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 | ||
00e2e675 DG |
326 | free(relayd); |
327 | } | |
328 | ||
329 | /* | |
330 | * Destroy and free relayd socket pair object. | |
00e2e675 | 331 | */ |
51230d70 | 332 | void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
333 | { |
334 | int ret; | |
335 | struct lttng_ht_iter iter; | |
336 | ||
173af62f DG |
337 | if (relayd == NULL) { |
338 | return; | |
339 | } | |
340 | ||
00e2e675 DG |
341 | DBG("Consumer destroy and close relayd socket pair"); |
342 | ||
343 | iter.iter.node = &relayd->node.node; | |
344 | ret = lttng_ht_del(consumer_data.relayd_ht, &iter); | |
173af62f | 345 | if (ret != 0) { |
8994307f | 346 | /* We assume the relayd is being or is destroyed */ |
173af62f DG |
347 | return; |
348 | } | |
00e2e675 | 349 | |
00e2e675 | 350 | /* RCU free() call */ |
ffe60014 DG |
351 | call_rcu(&relayd->node.head, free_relayd_rcu); |
352 | } | |
353 | ||
354 | /* | |
355 | * Remove a channel from the global list protected by a mutex. This function is | |
356 | * also responsible for freeing its data structures. | |
357 | */ | |
358 | void consumer_del_channel(struct lttng_consumer_channel *channel) | |
359 | { | |
360 | int ret; | |
361 | struct lttng_ht_iter iter; | |
362 | ||
d88aee68 | 363 | DBG("Consumer delete channel key %" PRIu64, channel->key); |
ffe60014 DG |
364 | |
365 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 366 | pthread_mutex_lock(&channel->lock); |
ffe60014 | 367 | |
212d67a2 DG |
368 | /* Destroy streams that might have been left in the stream list. */ |
369 | clean_channel_stream_list(channel); | |
51e762e5 | 370 | |
d3e2ba59 JD |
371 | if (channel->live_timer_enabled == 1) { |
372 | consumer_timer_live_stop(channel); | |
373 | } | |
e9404c27 JG |
374 | if (channel->monitor_timer_enabled == 1) { |
375 | consumer_timer_monitor_stop(channel); | |
376 | } | |
d3e2ba59 | 377 | |
ffe60014 DG |
378 | switch (consumer_data.type) { |
379 | case LTTNG_CONSUMER_KERNEL: | |
380 | break; | |
381 | case LTTNG_CONSUMER32_UST: | |
382 | case LTTNG_CONSUMER64_UST: | |
383 | lttng_ustconsumer_del_channel(channel); | |
384 | break; | |
385 | default: | |
386 | ERR("Unknown consumer_data type"); | |
387 | assert(0); | |
388 | goto end; | |
389 | } | |
390 | ||
391 | rcu_read_lock(); | |
392 | iter.iter.node = &channel->node.node; | |
393 | ret = lttng_ht_del(consumer_data.channel_ht, &iter); | |
394 | assert(!ret); | |
395 | rcu_read_unlock(); | |
396 | ||
397 | call_rcu(&channel->node.head, free_channel_rcu); | |
398 | end: | |
a9838785 | 399 | pthread_mutex_unlock(&channel->lock); |
ffe60014 | 400 | pthread_mutex_unlock(&consumer_data.lock); |
00e2e675 DG |
401 | } |
402 | ||
228b5bf7 DG |
403 | /* |
404 | * Iterate over the relayd hash table and destroy each element. Finally, | |
405 | * destroy the whole hash table. | |
406 | */ | |
407 | static void cleanup_relayd_ht(void) | |
408 | { | |
409 | struct lttng_ht_iter iter; | |
410 | struct consumer_relayd_sock_pair *relayd; | |
411 | ||
412 | rcu_read_lock(); | |
413 | ||
414 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
415 | node.node) { | |
51230d70 | 416 | consumer_destroy_relayd(relayd); |
228b5bf7 DG |
417 | } |
418 | ||
228b5bf7 | 419 | rcu_read_unlock(); |
36b588ed MD |
420 | |
421 | lttng_ht_destroy(consumer_data.relayd_ht); | |
228b5bf7 DG |
422 | } |
423 | ||
8994307f DG |
424 | /* |
425 | * Update the end point status of all streams having the given network sequence | |
426 | * index (relayd index). | |
427 | * | |
428 | * It's atomically set without having the stream mutex locked which is fine | |
429 | * because we handle the write/read race with a pipe wakeup for each thread. | |
430 | */ | |
da009f2c | 431 | static void update_endpoint_status_by_netidx(uint64_t net_seq_idx, |
8994307f DG |
432 | enum consumer_endpoint_status status) |
433 | { | |
434 | struct lttng_ht_iter iter; | |
435 | struct lttng_consumer_stream *stream; | |
436 | ||
da009f2c | 437 | DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx); |
8994307f DG |
438 | |
439 | rcu_read_lock(); | |
440 | ||
441 | /* Let's begin with metadata */ | |
442 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
443 | if (stream->net_seq_idx == net_seq_idx) { | |
444 | uatomic_set(&stream->endpoint_status, status); | |
445 | DBG("Delete flag set to metadata stream %d", stream->wait_fd); | |
446 | } | |
447 | } | |
448 | ||
449 | /* Follow up by the data streams */ | |
450 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
451 | if (stream->net_seq_idx == net_seq_idx) { | |
452 | uatomic_set(&stream->endpoint_status, status); | |
453 | DBG("Delete flag set to data stream %d", stream->wait_fd); | |
454 | } | |
455 | } | |
456 | rcu_read_unlock(); | |
457 | } | |
458 | ||
459 | /* | |
460 | * Cleanup a relayd object by flagging every associated streams for deletion, | |
461 | * destroying the object meaning removing it from the relayd hash table, | |
462 | * closing the sockets and freeing the memory in a RCU call. | |
463 | * | |
464 | * If a local data context is available, notify the threads that the streams' | |
465 | * state have changed. | |
466 | */ | |
467 | static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd, | |
468 | struct lttng_consumer_local_data *ctx) | |
469 | { | |
da009f2c | 470 | uint64_t netidx; |
8994307f DG |
471 | |
472 | assert(relayd); | |
473 | ||
9617607b DG |
474 | DBG("Cleaning up relayd sockets"); |
475 | ||
8994307f DG |
476 | /* Save the net sequence index before destroying the object */ |
477 | netidx = relayd->net_seq_idx; | |
478 | ||
479 | /* | |
480 | * Delete the relayd from the relayd hash table, close the sockets and free | |
481 | * the object in a RCU call. | |
482 | */ | |
51230d70 | 483 | consumer_destroy_relayd(relayd); |
8994307f DG |
484 | |
485 | /* Set inactive endpoint to all streams */ | |
486 | update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE); | |
487 | ||
488 | /* | |
489 | * With a local data context, notify the threads that the streams' state | |
490 | * have changed. The write() action on the pipe acts as an "implicit" | |
491 | * memory barrier ordering the updates of the end point status from the | |
492 | * read of this status which happens AFTER receiving this notify. | |
493 | */ | |
494 | if (ctx) { | |
acdb9057 | 495 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
13886d2d | 496 | notify_thread_lttng_pipe(ctx->consumer_metadata_pipe); |
8994307f DG |
497 | } |
498 | } | |
499 | ||
a6ba4fe1 DG |
500 | /* |
501 | * Flag a relayd socket pair for destruction. Destroy it if the refcount | |
502 | * reaches zero. | |
503 | * | |
504 | * RCU read side lock MUST be aquired before calling this function. | |
505 | */ | |
506 | void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) | |
507 | { | |
508 | assert(relayd); | |
509 | ||
510 | /* Set destroy flag for this object */ | |
511 | uatomic_set(&relayd->destroy_flag, 1); | |
512 | ||
513 | /* Destroy the relayd if refcount is 0 */ | |
514 | if (uatomic_read(&relayd->refcount) == 0) { | |
51230d70 | 515 | consumer_destroy_relayd(relayd); |
a6ba4fe1 DG |
516 | } |
517 | } | |
518 | ||
3bd1e081 | 519 | /* |
1d1a276c DG |
520 | * Completly destroy stream from every visiable data structure and the given |
521 | * hash table if one. | |
522 | * | |
523 | * One this call returns, the stream object is not longer usable nor visible. | |
3bd1e081 | 524 | */ |
e316aad5 DG |
525 | void consumer_del_stream(struct lttng_consumer_stream *stream, |
526 | struct lttng_ht *ht) | |
3bd1e081 | 527 | { |
1d1a276c | 528 | consumer_stream_destroy(stream, ht); |
3bd1e081 MD |
529 | } |
530 | ||
5ab66908 MD |
531 | /* |
532 | * XXX naming of del vs destroy is all mixed up. | |
533 | */ | |
534 | void consumer_del_stream_for_data(struct lttng_consumer_stream *stream) | |
535 | { | |
536 | consumer_stream_destroy(stream, data_ht); | |
537 | } | |
538 | ||
539 | void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream) | |
540 | { | |
541 | consumer_stream_destroy(stream, metadata_ht); | |
542 | } | |
543 | ||
6af2ba3b JD |
544 | void consumer_stream_copy_ro_channel_values(struct lttng_consumer_stream *stream, |
545 | struct lttng_consumer_channel *channel) | |
546 | { | |
547 | stream->channel_ro_tracefile_size = channel->tracefile_size; | |
548 | memcpy(stream->channel_ro_pathname, channel->pathname, PATH_MAX); | |
549 | } | |
550 | ||
d88aee68 DG |
551 | struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, |
552 | uint64_t stream_key, | |
3bd1e081 | 553 | enum lttng_consumer_stream_state state, |
ffe60014 | 554 | const char *channel_name, |
6df2e2c9 | 555 | uid_t uid, |
00e2e675 | 556 | gid_t gid, |
57a269f2 | 557 | uint64_t relayd_id, |
53632229 | 558 | uint64_t session_id, |
ffe60014 DG |
559 | int cpu, |
560 | int *alloc_ret, | |
4891ece8 DG |
561 | enum consumer_channel_type type, |
562 | unsigned int monitor) | |
3bd1e081 | 563 | { |
ffe60014 | 564 | int ret; |
3bd1e081 | 565 | struct lttng_consumer_stream *stream; |
3bd1e081 | 566 | |
effcf122 | 567 | stream = zmalloc(sizeof(*stream)); |
3bd1e081 | 568 | if (stream == NULL) { |
7a57cf92 | 569 | PERROR("malloc struct lttng_consumer_stream"); |
ffe60014 | 570 | ret = -ENOMEM; |
7a57cf92 | 571 | goto end; |
3bd1e081 | 572 | } |
7a57cf92 | 573 | |
d56db448 DG |
574 | rcu_read_lock(); |
575 | ||
3bd1e081 | 576 | stream->key = stream_key; |
3bd1e081 MD |
577 | stream->out_fd = -1; |
578 | stream->out_fd_offset = 0; | |
e5d1a9b3 | 579 | stream->output_written = 0; |
3bd1e081 | 580 | stream->state = state; |
6df2e2c9 MD |
581 | stream->uid = uid; |
582 | stream->gid = gid; | |
ffe60014 | 583 | stream->net_seq_idx = relayd_id; |
53632229 | 584 | stream->session_id = session_id; |
4891ece8 | 585 | stream->monitor = monitor; |
774d490c | 586 | stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE; |
f8f3885c | 587 | stream->index_file = NULL; |
fb83fe64 | 588 | stream->last_sequence_number = -1ULL; |
53632229 | 589 | pthread_mutex_init(&stream->lock, NULL); |
c585821b | 590 | pthread_mutex_init(&stream->metadata_timer_lock, NULL); |
58b1f425 | 591 | |
ffe60014 DG |
592 | /* If channel is the metadata, flag this stream as metadata. */ |
593 | if (type == CONSUMER_CHANNEL_TYPE_METADATA) { | |
594 | stream->metadata_flag = 1; | |
595 | /* Metadata is flat out. */ | |
596 | strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name)); | |
94d49140 JD |
597 | /* Live rendez-vous point. */ |
598 | pthread_cond_init(&stream->metadata_rdv, NULL); | |
599 | pthread_mutex_init(&stream->metadata_rdv_lock, NULL); | |
58b1f425 | 600 | } else { |
ffe60014 DG |
601 | /* Format stream name to <channel_name>_<cpu_number> */ |
602 | ret = snprintf(stream->name, sizeof(stream->name), "%s_%d", | |
603 | channel_name, cpu); | |
604 | if (ret < 0) { | |
605 | PERROR("snprintf stream name"); | |
606 | goto error; | |
607 | } | |
58b1f425 | 608 | } |
c30aaa51 | 609 | |
ffe60014 | 610 | /* Key is always the wait_fd for streams. */ |
d88aee68 | 611 | lttng_ht_node_init_u64(&stream->node, stream->key); |
ffe60014 | 612 | |
d8ef542d MD |
613 | /* Init node per channel id key */ |
614 | lttng_ht_node_init_u64(&stream->node_channel_id, channel_key); | |
615 | ||
53632229 | 616 | /* Init session id node with the stream session id */ |
d88aee68 | 617 | lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id); |
53632229 | 618 | |
07b86b52 JD |
619 | DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64 |
620 | " relayd_id %" PRIu64 ", session_id %" PRIu64, | |
621 | stream->name, stream->key, channel_key, | |
622 | stream->net_seq_idx, stream->session_id); | |
d56db448 DG |
623 | |
624 | rcu_read_unlock(); | |
3bd1e081 | 625 | return stream; |
c80048c6 MD |
626 | |
627 | error: | |
d56db448 | 628 | rcu_read_unlock(); |
c80048c6 | 629 | free(stream); |
7a57cf92 | 630 | end: |
ffe60014 DG |
631 | if (alloc_ret) { |
632 | *alloc_ret = ret; | |
633 | } | |
c80048c6 | 634 | return NULL; |
3bd1e081 MD |
635 | } |
636 | ||
637 | /* | |
638 | * Add a stream to the global list protected by a mutex. | |
639 | */ | |
5ab66908 | 640 | int consumer_add_data_stream(struct lttng_consumer_stream *stream) |
3bd1e081 | 641 | { |
5ab66908 | 642 | struct lttng_ht *ht = data_ht; |
3bd1e081 MD |
643 | int ret = 0; |
644 | ||
e316aad5 | 645 | assert(stream); |
43c34bc3 | 646 | assert(ht); |
c77fc10a | 647 | |
d88aee68 | 648 | DBG3("Adding consumer stream %" PRIu64, stream->key); |
e316aad5 DG |
649 | |
650 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 651 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 652 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 653 | pthread_mutex_lock(&stream->lock); |
b0b335c8 | 654 | rcu_read_lock(); |
e316aad5 | 655 | |
43c34bc3 | 656 | /* Steal stream identifier to avoid having streams with the same key */ |
ffe60014 | 657 | steal_stream_key(stream->key, ht); |
43c34bc3 | 658 | |
d88aee68 | 659 | lttng_ht_add_unique_u64(ht, &stream->node); |
00e2e675 | 660 | |
d8ef542d MD |
661 | lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, |
662 | &stream->node_channel_id); | |
663 | ||
ca22feea DG |
664 | /* |
665 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
666 | * the key since the HT does not use it and we allow to add redundant keys | |
667 | * into this table. | |
668 | */ | |
d88aee68 | 669 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 670 | |
e316aad5 | 671 | /* |
ffe60014 DG |
672 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
673 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
674 | * causes the count to become 0 also causes a stream to be added. The |
675 | * channel deletion will thus be triggered by the following removal of this | |
676 | * stream. | |
677 | */ | |
ffe60014 | 678 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
679 | /* Increment refcount before decrementing nb_init_stream_left */ |
680 | cmm_smp_wmb(); | |
ffe60014 | 681 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
682 | } |
683 | ||
684 | /* Update consumer data once the node is inserted. */ | |
3bd1e081 MD |
685 | consumer_data.stream_count++; |
686 | consumer_data.need_update = 1; | |
687 | ||
e316aad5 | 688 | rcu_read_unlock(); |
2e818a6a | 689 | pthread_mutex_unlock(&stream->lock); |
ec6ea7d0 | 690 | pthread_mutex_unlock(&stream->chan->timer_lock); |
a9838785 | 691 | pthread_mutex_unlock(&stream->chan->lock); |
3bd1e081 | 692 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 693 | |
3bd1e081 MD |
694 | return ret; |
695 | } | |
696 | ||
5ab66908 MD |
697 | void consumer_del_data_stream(struct lttng_consumer_stream *stream) |
698 | { | |
699 | consumer_del_stream(stream, data_ht); | |
700 | } | |
701 | ||
00e2e675 | 702 | /* |
3f8e211f DG |
703 | * Add relayd socket to global consumer data hashtable. RCU read side lock MUST |
704 | * be acquired before calling this. | |
00e2e675 | 705 | */ |
d09e1200 | 706 | static int add_relayd(struct consumer_relayd_sock_pair *relayd) |
00e2e675 DG |
707 | { |
708 | int ret = 0; | |
d88aee68 | 709 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
710 | struct lttng_ht_iter iter; |
711 | ||
ffe60014 | 712 | assert(relayd); |
00e2e675 | 713 | |
00e2e675 | 714 | lttng_ht_lookup(consumer_data.relayd_ht, |
d88aee68 DG |
715 | &relayd->net_seq_idx, &iter); |
716 | node = lttng_ht_iter_get_node_u64(&iter); | |
00e2e675 | 717 | if (node != NULL) { |
00e2e675 DG |
718 | goto end; |
719 | } | |
d88aee68 | 720 | lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node); |
00e2e675 | 721 | |
00e2e675 DG |
722 | end: |
723 | return ret; | |
724 | } | |
725 | ||
726 | /* | |
727 | * Allocate and return a consumer relayd socket. | |
728 | */ | |
027a694f | 729 | static struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( |
da009f2c | 730 | uint64_t net_seq_idx) |
00e2e675 DG |
731 | { |
732 | struct consumer_relayd_sock_pair *obj = NULL; | |
733 | ||
da009f2c MD |
734 | /* net sequence index of -1 is a failure */ |
735 | if (net_seq_idx == (uint64_t) -1ULL) { | |
00e2e675 DG |
736 | goto error; |
737 | } | |
738 | ||
739 | obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); | |
740 | if (obj == NULL) { | |
741 | PERROR("zmalloc relayd sock"); | |
742 | goto error; | |
743 | } | |
744 | ||
745 | obj->net_seq_idx = net_seq_idx; | |
746 | obj->refcount = 0; | |
173af62f | 747 | obj->destroy_flag = 0; |
f96e4545 MD |
748 | obj->control_sock.sock.fd = -1; |
749 | obj->data_sock.sock.fd = -1; | |
d88aee68 | 750 | lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx); |
00e2e675 DG |
751 | pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); |
752 | ||
753 | error: | |
754 | return obj; | |
755 | } | |
756 | ||
757 | /* | |
758 | * Find a relayd socket pair in the global consumer data. | |
759 | * | |
760 | * Return the object if found else NULL. | |
b0b335c8 MD |
761 | * RCU read-side lock must be held across this call and while using the |
762 | * returned object. | |
00e2e675 | 763 | */ |
d88aee68 | 764 | struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key) |
00e2e675 DG |
765 | { |
766 | struct lttng_ht_iter iter; | |
d88aee68 | 767 | struct lttng_ht_node_u64 *node; |
00e2e675 DG |
768 | struct consumer_relayd_sock_pair *relayd = NULL; |
769 | ||
770 | /* Negative keys are lookup failures */ | |
d88aee68 | 771 | if (key == (uint64_t) -1ULL) { |
00e2e675 DG |
772 | goto error; |
773 | } | |
774 | ||
d88aee68 | 775 | lttng_ht_lookup(consumer_data.relayd_ht, &key, |
00e2e675 | 776 | &iter); |
d88aee68 | 777 | node = lttng_ht_iter_get_node_u64(&iter); |
00e2e675 DG |
778 | if (node != NULL) { |
779 | relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); | |
780 | } | |
781 | ||
00e2e675 DG |
782 | error: |
783 | return relayd; | |
784 | } | |
785 | ||
10a50311 JD |
786 | /* |
787 | * Find a relayd and send the stream | |
788 | * | |
789 | * Returns 0 on success, < 0 on error | |
790 | */ | |
791 | int consumer_send_relayd_stream(struct lttng_consumer_stream *stream, | |
ccb5d350 | 792 | char *path, enum lttng_domain_type domain) |
10a50311 JD |
793 | { |
794 | int ret = 0; | |
795 | struct consumer_relayd_sock_pair *relayd; | |
796 | ||
797 | assert(stream); | |
798 | assert(stream->net_seq_idx != -1ULL); | |
799 | assert(path); | |
800 | ||
801 | /* The stream is not metadata. Get relayd reference if exists. */ | |
802 | rcu_read_lock(); | |
803 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
804 | if (relayd != NULL) { | |
805 | /* Add stream on the relayd */ | |
806 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
807 | ret = relayd_add_stream(&relayd->control_sock, stream->name, | |
808 | path, &stream->relayd_stream_id, | |
ccb5d350 JD |
809 | stream->chan->tracefile_size, stream->chan->tracefile_count, |
810 | domain); | |
10a50311 JD |
811 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
812 | if (ret < 0) { | |
813 | goto end; | |
814 | } | |
1c20f0e2 | 815 | |
10a50311 | 816 | uatomic_inc(&relayd->refcount); |
d01178b6 | 817 | stream->sent_to_relayd = 1; |
10a50311 JD |
818 | } else { |
819 | ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.", | |
820 | stream->key, stream->net_seq_idx); | |
821 | ret = -1; | |
822 | goto end; | |
823 | } | |
824 | ||
825 | DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64, | |
