_IRWXU | S_IRWXG, uid, gid); | |
4297 | if (ret < 0) { | |
4298 | ERR("Create directory on rotate"); | |
4299 | goto end; | |
4300 | } | |
4301 | ||
4302 | ret = rename(old_path, new_path); | |
4303 | if (ret < 0 && errno != ENOENT) { | |
4304 | PERROR("Rename completed rotation chunk"); | |
4305 | goto end; | |
4306 | } | |
4307 | ||
4308 | ret = 0; | |
4309 | end: | |
4310 | return ret; | |
4311 | } | |
4312 | ||
4313 | static | |
4314 | int rotate_rename_relay(const char *old_path, const char *new_path, | |
4315 | uint64_t relayd_id) | |
4316 | { | |
4317 | int ret; | |
4318 | struct consumer_relayd_sock_pair *relayd; | |
4319 | ||
4320 | relayd = consumer_find_relayd(relayd_id); | |
4321 | if (!relayd) { | |
4322 | ERR("Failed to find relayd while running rotate_rename_relay command"); | |
4323 | ret = -1; | |
4324 | goto end; | |
4325 | } | |
4326 | ||
4327 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
4328 | ret = relayd_rotate_rename(&relayd->control_sock, old_path, new_path); | |
768f6a48 JR |
4329 | if (ret < 0) { |
4330 | ERR("Relayd rotate rename failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); | |
4331 | lttng_consumer_cleanup_relayd(relayd); | |
4332 | } | |
00fb02ac JD |
4333 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
4334 | end: | |
4335 | return ret; | |
4336 | } | |
4337 | ||
4338 | int lttng_consumer_rotate_rename(const char *old_path, const char *new_path, | |
4339 | uid_t uid, gid_t gid, uint64_t relayd_id) | |
4340 | { | |
4341 | if (relayd_id != -1ULL) { | |
4342 | return rotate_rename_relay(old_path, new_path, relayd_id); | |
4343 | } else { | |
4344 | return rotate_rename_local(old_path, new_path, uid, gid); | |
4345 | } | |
4346 | } | |
4347 | ||
82528808 JG |
4348 | /* Stream lock must be acquired by the caller. */ |
4349 | static | |
4350 | bool check_stream_rotation_pending(const struct lttng_consumer_stream *stream, | |
4351 | uint64_t session_id, uint64_t chunk_id) | |
4352 | { | |
4353 | bool pending = false; | |
4354 | ||
4355 | if (stream->session_id != session_id) { | |
4356 | /* Skip. */ | |
4357 | goto end; | |
4358 | } | |
4359 | ||
4360 | /* | |
4361 | * If the stream's archive_id belongs to the chunk being rotated (or an | |
4362 | * even older one), it means that the consumer has not consumed all the | |
4363 | * buffers that belong to the chunk being rotated. Therefore, the | |
4364 | * rotation is considered as ongoing/pending. | |
4365 | */ | |
4366 | pending = stream->trace_archive_id <= chunk_id; | |
4367 | end: | |
4368 | return pending; | |
4369 | } | |
4370 | ||
4371 | /* RCU read lock must be acquired by the caller. */ | |
4372 | int lttng_consumer_check_rotation_pending_local(uint64_t session_id, | |
4373 | uint64_t chunk_id) | |
4374 | { | |
4375 | struct lttng_ht_iter iter; | |
4376 | struct lttng_consumer_stream *stream; | |
4377 | bool rotation_pending = false; | |
4378 | ||
4379 | /* Start with the metadata streams... */ | |
4380 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
4381 | pthread_mutex_lock(&stream->lock); | |
4382 | rotation_pending = check_stream_rotation_pending(stream, | |
4383 | session_id, chunk_id); | |
4384 | pthread_mutex_unlock(&stream->lock); | |
4385 | if (rotation_pending) { | |
4386 | goto end; | |
4387 | } | |
4388 | } | |
4389 | ||
4390 | /* ... followed by the data streams. */ | |
4391 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
4392 | pthread_mutex_lock(&stream->lock); | |
4393 | rotation_pending = check_stream_rotation_pending(stream, | |
4394 | session_id, chunk_id); | |
4395 | pthread_mutex_unlock(&stream->lock); | |
4396 | if (rotation_pending) { | |
4397 | goto end; | |
4398 | } | |
4399 | } | |
4400 | ||
4401 | end: | |
4402 | return !!rotation_pending; | |
4403 | } | |
4404 | ||
4405 | int lttng_consumer_check_rotation_pending_relay(uint64_t session_id, | |
d88744a4 JD |
4406 | uint64_t relayd_id, uint64_t chunk_id) |
4407 | { | |
4408 | int ret; | |
4409 | struct consumer_relayd_sock_pair *relayd; | |
4410 | ||
4411 | relayd = consumer_find_relayd(relayd_id); | |
4412 | if (!relayd) { | |
82528808 | 4413 | ERR("Failed to find relayd id %" PRIu64, relayd_id); |
d88744a4 JD |
4414 | ret = -1; |
4415 | goto end; | |
4416 | } | |
4417 | ||
4418 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
4419 | ret = relayd_rotate_pending(&relayd->control_sock, chunk_id); | |
768f6a48 JR |
4420 | if (ret < 0) { |
4421 | ERR("Relayd rotate pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); | |
4422 | lttng_consumer_cleanup_relayd(relayd); | |
4423 | } | |
d88744a4 JD |
4424 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
4425 | ||
4426 | end: | |
4427 | return ret; | |
4428 | } | |
4429 | ||
a1ae2ea5 JD |
4430 | static |
4431 | int mkdir_local(const char *path, uid_t uid, gid_t gid) | |
4432 | { | |
4433 | int ret; | |
4434 | ||
4435 | ret = utils_mkdir_recursive(path, S_IRWXU | S_IRWXG, uid, gid); | |
4436 | if (ret < 0) { | |
4437 | /* utils_mkdir_recursive logs an error. */ | |
4438 | goto end; | |
4439 | } | |
4440 | ||
4441 | ret = 0; | |
4442 | end: | |
4443 | return ret; | |
4444 | } | |
4445 | ||
4446 | static | |
4447 | int mkdir_relay(const char *path, uint64_t relayd_id) | |
4448 | { | |
4449 | int ret; | |
4450 | struct consumer_relayd_sock_pair *relayd; | |
4451 | ||
4452 | relayd = consumer_find_relayd(relayd_id); | |
4453 | if (!relayd) { | |
4454 | ERR("Failed to find relayd"); | |
4455 | ret = -1; | |
4456 | goto end; | |
4457 | } | |
4458 | ||
4459 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
4460 | ret = relayd_mkdir(&relayd->control_sock, path); | |
768f6a48 JR |
4461 | if (ret < 0) { |
4462 | ERR("Relayd mkdir failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); | |
4463 | lttng_consumer_cleanup_relayd(relayd); | |
4464 | } | |
a1ae2ea5 JD |
4465 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
4466 | ||
4467 | end: | |
4468 | return ret; | |
4469 | ||
4470 | } | |
4471 | ||
4472 | int lttng_consumer_mkdir(const char *path, uid_t uid, gid_t gid, | |
4473 | uint64_t relayd_id) | |
4474 | { | |
4475 | if (relayd_id != -1ULL) { | |
4476 | return mkdir_relay(path, relayd_id); | |
4477 | } else { | |
4478 | return mkdir_local(path, uid, gid); | |
4479 | } | |
4480 | } |