826 | stream->name, stream->key, stream->net_seq_idx); | |
827 | ||
828 | end: | |
829 | rcu_read_unlock(); | |
830 | return ret; | |
831 | } | |
832 | ||
a4baae1b JD |
833 | /* |
834 | * Find a relayd and send the streams sent message | |
835 | * | |
836 | * Returns 0 on success, < 0 on error | |
837 | */ | |
838 | int consumer_send_relayd_streams_sent(uint64_t net_seq_idx) | |
839 | { | |
840 | int ret = 0; | |
841 | struct consumer_relayd_sock_pair *relayd; | |
842 | ||
843 | assert(net_seq_idx != -1ULL); | |
844 | ||
845 | /* The stream is not metadata. Get relayd reference if exists. */ | |
846 | rcu_read_lock(); | |
847 | relayd = consumer_find_relayd(net_seq_idx); | |
848 | if (relayd != NULL) { | |
849 | /* Add stream on the relayd */ | |
850 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
851 | ret = relayd_streams_sent(&relayd->control_sock); | |
852 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
853 | if (ret < 0) { | |
854 | goto end; | |
855 | } | |
856 | } else { | |
857 | ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.", | |
858 | net_seq_idx); | |
859 | ret = -1; | |
860 | goto end; | |
861 | } | |
862 | ||
863 | ret = 0; | |
864 | DBG("All streams sent relayd id %" PRIu64, net_seq_idx); | |
865 | ||
866 | end: | |
867 | rcu_read_unlock(); | |
868 | return ret; | |
869 | } | |
870 | ||
10a50311 JD |
871 | /* |
872 | * Find a relayd and close the stream | |
873 | */ | |
874 | void close_relayd_stream(struct lttng_consumer_stream *stream) | |
875 | { | |
876 | struct consumer_relayd_sock_pair *relayd; | |
877 | ||
878 | /* The stream is not metadata. Get relayd reference if exists. */ | |
879 | rcu_read_lock(); | |
880 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
881 | if (relayd) { | |
882 | consumer_stream_relayd_close(stream, relayd); | |
883 | } | |
884 | rcu_read_unlock(); | |
885 | } | |
886 | ||
00e2e675 DG |
887 | /* |
888 | * Handle stream for relayd transmission if the stream applies for network | |
889 | * streaming where the net sequence index is set. | |
890 | * | |
891 | * Return destination file descriptor or negative value on error. | |
892 | */ | |
6197aea7 | 893 | static int write_relayd_stream_header(struct lttng_consumer_stream *stream, |
1d4dfdef DG |
894 | size_t data_size, unsigned long padding, |
895 | struct consumer_relayd_sock_pair *relayd) | |
00e2e675 DG |
896 | { |
897 | int outfd = -1, ret; | |
00e2e675 DG |
898 | struct lttcomm_relayd_data_hdr data_hdr; |
899 | ||
900 | /* Safety net */ | |
901 | assert(stream); | |
6197aea7 | 902 | assert(relayd); |
00e2e675 DG |
903 | |
904 | /* Reset data header */ | |
905 | memset(&data_hdr, 0, sizeof(data_hdr)); | |
906 | ||
00e2e675 DG |
907 | if (stream->metadata_flag) { |
908 | /* Caller MUST acquire the relayd control socket lock */ | |
909 | ret = relayd_send_metadata(&relayd->control_sock, data_size); | |
910 | if (ret < 0) { | |
911 | goto error; | |
912 | } | |
913 | ||
914 | /* Metadata are always sent on the control socket. */ | |
6151a90f | 915 | outfd = relayd->control_sock.sock.fd; |
00e2e675 DG |
916 | } else { |
917 | /* Set header with stream information */ | |
918 | data_hdr.stream_id = htobe64(stream->relayd_stream_id); | |
919 | data_hdr.data_size = htobe32(data_size); | |
1d4dfdef | 920 | data_hdr.padding_size = htobe32(padding); |
39df6d9f DG |
921 | /* |
922 | * Note that net_seq_num below is assigned with the *current* value of | |
923 | * next_net_seq_num and only after that the next_net_seq_num will be | |
924 | * increment. This is why when issuing a command on the relayd using | |
925 | * this next value, 1 should always be substracted in order to compare | |
926 | * the last seen sequence number on the relayd side to the last sent. | |
927 | */ | |
3604f373 | 928 | data_hdr.net_seq_num = htobe64(stream->next_net_seq_num); |
00e2e675 DG |
929 | /* Other fields are zeroed previously */ |
930 | ||
931 | ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, | |
932 | sizeof(data_hdr)); | |
933 | if (ret < 0) { | |
934 | goto error; | |
935 | } | |
936 | ||
3604f373 DG |
937 | ++stream->next_net_seq_num; |
938 | ||
00e2e675 | 939 | /* Set to go on data socket */ |
6151a90f | 940 | outfd = relayd->data_sock.sock.fd; |
00e2e675 DG |
941 | } |
942 | ||
943 | error: | |
944 | return outfd; | |
945 | } | |
946 | ||
3bd1e081 | 947 | /* |
ffe60014 DG |
948 | * Allocate and return a new lttng_consumer_channel object using the given key |
949 | * to initialize the hash table node. | |
950 | * | |
951 | * On error, return NULL. | |
3bd1e081 | 952 | */ |
886224ff | 953 | struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, |
ffe60014 DG |
954 | uint64_t session_id, |
955 | const char *pathname, | |
956 | const char *name, | |
957 | uid_t uid, | |
958 | gid_t gid, | |
57a269f2 | 959 | uint64_t relayd_id, |
1624d5b7 JD |
960 | enum lttng_event_output output, |
961 | uint64_t tracefile_size, | |
2bba9e53 | 962 | uint64_t tracefile_count, |
1950109e | 963 | uint64_t session_id_per_pid, |
ecc48a90 | 964 | unsigned int monitor, |
d7ba1388 | 965 | unsigned int live_timer_interval, |
3d071855 | 966 | const char *root_shm_path, |
d7ba1388 | 967 | const char *shm_path) |
3bd1e081 MD |
968 | { |
969 | struct lttng_consumer_channel *channel; | |
3bd1e081 | 970 | |
276b26d1 | 971 | channel = zmalloc(sizeof(*channel)); |
3bd1e081 | 972 | if (channel == NULL) { |
7a57cf92 | 973 | PERROR("malloc struct lttng_consumer_channel"); |
3bd1e081 MD |
974 | goto end; |
975 | } | |
ffe60014 DG |
976 | |
977 | channel->key = key; | |
3bd1e081 | 978 | channel->refcount = 0; |
ffe60014 | 979 | channel->session_id = session_id; |
1950109e | 980 | channel->session_id_per_pid = session_id_per_pid; |
ffe60014 DG |
981 | channel->uid = uid; |
982 | channel->gid = gid; | |
983 | channel->relayd_id = relayd_id; | |
1624d5b7 JD |
984 | channel->tracefile_size = tracefile_size; |
985 | channel->tracefile_count = tracefile_count; | |
2bba9e53 | 986 | channel->monitor = monitor; |
ecc48a90 | 987 | channel->live_timer_interval = live_timer_interval; |
a9838785 | 988 | pthread_mutex_init(&channel->lock, NULL); |
ec6ea7d0 | 989 | pthread_mutex_init(&channel->timer_lock, NULL); |
ffe60014 | 990 | |
0c759fc9 DG |
991 | switch (output) { |
992 | case LTTNG_EVENT_SPLICE: | |
993 | channel->output = CONSUMER_CHANNEL_SPLICE; | |
994 | break; | |
995 | case LTTNG_EVENT_MMAP: | |
996 | channel->output = CONSUMER_CHANNEL_MMAP; | |
997 | break; | |
998 | default: | |
999 | assert(0); | |
1000 | free(channel); | |
1001 | channel = NULL; | |
1002 | goto end; | |
1003 | } | |
1004 | ||
07b86b52 JD |
1005 | /* |
1006 | * In monitor mode, the streams associated with the channel will be put in | |
1007 | * a special list ONLY owned by this channel. So, the refcount is set to 1 | |
1008 | * here meaning that the channel itself has streams that are referenced. | |
1009 | * | |
1010 | * On a channel deletion, once the channel is no longer visible, the | |
1011 | * refcount is decremented and checked for a zero value to delete it. With | |
1012 | * streams in no monitor mode, it will now be safe to destroy the channel. | |
1013 | */ | |
1014 | if (!channel->monitor) { | |
1015 | channel->refcount = 1; | |
1016 | } | |
1017 | ||
ffe60014 DG |
1018 | strncpy(channel->pathname, pathname, sizeof(channel->pathname)); |
1019 | channel->pathname[sizeof(channel->pathname) - 1] = '\0'; | |
1020 | ||
1021 | strncpy(channel->name, name, sizeof(channel->name)); | |
1022 | channel->name[sizeof(channel->name) - 1] = '\0'; | |
1023 | ||
3d071855 MD |
1024 | if (root_shm_path) { |
1025 | strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path)); | |
1026 | channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0'; | |
1027 | } | |
d7ba1388 MD |
1028 | if (shm_path) { |
1029 | strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path)); | |
1030 | channel->shm_path[sizeof(channel->shm_path) - 1] = '\0'; | |
1031 | } | |
1032 | ||
d88aee68 | 1033 | lttng_ht_node_init_u64(&channel->node, channel->key); |
d8ef542d MD |
1034 | |
1035 | channel->wait_fd = -1; | |
1036 | ||
ffe60014 DG |
1037 | CDS_INIT_LIST_HEAD(&channel->streams.head); |
1038 | ||
62a7b8ed | 1039 | DBG("Allocated channel (key %" PRIu64 ")", channel->key); |
3bd1e081 | 1040 | |
3bd1e081 MD |
1041 | end: |
1042 | return channel; | |
1043 | } | |
1044 | ||
1045 | /* | |
1046 | * Add a channel to the global list protected by a mutex. | |
821fffb2 | 1047 | * |
b5a6470f | 1048 | * Always return 0 indicating success. |
3bd1e081 | 1049 | */ |
d8ef542d MD |
1050 | int consumer_add_channel(struct lttng_consumer_channel *channel, |
1051 | struct lttng_consumer_local_data *ctx) | |
3bd1e081 | 1052 | { |
3bd1e081 | 1053 | pthread_mutex_lock(&consumer_data.lock); |
a9838785 | 1054 | pthread_mutex_lock(&channel->lock); |
ec6ea7d0 | 1055 | pthread_mutex_lock(&channel->timer_lock); |
c77fc10a | 1056 | |
b5a6470f DG |
1057 | /* |
1058 | * This gives us a guarantee that the channel we are about to add to the | |
1059 | * channel hash table will be unique. See this function comment on the why | |
1060 | * we need to steel the channel key at this stage. | |
1061 | */ | |
1062 | steal_channel_key(channel->key); | |
c77fc10a | 1063 | |
b5a6470f | 1064 | rcu_read_lock(); |
d88aee68 | 1065 | lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node); |
6065ceec | 1066 | rcu_read_unlock(); |
b5a6470f | 1067 | |
ec6ea7d0 | 1068 | pthread_mutex_unlock(&channel->timer_lock); |
a9838785 | 1069 | pthread_mutex_unlock(&channel->lock); |
3bd1e081 | 1070 | pthread_mutex_unlock(&consumer_data.lock); |
702b1ea4 | 1071 | |
b5a6470f | 1072 | if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) { |
a0cbdd2e | 1073 | notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD); |
d8ef542d | 1074 | } |
b5a6470f DG |
1075 | |
1076 | return 0; | |
3bd1e081 MD |
1077 | } |
1078 | ||
1079 | /* | |
1080 | * Allocate the pollfd structure and the local view of the out fds to avoid | |
1081 | * doing a lookup in the linked list and concurrency issues when writing is | |
1082 | * needed. Called with consumer_data.lock held. | |
1083 | * | |
1084 | * Returns the number of fds in the structures. | |
1085 | */ | |
ffe60014 DG |
1086 | static int update_poll_array(struct lttng_consumer_local_data *ctx, |
1087 | struct pollfd **pollfd, struct lttng_consumer_stream **local_stream, | |
1088 | struct lttng_ht *ht) | |
3bd1e081 | 1089 | { |
3bd1e081 | 1090 | int i = 0; |
e4421fec DG |
1091 | struct lttng_ht_iter iter; |
1092 | struct lttng_consumer_stream *stream; | |
3bd1e081 | 1093 | |
ffe60014 DG |
1094 | assert(ctx); |
1095 | assert(ht); | |
1096 | assert(pollfd); | |
1097 | assert(local_stream); | |
1098 | ||
3bd1e081 | 1099 | DBG("Updating poll fd array"); |
481d6c57 | 1100 | rcu_read_lock(); |
43c34bc3 | 1101 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
8994307f DG |
1102 | /* |
1103 | * Only active streams with an active end point can be added to the | |
1104 | * poll set and local stream storage of the thread. | |
1105 | * | |
1106 | * There is a potential race here for endpoint_status to be updated | |
1107 | * just after the check. However, this is OK since the stream(s) will | |
1108 | * be deleted once the thread is notified that the end point state has | |
1109 | * changed where this function will be called back again. | |
1110 | */ | |
1111 | if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM || | |
79d4ffb7 | 1112 | stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { |
3bd1e081 MD |
1113 | continue; |
1114 | } | |
7972aab2 DG |
1115 | /* |
1116 | * This clobbers way too much the debug output. Uncomment that if you | |
1117 | * need it for debugging purposes. | |
1118 | * | |
1119 | * DBG("Active FD %d", stream->wait_fd); | |
1120 | */ | |
e4421fec | 1121 | (*pollfd)[i].fd = stream->wait_fd; |
3bd1e081 | 1122 | (*pollfd)[i].events = POLLIN | POLLPRI; |
e4421fec | 1123 | local_stream[i] = stream; |
3bd1e081 MD |
1124 | i++; |
1125 | } | |
481d6c57 | 1126 | rcu_read_unlock(); |
3bd1e081 MD |
1127 | |
1128 | /* | |
50f8ae69 | 1129 | * Insert the consumer_data_pipe at the end of the array and don't |
3bd1e081 MD |
1130 | * increment i so nb_fd is the number of real FD. |
1131 | */ | |
acdb9057 | 1132 | (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe); |
509bb1cf | 1133 | (*pollfd)[i].events = POLLIN | POLLPRI; |
02b3d176 DG |
1134 | |
1135 | (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe); | |
1136 | (*pollfd)[i + 1].events = POLLIN | POLLPRI; | |
3bd1e081 MD |
1137 | return i; |
1138 | } | |
1139 | ||
1140 | /* | |
84382d49 MD |
1141 | * Poll on the should_quit pipe and the command socket return -1 on |
1142 | * error, 1 if should exit, 0 if data is available on the command socket | |
3bd1e081 MD |
1143 | */ |
1144 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) | |
1145 | { | |
1146 | int num_rdy; | |
1147 | ||
88f2b785 | 1148 | restart: |
3bd1e081 MD |
1149 | num_rdy = poll(consumer_sockpoll, 2, -1); |
1150 | if (num_rdy == -1) { | |
88f2b785 MD |
1151 | /* |
1152 | * Restart interrupted system call. | |
1153 | */ | |
1154 | if (errno == EINTR) { | |
1155 | goto restart; | |
1156 | } | |
7a57cf92 | 1157 | PERROR("Poll error"); |
84382d49 | 1158 | return -1; |
3bd1e081 | 1159 | } |
509bb1cf | 1160 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
3bd1e081 | 1161 | DBG("consumer_should_quit wake up"); |
84382d49 | 1162 | return 1; |
3bd1e081 MD |
1163 | } |
1164 | return 0; | |
3bd1e081 MD |
1165 | } |
1166 | ||
1167 | /* | |
1168 | * Set the error socket. | |
1169 | */ | |
ffe60014 DG |
1170 | void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx, |
1171 | int sock) | |
3bd1e081 MD |
1172 | { |
1173 | ctx->consumer_error_socket = sock; | |
1174 | } | |
1175 | ||
1176 | /* | |
1177 | * Set the command socket path. | |
1178 | */ | |
3bd1e081 MD |
1179 | void lttng_consumer_set_command_sock_path( |
1180 | struct lttng_consumer_local_data *ctx, char *sock) | |
1181 | { | |
1182 | ctx->consumer_command_sock_path = sock; | |
1183 | } | |
1184 | ||
1185 | /* | |
1186 | * Send return code to the session daemon. | |
1187 | * If the socket is not defined, we return 0, it is not a fatal error | |
1188 | */ | |
ffe60014 | 1189 | int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd) |
3bd1e081 MD |
1190 | { |
1191 | if (ctx->consumer_error_socket > 0) { | |
1192 | return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd, | |
1193 | sizeof(enum lttcomm_sessiond_command)); | |
1194 | } | |
1195 | ||
1196 | return 0; | |
1197 | } | |
1198 | ||
1199 | /* | |
228b5bf7 DG |
1200 | * Close all the tracefiles and stream fds and MUST be called when all |
1201 | * instances are destroyed i.e. when all threads were joined and are ended. | |
3bd1e081 MD |
1202 | */ |
1203 | void lttng_consumer_cleanup(void) | |
1204 | { | |
e4421fec | 1205 | struct lttng_ht_iter iter; |
ffe60014 | 1206 | struct lttng_consumer_channel *channel; |
6065ceec DG |
1207 | |
1208 | rcu_read_lock(); | |
3bd1e081 | 1209 | |
ffe60014 DG |
1210 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel, |
1211 | node.node) { | |
702b1ea4 | 1212 | consumer_del_channel(channel); |
3bd1e081 | 1213 | } |
6065ceec DG |
1214 | |
1215 | rcu_read_unlock(); | |
d6ce1df2 | 1216 | |
d6ce1df2 | 1217 | lttng_ht_destroy(consumer_data.channel_ht); |
228b5bf7 DG |
1218 | |
1219 | cleanup_relayd_ht(); | |
1220 | ||
d8ef542d MD |
1221 | lttng_ht_destroy(consumer_data.stream_per_chan_id_ht); |
1222 | ||
228b5bf7 DG |
1223 | /* |
1224 | * This HT contains streams that are freed by either the metadata thread or | |
1225 | * the data thread so we do *nothing* on the hash table and simply destroy | |
1226 | * it. | |
1227 | */ | |
1228 | lttng_ht_destroy(consumer_data.stream_list_ht); | |
3bd1e081 MD |
1229 | } |
1230 | ||
1231 | /* | |
1232 | * Called from signal handler. | |
1233 | */ | |
1234 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) | |
1235 | { | |
6cd525e8 MD |
1236 | ssize_t ret; |
1237 | ||
10211f5c | 1238 | CMM_STORE_SHARED(consumer_quit, 1); |
6cd525e8 MD |
1239 | ret = lttng_write(ctx->consumer_should_quit[1], "4", 1); |
1240 | if (ret < 1) { | |
7a57cf92 | 1241 | PERROR("write consumer quit"); |
3bd1e081 | 1242 | } |
ab1027f4 DG |
1243 | |
1244 | DBG("Consumer flag that it should quit"); | |
3bd1e081 MD |
1245 | } |
1246 | ||
5199ffc4 JG |
1247 | |
1248 | /* | |
1249 | * Flush pending writes to trace output disk file. | |
1250 | */ | |
1251 | static | |
00e2e675 DG |
1252 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
1253 | off_t orig_offset) | |
3bd1e081 | 1254 | { |
c7a78aab | 1255 | int ret; |
3bd1e081 MD |
1256 | int outfd = stream->out_fd; |
1257 | ||
1258 | /* | |
1259 | * This does a blocking write-and-wait on any page that belongs to the | |
1260 | * subbuffer prior to the one we just wrote. | |
1261 | * Don't care about error values, as these are just hints and ways to | |
1262 | * limit the amount of page cache used. | |
1263 | */ | |
ffe60014 | 1264 | if (orig_offset < stream->max_sb_size) { |
3bd1e081 MD |
1265 | return; |
1266 | } | |
ffe60014 DG |
1267 | lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size, |
1268 | stream->max_sb_size, | |
3bd1e081 MD |
1269 | SYNC_FILE_RANGE_WAIT_BEFORE |
1270 | | SYNC_FILE_RANGE_WRITE | |
1271 | | SYNC_FILE_RANGE_WAIT_AFTER); | |
1272 | /* | |
1273 | * Give hints to the kernel about how we access the file: | |
1274 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after | |
1275 | * we write it. | |
1276 | * | |
1277 | * We need to call fadvise again after the file grows because the | |
1278 | * kernel does not seem to apply fadvise to non-existing parts of the | |
1279 | * file. | |
1280 | * | |
1281 | * Call fadvise _after_ having waited for the page writeback to | |
1282 | * complete because the dirty page writeback semantic is not well | |
1283 | * defined. So it can be expected to lead to lower throughput in | |
1284 | * streaming. | |
1285 | */ | |
c7a78aab | 1286 | ret = posix_fadvise(outfd, orig_offset - stream->max_sb_size, |
ffe60014 | 1287 | stream->max_sb_size, POSIX_FADV_DONTNEED); |
a0d0e127 | 1288 | if (ret && ret != -ENOSYS) { |
a74a5f4a JG |
1289 | errno = ret; |
1290 | PERROR("posix_fadvise on fd %i", outfd); | |
c7a78aab | 1291 | } |
3bd1e081 MD |
1292 | } |
1293 | ||
1294 | /* | |
1295 | * Initialise the necessary environnement : | |
1296 | * - create a new context | |
1297 | * - create the poll_pipe | |
1298 | * - create the should_quit pipe (for signal handler) | |
1299 | * - create the thread pipe (for splice) | |
1300 | * | |
1301 | * Takes a function pointer as argument, this function is called when data is | |
1302 | * available on a buffer. This function is responsible to do the | |
1303 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the | |
1304 | * buffer configuration and then kernctl_put_next_subbuf at the end. | |
1305 | * | |
1306 | * Returns a pointer to the new context or NULL on error. | |
1307 | */ | |
1308 | struct lttng_consumer_local_data *lttng_consumer_create( | |
1309 | enum lttng_consumer_type type, | |
4078b776 | 1310 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
d41f73b7 | 1311 | struct lttng_consumer_local_data *ctx), |
3bd1e081 MD |
1312 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
1313 | int (*recv_stream)(struct lttng_consumer_stream *stream), | |
30319bcb | 1314 | int (*update_stream)(uint64_t stream_key, uint32_t state)) |
3bd1e081 | 1315 | { |
d8ef542d | 1316 | int ret; |
3bd1e081 MD |
1317 | struct lttng_consumer_local_data *ctx; |
1318 | ||
1319 | assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || | |
1320 | consumer_data.type == type); | |
1321 | consumer_data.type = type; | |
1322 | ||
effcf122 | 1323 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
3bd1e081 | 1324 | if (ctx == NULL) { |
7a57cf92 | 1325 | PERROR("allocating context"); |
3bd1e081 MD |
1326 | goto error; |
1327 | } | |
1328 | ||
1329 | ctx->consumer_error_socket = -1; | |
331744e3 | 1330 | ctx->consumer_metadata_socket = -1; |
75d83e50 | 1331 | pthread_mutex_init(&ctx->metadata_socket_lock, NULL); |
3bd1e081 MD |
1332 | /* assign the callbacks */ |
1333 | ctx->on_buffer_ready = buffer_ready; | |
1334 | ctx->on_recv_channel = recv_channel; | |
1335 | ctx->on_recv_stream = recv_stream; | |
1336 | ctx->on_update_stream = update_stream; | |
1337 | ||
acdb9057 DG |
1338 | ctx->consumer_data_pipe = lttng_pipe_open(0); |
1339 | if (!ctx->consumer_data_pipe) { | |
3bd1e081 MD |
1340 | goto error_poll_pipe; |
1341 | } | |
1342 | ||
02b3d176 DG |
1343 | ctx->consumer_wakeup_pipe = lttng_pipe_open(0); |
1344 | if (!ctx->consumer_wakeup_pipe) { | |
1345 | goto error_wakeup_pipe; | |
1346 | } | |
1347 | ||
3bd1e081 MD |
1348 | ret = pipe(ctx->consumer_should_quit); |
1349 | if (ret < 0) { | |
7a57cf92 | 1350 | PERROR("Error creating recv pipe"); |
3bd1e081 MD |
1351 | goto error_quit_pipe; |
1352 | } | |
1353 | ||
d8ef542d MD |
1354 | ret = pipe(ctx->consumer_channel_pipe); |
1355 | if (ret < 0) { | |
1356 | PERROR("Error creating channel pipe"); | |
1357 | goto error_channel_pipe; | |
1358 | } | |
1359 | ||
13886d2d DG |
1360 | ctx->consumer_metadata_pipe = lttng_pipe_open(0); |
1361 | if (!ctx->consumer_metadata_pipe) { | |
fb3a43a9 DG |
1362 | goto error_metadata_pipe; |
1363 | } | |
3bd1e081 | 1364 | |
e9404c27 JG |
1365 | ctx->channel_monitor_pipe = -1; |
1366 | ||
fb3a43a9 | 1367 | return ctx; |
3bd1e081 | 1368 | |
fb3a43a9 | 1369 | error_metadata_pipe: |
d8ef542d MD |
1370 | utils_close_pipe(ctx->consumer_channel_pipe); |
1371 | error_channel_pipe: | |
d8ef542d | 1372 | utils_close_pipe(ctx->consumer_should_quit); |
3bd1e081 | 1373 | error_quit_pipe: |
02b3d176 DG |
1374 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
1375 | error_wakeup_pipe: | |
acdb9057 | 1376 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
3bd1e081 MD |
1377 | error_poll_pipe: |
1378 | free(ctx); | |
1379 | error: | |
1380 | return NULL; | |
1381 | } | |
1382 | ||
282dadbc MD |
1383 | /* |
1384 | * Iterate over all streams of the hashtable and free them properly. | |
1385 | */ | |
1386 | static void destroy_data_stream_ht(struct lttng_ht *ht) | |
1387 | { | |
1388 | struct lttng_ht_iter iter; | |
1389 | struct lttng_consumer_stream *stream; | |
1390 | ||
1391 | if (ht == NULL) { | |
1392 | return; | |
1393 | } | |
1394 | ||
1395 | rcu_read_lock(); | |
1396 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1397 | /* | |
1398 | * Ignore return value since we are currently cleaning up so any error | |
1399 | * can't be handled. | |
1400 | */ | |
1401 | (void) consumer_del_stream(stream, ht); | |
1402 | } | |
1403 | rcu_read_unlock(); | |
1404 | ||
1405 | lttng_ht_destroy(ht); | |
1406 | } | |
1407 | ||
1408 | /* | |
1409 | * Iterate over all streams of the metadata hashtable and free them | |
1410 | * properly. | |
1411 | */ | |
1412 | static void destroy_metadata_stream_ht(struct lttng_ht *ht) | |
1413 | { | |
1414 | struct lttng_ht_iter iter; | |
1415 | struct lttng_consumer_stream *stream; | |
1416 | ||
1417 | if (ht == NULL) { | |
1418 | return; | |
1419 | } | |
1420 | ||
1421 | rcu_read_lock(); | |
1422 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { | |
1423 | /* | |
1424 | * Ignore return value since we are currently cleaning up so any error | |
1425 | * can't be handled. | |
1426 | */ | |
1427 | (void) consumer_del_metadata_stream(stream, ht); | |
1428 | } | |
1429 | rcu_read_unlock(); | |
1430 | ||
1431 | lttng_ht_destroy(ht); | |
1432 | } | |
1433 | ||
3bd1e081 MD |
1434 | /* |
1435 | * Close all fds associated with the instance and free the context. | |
1436 | */ | |
1437 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) | |
1438 | { | |
4c462e79 MD |
1439 | int ret; |
1440 | ||
ab1027f4 DG |
1441 | DBG("Consumer destroying it. Closing everything."); |
1442 | ||
4f2e75b9 DG |
1443 | if (!ctx) { |
1444 | return; | |
1445 | } | |
1446 | ||
282dadbc MD |
1447 | destroy_data_stream_ht(data_ht); |
1448 | destroy_metadata_stream_ht(metadata_ht); | |
1449 | ||
4c462e79 MD |
1450 | ret = close(ctx->consumer_error_socket); |
1451 | if (ret) { | |
1452 | PERROR("close"); | |
1453 | } | |
331744e3 JD |
1454 | ret = close(ctx->consumer_metadata_socket); |
1455 | if (ret) { | |
1456 | PERROR("close"); | |
1457 | } | |
d8ef542d | 1458 | utils_close_pipe(ctx->consumer_channel_pipe); |
acdb9057 | 1459 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
13886d2d | 1460 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
02b3d176 | 1461 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
d8ef542d | 1462 | utils_close_pipe(ctx->consumer_should_quit); |
fb3a43a9 | 1463 | |
3bd1e081 MD |
1464 | unlink(ctx->consumer_command_sock_path); |
1465 | free(ctx); | |
1466 | } | |
1467 | ||
6197aea7 DG |
1468 | /* |
1469 | * Write the metadata stream id on the specified file descriptor. | |
1470 | */ | |
1471 | static int write_relayd_metadata_id(int fd, | |
1472 | struct lttng_consumer_stream *stream, | |
ffe60014 | 1473 | struct consumer_relayd_sock_pair *relayd, unsigned long padding) |
6197aea7 | 1474 | { |
6cd525e8 | 1475 | ssize_t ret; |
1d4dfdef | 1476 | struct lttcomm_relayd_metadata_payload hdr; |
6197aea7 | 1477 | |
1d4dfdef DG |
1478 | hdr.stream_id = htobe64(stream->relayd_stream_id); |
1479 | hdr.padding_size = htobe32(padding); | |
6cd525e8 MD |
1480 | ret = lttng_write(fd, (void *) &hdr, sizeof(hdr)); |
1481 | if (ret < sizeof(hdr)) { | |
d7b75ec8 | 1482 | /* |
6f04ed72 | 1483 | * This error means that the fd's end is closed so ignore the PERROR |
d7b75ec8 DG |
1484 | * not to clubber the error output since this can happen in a normal |
1485 | * code path. | |
1486 | */ | |
1487 | if (errno != EPIPE) { | |
1488 | PERROR("write metadata stream id"); | |
1489 | } | |
1490 | DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno); | |
534d2592 DG |
1491 | /* |
1492 | * Set ret to a negative value because if ret != sizeof(hdr), we don't | |
1493 | * handle writting the missing part so report that as an error and | |
1494 | * don't lie to the caller. | |
1495 | */ | |
1496 | ret = -1; | |
6197aea7 DG |
1497 | goto end; |
1498 | } | |
1d4dfdef DG |
1499 | DBG("Metadata stream id %" PRIu64 " with padding %lu written before data", |
1500 | stream->relayd_stream_id, padding); | |
6197aea7 DG |
1501 | |
1502 | end: | |
6cd525e8 | 1503 | return (int) ret; |
6197aea7 DG |
1504 | } |
1505 | ||
3bd1e081 | 1506 | /* |
09e26845 DG |
1507 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
1508 | * core function for writing trace buffers to either the local filesystem or | |
1509 | * the network. | |
1510 | * | |
79d4ffb7 DG |
1511 | * It must be called with the stream lock held. |
1512 | * | |
09e26845 | 1513 | * Careful review MUST be put if any changes occur! |
3bd1e081 MD |
1514 | * |
1515 | * Returns the number of bytes written | |
1516 | */ | |
4078b776 | 1517 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
3bd1e081 | 1518 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1519 | struct lttng_consumer_stream *stream, unsigned long len, |
309167d2 | 1520 | unsigned long padding, |
50adc264 | 1521 | struct ctf_packet_index *index) |
3bd1e081 | 1522 | { |
f02e1e8a | 1523 | unsigned long mmap_offset; |
ffe60014 | 1524 | void *mmap_base; |
994ab360 | 1525 | ssize_t ret = 0; |
f02e1e8a DG |
1526 | off_t orig_offset = stream->out_fd_offset; |
1527 | /* Default is on the disk */ | |
1528 | int outfd = stream->out_fd; | |
f02e1e8a | 1529 | struct consumer_relayd_sock_pair *relayd = NULL; |
8994307f | 1530 | unsigned int relayd_hang_up = 0; |
f02e1e8a DG |
1531 | |
1532 | /* RCU lock for the relayd pointer */ | |
1533 | rcu_read_lock(); | |
1534 | ||
1535 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1536 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1537 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1538 | if (relayd == NULL) { | |
56591bac | 1539 | ret = -EPIPE; |
f02e1e8a DG |
1540 | goto end; |
1541 | } | |
1542 | } | |
1543 | ||
1544 | /* get the offset inside the fd to mmap */ | |
3bd1e081 MD |
1545 | switch (consumer_data.type) { |
1546 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1547 | mmap_base = stream->mmap_base; |
f02e1e8a | 1548 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); |
994ab360 | 1549 | if (ret < 0) { |
56591bac | 1550 | PERROR("tracer ctl get_mmap_read_offset"); |
56591bac MD |
1551 | goto end; |
1552 | } | |
f02e1e8a | 1553 | break; |
7753dea8 MD |
1554 | case LTTNG_CONSUMER32_UST: |
1555 | case LTTNG_CONSUMER64_UST: | |
ffe60014 DG |
1556 | mmap_base = lttng_ustctl_get_mmap_base(stream); |
1557 | if (!mmap_base) { | |
1558 | ERR("read mmap get mmap base for stream %s", stream->name); | |
994ab360 | 1559 | ret = -EPERM; |
ffe60014 DG |
1560 | goto end; |
1561 | } | |
1562 | ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset); | |
56591bac MD |
1563 | if (ret != 0) { |
1564 | PERROR("tracer ctl get_mmap_read_offset"); | |
994ab360 | 1565 | ret = -EINVAL; |
56591bac MD |
1566 | goto end; |
1567 | } | |
f02e1e8a | 1568 | break; |
3bd1e081 MD |
1569 | default: |
1570 | ERR("Unknown consumer_data type"); | |
1571 | assert(0); | |
1572 | } | |
b9182dd9 | 1573 | |
f02e1e8a DG |
1574 | /* Handle stream on the relayd if the output is on the network */ |
1575 | if (relayd) { | |
1576 | unsigned long netlen = len; | |
1577 | ||
1578 | /* | |
1579 | * Lock the control socket for the complete duration of the function | |
1580 | * since from this point on we will use the socket. | |
1581 | */ | |
1582 | if (stream->metadata_flag) { | |
1583 | /* Metadata requires the control socket. */ | |
1584 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
93ec662e JD |
1585 | if (stream->reset_metadata_flag) { |
1586 | ret = relayd_reset_metadata(&relayd->control_sock, | |
1587 | stream->relayd_stream_id, | |
1588 | stream->metadata_version); | |
1589 | if (ret < 0) { | |
1590 | relayd_hang_up = 1; | |
1591 | goto write_error; | |
1592 | } | |
1593 | stream->reset_metadata_flag = 0; | |
1594 | } | |
1d4dfdef | 1595 | netlen += sizeof(struct lttcomm_relayd_metadata_payload); |
f02e1e8a DG |
1596 | } |
1597 | ||
1d4dfdef | 1598 | ret = write_relayd_stream_header(stream, netlen, padding, relayd); |
994ab360 DG |
1599 | if (ret < 0) { |
1600 | relayd_hang_up = 1; | |
1601 | goto write_error; | |
1602 | } | |
1603 | /* Use the returned socket. */ | |
1604 | outfd = ret; | |
f02e1e8a | 1605 | |
994ab360 DG |
1606 | /* Write metadata stream id before payload */ |
1607 | if (stream->metadata_flag) { | |
1608 | ret = write_relayd_metadata_id(outfd, stream, relayd, padding); | |
1609 | if (ret < 0) { | |
8994307f DG |
1610 | relayd_hang_up = 1; |
1611 | goto write_error; | |
1612 | } | |
f02e1e8a | 1613 | } |
1d4dfdef DG |
1614 | } else { |
1615 | /* No streaming, we have to set the len with the full padding */ | |
1616 | len += padding; | |
1624d5b7 | 1617 | |
93ec662e JD |
1618 | if (stream->metadata_flag && stream->reset_metadata_flag) { |
1619 | ret = utils_truncate_stream_file(stream->out_fd, 0); | |
1620 | if (ret < 0) { | |
1621 | ERR("Reset metadata file"); | |
1622 | goto end; | |
1623 | } | |
1624 | stream->reset_metadata_flag = 0; | |
1625 | } | |
1626 | ||
1624d5b7 JD |
1627 | /* |
1628 | * Check if we need to change the tracefile before writing the packet. | |
1629 | */ | |
1630 | if (stream->chan->tracefile_size > 0 && | |
1631 | (stream->tracefile_size_current + len) > | |
1632 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1633 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1634 | stream->name, stream->chan->tracefile_size, | |
1635 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
309167d2 JD |
1636 | stream->out_fd, &(stream->tracefile_count_current), |
1637 | &stream->out_fd); | |
1624d5b7 JD |
1638 | if (ret < 0) { |
1639 | ERR("Rotating output file"); | |
1640 | goto end; | |
1641 | } | |
309167d2 JD |
1642 | outfd = stream->out_fd; |
1643 | ||
f8f3885c MD |
1644 | if (stream->index_file) { |
1645 | lttng_index_file_put(stream->index_file); | |
1646 | stream->index_file = lttng_index_file_create(stream->chan->pathname, | |
309167d2 JD |
1647 | stream->name, stream->uid, stream->gid, |
1648 | stream->chan->tracefile_size, | |
f8f3885c MD |
1649 | stream->tracefile_count_current, |
1650 | CTF_INDEX_MAJOR, CTF_INDEX_MINOR); | |
1651 | if (!stream->index_file) { | |
309167d2 JD |
1652 | goto end; |
1653 | } | |
309167d2 JD |
1654 | } |
1655 | ||
a6976990 DG |
1656 | /* Reset current size because we just perform a rotation. */ |
1657 | stream->tracefile_size_current = 0; | |
a1ae300f JD |
1658 | stream->out_fd_offset = 0; |
1659 | orig_offset = 0; | |
1624d5b7 JD |
1660 | } |
1661 | stream->tracefile_size_current += len; | |
309167d2 JD |
1662 | if (index) { |
1663 | index->offset = htobe64(stream->out_fd_offset); | |
1664 | } | |
f02e1e8a DG |
1665 | } |
1666 | ||
d02b8372 DG |
1667 | /* |
1668 | * This call guarantee that len or less is returned. It's impossible to | |
1669 | * receive a ret value that is bigger than len. | |
1670 | */ | |
1671 | ret = lttng_write(outfd, mmap_base + mmap_offset, len); | |
1672 | DBG("Consumer mmap write() ret %zd (len %lu)", ret, len); | |
1673 | if (ret < 0 || ((size_t) ret != len)) { | |
1674 | /* | |
1675 | * Report error to caller if nothing was written else at least send the | |
1676 | * amount written. | |
1677 | */ | |
1678 | if (ret < 0) { | |
994ab360 | 1679 | ret = -errno; |
f02e1e8a | 1680 | } |
994ab360 | 1681 | relayd_hang_up = 1; |
f02e1e8a | 1682 | |
d02b8372 | 1683 | /* Socket operation failed. We consider the relayd dead */ |
994ab360 | 1684 | if (errno == EPIPE || errno == EINVAL || errno == EBADF) { |
d02b8372 DG |
1685 | /* |
1686 | * This is possible if the fd is closed on the other side | |
1687 | * (outfd) or any write problem. It can be verbose a bit for a | |
1688 | * normal execution if for instance the relayd is stopped | |
1689 | * abruptly. This can happen so set this to a DBG statement. | |
1690 | */ | |
1691 | DBG("Consumer mmap write detected relayd hang up"); | |
994ab360 DG |
1692 | } else { |
1693 | /* Unhandled error, print it and stop function right now. */ | |
1694 | PERROR("Error in write mmap (ret %zd != len %lu)", ret, len); | |
f02e1e8a | 1695 | } |
994ab360 | 1696 | goto write_error; |
d02b8372 DG |
1697 | } |
1698 | stream->output_written += ret; | |
d02b8372 DG |
1699 | |
1700 | /* This call is useless on a socket so better save a syscall. */ | |
1701 | if (!relayd) { | |
1702 | /* This won't block, but will start writeout asynchronously */ | |
1703 | lttng_sync_file_range(outfd, stream->out_fd_offset, len, | |
1704 | SYNC_FILE_RANGE_WRITE); | |
1705 | stream->out_fd_offset += len; | |
f5dbe415 | 1706 | lttng_consumer_sync_trace_file(stream, orig_offset); |
f02e1e8a | 1707 | } |
f02e1e8a | 1708 | |
8994307f DG |
1709 | write_error: |
1710 | /* | |
1711 | * This is a special case that the relayd has closed its socket. Let's | |
1712 | * cleanup the relayd object and all associated streams. | |
1713 | */ | |
1714 | if (relayd && relayd_hang_up) { | |
1715 | cleanup_relayd(relayd, ctx); | |
1716 | } | |
1717 | ||
f02e1e8a DG |
1718 | end: |
1719 | /* Unlock only if ctrl socket used */ | |
1720 | if (relayd && stream->metadata_flag) { | |
1721 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1722 | } | |
1723 | ||
1724 | rcu_read_unlock(); | |
994ab360 | 1725 | return ret; |
3bd1e081 MD |
1726 | } |
1727 | ||
1728 | /* | |
1729 | * Splice the data from the ring buffer to the tracefile. | |
1730 | * | |
79d4ffb7 DG |
1731 | * It must be called with the stream lock held. |
1732 | * | |
3bd1e081 MD |
1733 | * Returns the number of bytes spliced. |
1734 | */ | |
4078b776 | 1735 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
3bd1e081 | 1736 | struct lttng_consumer_local_data *ctx, |
1d4dfdef | 1737 | struct lttng_consumer_stream *stream, unsigned long len, |
309167d2 | 1738 | unsigned long padding, |
50adc264 | 1739 | struct ctf_packet_index *index) |
3bd1e081 | 1740 | { |
f02e1e8a DG |
1741 | ssize_t ret = 0, written = 0, ret_splice = 0; |
1742 | loff_t offset = 0; | |
1743 | off_t orig_offset = stream->out_fd_offset; | |
1744 | int fd = stream->wait_fd; | |
1745 | /* Default is on the disk */ | |
1746 | int outfd = stream->out_fd; | |
f02e1e8a | 1747 | struct consumer_relayd_sock_pair *relayd = NULL; |
fb3a43a9 | 1748 | int *splice_pipe; |
8994307f | 1749 | unsigned int relayd_hang_up = 0; |
f02e1e8a | 1750 | |
3bd1e081 MD |
1751 | switch (consumer_data.type) { |
1752 | case LTTNG_CONSUMER_KERNEL: | |
f02e1e8a | 1753 | break; |
7753dea8 MD |
1754 | case LTTNG_CONSUMER32_UST: |
1755 | case LTTNG_CONSUMER64_UST: | |
f02e1e8a | 1756 | /* Not supported for user space tracing */ |
3bd1e081 MD |
1757 | return -ENOSYS; |
1758 | default: | |
1759 | ERR("Unknown consumer_data type"); | |
1760 | assert(0); | |
3bd1e081 MD |
1761 | } |
1762 | ||
f02e1e8a DG |
1763 | /* RCU lock for the relayd pointer */ |
1764 | rcu_read_lock(); | |
1765 | ||
1766 | /* Flag that the current stream if set for network streaming. */ | |
da009f2c | 1767 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
f02e1e8a DG |
1768 | relayd = consumer_find_relayd(stream->net_seq_idx); |
1769 | if (relayd == NULL) { | |
ad0b0d23 | 1770 | written = -ret; |
f02e1e8a DG |
1771 | goto end; |
1772 | } | |
1773 | } | |
a2361a61 | 1774 | splice_pipe = stream->splice_pipe; |
fb3a43a9 | 1775 | |
f02e1e8a | 1776 | /* Write metadata stream id before payload */ |
1d4dfdef | 1777 | if (relayd) { |
ad0b0d23 | 1778 | unsigned long total_len = len; |
f02e1e8a | 1779 | |
1d4dfdef DG |
1780 | if (stream->metadata_flag) { |
1781 | /* | |
1782 | * Lock the control socket for the complete duration of the function | |
1783 | * since from this point on we will use the socket. | |
1784 | */ | |
1785 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
1786 | ||
93ec662e JD |
1787 | if (stream->reset_metadata_flag) { |
1788 | ret = relayd_reset_metadata(&relayd->control_sock, | |
1789 | stream->relayd_stream_id, | |
1790 | stream->metadata_version); | |
1791 | if (ret < 0) { | |
1792 | relayd_hang_up = 1; | |
1793 | goto write_error; | |
1794 | } | |
1795 | stream->reset_metadata_flag = 0; | |
1796 | } | |
1d4dfdef DG |
1797 | ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd, |
1798 | padding); | |
1799 | if (ret < 0) { | |
1800 | written = ret; | |
ad0b0d23 DG |
1801 | relayd_hang_up = 1; |
1802 | goto write_error; | |
1d4dfdef DG |
1803 | } |
1804 | ||
1805 | total_len += sizeof(struct lttcomm_relayd_metadata_payload); | |
1806 | } | |
1807 | ||
1808 | ret = write_relayd_stream_header(stream, total_len, padding, relayd); | |
ad0b0d23 DG |
1809 | if (ret < 0) { |
1810 | written = ret; | |
1811 | relayd_hang_up = 1; | |
1812 | goto write_error; | |
f02e1e8a | 1813 | } |
ad0b0d23 DG |
1814 | /* Use the returned socket. */ |
1815 | outfd = ret; | |
1d4dfdef DG |
1816 | } else { |
1817 | /* No streaming, we have to set the len with the full padding */ | |
1818 | len += padding; | |
1624d5b7 | 1819 | |
93ec662e JD |
1820 | if (stream->metadata_flag && stream->reset_metadata_flag) { |
1821 | ret = utils_truncate_stream_file(stream->out_fd, 0); | |
1822 | if (ret < 0) { | |
1823 | ERR("Reset metadata file"); | |
1824 | goto end; | |
1825 | } | |
1826 | stream->reset_metadata_flag = 0; | |
1827 | } | |
1624d5b7 JD |
1828 | /* |
1829 | * Check if we need to change the tracefile before writing the packet. | |
1830 | */ | |
1831 | if (stream->chan->tracefile_size > 0 && | |
1832 | (stream->tracefile_size_current + len) > | |
1833 | stream->chan->tracefile_size) { | |
fe4477ee JD |
1834 | ret = utils_rotate_stream_file(stream->chan->pathname, |
1835 | stream->name, stream->chan->tracefile_size, | |
1836 | stream->chan->tracefile_count, stream->uid, stream->gid, | |
309167d2 JD |
1837 | stream->out_fd, &(stream->tracefile_count_current), |
1838 | &stream->out_fd); | |
1624d5b7 | 1839 | if (ret < 0) { |
ad0b0d23 | 1840 | written = ret; |
1624d5b7 JD |
1841 | ERR("Rotating output file"); |
1842 | goto end; | |
1843 | } | |
309167d2 JD |
1844 | outfd = stream->out_fd; |
1845 | ||
f8f3885c MD |
1846 | if (stream->index_file) { |
1847 | lttng_index_file_put(stream->index_file); | |
1848 | stream->index_file = lttng_index_file_create(stream->chan->pathname, | |
309167d2 JD |
1849 | stream->name, stream->uid, stream->gid, |
1850 | stream->chan->tracefile_size, | |
f8f3885c MD |
1851 | stream->tracefile_count_current, |
1852 | CTF_INDEX_MAJOR, CTF_INDEX_MINOR); | |
1853 | if (!stream->index_file) { | |
309167d2 JD |
1854 | goto end; |
1855 | } | |
309167d2 JD |
1856 | } |
1857 | ||
a6976990 DG |
1858 | /* Reset current size because we just perform a rotation. */ |
1859 | stream->tracefile_size_current = 0; | |
a1ae300f JD |
1860 | stream->out_fd_offset = 0; |
1861 | orig_offset = 0; | |
1624d5b7 JD |
1862 | } |
1863 | stream->tracefile_size_current += len; | |
309167d2 | 1864 | index->offset = htobe64(stream->out_fd_offset); |
f02e1e8a DG |
1865 | } |
1866 | ||
1867 | while (len > 0) { | |
1d4dfdef DG |
1868 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)", |
1869 | (unsigned long)offset, len, fd, splice_pipe[1]); | |
fb3a43a9 | 1870 | ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len, |
f02e1e8a DG |
1871 | SPLICE_F_MOVE | SPLICE_F_MORE); |
1872 | DBG("splice chan to pipe, ret %zd", ret_splice); | |
1873 | if (ret_splice < 0) { | |
d02b8372 | 1874 | ret = errno; |
ad0b0d23 | 1875 | written = -ret; |
d02b8372 | 1876 | PERROR("Error in relay splice"); |
f02e1e8a DG |
1877 | goto splice_error; |
1878 | } | |
1879 | ||
1880 | /* Handle stream on the relayd if the output is on the network */ | |
ad0b0d23 DG |
1881 | if (relayd && stream->metadata_flag) { |
1882 | size_t metadata_payload_size = | |
1883 | sizeof(struct lttcomm_relayd_metadata_payload); | |
1884 | ||
1885 | /* Update counter to fit the spliced data */ | |
1886 | ret_splice += metadata_payload_size; | |
1887 | len += metadata_payload_size; | |
1888 | /* | |
1889 | * We do this so the return value can match the len passed as | |
1890 | * argument to this function. | |
1891 | */ | |
1892 | written -= metadata_payload_size; | |
f02e1e8a DG |
1893 | } |
1894 | ||
1895 | /* Splice data out */ | |
fb3a43a9 | 1896 | ret_splice = splice(splice_pipe[0], NULL, outfd, NULL, |
f02e1e8a | 1897 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); |
a2361a61 JD |
1898 | DBG("Consumer splice pipe to file (out_fd: %d), ret %zd", |
1899 | outfd, ret_splice); | |
f02e1e8a | 1900 | if (ret_splice < 0) { |
d02b8372 | 1901 | ret = errno; |
ad0b0d23 DG |
1902 | written = -ret; |
1903 | relayd_hang_up = 1; | |
1904 | goto write_error; | |
f02e1e8a | 1905 | } else if (ret_splice > len) { |
d02b8372 DG |
1906 | /* |
1907 | * We don't expect this code path to be executed but you never know | |
1908 | * so this is an extra protection agains a buggy splice(). | |
1909 | */ | |
f02e1e8a | 1910 | ret = errno; |
ad0b0d23 | 1911 | written += ret_splice; |
d02b8372 DG |
1912 | PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice, |
1913 | len); | |
f02e1e8a | 1914 | goto splice_error; |
d02b8372 DG |
1915 | } else { |
1916 | /* All good, update current len and continue. */ | |
1917 | len -= ret_splice; | |
f02e1e8a | 1918 | } |
f02e1e8a DG |
1919 | |
1920 | /* This call is useless on a socket so better save a syscall. */ | |
1921 | if (!relayd) { | |
1922 | /* This won't block, but will start writeout asynchronously */ | |
1923 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, | |
1924 | SYNC_FILE_RANGE_WRITE); | |
1925 | stream->out_fd_offset += ret_splice; | |
1926 | } | |
e5d1a9b3 | 1927 | stream->output_written += ret_splice; |
f02e1e8a DG |
1928 | written += ret_splice; |
1929 | } | |
f5dbe415 JG |
1930 | if (!relayd) { |
1931 | lttng_consumer_sync_trace_file(stream, orig_offset); | |
1932 | } | |
f02e1e8a DG |
1933 | goto end; |
1934 | ||
8994307f DG |
1935 | write_error: |
1936 | /* | |
1937 | * This is a special case that the relayd has closed its socket. Let's | |
1938 | * cleanup the relayd object and all associated streams. | |
1939 | */ | |
1940 | if (relayd && relayd_hang_up) { | |
1941 | cleanup_relayd(relayd, ctx); | |
1942 | /* Skip splice error so the consumer does not fail */ | |
1943 | goto end; | |
1944 | } | |
1945 | ||
f02e1e8a DG |
1946 | splice_error: |
1947 | /* send the appropriate error description to sessiond */ | |
1948 | switch (ret) { | |
f02e1e8a | 1949 | case EINVAL: |
f73fabfd | 1950 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); |
f02e1e8a DG |
1951 | break; |
1952 | case ENOMEM: | |
f73fabfd | 1953 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); |
f02e1e8a DG |
1954 | break; |
1955 | case ESPIPE: | |
f73fabfd | 1956 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); |
f02e1e8a DG |
1957 | break; |
1958 | } | |
1959 | ||
1960 | end: | |
1961 | if (relayd && stream->metadata_flag) { | |
1962 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
1963 | } | |
1964 | ||
1965 | rcu_read_unlock(); | |
1966 | return written; | |
3bd1e081 MD |
1967 | } |
1968 | ||
6af2ba3b JD |
1969 | /* |
1970 | * Sample the snapshot positions for a specific fd | |
1971 | * | |
1972 | * Returns 0 on success, < 0 on error | |
1973 | */ | |
1974 | int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream *stream) | |
1975 | { | |
1976 | switch (consumer_data.type) { | |
1977 | case LTTNG_CONSUMER_KERNEL: | |
1978 | return lttng_kconsumer_sample_snapshot_positions(stream); | |
1979 | case LTTNG_CONSUMER32_UST: | |
1980 | case LTTNG_CONSUMER64_UST: | |
1981 | return lttng_ustconsumer_sample_snapshot_positions(stream); | |
1982 | default: | |
1983 | ERR("Unknown consumer_data type"); | |
1984 | assert(0); | |
1985 | return -ENOSYS; | |
1986 | } | |
1987 | } | |
3bd1e081 MD |
1988 | /* |
1989 | * Take a snapshot for a specific fd | |
1990 | * | |
1991 | * Returns 0 on success, < 0 on error | |
1992 | */ | |
ffe60014 | 1993 | int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) |
3bd1e081 MD |
1994 | { |
1995 | switch (consumer_data.type) { | |
1996 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 1997 | return lttng_kconsumer_take_snapshot(stream); |
7753dea8 MD |
1998 | case LTTNG_CONSUMER32_UST: |
1999 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 2000 | return lttng_ustconsumer_take_snapshot(stream); |
3bd1e081 MD |
2001 | default: |
2002 | ERR("Unknown consumer_data type"); | |
2003 | assert(0); | |
2004 | return -ENOSYS; | |
2005 | } | |
3bd1e081 MD |
2006 | } |
2007 | ||
2008 | /* | |
2009 | * Get the produced position | |
2010 | * | |
2011 | * Returns 0 on success, < 0 on error | |
2012 | */ | |
ffe60014 | 2013 | int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, |
3bd1e081 MD |
2014 | unsigned long *pos) |
2015 | { | |
2016 | switch (consumer_data.type) { | |
2017 | case LTTNG_CONSUMER_KERNEL: | |
ffe60014 | 2018 | return lttng_kconsumer_get_produced_snapshot(stream, pos); |
7753dea8 MD |
2019 | case LTTNG_CONSUMER32_UST: |
2020 | case LTTNG_CONSUMER64_UST: | |
ffe60014 | 2021 | return lttng_ustconsumer_get_produced_snapshot(stream, pos); |
3bd1e081 MD |
2022 | default: |
2023 | ERR("Unknown consumer_data type"); | |
2024 | assert(0); | |
2025 | return -ENOSYS; | |
2026 | } | |
2027 | } | |
2028 | ||
6af2ba3b JD |
2029 | /* |
2030 | * Get the consumed position | |
2031 | * | |
2032 | * Returns 0 on success, < 0 on error | |
2033 | */ | |
2034 | int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream *stream, | |
2035 | unsigned long *pos) | |
2036 | { | |
2037 | switch (consumer_data.type) { | |
2038 | case LTTNG_CONSUMER_KERNEL: | |
2039 | return lttng_kconsumer_get_consumed_snapshot(stream, pos); | |
2040 | case LTTNG_CONSUMER32_UST: | |
2041 | case LTTNG_CONSUMER64_UST: | |
2042 | return lttng_ustconsumer_get_consumed_snapshot(stream, pos); | |
2043 | default: | |
2044 | ERR("Unknown consumer_data type"); | |
2045 | assert(0); | |
2046 | return -ENOSYS; | |
2047 | } | |
2048 | } | |
2049 | ||
3bd1e081 MD |
2050 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, |
2051 | int sock, struct pollfd *consumer_sockpoll) | |
2052 | { | |
2053 | switch (consumer_data.type) { | |
2054 | case LTTNG_CONSUMER_KERNEL: | |
2055 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
7753dea8 MD |
2056 | case LTTNG_CONSUMER32_UST: |
2057 | case LTTNG_CONSUMER64_UST: | |
3bd1e081 MD |
2058 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
2059 | default: | |
2060 | ERR("Unknown consumer_data type"); | |
2061 | assert(0); | |
2062 | return -ENOSYS; | |
2063 | } | |
2064 | } | |
2065 | ||
6d574024 | 2066 | void lttng_consumer_close_all_metadata(void) |
d88aee68 DG |
2067 | { |
2068 | switch (consumer_data.type) { | |
2069 | case LTTNG_CONSUMER_KERNEL: | |
2070 | /* | |
2071 | * The Kernel consumer has a different metadata scheme so we don't | |
2072 | * close anything because the stream will be closed by the session | |
2073 | * daemon. | |
2074 | */ | |
2075 | break; | |
2076 | case LTTNG_CONSUMER32_UST: | |
2077 | case LTTNG_CONSUMER64_UST: | |
2078 | /* | |
2079 | * Close all metadata streams. The metadata hash table is passed and | |
2080 | * this call iterates over it by closing all wakeup fd. This is safe | |
2081 | * because at this point we are sure that the metadata producer is | |
2082 | * either dead or blocked. | |
2083 | */ | |
6d574024 | 2084 | lttng_ustconsumer_close_all_metadata(metadata_ht); |
d88aee68 DG |
2085 | break; |
2086 | default: | |
2087 | ERR("Unknown consumer_data type"); | |
2088 | assert(0); | |
2089 | } | |
2090 | } | |
2091 | ||
fb3a43a9 DG |
2092 | /* |
2093 | * Clean up a metadata stream and free its memory. | |
2094 | */ | |
e316aad5 DG |
2095 | void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, |
2096 | struct lttng_ht *ht) | |
fb3a43a9 | 2097 | { |
e316aad5 | 2098 | struct lttng_consumer_channel *free_chan = NULL; |
fb3a43a9 DG |
2099 | |
2100 | assert(stream); | |
2101 | /* | |
2102 | * This call should NEVER receive regular stream. It must always be | |
2103 | * metadata stream and this is crucial for data structure synchronization. | |
2104 | */ | |
2105 | assert(stream->metadata_flag); | |
2106 | ||
e316aad5 DG |
2107 | DBG3("Consumer delete metadata stream %d", stream->wait_fd); |
2108 | ||
74251bb8 | 2109 | pthread_mutex_lock(&consumer_data.lock); |
fb549e7a | 2110 | pthread_mutex_lock(&stream->chan->lock); |
3dad2c0f | 2111 | pthread_mutex_lock(&stream->lock); |
081424af JG |
2112 | if (stream->chan->metadata_cache) { |
2113 | /* Only applicable to userspace consumers. */ | |
2114 | pthread_mutex_lock(&stream->chan->metadata_cache->lock); | |
2115 | } | |
8994307f | 2116 | |
6d574024 DG |
2117 | /* Remove any reference to that stream. */ |
2118 | consumer_stream_delete(stream, ht); | |
ca22feea | 2119 | |
6d574024 DG |
2120 | /* Close down everything including the relayd if one. */ |
2121 | consumer_stream_close(stream); | |
2122 | /* Destroy tracer buffers of the stream. */ | |
2123 | consumer_stream_destroy_buffers(stream); | |
fb3a43a9 DG |
2124 | |
2125 | /* Atomically decrement channel refcount since other threads can use it. */ | |
f2ad556d | 2126 | if (!uatomic_sub_return(&stream->chan->refcount, 1) |
ffe60014 | 2127 | && !uatomic_read(&stream->chan->nb_init_stream_left)) { |
c30aaa51 | 2128 | /* Go for channel deletion! */ |
e316aad5 | 2129 | free_chan = stream->chan; |
fb3a43a9 DG |
2130 | } |
2131 | ||
73811ecc DG |
2132 | /* |
2133 | * Nullify the stream reference so it is not used after deletion. The | |
6d574024 DG |
2134 | * channel lock MUST be acquired before being able to check for a NULL |
2135 | * pointer value. | |
73811ecc DG |
2136 | */ |
2137 | stream->chan->metadata_stream = NULL; | |
2138 | ||
081424af JG |
2139 | if (stream->chan->metadata_cache) { |
2140 | pthread_mutex_unlock(&stream->chan->metadata_cache->lock); | |
2141 | } | |
3dad2c0f | 2142 | pthread_mutex_unlock(&stream->lock); |
fb549e7a | 2143 | pthread_mutex_unlock(&stream->chan->lock); |
74251bb8 | 2144 | pthread_mutex_unlock(&consumer_data.lock); |
e316aad5 DG |
2145 | |
2146 | if (free_chan) { | |
2147 | consumer_del_channel(free_chan); | |
2148 | } | |
2149 | ||
6d574024 | 2150 | consumer_stream_free(stream); |
fb3a43a9 DG |
2151 | } |
2152 | ||
2153 | /* | |
2154 | * Action done with the metadata stream when adding it to the consumer internal | |
2155 | * data structures to handle it. | |
2156 | */ | |
5ab66908 | 2157 | int consumer_add_metadata_stream(struct lttng_consumer_stream *stream) |
fb3a43a9 | 2158 | { |
5ab66908 | 2159 | struct lttng_ht *ht = metadata_ht; |
e316aad5 | 2160 | int ret = 0; |
76082088 | 2161 | struct lttng_ht_iter iter; |
d88aee68 | 2162 | struct lttng_ht_node_u64 *node; |
fb3a43a9 | 2163 | |
e316aad5 DG |
2164 | assert(stream); |
2165 | assert(ht); | |
2166 | ||
d88aee68 | 2167 | DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key); |
e316aad5 DG |
2168 | |
2169 | pthread_mutex_lock(&consumer_data.lock); | |
a9838785 | 2170 | pthread_mutex_lock(&stream->chan->lock); |
ec6ea7d0 | 2171 | pthread_mutex_lock(&stream->chan->timer_lock); |
2e818a6a | 2172 | pthread_mutex_lock(&stream->lock); |
e316aad5 | 2173 | |
e316aad5 DG |
2174 | /* |
2175 | * From here, refcounts are updated so be _careful_ when returning an error | |
2176 | * after this point. | |
2177 | */ | |
2178 | ||
fb3a43a9 | 2179 | rcu_read_lock(); |
76082088 DG |
2180 | |
2181 | /* | |
2182 | * Lookup the stream just to make sure it does not exist in our internal | |
2183 | * state. This should NEVER happen. | |
2184 | */ | |
d88aee68 DG |
2185 | lttng_ht_lookup(ht, &stream->key, &iter); |
2186 | node = lttng_ht_iter_get_node_u64(&iter); | |
76082088 DG |
2187 | assert(!node); |
2188 | ||
e316aad5 | 2189 | /* |
ffe60014 DG |
2190 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
2191 | * in terms of destroying the associated channel, because the action that | |
e316aad5 DG |
2192 | * causes the count to become 0 also causes a stream to be added. The |
2193 | * channel deletion will thus be triggered by the following removal of this | |
2194 | * stream. | |
2195 | */ | |
ffe60014 | 2196 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
f2ad556d MD |
2197 | /* Increment refcount before decrementing nb_init_stream_left */ |
2198 | cmm_smp_wmb(); | |
ffe60014 | 2199 | uatomic_dec(&stream->chan->nb_init_stream_left); |
e316aad5 DG |
2200 | } |
2201 | ||
d88aee68 | 2202 | lttng_ht_add_unique_u64(ht, &stream->node); |
ca22feea | 2203 | |
d8ef542d MD |
2204 | lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht, |
2205 | &stream->node_channel_id); | |
2206 | ||
ca22feea DG |
2207 | /* |
2208 | * Add stream to the stream_list_ht of the consumer data. No need to steal | |
2209 | * the key since the HT does not use it and we allow to add redundant keys | |
2210 | * into this table. | |
2211 | */ | |
d88aee68 | 2212 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
ca22feea | 2213 | |
fb3a43a9 | 2214 | rcu_read_unlock(); |
e316aad5 | 2215 | |
2e818a6a | 2216 | pthread_mutex_unlock(&stream->lock); |
a9838785 | 2217 | pthread_mutex_unlock(&stream->chan->lock); |
ec6ea7d0 | 2218 | pthread_mutex_unlock(&stream->chan->timer_lock); |
e316aad5 DG |
2219 | pthread_mutex_unlock(&consumer_data.lock); |
2220 | return ret; | |
fb3a43a9 DG |
2221 | } |
2222 | ||
8994307f DG |
2223 | /* |
2224 | * Delete data stream that are flagged for deletion (endpoint_status). | |
2225 | */ | |
2226 | static void validate_endpoint_status_data_stream(void) | |
2227 | { | |
2228 | struct lttng_ht_iter iter; | |
2229 | struct lttng_consumer_stream *stream; | |
2230 | ||
2231 | DBG("Consumer delete flagged data stream"); | |
2232 | ||
2233 | rcu_read_lock(); | |
2234 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { | |
2235 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2236 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2237 | continue; |
2238 | } | |
2239 | /* Delete it right now */ | |
2240 | consumer_del_stream(stream, data_ht); | |
2241 | } | |
2242 | rcu_read_unlock(); | |
2243 | } | |
2244 | ||
2245 | /* | |
2246 | * Delete metadata stream that are flagged for deletion (endpoint_status). | |
2247 | */ | |
2248 | static void validate_endpoint_status_metadata_stream( | |
2249 | struct lttng_poll_event *pollset) | |
2250 | { | |
2251 | struct lttng_ht_iter iter; | |
2252 | struct lttng_consumer_stream *stream; | |
2253 | ||
2254 | DBG("Consumer delete flagged metadata stream"); | |
2255 | ||
2256 | assert(pollset); | |
2257 | ||
2258 | rcu_read_lock(); | |
2259 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { | |
2260 | /* Validate delete flag of the stream */ | |
79d4ffb7 | 2261 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
8994307f DG |
2262 | continue; |
2263 | } | |
2264 | /* | |
2265 | * Remove from pollset so the metadata thread can continue without | |
2266 | * blocking on a deleted stream. | |
2267 | */ | |
2268 | lttng_poll_del(pollset, stream->wait_fd); | |
2269 | ||
2270 | /* Delete it right now */ | |
2271 | consumer_del_metadata_stream(stream, metadata_ht); | |
2272 | } | |
2273 | rcu_read_unlock(); | |
2274 | } | |
2275 | ||
fb3a43a9 DG |
2276 | /* |
2277 | * Thread polls on metadata file descriptor and write them on disk or on the | |
2278 | * network. | |
2279 | */ | |
7d980def | 2280 | void *consumer_thread_metadata_poll(void *data) |
fb3a43a9 | 2281 | { |
1fc79fb4 | 2282 | int ret, i, pollfd, err = -1; |
fb3a43a9 | 2283 | uint32_t revents, nb_fd; |
e316aad5 | 2284 | struct lttng_consumer_stream *stream = NULL; |
fb3a43a9 | 2285 | struct lttng_ht_iter iter; |
d88aee68 | 2286 | struct lttng_ht_node_u64 *node; |
fb3a43a9 DG |
2287 | struct lttng_poll_event events; |
2288 | struct lttng_consumer_local_data *ctx = data; | |
2289 | ssize_t len; | |
2290 | ||
2291 | rcu_register_thread(); | |
2292 | ||
1fc79fb4 MD |
2293 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA); |
2294 | ||
2d57de81 MD |
2295 | if (testpoint(consumerd_thread_metadata)) { |
2296 | goto error_testpoint; | |
2297 | } | |
2298 | ||
9ce5646a MD |
2299 | health_code_update(); |
2300 | ||
fb3a43a9 DG |
2301 | DBG("Thread metadata poll started"); |
2302 | ||
fb3a43a9 DG |
2303 | /* Size is set to 1 for the consumer_metadata pipe */ |
2304 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2305 | if (ret < 0) { | |
2306 | ERR("Poll set creation failed"); | |
d8ef542d | 2307 | goto end_poll; |
fb3a43a9 DG |
2308 | } |
2309 | ||
13886d2d DG |
2310 | ret = lttng_poll_add(&events, |
2311 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN); | |
fb3a43a9 DG |
2312 | if (ret < 0) { |
2313 | goto end; | |
2314 | } | |
2315 | ||
2316 | /* Main loop */ | |
2317 | DBG("Metadata main loop started"); | |
2318 | ||
2319 | while (1) { | |
fb3a43a9 | 2320 | restart: |
7fa2082e | 2321 | health_code_update(); |
9ce5646a | 2322 | health_poll_entry(); |
7fa2082e | 2323 | DBG("Metadata poll wait"); |
fb3a43a9 | 2324 | ret = lttng_poll_wait(&events, -1); |
7fa2082e MD |
2325 | DBG("Metadata poll return from wait with %d fd(s)", |
2326 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2327 | health_poll_exit(); |
40063ead | 2328 | DBG("Metadata event caught in thread"); |
fb3a43a9 DG |
2329 | if (ret < 0) { |
2330 | if (errno == EINTR) { | |
40063ead | 2331 | ERR("Poll EINTR caught"); |
fb3a43a9 DG |
2332 | goto restart; |
2333 | } | |
d9607cd7 MD |
2334 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2335 | err = 0; /* All is OK */ | |
2336 | } | |
2337 | goto end; | |
fb3a43a9 DG |
2338 | } |
2339 | ||
0d9c5d77 DG |
2340 | nb_fd = ret; |
2341 | ||
e316aad5 | 2342 | /* From here, the event is a metadata wait fd */ |
fb3a43a9 | 2343 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2344 | health_code_update(); |
2345 | ||
fb3a43a9 DG |
2346 | revents = LTTNG_POLL_GETEV(&events, i); |
2347 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2348 | ||
fd20dac9 MD |
2349 | if (!revents) { |
2350 | /* No activity for this FD (poll implementation). */ | |
2351 | continue; | |
2352 | } | |
2353 | ||
13886d2d | 2354 | if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) { |
03e43155 | 2355 | if (revents & LPOLLIN) { |
13886d2d DG |
2356 | ssize_t pipe_len; |
2357 | ||
2358 | pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe, | |
2359 | &stream, sizeof(stream)); | |
6cd525e8 | 2360 | if (pipe_len < sizeof(stream)) { |
03e43155 MD |
2361 | if (pipe_len < 0) { |
2362 | PERROR("read metadata stream"); | |
2363 | } | |
fb3a43a9 | 2364 | /* |
03e43155 MD |
2365 | * Remove the pipe from the poll set and continue the loop |
2366 | * since their might be data to consume. | |
fb3a43a9 | 2367 | */ |
03e43155 MD |
2368 | lttng_poll_del(&events, |
2369 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2370 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
fb3a43a9 DG |
2371 | continue; |
2372 | } | |
2373 | ||
8994307f DG |
2374 | /* A NULL stream means that the state has changed. */ |
2375 | if (stream == NULL) { | |
2376 | /* Check for deleted streams. */ | |
2377 | validate_endpoint_status_metadata_stream(&events); | |
3714380f | 2378 | goto restart; |
8994307f DG |
2379 | } |
2380 | ||
fb3a43a9 DG |
2381 | DBG("Adding metadata stream %d to poll set", |
2382 | stream->wait_fd); | |
2383 | ||
fb3a43a9 DG |
2384 | /* Add metadata stream to the global poll events list */ |
2385 | lttng_poll_add(&events, stream->wait_fd, | |
6d574024 | 2386 | LPOLLIN | LPOLLPRI | LPOLLHUP); |
03e43155 MD |
2387 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
2388 | DBG("Metadata thread pipe hung up"); | |
2389 | /* | |
2390 | * Remove the pipe from the poll set and continue the loop | |
2391 | * since their might be data to consume. | |
2392 | */ | |
2393 | lttng_poll_del(&events, | |
2394 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); | |
2395 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); | |
2396 | continue; | |
2397 | } else { | |
2398 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
2399 | goto end; | |
fb3a43a9 DG |
2400 | } |
2401 | ||
e316aad5 | 2402 | /* Handle other stream */ |
fb3a43a9 DG |
2403 | continue; |
2404 | } | |
2405 | ||
d09e1200 | 2406 | rcu_read_lock(); |
d88aee68 DG |
2407 | { |
2408 | uint64_t tmp_id = (uint64_t) pollfd; | |
2409 | ||
2410 | lttng_ht_lookup(metadata_ht, &tmp_id, &iter); | |
2411 | } | |
2412 | node = lttng_ht_iter_get_node_u64(&iter); | |
e316aad5 | 2413 | assert(node); |
fb3a43a9 DG |
2414 | |
2415 | stream = caa_container_of(node, struct lttng_consumer_stream, | |
58b1f425 | 2416 | node); |
fb3a43a9 | 2417 | |
03e43155 MD |
2418 | if (revents & (LPOLLIN | LPOLLPRI)) { |
2419 | /* Get the data out of the metadata file descriptor */ | |
2420 | DBG("Metadata available on fd %d", pollfd); | |
2421 | assert(stream->wait_fd == pollfd); | |
2422 | ||
2423 | do { | |
2424 | health_code_update(); | |
2425 | ||
2426 | len = ctx->on_buffer_ready(stream, ctx); | |
2427 | /* | |
2428 | * We don't check the return value here since if we get | |
83f4233d | 2429 | * a negative len, it means an error occurred thus we |
03e43155 MD |
2430 | * simply remove it from the poll set and free the |
2431 | * stream. | |
2432 | */ | |
2433 | } while (len > 0); | |
2434 | ||
2435 | /* It's ok to have an unavailable sub-buffer */ | |
2436 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { | |
2437 | /* Clean up stream from consumer and free it. */ | |
2438 | lttng_poll_del(&events, stream->wait_fd); | |
2439 | consumer_del_metadata_stream(stream, metadata_ht); | |
2440 | } | |
2441 | } else if (revents & (LPOLLERR | LPOLLHUP)) { | |
e316aad5 | 2442 | DBG("Metadata fd %d is hup|err.", pollfd); |
fb3a43a9 DG |
2443 | if (!stream->hangup_flush_done |
2444 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2445 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2446 | DBG("Attempting to flush and consume the UST buffers"); | |
2447 | lttng_ustconsumer_on_stream_hangup(stream); | |
2448 | ||
2449 | /* We just flushed the stream now read it. */ | |
4bb94b75 | 2450 | do { |
9ce5646a MD |
2451 | health_code_update(); |
2452 | ||
4bb94b75 DG |
2453 | len = ctx->on_buffer_ready(stream, ctx); |
2454 | /* | |
2455 | * We don't check the return value here since if we get | |
83f4233d | 2456 | * a negative len, it means an error occurred thus we |
4bb94b75 DG |
2457 | * simply remove it from the poll set and free the |
2458 | * stream. | |
2459 | */ | |
2460 | } while (len > 0); | |
fb3a43a9 DG |
2461 | } |
2462 | ||
fb3a43a9 | 2463 | lttng_poll_del(&events, stream->wait_fd); |
e316aad5 DG |
2464 | /* |
2465 | * This call update the channel states, closes file descriptors | |
2466 | * and securely free the stream. | |
2467 | */ | |
2468 | consumer_del_metadata_stream(stream, metadata_ht); | |
03e43155 MD |
2469 | } else { |
2470 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
6f2f1a70 | 2471 | rcu_read_unlock(); |
03e43155 | 2472 | goto end; |
fb3a43a9 | 2473 | } |
e316aad5 | 2474 | /* Release RCU lock for the stream looked up */ |
d09e1200 | 2475 | rcu_read_unlock(); |
fb3a43a9 DG |
2476 | } |
2477 | } | |
2478 | ||
1fc79fb4 MD |
2479 | /* All is OK */ |
2480 | err = 0; | |
fb3a43a9 DG |
2481 | end: |
2482 | DBG("Metadata poll thread exiting"); | |
fb3a43a9 | 2483 | |
d8ef542d MD |
2484 | lttng_poll_clean(&events); |
2485 | end_poll: | |
2d57de81 | 2486 | error_testpoint: |
1fc79fb4 MD |
2487 | if (err) { |
2488 | health_error(); | |
2489 | ERR("Health error occurred in %s", __func__); | |
2490 | } | |
2491 | health_unregister(health_consumerd); | |
fb3a43a9 DG |
2492 | rcu_unregister_thread(); |
2493 | return NULL; | |
2494 | } | |
2495 | ||
3bd1e081 | 2496 | /* |
e4421fec | 2497 | * This thread polls the fds in the set to consume the data and write |
3bd1e081 MD |
2498 | * it to tracefile if necessary. |
2499 | */ | |
7d980def | 2500 | void *consumer_thread_data_poll(void *data) |
3bd1e081 | 2501 | { |
1fc79fb4 | 2502 | int num_rdy, num_hup, high_prio, ret, i, err = -1; |
3bd1e081 MD |
2503 | struct pollfd *pollfd = NULL; |
2504 | /* local view of the streams */ | |
c869f647 | 2505 | struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; |
3bd1e081 MD |
2506 | /* local view of consumer_data.fds_count */ |
2507 | int nb_fd = 0; | |
3bd1e081 | 2508 | struct lttng_consumer_local_data *ctx = data; |
00e2e675 | 2509 | ssize_t len; |
3bd1e081 | 2510 | |
e7b994a3 DG |
2511 | rcu_register_thread(); |
2512 | ||
1fc79fb4 MD |
2513 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA); |
2514 | ||
2d57de81 MD |
2515 | if (testpoint(consumerd_thread_data)) { |
2516 | goto error_testpoint; | |
2517 | } | |
2518 | ||
9ce5646a MD |
2519 | health_code_update(); |
2520 | ||
4df6c8cb MD |
2521 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream *)); |
2522 | if (local_stream == NULL) { | |
2523 | PERROR("local_stream malloc"); | |
2524 | goto end; | |
2525 | } | |
3bd1e081 MD |
2526 | |
2527 | while (1) { | |
9ce5646a MD |
2528 | health_code_update(); |
2529 | ||
3bd1e081 MD |
2530 | high_prio = 0; |
2531 | num_hup = 0; | |
2532 | ||
2533 | /* | |
e4421fec | 2534 | * the fds set has been updated, we need to update our |
3bd1e081 MD |
2535 | * local array as well |
2536 | */ | |
2537 | pthread_mutex_lock(&consumer_data.lock); | |
2538 | if (consumer_data.need_update) { | |
0e428499 DG |
2539 | free(pollfd); |
2540 | pollfd = NULL; | |
2541 | ||
2542 | free(local_stream); | |
2543 | local_stream = NULL; | |
3bd1e081 | 2544 | |
02b3d176 DG |
2545 | /* |
2546 | * Allocate for all fds +1 for the consumer_data_pipe and +1 for | |
2547 | * wake up pipe. | |
2548 | */ | |
2549 | pollfd = zmalloc((consumer_data.stream_count + 2) * sizeof(struct pollfd)); | |
3bd1e081 | 2550 | if (pollfd == NULL) { |
7a57cf92 | 2551 | PERROR("pollfd malloc"); |
3bd1e081 MD |
2552 | pthread_mutex_unlock(&consumer_data.lock); |
2553 | goto end; | |
2554 | } | |
2555 | ||
02b3d176 | 2556 | local_stream = zmalloc((consumer_data.stream_count + 2) * |
747f8642 | 2557 | sizeof(struct lttng_consumer_stream *)); |
3bd1e081 | 2558 | if (local_stream == NULL) { |
7a57cf92 | 2559 | PERROR("local_stream malloc"); |
3bd1e081 MD |
2560 | pthread_mutex_unlock(&consumer_data.lock); |
2561 | goto end; | |
2562 | } | |
ffe60014 | 2563 | ret = update_poll_array(ctx, &pollfd, local_stream, |
43c34bc3 | 2564 | data_ht); |
3bd1e081 MD |
2565 | if (ret < 0) { |
2566 | ERR("Error in allocating pollfd or local_outfds"); | |
f73fabfd | 2567 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2568 | pthread_mutex_unlock(&consumer_data.lock); |
2569 | goto end; | |
2570 | } | |
2571 | nb_fd = ret; | |
2572 | consumer_data.need_update = 0; | |
2573 | } | |
2574 | pthread_mutex_unlock(&consumer_data.lock); | |
2575 | ||
4078b776 | 2576 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
10211f5c | 2577 | if (nb_fd == 0 && CMM_LOAD_SHARED(consumer_quit) == 1) { |
1fc79fb4 | 2578 | err = 0; /* All is OK */ |
4078b776 MD |
2579 | goto end; |
2580 | } | |
3bd1e081 | 2581 | /* poll on the array of fds */ |
88f2b785 | 2582 | restart: |
02b3d176 | 2583 | DBG("polling on %d fd", nb_fd + 2); |
cf0bcb51 JG |
2584 | if (testpoint(consumerd_thread_data_poll)) { |
2585 | goto end; | |
2586 | } | |
9ce5646a | 2587 | health_poll_entry(); |
02b3d176 | 2588 | num_rdy = poll(pollfd, nb_fd + 2, -1); |
9ce5646a | 2589 | health_poll_exit(); |
3bd1e081 MD |
2590 | DBG("poll num_rdy : %d", num_rdy); |
2591 | if (num_rdy == -1) { | |
88f2b785 MD |
2592 | /* |
2593 | * Restart interrupted system call. | |
2594 | */ | |
2595 | if (errno == EINTR) { | |
2596 | goto restart; | |
2597 | } | |
7a57cf92 | 2598 | PERROR("Poll error"); |
f73fabfd | 2599 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
3bd1e081 MD |
2600 | goto end; |
2601 | } else if (num_rdy == 0) { | |
2602 | DBG("Polling thread timed out"); | |
2603 | goto end; | |
2604 | } | |
2605 | ||
80957876 JG |
2606 | if (caa_unlikely(data_consumption_paused)) { |
2607 | DBG("Data consumption paused, sleeping..."); | |
2608 | sleep(1); | |
2609 | goto restart; | |
2610 | } | |
2611 | ||
3bd1e081 | 2612 | /* |
50f8ae69 | 2613 | * If the consumer_data_pipe triggered poll go directly to the |
00e2e675 DG |
2614 | * beginning of the loop to update the array. We want to prioritize |
2615 | * array update over low-priority reads. | |
3bd1e081 | 2616 | */ |
509bb1cf | 2617 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
ab30f567 | 2618 | ssize_t pipe_readlen; |
04fdd819 | 2619 | |
50f8ae69 | 2620 | DBG("consumer_data_pipe wake up"); |
acdb9057 DG |
2621 | pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe, |
2622 | &new_stream, sizeof(new_stream)); | |
6cd525e8 MD |
2623 | if (pipe_readlen < sizeof(new_stream)) { |
2624 | PERROR("Consumer data pipe"); | |
23f5f35d DG |
2625 | /* Continue so we can at least handle the current stream(s). */ |
2626 | continue; | |
2627 | } | |
c869f647 DG |
2628 | |
2629 | /* | |
2630 | * If the stream is NULL, just ignore it. It's also possible that | |
2631 | * the sessiond poll thread changed the consumer_quit state and is | |
2632 | * waking us up to test it. | |
2633 | */ | |
2634 | if (new_stream == NULL) { | |
8994307f | 2635 | validate_endpoint_status_data_stream(); |
c869f647 DG |
2636 | continue; |
2637 | } | |
2638 | ||
c869f647 | 2639 | /* Continue to update the local streams and handle prio ones */ |
3bd1e081 MD |
2640 | continue; |
2641 | } | |
2642 | ||
02b3d176 DG |
2643 | /* Handle wakeup pipe. */ |
2644 | if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) { | |
2645 | char dummy; | |
2646 | ssize_t pipe_readlen; | |
2647 | ||
2648 | pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy, | |
2649 | sizeof(dummy)); | |
2650 | if (pipe_readlen < 0) { | |
2651 | PERROR("Consumer data wakeup pipe"); | |
2652 | } | |
2653 | /* We've been awakened to handle stream(s). */ | |
2654 | ctx->has_wakeup = 0; | |
2655 | } | |
2656 | ||
3bd1e081 MD |
2657 | /* Take care of high priority channels first. */ |
2658 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2659 | health_code_update(); |
2660 | ||
9617607b DG |
2661 | if (local_stream[i] == NULL) { |
2662 | continue; | |
2663 | } | |
fb3a43a9 | 2664 | if (pollfd[i].revents & POLLPRI) { |
d41f73b7 MD |
2665 | DBG("Urgent read on fd %d", pollfd[i].fd); |
2666 | high_prio = 1; | |
4078b776 | 2667 | len = ctx->on_buffer_ready(local_stream[i], ctx); |
d41f73b7 | 2668 | /* it's ok to have an unavailable sub-buffer */ |
b64403e3 | 2669 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2670 | /* Clean the stream and free it. */ |
2671 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2672 | local_stream[i] = NULL; |
4078b776 MD |
2673 | } else if (len > 0) { |
2674 | local_stream[i]->data_read = 1; | |
d41f73b7 | 2675 | } |
3bd1e081 MD |
2676 | } |
2677 | } | |
2678 | ||
4078b776 MD |
2679 | /* |
2680 | * If we read high prio channel in this loop, try again | |
2681 | * for more high prio data. | |
2682 | */ | |
2683 | if (high_prio) { | |
3bd1e081 MD |
2684 | continue; |
2685 | } | |
2686 | ||
2687 | /* Take care of low priority channels. */ | |
4078b776 | 2688 | for (i = 0; i < nb_fd; i++) { |
9ce5646a MD |
2689 | health_code_update(); |
2690 | ||
9617607b DG |
2691 | if (local_stream[i] == NULL) { |
2692 | continue; | |
2693 | } | |
4078b776 | 2694 | if ((pollfd[i].revents & POLLIN) || |
02b3d176 DG |
2695 | local_stream[i]->hangup_flush_done || |
2696 | local_stream[i]->has_data) { | |
4078b776 MD |
2697 | DBG("Normal read on fd %d", pollfd[i].fd); |
2698 | len = ctx->on_buffer_ready(local_stream[i], ctx); | |
2699 | /* it's ok to have an unavailable sub-buffer */ | |
b64403e3 | 2700 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
ab1027f4 DG |
2701 | /* Clean the stream and free it. */ |
2702 | consumer_del_stream(local_stream[i], data_ht); | |
9617607b | 2703 | local_stream[i] = NULL; |
4078b776 MD |
2704 | } else if (len > 0) { |
2705 | local_stream[i]->data_read = 1; | |
2706 | } | |
2707 | } | |
2708 | } | |
2709 | ||
2710 | /* Handle hangup and errors */ | |
2711 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2712 | health_code_update(); |
2713 | ||
9617607b DG |
2714 | if (local_stream[i] == NULL) { |
2715 | continue; | |
2716 | } | |
4078b776 MD |
2717 | if (!local_stream[i]->hangup_flush_done |
2718 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) | |
2719 | && (consumer_data.type == LTTNG_CONSUMER32_UST | |
2720 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { | |
2721 | DBG("fd %d is hup|err|nval. Attempting flush and read.", | |
9617607b | 2722 | pollfd[i].fd); |
4078b776 MD |
2723 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); |
2724 | /* Attempt read again, for the data we just flushed. */ | |
2725 | local_stream[i]->data_read = 1; | |
2726 | } | |
2727 | /* | |
2728 | * If the poll flag is HUP/ERR/NVAL and we have | |
2729 | * read no data in this pass, we can remove the | |
2730 | * stream from its hash table. | |
2731 | */ | |
2732 | if ((pollfd[i].revents & POLLHUP)) { | |
2733 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); | |
2734 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2735 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2736 | local_stream[i] = NULL; |
4078b776 MD |
2737 | num_hup++; |
2738 | } | |
2739 | } else if (pollfd[i].revents & POLLERR) { | |
2740 | ERR("Error returned in polling fd %d.", pollfd[i].fd); | |
2741 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2742 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2743 | local_stream[i] = NULL; |
4078b776 MD |
2744 | num_hup++; |
2745 | } | |
2746 | } else if (pollfd[i].revents & POLLNVAL) { | |
2747 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); | |
2748 | if (!local_stream[i]->data_read) { | |
43c34bc3 | 2749 | consumer_del_stream(local_stream[i], data_ht); |
9617607b | 2750 | local_stream[i] = NULL; |
4078b776 | 2751 | num_hup++; |
3bd1e081 MD |
2752 | } |
2753 | } | |
9617607b DG |
2754 | if (local_stream[i] != NULL) { |
2755 | local_stream[i]->data_read = 0; | |
2756 | } | |
3bd1e081 MD |
2757 | } |
2758 | } | |
1fc79fb4 MD |
2759 | /* All is OK */ |
2760 | err = 0; | |
3bd1e081 MD |
2761 | end: |
2762 | DBG("polling thread exiting"); | |
0e428499 DG |
2763 | free(pollfd); |
2764 | free(local_stream); | |
fb3a43a9 DG |
2765 | |
2766 | /* | |
2767 | * Close the write side of the pipe so epoll_wait() in | |
7d980def DG |
2768 | * consumer_thread_metadata_poll can catch it. The thread is monitoring the |
2769 | * read side of the pipe. If we close them both, epoll_wait strangely does | |
2770 | * not return and could create a endless wait period if the pipe is the | |
2771 | * only tracked fd in the poll set. The thread will take care of closing | |
2772 | * the read side. | |
fb3a43a9 | 2773 | */ |
13886d2d | 2774 | (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe); |
fb3a43a9 | 2775 | |
2d57de81 | 2776 | error_testpoint: |
1fc79fb4 MD |
2777 | if (err) { |
2778 | health_error(); | |
2779 | ERR("Health error occurred in %s", __func__); | |
2780 | } | |
2781 | health_unregister(health_consumerd); | |
2782 | ||
e7b994a3 | 2783 | rcu_unregister_thread(); |
3bd1e081 MD |
2784 | return NULL; |
2785 | } | |
2786 | ||
d8ef542d MD |
2787 | /* |
2788 | * Close wake-up end of each stream belonging to the channel. This will | |
2789 | * allow the poll() on the stream read-side to detect when the | |
2790 | * write-side (application) finally closes them. | |
2791 | */ | |
2792 | static | |
2793 | void consumer_close_channel_streams(struct lttng_consumer_channel *channel) | |
2794 | { | |
2795 | struct lttng_ht *ht; | |
2796 | struct lttng_consumer_stream *stream; | |
2797 | struct lttng_ht_iter iter; | |
2798 | ||
2799 | ht = consumer_data.stream_per_chan_id_ht; | |
2800 | ||
2801 | rcu_read_lock(); | |
2802 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
2803 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
2804 | ht->match_fct, &channel->key, | |
2805 | &iter.iter, stream, node_channel_id.node) { | |
f2ad556d MD |
2806 | /* |
2807 | * Protect against teardown with mutex. | |
2808 | */ | |
2809 | pthread_mutex_lock(&stream->lock); | |
2810 | if (cds_lfht_is_node_deleted(&stream->node.node)) { | |
2811 | goto next; | |
2812 | } | |
d8ef542d MD |
2813 | switch (consumer_data.type) { |
2814 | case LTTNG_CONSUMER_KERNEL: | |
2815 | break; | |
2816 | case LTTNG_CONSUMER32_UST: | |
2817 | case LTTNG_CONSUMER64_UST: | |
b4a650f3 DG |
2818 | if (stream->metadata_flag) { |
2819 | /* Safe and protected by the stream lock. */ | |
2820 | lttng_ustconsumer_close_metadata(stream->chan); | |
2821 | } else { | |
2822 | /* | |
2823 | * Note: a mutex is taken internally within | |
2824 | * liblttng-ust-ctl to protect timer wakeup_fd | |
2825 | * use from concurrent close. | |
2826 | */ | |
2827 | lttng_ustconsumer_close_stream_wakeup(stream); | |
2828 | } | |
d8ef542d MD |
2829 | break; |
2830 | default: | |
2831 | ERR("Unknown consumer_data type"); | |
2832 | assert(0); | |
2833 | } | |
f2ad556d MD |
2834 | next: |
2835 | pthread_mutex_unlock(&stream->lock); | |
d8ef542d MD |
2836 | } |
2837 | rcu_read_unlock(); | |
2838 | } | |
2839 | ||
2840 | static void destroy_channel_ht(struct lttng_ht *ht) | |
2841 | { | |
2842 | struct lttng_ht_iter iter; | |
2843 | struct lttng_consumer_channel *channel; | |
2844 | int ret; | |
2845 | ||
2846 | if (ht == NULL) { | |
2847 | return; | |
2848 | } | |
2849 | ||
2850 | rcu_read_lock(); | |
2851 | cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) { | |
2852 | ret = lttng_ht_del(ht, &iter); | |
2853 | assert(ret != 0); | |
2854 | } | |
2855 | rcu_read_unlock(); | |
2856 | ||
2857 | lttng_ht_destroy(ht); | |
2858 | } | |
2859 | ||
2860 | /* | |
2861 | * This thread polls the channel fds to detect when they are being | |
2862 | * closed. It closes all related streams if the channel is detected as | |
2863 | * closed. It is currently only used as a shim layer for UST because the | |
2864 | * consumerd needs to keep the per-stream wakeup end of pipes open for | |
2865 | * periodical flush. | |
2866 | */ | |
2867 | void *consumer_thread_channel_poll(void *data) | |
2868 | { | |
1fc79fb4 | 2869 | int ret, i, pollfd, err = -1; |
d8ef542d MD |
2870 | uint32_t revents, nb_fd; |
2871 | struct lttng_consumer_channel *chan = NULL; | |
2872 | struct lttng_ht_iter iter; | |
2873 | struct lttng_ht_node_u64 *node; | |
2874 | struct lttng_poll_event events; | |
2875 | struct lttng_consumer_local_data *ctx = data; | |
2876 | struct lttng_ht *channel_ht; | |
2877 | ||
2878 | rcu_register_thread(); | |
2879 | ||
1fc79fb4 MD |
2880 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL); |
2881 | ||
2d57de81 MD |
2882 | if (testpoint(consumerd_thread_channel)) { |
2883 | goto error_testpoint; | |
2884 | } | |
2885 | ||
9ce5646a MD |
2886 | health_code_update(); |
2887 | ||
d8ef542d MD |
2888 | channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
2889 | if (!channel_ht) { | |
2890 | /* ENOMEM at this point. Better to bail out. */ | |
2891 | goto end_ht; | |
2892 | } | |
2893 | ||
2894 | DBG("Thread channel poll started"); | |
2895 | ||
2896 | /* Size is set to 1 for the consumer_channel pipe */ | |
2897 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); | |
2898 | if (ret < 0) { | |
2899 | ERR("Poll set creation failed"); | |
2900 | goto end_poll; | |
2901 | } | |
2902 | ||
2903 | ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN); | |
2904 | if (ret < 0) { | |
2905 | goto end; | |
2906 | } | |
2907 | ||
2908 | /* Main loop */ | |
2909 | DBG("Channel main loop started"); | |
2910 | ||
2911 | while (1) { | |
d8ef542d | 2912 | restart: |
7fa2082e MD |
2913 | health_code_update(); |
2914 | DBG("Channel poll wait"); | |
9ce5646a | 2915 | health_poll_entry(); |
d8ef542d | 2916 | ret = lttng_poll_wait(&events, -1); |
7fa2082e MD |
2917 | DBG("Channel poll return from wait with %d fd(s)", |
2918 | LTTNG_POLL_GETNB(&events)); | |
9ce5646a | 2919 | health_poll_exit(); |
40063ead | 2920 | DBG("Channel event caught in thread"); |
d8ef542d MD |
2921 | if (ret < 0) { |
2922 | if (errno == EINTR) { | |
40063ead | 2923 | ERR("Poll EINTR caught"); |
d8ef542d MD |
2924 | goto restart; |
2925 | } | |
d9607cd7 MD |
2926 | if (LTTNG_POLL_GETNB(&events) == 0) { |
2927 | err = 0; /* All is OK */ | |
2928 | } | |
d8ef542d MD |
2929 | goto end; |
2930 | } | |
2931 | ||
2932 | nb_fd = ret; | |
2933 | ||
2934 | /* From here, the event is a channel wait fd */ | |
2935 | for (i = 0; i < nb_fd; i++) { | |
9ce5646a MD |
2936 | health_code_update(); |
2937 | ||
d8ef542d MD |
2938 | revents = LTTNG_POLL_GETEV(&events, i); |
2939 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
2940 | ||
d8ef542d | 2941 | if (!revents) { |
fd20dac9 | 2942 | /* No activity for this FD (poll implementation). */ |
d8ef542d MD |
2943 | continue; |
2944 | } | |
fd20dac9 | 2945 | |
d8ef542d | 2946 | if (pollfd == ctx->consumer_channel_pipe[0]) { |
03e43155 | 2947 | if (revents & LPOLLIN) { |
d8ef542d | 2948 | enum consumer_channel_action action; |
a0cbdd2e | 2949 | uint64_t key; |
d8ef542d | 2950 | |
a0cbdd2e | 2951 | ret = read_channel_pipe(ctx, &chan, &key, &action); |
d8ef542d | 2952 | if (ret <= 0) { |
03e43155 MD |
2953 | if (ret < 0) { |
2954 | ERR("Error reading channel pipe"); | |
2955 | } | |
2956 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
d8ef542d MD |
2957 | continue; |
2958 | } | |
2959 | ||
2960 | switch (action) { | |
2961 | case CONSUMER_CHANNEL_ADD: | |
2962 | DBG("Adding channel %d to poll set", | |
2963 | chan->wait_fd); | |
2964 | ||
2965 | lttng_ht_node_init_u64(&chan->wait_fd_node, | |
2966 | chan->wait_fd); | |
c7260a81 | 2967 | rcu_read_lock(); |
d8ef542d MD |
2968 | lttng_ht_add_unique_u64(channel_ht, |
2969 | &chan->wait_fd_node); | |
c7260a81 | 2970 | rcu_read_unlock(); |
d8ef542d MD |
2971 | /* Add channel to the global poll events list */ |
2972 | lttng_poll_add(&events, chan->wait_fd, | |
03e43155 | 2973 | LPOLLERR | LPOLLHUP); |
d8ef542d | 2974 | break; |
a0cbdd2e MD |
2975 | case CONSUMER_CHANNEL_DEL: |
2976 | { | |
b4a650f3 DG |
2977 | /* |
2978 | * This command should never be called if the channel | |
2979 | * has streams monitored by either the data or metadata | |
2980 | * thread. The consumer only notify this thread with a | |
2981 | * channel del. command if it receives a destroy | |
2982 | * channel command from the session daemon that send it | |
2983 | * if a command prior to the GET_CHANNEL failed. | |
2984 | */ | |
2985 | ||
c7260a81 | 2986 | rcu_read_lock(); |
a0cbdd2e MD |
2987 | chan = consumer_find_channel(key); |
2988 | if (!chan) { | |
c7260a81 | 2989 | rcu_read_unlock(); |
a0cbdd2e MD |
2990 | ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key); |
2991 | break; | |
2992 | } | |
2993 | lttng_poll_del(&events, chan->wait_fd); | |
f623cc0b | 2994 | iter.iter.node = &chan->wait_fd_node.node; |
a0cbdd2e MD |
2995 | ret = lttng_ht_del(channel_ht, &iter); |
2996 | assert(ret == 0); | |
a0cbdd2e | 2997 | |
f2a444f1 DG |
2998 | switch (consumer_data.type) { |
2999 | case LTTNG_CONSUMER_KERNEL: | |
3000 | break; | |
3001 | case LTTNG_CONSUMER32_UST: | |
3002 | case LTTNG_CONSUMER64_UST: | |
212d67a2 DG |
3003 | health_code_update(); |
3004 | /* Destroy streams that might have been left in the stream list. */ | |
3005 | clean_channel_stream_list(chan); | |
f2a444f1 DG |
3006 | break; |
3007 | default: | |
3008 | ERR("Unknown consumer_data type"); | |
3009 | assert(0); | |
3010 | } | |
3011 | ||
a0cbdd2e MD |
3012 | /* |
3013 | * Release our own refcount. Force channel deletion even if | |
3014 | * streams were not initialized. | |
3015 | */ | |
3016 | if (!uatomic_sub_return(&chan->refcount, 1)) { | |
3017 | consumer_del_channel(chan); | |
3018 | } | |
c7260a81 | 3019 | rcu_read_unlock(); |
a0cbdd2e MD |
3020 | goto restart; |
3021 | } | |
d8ef542d MD |
3022 | case CONSUMER_CHANNEL_QUIT: |
3023 | /* | |
3024 | * Remove the pipe from the poll set and continue the loop | |
3025 | * since their might be data to consume. | |
3026 | */ | |
3027 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
3028 | continue; | |
3029 | default: | |
3030 | ERR("Unknown action"); | |
3031 | break; | |
3032 | } | |
03e43155 MD |
3033 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
3034 | DBG("Channel thread pipe hung up"); | |
3035 | /* | |
3036 | * Remove the pipe from the poll set and continue the loop | |
3037 | * since their might be data to consume. | |
3038 | */ | |
3039 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); | |
3040 | continue; | |
3041 | } else { | |
3042 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
3043 | goto end; | |
d8ef542d MD |
3044 | } |
3045 | ||
3046 | /* Handle other stream */ | |
3047 | continue; | |
3048 | } | |
3049 | ||
3050 | rcu_read_lock(); | |
3051 | { | |
3052 | uint64_t tmp_id = (uint64_t) pollfd; | |
3053 | ||
3054 | lttng_ht_lookup(channel_ht, &tmp_id, &iter); | |
3055 | } | |
3056 | node = lttng_ht_iter_get_node_u64(&iter); | |
3057 | assert(node); | |
3058 | ||
3059 | chan = caa_container_of(node, struct lttng_consumer_channel, | |
3060 | wait_fd_node); | |
3061 | ||
3062 | /* Check for error event */ | |
3063 | if (revents & (LPOLLERR | LPOLLHUP)) { | |
3064 | DBG("Channel fd %d is hup|err.", pollfd); | |
3065 | ||
3066 | lttng_poll_del(&events, chan->wait_fd); | |
3067 | ret = lttng_ht_del(channel_ht, &iter); | |
3068 | assert(ret == 0); | |
b4a650f3 DG |
3069 | |
3070 | /* | |
3071 | * This will close the wait fd for each stream associated to | |
3072 | * this channel AND monitored by the data/metadata thread thus | |
3073 | * will be clean by the right thread. | |
3074 | */ | |
d8ef542d | 3075 | consumer_close_channel_streams(chan); |
f2ad556d MD |
3076 | |
3077 | /* Release our own refcount */ | |
3078 | if (!uatomic_sub_return(&chan->refcount, 1) | |
3079 | && !uatomic_read(&chan->nb_init_stream_left)) { | |
3080 | consumer_del_channel(chan); | |
3081 | } | |
03e43155 MD |
3082 | } else { |
3083 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); | |
3084 | rcu_read_unlock(); | |
3085 | goto end; | |
d8ef542d MD |
3086 | } |
3087 | ||
3088 | /* Release RCU lock for the channel looked up */ | |
3089 | rcu_read_unlock(); | |
3090 | } | |
3091 | } | |
3092 | ||
1fc79fb4 MD |
3093 | /* All is OK */ |
3094 | err = 0; | |
d8ef542d MD |
3095 | end: |
3096 | lttng_poll_clean(&events); | |
3097 | end_poll: | |
3098 | destroy_channel_ht(channel_ht); | |
3099 | end_ht: | |
2d57de81 | 3100 | error_testpoint: |
d8ef542d | 3101 | DBG("Channel poll thread exiting"); |
1fc79fb4 MD |
3102 | if (err) { |
3103 | health_error(); | |
3104 | ERR("Health error occurred in %s", __func__); | |
3105 | } | |
3106 | health_unregister(health_consumerd); | |
d8ef542d MD |
3107 | rcu_unregister_thread(); |
3108 | return NULL; | |
3109 | } | |
3110 | ||
331744e3 JD |
3111 | static int set_metadata_socket(struct lttng_consumer_local_data *ctx, |
3112 | struct pollfd *sockpoll, int client_socket) | |
3113 | { | |
3114 | int ret; | |
3115 | ||
3116 | assert(ctx); | |
3117 | assert(sockpoll); | |
3118 | ||
84382d49 MD |
3119 | ret = lttng_consumer_poll_socket(sockpoll); |
3120 | if (ret) { | |
331744e3 JD |
3121 | goto error; |
3122 | } | |
3123 | DBG("Metadata connection on client_socket"); | |
3124 | ||
3125 | /* Blocking call, waiting for transmission */ | |
3126 | ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket); | |
3127 | if (ctx->consumer_metadata_socket < 0) { | |
3128 | WARN("On accept metadata"); | |
3129 | ret = -1; | |
3130 | goto error; | |
3131 | } | |
3132 | ret = 0; | |
3133 | ||
3134 | error: | |
3135 | return ret; | |
3136 | } | |
3137 | ||
3bd1e081 MD |
3138 | /* |
3139 | * This thread listens on the consumerd socket and receives the file | |
3140 | * descriptors from the session daemon. | |
3141 | */ | |
7d980def | 3142 | void *consumer_thread_sessiond_poll(void *data) |
3bd1e081 | 3143 | { |
1fc79fb4 | 3144 | int sock = -1, client_socket, ret, err = -1; |
3bd1e081 MD |
3145 | /* |
3146 | * structure to poll for incoming data on communication socket avoids | |
3147 | * making blocking sockets. | |
3148 | */ | |
3149 | struct pollfd consumer_sockpoll[2]; | |
3150 | struct lttng_consumer_local_data *ctx = data; | |
3151 | ||
e7b994a3 DG |
3152 | rcu_register_thread(); |
3153 | ||
1fc79fb4 MD |
3154 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND); |
3155 | ||
2d57de81 MD |
3156 | if (testpoint(consumerd_thread_sessiond)) { |
3157 | goto error_testpoint; | |
3158 | } | |
3159 | ||
9ce5646a MD |
3160 | health_code_update(); |
3161 | ||
3bd1e081 MD |
3162 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
3163 | unlink(ctx->consumer_command_sock_path); | |
3164 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); | |
3165 | if (client_socket < 0) { | |
3166 | ERR("Cannot create command socket"); | |
3167 | goto end; | |
3168 | } | |
3169 | ||
3170 | ret = lttcomm_listen_unix_sock(client_socket); | |
3171 | if (ret < 0) { | |
3172 | goto end; | |
3173 | } | |
3174 | ||
32258573 | 3175 | DBG("Sending ready command to lttng-sessiond"); |
f73fabfd | 3176 | ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); |
3bd1e081 MD |
3177 | /* return < 0 on error, but == 0 is not fatal */ |
3178 | if (ret < 0) { | |
32258573 | 3179 | ERR("Error sending ready command to lttng-sessiond"); |
3bd1e081 MD |
3180 | goto end; |
3181 | } | |
3182 | ||
3bd1e081 MD |
3183 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ |
3184 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; | |
3185 | consumer_sockpoll[0].events = POLLIN | POLLPRI; | |
3186 | consumer_sockpoll[1].fd = client_socket; | |
3187 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3188 | ||
84382d49 MD |
3189 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3190 | if (ret) { | |
3191 | if (ret > 0) { | |
3192 | /* should exit */ | |
3193 | err = 0; | |
3194 | } | |
3bd1e081 MD |
3195 | goto end; |
3196 | } | |
3197 | DBG("Connection on client_socket"); | |
3198 | ||
3199 | /* Blocking call, waiting for transmission */ | |
3200 | sock = lttcomm_accept_unix_sock(client_socket); | |
534d2592 | 3201 | if (sock < 0) { |
3bd1e081 MD |
3202 | WARN("On accept"); |
3203 | goto end; | |
3204 | } | |
3bd1e081 | 3205 | |
331744e3 JD |
3206 | /* |
3207 | * Setup metadata socket which is the second socket connection on the | |
3208 | * command unix socket. | |
3209 | */ | |
3210 | ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket); | |
84382d49 MD |
3211 | if (ret) { |
3212 | if (ret > 0) { | |
3213 | /* should exit */ | |
3214 | err = 0; | |
3215 | } | |
331744e3 JD |
3216 | goto end; |
3217 | } | |
3218 | ||
d96f09c6 DG |
3219 | /* This socket is not useful anymore. */ |
3220 | ret = close(client_socket); | |
3221 | if (ret < 0) { | |
3222 | PERROR("close client_socket"); | |
3223 | } | |
3224 | client_socket = -1; | |
3225 | ||
3bd1e081 MD |
3226 | /* update the polling structure to poll on the established socket */ |
3227 | consumer_sockpoll[1].fd = sock; | |
3228 | consumer_sockpoll[1].events = POLLIN | POLLPRI; | |
3229 | ||
3230 | while (1) { | |
9ce5646a MD |
3231 | health_code_update(); |
3232 | ||
3233 | health_poll_entry(); | |
3234 | ret = lttng_consumer_poll_socket(consumer_sockpoll); | |
3235 | health_poll_exit(); | |
84382d49 MD |
3236 | if (ret) { |
3237 | if (ret > 0) { | |
3238 | /* should exit */ | |
3239 | err = 0; | |
3240 | } | |
3bd1e081 MD |
3241 | goto end; |
3242 | } | |
3243 | DBG("Incoming command on sock"); | |
3244 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); | |
4cbc1a04 DG |
3245 | if (ret <= 0) { |
3246 | /* | |
3247 | * This could simply be a session daemon quitting. Don't output | |
3248 | * ERR() here. | |
3249 | */ | |
3250 | DBG("Communication interrupted on command socket"); | |
41ba6035 | 3251 | err = 0; |
3bd1e081 MD |
3252 | goto end; |
3253 | } | |
10211f5c | 3254 | if (CMM_LOAD_SHARED(consumer_quit)) { |
3bd1e081 | 3255 | DBG("consumer_thread_receive_fds received quit from signal"); |
1fc79fb4 | 3256 | err = 0; /* All is OK */ |
3bd1e081 MD |
3257 | goto end; |
3258 | } | |
ffe60014 | 3259 | DBG("received command on sock"); |
3bd1e081 | 3260 | } |
1fc79fb4 MD |
3261 | /* All is OK */ |
3262 | err = 0; | |
3263 | ||
3bd1e081 | 3264 | end: |
ffe60014 | 3265 | DBG("Consumer thread sessiond poll exiting"); |
3bd1e081 | 3266 | |
d88aee68 DG |
3267 | /* |
3268 | * Close metadata streams since the producer is the session daemon which | |
3269 | * just died. | |
3270 | * | |
3271 | * NOTE: for now, this only applies to the UST tracer. | |
3272 | */ | |
6d574024 | 3273 | lttng_consumer_close_all_metadata(); |
d88aee68 | 3274 | |
3bd1e081 MD |
3275 | /* |
3276 | * when all fds have hung up, the polling thread | |
3277 | * can exit cleanly | |
3278 | */ | |
10211f5c | 3279 | CMM_STORE_SHARED(consumer_quit, 1); |
3bd1e081 | 3280 | |
04fdd819 | 3281 | /* |
c869f647 | 3282 | * Notify the data poll thread to poll back again and test the |
8994307f | 3283 | * consumer_quit state that we just set so to quit gracefully. |
04fdd819 | 3284 | */ |
acdb9057 | 3285 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
c869f647 | 3286 | |
a0cbdd2e | 3287 | notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT); |
d8ef542d | 3288 | |
5c635c72 MD |
3289 | notify_health_quit_pipe(health_quit_pipe); |
3290 | ||
d96f09c6 DG |
3291 | /* Cleaning up possibly open sockets. */ |
3292 | if (sock >= 0) { | |
3293 | ret = close(sock); | |
3294 | if (ret < 0) { | |
3295 | PERROR("close sock sessiond poll"); | |
3296 | } | |
3297 | } | |
3298 | if (client_socket >= 0) { | |
38476d24 | 3299 | ret = close(client_socket); |
d96f09c6 DG |
3300 | if (ret < 0) { |
3301 | PERROR("close client_socket sessiond poll"); | |
3302 | } | |
3303 | } | |
3304 | ||
2d57de81 | 3305 | error_testpoint: |
1fc79fb4 MD |
3306 | if (err) { |
3307 | health_error(); | |
3308 | ERR("Health error occurred in %s", __func__); | |
3309 | } | |
3310 | health_unregister(health_consumerd); | |
3311 | ||
e7b994a3 | 3312 | rcu_unregister_thread(); |
3bd1e081 MD |
3313 | return NULL; |
3314 | } | |
d41f73b7 | 3315 | |
b9478f8d JD |
3316 | static |
3317 | int rotate_notify_sessiond(struct lttng_consumer_local_data *ctx, | |
3318 | uint64_t key) | |
3319 | { | |
3320 | int ret; | |
3321 | ||
3322 | do { | |
3323 | ret = write(ctx->channel_rotate_pipe, &key, sizeof(key)); | |
3324 | } while (ret == -1 && errno == EINTR); | |
3325 | if (ret == -1) { | |
3326 | PERROR("write to the channel rotate pipe"); | |
3327 | } else { | |
3328 | DBG("Sent channel rotation notification for channel key %" | |
3329 | PRIu64, key); | |
3330 | } | |
3331 | ||
3332 | return ret; | |
3333 | } | |
3334 | ||
3335 | /* | |
3336 | * Perform operations that need to be done after a stream has | |
3337 | * rotated and released the stream lock. | |
3338 | * | |
3339 | * Multiple rotations cannot occur simultaneously, so we know the state of the | |
3340 | * "rotated" stream flag cannot change. | |
3341 | * | |
3342 | * This MUST be called WITHOUT the stream lock held. | |
3343 | */ | |
3344 | static | |
3345 | int consumer_post_rotation(struct lttng_consumer_stream *stream, | |
3346 | struct lttng_consumer_local_data *ctx) | |
3347 | { | |
3348 | int ret = 0; | |
3349 | ||
3350 | if (!stream->rotated) { | |
3351 | goto end; | |
3352 | } | |
3353 | ||
3354 | pthread_mutex_lock(&stream->chan->lock); | |
3355 | switch (consumer_data.type) { | |
3356 | case LTTNG_CONSUMER_KERNEL: | |
3357 | break; | |
3358 | case LTTNG_CONSUMER32_UST: | |
3359 | case LTTNG_CONSUMER64_UST: | |
3360 | /* | |
72ae4ac9 JD |
3361 | * The ust_metadata_pushed counter has been reset to 0, so now |
3362 | * we can wakeup the metadata thread so it dumps the metadata | |
3363 | * cache to the new file. | |
b9478f8d | 3364 | */ |
72ae4ac9 JD |
3365 | if (stream->metadata_flag) { |
3366 | consumer_metadata_wakeup_pipe(stream->chan); | |
3367 | } | |
b9478f8d JD |
3368 | break; |
3369 | default: | |
3370 | ERR("Unknown consumer_data type"); | |
3371 | abort(); | |
3372 | } | |
3373 | ||
3374 | if (--stream->chan->nr_stream_rotate_pending == 0) { | |
3375 | ret = rotate_notify_sessiond(ctx, stream->chan->key); | |
3376 | } | |
3377 | pthread_mutex_unlock(&stream->chan->lock); | |
3378 | stream->rotated = 0; | |
3379 | ||
3380 | end: | |
3381 | return ret; | |
3382 | } | |
3383 | ||
4078b776 | 3384 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
d41f73b7 MD |
3385 | struct lttng_consumer_local_data *ctx) |
3386 | { | |
74251bb8 | 3387 | ssize_t ret; |
b9478f8d | 3388 | int rotate_ret; |
74251bb8 DG |
3389 | |
3390 | pthread_mutex_lock(&stream->lock); | |
94d49140 JD |
3391 | if (stream->metadata_flag) { |
3392 | pthread_mutex_lock(&stream->metadata_rdv_lock); | |
3393 | } | |
74251bb8 | 3394 | |
d41f73b7 MD |
3395 | switch (consumer_data.type) { |
3396 | case LTTNG_CONSUMER_KERNEL: | |
74251bb8 DG |
3397 | ret = lttng_kconsumer_read_subbuffer(stream, ctx); |
3398 | break; | |
7753dea8 MD |
3399 | case LTTNG_CONSUMER32_UST: |
3400 | case LTTNG_CONSUMER64_UST: | |
74251bb8 DG |
3401 | ret = lttng_ustconsumer_read_subbuffer(stream, ctx); |
3402 | break; | |
d41f73b7 MD |
3403 | default: |
3404 | ERR("Unknown consumer_data type"); | |
3405 | assert(0); | |
74251bb8 DG |
3406 | ret = -ENOSYS; |
3407 | break; | |
d41f73b7 | 3408 | } |
74251bb8 | 3409 | |
94d49140 JD |
3410 | if (stream->metadata_flag) { |
3411 | pthread_cond_broadcast(&stream->metadata_rdv); | |
3412 | pthread_mutex_unlock(&stream->metadata_rdv_lock); | |
3413 | } | |
74251bb8 | 3414 | pthread_mutex_unlock(&stream->lock); |
b9478f8d JD |
3415 | |
3416 | rotate_ret = consumer_post_rotation(stream, ctx); | |
3417 | if (rotate_ret < 0) { | |
3418 | ERR("Failed after a rotation"); | |
3419 | ret = -1; | |
3420 | } | |
3421 | ||
74251bb8 | 3422 | return ret; |
d41f73b7 MD |
3423 | } |
3424 | ||
3425 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) | |
3426 | { | |
3427 | switch (consumer_data.type) { | |
3428 | case LTTNG_CONSUMER_KERNEL: | |
3429 | return lttng_kconsumer_on_recv_stream(stream); | |
7753dea8 MD |
3430 | case LTTNG_CONSUMER32_UST: |
3431 | case LTTNG_CONSUMER64_UST: | |
d41f73b7 MD |
3432 | return lttng_ustconsumer_on_recv_stream(stream); |
3433 | default: | |
3434 | ERR("Unknown consumer_data type"); | |
3435 | assert(0); | |
3436 | return -ENOSYS; | |
3437 | } | |
3438 | } | |
e4421fec DG |
3439 | |
3440 | /* | |
3441 | * Allocate and set consumer data hash tables. | |
3442 | */ | |
282dadbc | 3443 | int lttng_consumer_init(void) |
e4421fec | 3444 | { |
d88aee68 | 3445 | consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3446 | if (!consumer_data.channel_ht) { |
3447 | goto error; | |
3448 | } | |
3449 | ||
d88aee68 | 3450 | consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3451 | if (!consumer_data.relayd_ht) { |
3452 | goto error; | |
3453 | } | |
3454 | ||
d88aee68 | 3455 | consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3456 | if (!consumer_data.stream_list_ht) { |
3457 | goto error; | |
3458 | } | |
3459 | ||
d8ef542d | 3460 | consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
282dadbc MD |
3461 | if (!consumer_data.stream_per_chan_id_ht) { |
3462 | goto error; | |
3463 | } | |
3464 | ||
3465 | data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3466 | if (!data_ht) { | |
3467 | goto error; | |
3468 | } | |
3469 | ||
3470 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); | |
3471 | if (!metadata_ht) { | |
3472 | goto error; | |
3473 | } | |
3474 | ||
3475 | return 0; | |
3476 | ||
3477 | error: | |
3478 | return -1; | |
e4421fec | 3479 | } |
7735ef9e DG |
3480 | |
3481 | /* | |
3482 | * Process the ADD_RELAYD command receive by a consumer. | |
3483 | * | |
3484 | * This will create a relayd socket pair and add it to the relayd hash table. | |
3485 | * The caller MUST acquire a RCU read side lock before calling it. | |
3486 | */ | |
2527bf85 | 3487 | void consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type, |
7735ef9e | 3488 | struct lttng_consumer_local_data *ctx, int sock, |
6151a90f | 3489 | struct pollfd *consumer_sockpoll, |
d3e2ba59 JD |
3490 | struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id, |
3491 | uint64_t relayd_session_id) | |
7735ef9e | 3492 | { |
cd2b09ed | 3493 | int fd = -1, ret = -1, relayd_created = 0; |
0c759fc9 | 3494 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
d4298c99 | 3495 | struct consumer_relayd_sock_pair *relayd = NULL; |
7735ef9e | 3496 | |
6151a90f JD |
3497 | assert(ctx); |
3498 | assert(relayd_sock); | |
3499 | ||
da009f2c | 3500 | DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx); |
7735ef9e DG |
3501 | |
3502 | /* Get relayd reference if exists. */ | |
3503 | relayd = consumer_find_relayd(net_seq_idx); | |
3504 | if (relayd == NULL) { | |
da009f2c | 3505 | assert(sock_type == LTTNG_STREAM_CONTROL); |
7735ef9e DG |
3506 | /* Not found. Allocate one. */ |
3507 | relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); | |
3508 | if (relayd == NULL) { | |
618a6a28 MD |
3509 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
3510 | goto error; | |
0d08d75e | 3511 | } else { |
30319bcb | 3512 | relayd->sessiond_session_id = sessiond_id; |
0d08d75e | 3513 | relayd_created = 1; |
7735ef9e | 3514 | } |
0d08d75e DG |
3515 | |
3516 | /* | |
3517 | * This code path MUST continue to the consumer send status message to | |
3518 | * we can notify the session daemon and continue our work without | |
3519 | * killing everything. | |
3520 | */ | |
da009f2c MD |
3521 | } else { |
3522 | /* | |
3523 | * relayd key should never be found for control socket. | |
3524 | */ | |
3525 | assert(sock_type != LTTNG_STREAM_CONTROL); | |
0d08d75e DG |
3526 | } |
3527 | ||
3528 | /* First send a status message before receiving the fds. */ | |
0c759fc9 | 3529 | ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS); |
618a6a28 | 3530 | if (ret < 0) { |
0d08d75e | 3531 | /* Somehow, the session daemon is not responding anymore. */ |
618a6a28 MD |
3532 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); |
3533 | goto error_nosignal; | |
7735ef9e DG |
3534 | } |
3535 | ||
3536 | /* Poll on consumer socket. */ | |
84382d49 MD |
3537 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
3538 | if (ret) { | |
3539 | /* Needing to exit in the middle of a command: error. */ | |
0d08d75e | 3540 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
618a6a28 | 3541 | goto error_nosignal; |
7735ef9e DG |
3542 | } |
3543 | ||
3544 | /* Get relayd socket from session daemon */ | |
3545 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); | |
3546 | if (ret != sizeof(fd)) { | |
4028eeb9 | 3547 | fd = -1; /* Just in case it gets set with an invalid value. */ |
0d08d75e DG |
3548 | |
3549 | /* | |
3550 | * Failing to receive FDs might indicate a major problem such as | |
3551 | * reaching a fd limit during the receive where the kernel returns a | |
3552 | * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we | |
3553 | * don't take any chances and stop everything. | |
3554 | * | |
3555 | * XXX: Feature request #558 will fix that and avoid this possible | |
3556 | * issue when reaching the fd limit. | |
3557 | */ | |
3558 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); | |
618a6a28 | 3559 | ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD; |
f50f23d9 DG |
3560 | goto error; |
3561 | } | |
3562 | ||
7735ef9e DG |
3563 | /* Copy socket information and received FD */ |
3564 | switch (sock_type) { | |
3565 | case LTTNG_STREAM_CONTROL: | |
3566 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3567 | lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock); |
3568 | ret = lttcomm_create_sock(&relayd->control_sock.sock); | |
4028eeb9 | 3569 | /* Handle create_sock error. */ |
f66c074c | 3570 | if (ret < 0) { |
618a6a28 | 3571 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3572 | goto error; |
f66c074c | 3573 | } |
da009f2c MD |
3574 | /* |
3575 | * Close the socket created internally by | |
3576 | * lttcomm_create_sock, so we can replace it by the one | |
3577 | * received from sessiond. | |
3578 | */ | |
3579 | if (close(relayd->control_sock.sock.fd)) { | |
3580 | PERROR("close"); | |
3581 | } | |
7735ef9e DG |
3582 | |
3583 | /* Assign new file descriptor */ | |
6151a90f | 3584 | relayd->control_sock.sock.fd = fd; |
4b29f1ce | 3585 | fd = -1; /* For error path */ |
6151a90f JD |
3586 | /* Assign version values. */ |
3587 | relayd->control_sock.major = relayd_sock->major; | |
3588 | relayd->control_sock.minor = relayd_sock->minor; | |
c5b6f4f0 | 3589 | |
d3e2ba59 | 3590 | relayd->relayd_session_id = relayd_session_id; |
c5b6f4f0 | 3591 | |
7735ef9e DG |
3592 | break; |
3593 | case LTTNG_STREAM_DATA: | |
3594 | /* Copy received lttcomm socket */ | |
6151a90f JD |
3595 | lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock); |
3596 | ret = lttcomm_create_sock(&relayd->data_sock.sock); | |
4028eeb9 | 3597 | /* Handle create_sock error. */ |
f66c074c | 3598 | if (ret < 0) { |
618a6a28 | 3599 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
4028eeb9 | 3600 | goto error; |
f66c074c | 3601 | } |
da009f2c MD |
3602 | /* |
3603 | * Close the socket created internally by | |
3604 | * lttcomm_create_sock, so we can replace it by the one | |
3605 | * received from sessiond. | |
3606 | */ | |
3607 | if (close(relayd->data_sock.sock.fd)) { | |
3608 | PERROR("close"); | |
3609 | } | |
7735ef9e DG |
3610 | |
3611 | /* Assign new file descriptor */ | |
6151a90f | 3612 | relayd->data_sock.sock.fd = fd; |
4b29f1ce | 3613 | fd = -1; /* for eventual error paths */ |
6151a90f JD |
3614 | /* Assign version values. */ |
3615 | relayd->data_sock.major = relayd_sock->major; | |
3616 | relayd->data_sock.minor = relayd_sock->minor; | |
7735ef9e DG |
3617 | break; |
3618 | default: | |
3619 | ERR("Unknown relayd socket type (%d)", sock_type); | |
618a6a28 | 3620 | ret_code = LTTCOMM_CONSUMERD_FATAL; |
7735ef9e DG |
3621 | goto error; |
3622 | } | |
3623 | ||
d88aee68 | 3624 | DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)", |
7735ef9e DG |
3625 | sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", |
3626 | relayd->net_seq_idx, fd); | |
3627 | ||
618a6a28 MD |
3628 | /* We successfully added the socket. Send status back. */ |
3629 | ret = consumer_send_status_msg(sock, ret_code); | |
3630 | if (ret < 0) { | |
3631 | /* Somehow, the session daemon is not responding anymore. */ | |
3632 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3633 | goto error_nosignal; | |
3634 | } | |
3635 | ||
7735ef9e DG |
3636 | /* |
3637 | * Add relayd socket pair to consumer data hashtable. If object already | |
3638 | * exists or on error, the function gracefully returns. | |
3639 | */ | |
d09e1200 | 3640 | add_relayd(relayd); |
7735ef9e DG |
3641 | |
3642 | /* All good! */ | |
2527bf85 | 3643 | return; |
7735ef9e DG |
3644 | |
3645 | error: | |
618a6a28 MD |
3646 | if (consumer_send_status_msg(sock, ret_code) < 0) { |
3647 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); | |
3648 | } | |
3649 | ||
3650 | error_nosignal: | |
4028eeb9 DG |
3651 | /* Close received socket if valid. */ |
3652 | if (fd >= 0) { | |
3653 | if (close(fd)) { | |
3654 | PERROR("close received socket"); | |
3655 | } | |
3656 | } | |
cd2b09ed DG |
3657 | |
3658 | if (relayd_created) { | |
cd2b09ed DG |
3659 | free(relayd); |
3660 | } | |
7735ef9e | 3661 | } |
ca22feea | 3662 | |
4e9a4686 DG |
3663 | /* |
3664 | * Try to lock the stream mutex. | |
3665 | * | |
3666 | * On success, 1 is returned else 0 indicating that the mutex is NOT lock. | |
3667 | */ | |
3668 | static int stream_try_lock(struct lttng_consumer_stream *stream) | |
3669 | { | |
3670 | int ret; | |
3671 | ||
3672 | assert(stream); | |
3673 | ||
3674 | /* | |
3675 | * Try to lock the stream mutex. On failure, we know that the stream is | |
3676 | * being used else where hence there is data still being extracted. | |
3677 | */ | |
3678 | ret = pthread_mutex_trylock(&stream->lock); | |
3679 | if (ret) { | |
3680 | /* For both EBUSY and EINVAL error, the mutex is NOT locked. */ | |
3681 | ret = 0; | |
3682 | goto end; | |
3683 | } | |
3684 | ||
3685 | ret = 1; | |
3686 | ||
3687 | end: | |
3688 | return ret; | |
3689 | } | |
3690 | ||
f7079f67 DG |
3691 | /* |
3692 | * Search for a relayd associated to the session id and return the reference. | |
3693 | * | |
3694 | * A rcu read side lock MUST be acquire before calling this function and locked | |
3695 | * until the relayd object is no longer necessary. | |
3696 | */ | |
3697 | static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id) | |
3698 | { | |
3699 | struct lttng_ht_iter iter; | |
f7079f67 | 3700 | struct consumer_relayd_sock_pair *relayd = NULL; |
f7079f67 DG |
3701 | |
3702 | /* Iterate over all relayd since they are indexed by net_seq_idx. */ | |
3703 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, | |
3704 | node.node) { | |
18261bd1 DG |
3705 | /* |
3706 | * Check by sessiond id which is unique here where the relayd session | |
3707 | * id might not be when having multiple relayd. | |
3708 | */ | |
3709 | if (relayd->sessiond_session_id == id) { | |
f7079f67 | 3710 | /* Found the relayd. There can be only one per id. */ |
18261bd1 | 3711 | goto found; |
f7079f67 DG |
3712 | } |
3713 | } | |
3714 | ||
18261bd1 DG |
3715 | return NULL; |
3716 | ||
3717 | found: | |
f7079f67 DG |
3718 | return relayd; |
3719 | } | |
3720 | ||
ca22feea DG |
3721 | /* |
3722 | * Check if for a given session id there is still data needed to be extract | |
3723 | * from the buffers. | |
3724 | * | |
6d805429 | 3725 | * Return 1 if data is pending or else 0 meaning ready to be read. |
ca22feea | 3726 | */ |
6d805429 | 3727 | int consumer_data_pending(uint64_t id) |
ca22feea DG |
3728 | { |
3729 | int ret; | |
3730 | struct lttng_ht_iter iter; | |
3731 | struct lttng_ht *ht; | |
3732 | struct lttng_consumer_stream *stream; | |
f7079f67 | 3733 | struct consumer_relayd_sock_pair *relayd = NULL; |
6d805429 | 3734 | int (*data_pending)(struct lttng_consumer_stream *); |
ca22feea | 3735 | |
6d805429 | 3736 | DBG("Consumer data pending command on session id %" PRIu64, id); |
ca22feea | 3737 | |
6f6eda74 | 3738 | rcu_read_lock(); |
ca22feea DG |
3739 | pthread_mutex_lock(&consumer_data.lock); |
3740 | ||
3741 | switch (consumer_data.type) { | |
3742 | case LTTNG_CONSUMER_KERNEL: | |
6d805429 | 3743 | data_pending = lttng_kconsumer_data_pending; |
ca22feea DG |
3744 | break; |
3745 | case LTTNG_CONSUMER32_UST: | |
3746 | case LTTNG_CONSUMER64_UST: | |
6d805429 | 3747 | data_pending = lttng_ustconsumer_data_pending; |
ca22feea DG |
3748 | break; |
3749 | default: | |
3750 | ERR("Unknown consumer data type"); | |
3751 | assert(0); | |
3752 | } | |
3753 | ||
3754 | /* Ease our life a bit */ | |
3755 | ht = consumer_data.stream_list_ht; | |
3756 | ||
f7079f67 DG |
3757 | relayd = find_relayd_by_session_id(id); |
3758 | if (relayd) { | |
3759 | /* Send init command for data pending. */ | |
3760 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3761 | ret = relayd_begin_data_pending(&relayd->control_sock, | |
3762 | relayd->relayd_session_id); | |
3763 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
3764 | if (ret < 0) { | |
3765 | /* Communication error thus the relayd so no data pending. */ | |
3766 | goto data_not_pending; | |
3767 | } | |
3768 | } | |
3769 | ||
c8f59ee5 | 3770 | cds_lfht_for_each_entry_duplicate(ht->ht, |
d88aee68 DG |
3771 | ht->hash_fct(&id, lttng_ht_seed), |
3772 | ht->match_fct, &id, | |
ca22feea | 3773 | &iter.iter, stream, node_session_id.node) { |
4e9a4686 DG |
3774 | /* If this call fails, the stream is being used hence data pending. */ |
3775 | ret = stream_try_lock(stream); | |
3776 | if (!ret) { | |
f7079f67 | 3777 | goto data_pending; |
ca22feea | 3778 | } |
ca22feea | 3779 | |
4e9a4686 DG |
3780 | /* |
3781 | * A removed node from the hash table indicates that the stream has | |
3782 | * been deleted thus having a guarantee that the buffers are closed | |
3783 | * on the consumer side. However, data can still be transmitted | |
3784 | * over the network so don't skip the relayd check. | |
3785 | */ | |
3786 | ret = cds_lfht_is_node_deleted(&stream->node.node); | |
3787 | if (!ret) { | |
3788 | /* Check the stream if there is data in the buffers. */ | |
6d805429 DG |
3789 | ret = data_pending(stream); |
3790 | if (ret == 1) { | |
4e9a4686 | 3791 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3792 | goto data_pending; |
4e9a4686 DG |
3793 | } |
3794 | } | |
3795 | ||
3796 | /* Relayd check */ | |
f7079f67 | 3797 | if (relayd) { |
c8f59ee5 DG |
3798 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
3799 | if (stream->metadata_flag) { | |
ad7051c0 DG |
3800 | ret = relayd_quiescent_control(&relayd->control_sock, |
3801 | stream->relayd_stream_id); | |
c8f59ee5 | 3802 | } else { |
6d805429 | 3803 | ret = relayd_data_pending(&relayd->control_sock, |
39df6d9f DG |
3804 | stream->relayd_stream_id, |
3805 | stream->next_net_seq_num - 1); | |
c8f59ee5 DG |
3806 | } |
3807 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
6d805429 | 3808 | if (ret == 1) { |
4e9a4686 | 3809 | pthread_mutex_unlock(&stream->lock); |
f7079f67 | 3810 | goto data_pending; |
c8f59ee5 DG |
3811 | } |
3812 | } | |
4e9a4686 | 3813 | pthread_mutex_unlock(&stream->lock); |
c8f59ee5 | 3814 | } |
ca22feea | 3815 | |
f7079f67 DG |
3816 | if (relayd) { |
3817 | unsigned int is_data_inflight = 0; | |
3818 | ||
3819 | /* Send init command for data pending. */ | |
3820 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); | |
3821 | ret = relayd_end_data_pending(&relayd->control_sock, | |
3822 | relayd->relayd_session_id, &is_data_inflight); | |
3823 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); | |
bdd88757 | 3824 | if (ret < 0) { |
f7079f67 DG |
3825 | goto data_not_pending; |
3826 | } | |
bdd88757 DG |
3827 | if (is_data_inflight) { |
3828 | goto data_pending; | |
3829 | } | |
f7079f67 DG |
3830 | } |
3831 | ||
ca22feea | 3832 | /* |
f7079f67 DG |
3833 | * Finding _no_ node in the hash table and no inflight data means that the |
3834 | * stream(s) have been removed thus data is guaranteed to be available for | |
3835 | * analysis from the trace files. | |
ca22feea DG |
3836 | */ |
3837 | ||
f7079f67 | 3838 | data_not_pending: |
ca22feea DG |
3839 | /* Data is available to be read by a viewer. */ |
3840 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3841 | rcu_read_unlock(); |
6d805429 | 3842 | return 0; |
ca22feea | 3843 | |
f7079f67 | 3844 | data_pending: |
ca22feea DG |
3845 | /* Data is still being extracted from buffers. */ |
3846 | pthread_mutex_unlock(&consumer_data.lock); | |
c8f59ee5 | 3847 | rcu_read_unlock(); |
6d805429 | 3848 | return 1; |
ca22feea | 3849 | } |
f50f23d9 DG |
3850 | |
3851 | /* | |
3852 | * Send a ret code status message to the sessiond daemon. | |
3853 | * | |
3854 | * Return the sendmsg() return value. | |
3855 | */ | |
3856 | int consumer_send_status_msg(int sock, int ret_code) | |
3857 | { | |
3858 | struct lttcomm_consumer_status_msg msg; | |
3859 | ||
53efb85a | 3860 | memset(&msg, 0, sizeof(msg)); |
f50f23d9 DG |
3861 | msg.ret_code = ret_code; |
3862 | ||
3863 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3864 | } | |
ffe60014 DG |
3865 | |
3866 | /* | |
3867 | * Send a channel status message to the sessiond daemon. | |
3868 | * | |
3869 | * Return the sendmsg() return value. | |
3870 | */ | |
3871 | int consumer_send_status_channel(int sock, | |
3872 | struct lttng_consumer_channel *channel) | |
3873 | { | |
3874 | struct lttcomm_consumer_status_channel msg; | |
3875 | ||
3876 | assert(sock >= 0); | |
3877 | ||
53efb85a | 3878 | memset(&msg, 0, sizeof(msg)); |
ffe60014 | 3879 | if (!channel) { |
0c759fc9 | 3880 | msg.ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL; |
ffe60014 | 3881 | } else { |
0c759fc9 | 3882 | msg.ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
ffe60014 DG |
3883 | msg.key = channel->key; |
3884 | msg.stream_count = channel->streams.count; | |
3885 | } | |
3886 | ||
3887 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); | |
3888 | } | |
5c786ded | 3889 | |
d07ceecd MD |
3890 | unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos, |
3891 | unsigned long produced_pos, uint64_t nb_packets_per_stream, | |
3892 | uint64_t max_sb_size) | |
5c786ded | 3893 | { |
d07ceecd | 3894 | unsigned long start_pos; |
5c786ded | 3895 | |
d07ceecd MD |
3896 | if (!nb_packets_per_stream) { |
3897 | return consumed_pos; /* Grab everything */ | |
3898 | } | |
3899 | start_pos = produced_pos - offset_align_floor(produced_pos, max_sb_size); | |
3900 | start_pos -= max_sb_size * nb_packets_per_stream; | |
3901 | if ((long) (start_pos - consumed_pos) < 0) { | |
3902 | return consumed_pos; /* Grab everything */ | |
3903 | } | |
3904 | return start_pos; | |
5c786ded | 3905 | } |
6af2ba3b JD |
3906 | |
3907 | static | |
b9478f8d | 3908 | int consumer_flush_buffer(struct lttng_consumer_stream *stream, int producer_active) |
6af2ba3b JD |
3909 | { |
3910 | int ret = 0; | |
3911 | ||
3912 | switch (consumer_data.type) { | |
3913 | case LTTNG_CONSUMER_KERNEL: | |
3914 | ret = kernctl_buffer_flush(stream->wait_fd); | |
3915 | if (ret < 0) { | |
3916 | ERR("Failed to flush kernel stream"); | |
3917 | goto end; | |
3918 | } | |
3919 | break; | |
3920 | case LTTNG_CONSUMER32_UST: | |
3921 | case LTTNG_CONSUMER64_UST: | |
3922 | lttng_ustctl_flush_buffer(stream, producer_active); | |
3923 | break; | |
3924 | default: | |
3925 | ERR("Unknown consumer_data type"); | |
3926 | abort(); | |
3927 | } | |
3928 | ||
3929 | end: | |
3930 | return ret; | |
3931 | } | |
3932 | ||
3933 | /* | |
3934 | * Sample the rotate position for all the streams of a channel. | |
3935 | * | |
3936 | * Returns 0 on success, < 0 on error | |
3937 | */ | |
3938 | int lttng_consumer_rotate_channel(uint64_t key, char *path, | |
2d7d73bc | 3939 | uint64_t relayd_id, uint32_t metadata, uint64_t new_chunk_id, |
6af2ba3b JD |
3940 | struct lttng_consumer_local_data *ctx) |
3941 | { | |
3942 | int ret; | |
3943 | struct lttng_consumer_channel *channel; | |
3944 | struct lttng_consumer_stream *stream; | |
3945 | struct lttng_ht_iter iter; | |
3946 | struct lttng_ht *ht = consumer_data.stream_per_chan_id_ht; | |
3947 | ||
3948 | DBG("Kernel consumer sample rotate position for channel %" PRIu64, key); | |
3949 | ||
3950 | rcu_read_lock(); | |
3951 | ||
3952 | channel = consumer_find_channel(key); | |
3953 | if (!channel) { | |
3954 | ERR("No channel found for key %" PRIu64, key); | |
3955 | ret = -1; | |
3956 | goto end; | |
3957 | } | |
3958 | pthread_mutex_lock(&channel->lock); | |
2d7d73bc | 3959 | channel->current_chunk_id = new_chunk_id; |
6af2ba3b | 3960 | snprintf(channel->pathname, PATH_MAX, "%s", path); |
4a478f45 JD |
3961 | ret = utils_mkdir_recursive(channel->pathname, S_IRWXU | S_IRWXG, |
3962 | channel->uid, channel->gid); | |
3963 | if (ret < 0) { | |
3964 | ERR("Trace directory creation error"); | |
3965 | ret = -1; | |
5eecef35 JD |
3966 | pthread_mutex_unlock(&channel->lock); |
3967 | goto end; | |
4a478f45 | 3968 | } |
5eecef35 | 3969 | pthread_mutex_unlock(&channel->lock); |
6af2ba3b JD |
3970 | |
3971 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
3972 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
3973 | ht->match_fct, &channel->key, &iter.iter, | |
3974 | stream, node_channel_id.node) { | |
103373d7 JD |
3975 | uint64_t consumed_pos; |
3976 | ||
6af2ba3b JD |
3977 | health_code_update(); |
3978 | ||
3979 | /* | |
3980 | * Lock stream because we are about to change its state. | |
3981 | */ | |
3982 | pthread_mutex_lock(&stream->lock); | |
103373d7 | 3983 | |
b9478f8d | 3984 | memcpy(stream->channel_ro_pathname, channel->pathname, PATH_MAX); |
6af2ba3b JD |
3985 | ret = lttng_consumer_sample_snapshot_positions(stream); |
3986 | if (ret < 0) { | |
3987 | ERR("Taking kernel snapshot positions"); | |
3988 | goto end_unlock; | |
103373d7 | 3989 | } |
6af2ba3b | 3990 | |
103373d7 JD |
3991 | ret = lttng_consumer_get_produced_snapshot(stream, |
3992 | &stream->rotate_position); | |
3993 | if (ret < 0) { | |
3994 | ERR("Produced kernel snapshot position"); | |
3995 | goto end_unlock; | |
3996 | } | |
3997 | fprintf(stderr, "Stream %lu should rotate after %lu to %s\n", | |
3998 | stream->key, stream->rotate_position, | |
3999 | channel->pathname); | |
4000 | lttng_consumer_get_consumed_snapshot(stream, | |
4001 | &consumed_pos); | |
4002 | fprintf(stderr, "consumed %lu\n", consumed_pos); | |
4003 | if (consumed_pos == stream->rotate_position) { | |
4004 | stream->rotate_ready = 1; | |
4005 | fprintf(stderr, "Stream %lu ready to rotate to %s\n", | |
4006 | stream->key, channel->pathname); | |
6af2ba3b JD |
4007 | } |
4008 | channel->nr_stream_rotate_pending++; | |
4009 | ||
b9478f8d | 4010 | ret = consumer_flush_buffer(stream, 1); |
6af2ba3b JD |
4011 | if (ret < 0) { |
4012 | ERR("Failed to flush stream"); | |
4013 | goto end_unlock; | |
4014 | } | |
4015 | ||
4016 | pthread_mutex_unlock(&stream->lock); | |
4017 | } | |
4018 | ||
4019 | ret = 0; | |
5eecef35 | 4020 | goto end; |
6af2ba3b JD |
4021 | |
4022 | end_unlock: | |
4023 | pthread_mutex_unlock(&stream->lock); | |
6af2ba3b JD |
4024 | end: |
4025 | rcu_read_unlock(); | |
4026 | return ret; | |
4027 | } | |
4028 | ||
6af2ba3b | 4029 | /* |
c1b7c694 | 4030 | * Check if a stream is ready to be rotated after extracting it. |
6af2ba3b | 4031 | * |
c1b7c694 JD |
4032 | * When we are called between get_next_subbuf and put_next_subbuf, the len |
4033 | * parameter is the subbuf size of the current subbuffer being extracted. This | |
4034 | * len is with padding, so it is normal to see that the current position is | |
4035 | * farther than the expected rotate position. | |
4036 | * | |
4037 | * Return 1 if it is ready for rotation, 0 if it is not, a negative value on | |
4038 | * error. | |
6af2ba3b | 4039 | */ |
c1b7c694 JD |
4040 | int lttng_consumer_stream_is_rotate_ready(struct lttng_consumer_stream *stream, |
4041 | unsigned long len) | |
6af2ba3b JD |
4042 | { |
4043 | int ret; | |
4044 | unsigned long consumed_pos; | |
4045 | ||
4046 | if (!stream->rotate_position && !stream->rotate_ready) { | |
4047 | ret = 0; | |
4048 | goto end; | |
4049 | } | |
4050 | ||
e47608fa JD |
4051 | if (stream->rotate_ready) { |
4052 | fprintf(stderr, "Rotate position reached for stream %lu\n", | |
4053 | stream->key); | |
4054 | ret = 1; | |
4055 | goto end; | |
4056 | } | |
4057 | ||
6af2ba3b JD |
4058 | /* |
4059 | * If we don't have the rotate_ready flag, check the consumed position | |
4060 | * to determine if we need to rotate. | |
4061 | */ | |
e47608fa JD |
4062 | ret = lttng_consumer_sample_snapshot_positions(stream); |
4063 | if (ret < 0) { | |
4064 | ERR("Taking kernel snapshot positions"); | |
4065 | goto end; | |
4066 | } | |
6af2ba3b | 4067 | |
e47608fa JD |
4068 | ret = lttng_consumer_get_consumed_snapshot(stream, &consumed_pos); |
4069 | if (ret < 0) { | |
4070 | ERR("Produced kernel snapshot position"); | |
4071 | goto end; | |
4072 | } | |
6af2ba3b | 4073 | |
e47608fa JD |
4074 | fprintf(stderr, "packet %lu, pos %lu\n", stream->key, consumed_pos); |
4075 | /* Rotate position not reached yet. */ | |
4076 | if ((consumed_pos + len) < stream->rotate_position) { | |
4077 | ret = 0; | |
4078 | goto end; | |
6af2ba3b | 4079 | } |
e47608fa JD |
4080 | fprintf(stderr, "Rotate position %lu (expected %lu) reached for stream %lu\n", |
4081 | consumed_pos + len, stream->rotate_position, | |
4082 | stream->key); | |
4083 | ret = 1; | |
6af2ba3b | 4084 | |
c1b7c694 JD |
4085 | end: |
4086 | return ret; | |
4087 | } | |
4088 | ||
e47608fa JD |
4089 | /* |
4090 | * Reset the state for a stream after a rotation occurred. | |
4091 | */ | |
4092 | void lttng_consumer_reset_stream_rotate_state(struct lttng_consumer_stream *stream) | |
4093 | { | |
4094 | stream->rotate_position = 0; | |
4095 | stream->rotate_ready = 0; | |
4096 | stream->rotated = 1; | |
4097 | } | |
4098 | ||
c1b7c694 | 4099 | /* |
d26a0e48 | 4100 | * Perform the rotation a local stream file. |
c1b7c694 | 4101 | */ |
d26a0e48 | 4102 | int rotate_local_stream(struct lttng_consumer_local_data *ctx, |
c1b7c694 JD |
4103 | struct lttng_consumer_stream *stream) |
4104 | { | |
4105 | int ret; | |
4106 | ||
6af2ba3b JD |
4107 | ret = close(stream->out_fd); |
4108 | if (ret < 0) { | |
4109 | PERROR("Closing tracefile"); | |
4110 | goto error; | |
4111 | } | |
4112 | ||
4113 | fprintf(stderr, "Rotating stream %lu to %s/%s\n", stream->key, | |
b9478f8d JD |
4114 | stream->channel_ro_pathname, stream->name); |
4115 | ret = utils_create_stream_file(stream->channel_ro_pathname, stream->name, | |
4116 | stream->channel_ro_tracefile_size, stream->tracefile_count_current, | |
6af2ba3b JD |
4117 | stream->uid, stream->gid, NULL); |
4118 | if (ret < 0) { | |
4119 | goto error; | |
4120 | } | |
4121 | stream->out_fd = ret; | |
4122 | stream->tracefile_size_current = 0; | |
4123 | ||
4124 | if (!stream->metadata_flag) { | |
4125 | struct lttng_index_file *index_file; | |
4126 | ||
4127 | lttng_index_file_put(stream->index_file); | |
4128 | ||
b9478f8d | 4129 | index_file = lttng_index_file_create(stream->channel_ro_pathname, |
6af2ba3b | 4130 | stream->name, stream->uid, stream->gid, |
b9478f8d | 4131 | stream->channel_ro_tracefile_size, |
6af2ba3b JD |
4132 | stream->tracefile_count_current, |
4133 | CTF_INDEX_MAJOR, CTF_INDEX_MINOR); | |
4134 | if (!index_file) { | |
4135 | goto error; | |
4136 | } | |
4137 | stream->index_file = index_file; | |
4138 | stream->out_fd_offset = 0; | |
6af2ba3b JD |
4139 | } |
4140 | ||
6af2ba3b JD |
4141 | ret = 0; |
4142 | goto end; | |
4143 | ||
4144 | error: | |
4145 | ret = -1; | |
4146 | end: | |
4147 | return ret; | |
d26a0e48 JD |
4148 | |
4149 | } | |
4150 | ||
4151 | /* | |
4152 | * Perform the rotation a stream file on the relay. | |
4153 | */ | |
4154 | int rotate_relay_stream(struct lttng_consumer_local_data *ctx, | |
4155 | struct lttng_consumer_stream *stream) | |
4156 | { | |
4157 | int ret; | |
d26a0e48 JD |
4158 | struct consumer_relayd_sock_pair *relayd; |
4159 | ||
4160 | relayd = consumer_find_relayd(stream->net_seq_idx); | |
4161 | if (!relayd) { | |
4162 | ERR("Failed to find relayd"); | |
4163 | ret = -1; | |
4164 | goto end; | |
4165 | } | |
4166 | ||
2d7d73bc | 4167 | /* FIXME: chan_ro ? */ |
d26a0e48 | 4168 | ret = relayd_rotate_stream(&relayd->control_sock, |
2d7d73bc JD |
4169 | stream->relayd_stream_id, stream->channel_ro_pathname, |
4170 | stream->chan->current_chunk_id, | |
4171 | stream->last_sequence_number); | |
d26a0e48 JD |
4172 | |
4173 | end: | |
d26a0e48 JD |
4174 | return ret; |
4175 | } | |
4176 | ||
4177 | /* | |
4178 | * Performs the stream rotation for the rotate session feature if needed. | |
4179 | * It must be called with the stream lock held. | |
4180 | * | |
4181 | * Return 0 on success, a negative number of error. | |
4182 | */ | |
4183 | int lttng_consumer_rotate_stream(struct lttng_consumer_local_data *ctx, | |
4184 | struct lttng_consumer_stream *stream) | |
4185 | { | |
4186 | int ret; | |
4187 | ||
4188 | if (stream->net_seq_idx != (uint64_t) -1ULL) { | |
4189 | ret = rotate_relay_stream(ctx, stream); | |
4190 | } else { | |
4191 | ret = rotate_local_stream(ctx, stream); | |
4192 | } | |
a9b8a25d JD |
4193 | if (ret < 0) { |
4194 | goto error; | |
4195 | } | |
4196 | ||
4197 | if (stream->metadata_flag) { | |
4198 | switch (consumer_data.type) { | |
4199 | case LTTNG_CONSUMER_KERNEL: | |
4200 | /* | |
4201 | * Reset the position of what has been read from the metadata | |
4202 | * cache to 0 so we can dump it again. | |
4203 | */ | |
4204 | ret = kernctl_metadata_cache_dump(stream->wait_fd); | |
4205 | if (ret < 0) { | |
4206 | ERR("Failed to dump the kernel metadata cache after rotation"); | |
4207 | goto error; | |
4208 | } | |
4209 | break; | |
4210 | case LTTNG_CONSUMER32_UST: | |
4211 | case LTTNG_CONSUMER64_UST: | |
4212 | /* | |
4213 | * Reset the position pushed from the metadata cache so it | |
4214 | * will write from the beginning on the next push. | |
4215 | */ | |
4216 | stream->ust_metadata_pushed = 0; | |
4217 | break; | |
4218 | default: | |
4219 | ERR("Unknown consumer_data type"); | |
4220 | abort(); | |
4221 | } | |
4222 | } | |
d26a0e48 JD |
4223 | lttng_consumer_reset_stream_rotate_state(stream); |
4224 | ||
a9b8a25d JD |
4225 | ret = 0; |
4226 | ||
4227 | error: | |
d26a0e48 | 4228 | return ret; |
6af2ba3b JD |
4229 | } |
4230 | ||
4231 | /* | |
4232 | * Rotate all the ready streams. | |
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. | |
4237 | * | |
4238 | * Returns 0 on success, < 0 on error | |
4239 | */ | |
4240 | int lttng_consumer_rotate_ready_streams(uint64_t key, | |
4241 | struct lttng_consumer_local_data *ctx) | |
4242 | { | |
4243 | int ret; | |
4244 | struct lttng_consumer_channel *channel; | |
4245 | struct lttng_consumer_stream *stream; | |
4246 | struct lttng_ht_iter iter; | |
4247 | struct lttng_ht *ht = consumer_data.stream_per_chan_id_ht; | |
4248 | ||
4249 | rcu_read_lock(); | |
4250 | ||
4251 | channel = consumer_find_channel(key); | |
4252 | if (!channel) { | |
4253 | ERR("No channel found for key %" PRIu64, key); | |
4254 | ret = -1; | |
4255 | goto end; | |
4256 | } | |
4257 | cds_lfht_for_each_entry_duplicate(ht->ht, | |
4258 | ht->hash_fct(&channel->key, lttng_ht_seed), | |
4259 | ht->match_fct, &channel->key, &iter.iter, | |
4260 | stream, node_channel_id.node) { | |
4261 | health_code_update(); | |
4262 | ||
4263 | /* | |
4264 | * Lock stream because we are about to change its state. | |
4265 | */ | |
4266 | pthread_mutex_lock(&stream->lock); | |
4267 | if (stream->rotate_ready == 0) { | |
4268 | pthread_mutex_unlock(&stream->lock); | |
4269 | continue; | |
4270 | } | |
4271 | ret = lttng_consumer_rotate_stream(ctx, stream); | |
4272 | if (ret < 0) { | |
4273 | pthread_mutex_unlock(&stream->lock); | |
4274 | ERR("Stream rotation error"); | |
4275 | goto end; | |
4276 | } | |
4277 | ||
4278 | pthread_mutex_unlock(&stream->lock); | |
b9478f8d JD |
4279 | ret = consumer_post_rotation(stream, ctx); |
4280 | if (ret < 0) { | |
4281 | ERR("Failed after a rotation"); | |
4282 | goto end; | |
4283 | } | |
6af2ba3b JD |
4284 | } |
4285 | ||
4286 | ret = 0; | |
4287 | ||
4288 | end: | |
4289 | rcu_read_unlock(); | |
4290 | return ret; | |
4291 | } | |
e8ed1ed1 | 4292 | |
0ee5eef6 JD |
4293 | static |
4294 | int rotate_rename_local(char *current_path, char *new_path, | |
4295 | uid_t uid, gid_t gid) | |
e8ed1ed1 JD |
4296 | { |
4297 | int ret; | |
4298 | ||
0ee5eef6 JD |
4299 | ret = utils_mkdir_recursive(new_path, S_IRWXU | S_IRWXG, |
4300 | uid, gid); | |
4301 | if (ret < 0) { | |
4302 | ERR("Create directory on rotate"); | |
4303 | goto end; | |
e8ed1ed1 | 4304 | } |
0ee5eef6 | 4305 | |
e8ed1ed1 | 4306 | ret = rename(current_path, new_path); |
4a478f45 JD |
4307 | /* |
4308 | * If a domain has not yet created its channel, the domain-specific | |
4309 | * folder might not exist, but this is not an error. | |
4310 | */ | |
4311 | if (ret < 0 && errno != ENOENT) { | |
e8ed1ed1 | 4312 | PERROR("Rename completed rotation chunk"); |
4a478f45 | 4313 | goto end; |
e8ed1ed1 JD |
4314 | } |
4315 | ||
4a478f45 JD |
4316 | ret = 0; |
4317 | ||
e8ed1ed1 JD |
4318 | end: |
4319 | return ret; | |
4320 | } | |
0ee5eef6 JD |
4321 | |
4322 | static | |
4323 | int rotate_rename_relay(char *current_path, char *new_path, uint64_t relayd_id) | |
4324 | { | |
4325 | int ret; | |
4326 | struct consumer_relayd_sock_pair *relayd; | |
4327 | ||
4328 | relayd = consumer_find_relayd(relayd_id); | |
4329 | if (!relayd) { | |
4330 | ERR("Failed to find relayd"); | |
4331 | ret = -1; | |
4332 | goto end; | |
4333 | } | |
4334 | ||
4335 | ret = relayd_rotate_rename(&relayd->control_sock, current_path, new_path); | |
4336 | ||
4337 | end: | |
4338 | return ret; | |
4339 | ||
4340 | } | |
4341 | ||
4342 | int lttng_consumer_rotate_rename(char *current_path, char *new_path, | |
4343 | uid_t uid, gid_t gid, uint64_t relayd_id) | |
4344 | { | |
4345 | if (relayd_id != (uint64_t) -1ULL) { | |
4346 | return rotate_rename_relay(current_path, new_path, relayd_id); | |
4347 | } else { | |
4348 | return rotate_rename_local(current_path, new_path, uid, gid); | |
4349 | } | |
4350 | } | |
d3dedf27 JD |
4351 | |
4352 | int lttng_consumer_rotate_pending_relay(uint64_t session_id, | |
4353 | uint64_t relayd_id, uint64_t chunk_id) | |
4354 | { | |
4355 | int ret; | |
4356 | struct consumer_relayd_sock_pair *relayd; | |
4357 | ||
4358 | relayd = consumer_find_relayd(relayd_id); | |
4359 | if (!relayd) { | |
4360 | ERR("Failed to find relayd"); | |
4361 | ret = -1; | |
4362 | goto end; | |
4363 | } | |
4364 | ||
4365 | ret = relayd_rotate_pending(&relayd->control_sock, chunk_id); | |
4366 | ||
4367 | end: | |
4368 | return ret; | |
4369 | ||
4370 | } |